Linux Audio

Check our new training course

Loading...
v6.13.7
   1/*
   2 * Copyright 2014 Advanced Micro Devices, Inc.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20 * OTHER DEALINGS IN THE SOFTWARE.
  21 *
  22 */
  23
  24#include <linux/delay.h>
  25#include <linux/kernel.h>
  26#include <linux/firmware.h>
  27#include <linux/module.h>
  28#include <linux/pci.h>
  29
  30#include "amdgpu.h"
  31#include "amdgpu_gfx.h"
  32#include "amdgpu_ring.h"
  33#include "vi.h"
  34#include "vi_structs.h"
  35#include "vid.h"
  36#include "amdgpu_ucode.h"
  37#include "amdgpu_atombios.h"
  38#include "atombios_i2c.h"
  39#include "clearstate_vi.h"
  40
  41#include "gmc/gmc_8_2_d.h"
  42#include "gmc/gmc_8_2_sh_mask.h"
  43
  44#include "oss/oss_3_0_d.h"
  45#include "oss/oss_3_0_sh_mask.h"
  46
  47#include "bif/bif_5_0_d.h"
  48#include "bif/bif_5_0_sh_mask.h"
  49#include "gca/gfx_8_0_d.h"
  50#include "gca/gfx_8_0_enum.h"
  51#include "gca/gfx_8_0_sh_mask.h"
  52
  53#include "dce/dce_10_0_d.h"
  54#include "dce/dce_10_0_sh_mask.h"
  55
  56#include "smu/smu_7_1_3_d.h"
  57
  58#include "ivsrcid/ivsrcid_vislands30.h"
  59
  60#define GFX8_NUM_GFX_RINGS     1
  61#define GFX8_MEC_HPD_SIZE 4096
  62
  63#define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001
  64#define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001
  65#define POLARIS11_GB_ADDR_CONFIG_GOLDEN 0x22011002
  66#define TONGA_GB_ADDR_CONFIG_GOLDEN 0x22011003
  67
  68#define ARRAY_MODE(x)					((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT)
  69#define PIPE_CONFIG(x)					((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT)
  70#define TILE_SPLIT(x)					((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT)
  71#define MICRO_TILE_MODE_NEW(x)				((x) << GB_TILE_MODE0__MICRO_TILE_MODE_NEW__SHIFT)
  72#define SAMPLE_SPLIT(x)					((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT)
  73#define BANK_WIDTH(x)					((x) << GB_MACROTILE_MODE0__BANK_WIDTH__SHIFT)
  74#define BANK_HEIGHT(x)					((x) << GB_MACROTILE_MODE0__BANK_HEIGHT__SHIFT)
  75#define MACRO_TILE_ASPECT(x)				((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT)
  76#define NUM_BANKS(x)					((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT)
  77
  78#define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK            0x00000001L
  79#define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK            0x00000002L
  80#define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK           0x00000004L
  81#define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK           0x00000008L
  82#define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK           0x00000010L
  83#define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK           0x00000020L
  84
  85/* BPM SERDES CMD */
  86#define SET_BPM_SERDES_CMD    1
  87#define CLE_BPM_SERDES_CMD    0
  88
  89/* BPM Register Address*/
  90enum {
  91	BPM_REG_CGLS_EN = 0,        /* Enable/Disable CGLS */
  92	BPM_REG_CGLS_ON,            /* ON/OFF CGLS: shall be controlled by RLC FW */
  93	BPM_REG_CGCG_OVERRIDE,      /* Set/Clear CGCG Override */
  94	BPM_REG_MGCG_OVERRIDE,      /* Set/Clear MGCG Override */
  95	BPM_REG_FGCG_OVERRIDE,      /* Set/Clear FGCG Override */
  96	BPM_REG_FGCG_MAX
  97};
  98
  99#define RLC_FormatDirectRegListLength        14
 100
 101MODULE_FIRMWARE("amdgpu/carrizo_ce.bin");
 102MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin");
 103MODULE_FIRMWARE("amdgpu/carrizo_me.bin");
 104MODULE_FIRMWARE("amdgpu/carrizo_mec.bin");
 105MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin");
 106MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin");
 107
 108MODULE_FIRMWARE("amdgpu/stoney_ce.bin");
 109MODULE_FIRMWARE("amdgpu/stoney_pfp.bin");
 110MODULE_FIRMWARE("amdgpu/stoney_me.bin");
 111MODULE_FIRMWARE("amdgpu/stoney_mec.bin");
 112MODULE_FIRMWARE("amdgpu/stoney_rlc.bin");
 113
 114MODULE_FIRMWARE("amdgpu/tonga_ce.bin");
 115MODULE_FIRMWARE("amdgpu/tonga_pfp.bin");
 116MODULE_FIRMWARE("amdgpu/tonga_me.bin");
 117MODULE_FIRMWARE("amdgpu/tonga_mec.bin");
 118MODULE_FIRMWARE("amdgpu/tonga_mec2.bin");
 119MODULE_FIRMWARE("amdgpu/tonga_rlc.bin");
 120
 121MODULE_FIRMWARE("amdgpu/topaz_ce.bin");
 122MODULE_FIRMWARE("amdgpu/topaz_pfp.bin");
 123MODULE_FIRMWARE("amdgpu/topaz_me.bin");
 124MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
 125MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
 126
 127MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
 128MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
 129MODULE_FIRMWARE("amdgpu/fiji_me.bin");
 130MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
 131MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
 132MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
 133
 134MODULE_FIRMWARE("amdgpu/polaris10_ce.bin");
 135MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin");
 136MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin");
 137MODULE_FIRMWARE("amdgpu/polaris10_pfp_2.bin");
 138MODULE_FIRMWARE("amdgpu/polaris10_me.bin");
 139MODULE_FIRMWARE("amdgpu/polaris10_me_2.bin");
 140MODULE_FIRMWARE("amdgpu/polaris10_mec.bin");
 141MODULE_FIRMWARE("amdgpu/polaris10_mec_2.bin");
 142MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin");
 143MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin");
 144MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin");
 145
 146MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
 147MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
 148MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
 149MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
 150MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
 151MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
 152MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
 153MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
 154MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
 155MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
 156MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
 157
 158MODULE_FIRMWARE("amdgpu/polaris12_ce.bin");
 159MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin");
 160MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin");
 161MODULE_FIRMWARE("amdgpu/polaris12_pfp_2.bin");
 162MODULE_FIRMWARE("amdgpu/polaris12_me.bin");
 163MODULE_FIRMWARE("amdgpu/polaris12_me_2.bin");
 164MODULE_FIRMWARE("amdgpu/polaris12_mec.bin");
 165MODULE_FIRMWARE("amdgpu/polaris12_mec_2.bin");
 166MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin");
 167MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin");
 168MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin");
 169
 170MODULE_FIRMWARE("amdgpu/vegam_ce.bin");
 171MODULE_FIRMWARE("amdgpu/vegam_pfp.bin");
 172MODULE_FIRMWARE("amdgpu/vegam_me.bin");
 173MODULE_FIRMWARE("amdgpu/vegam_mec.bin");
 174MODULE_FIRMWARE("amdgpu/vegam_mec2.bin");
 175MODULE_FIRMWARE("amdgpu/vegam_rlc.bin");
 176
 177static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
 178{
 179	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
 180	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
 181	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
 182	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
 183	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
 184	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
 185	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
 186	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
 187	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
 188	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
 189	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
 190	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
 191	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
 192	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
 193	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
 194	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
 195};
 196
 197static const u32 golden_settings_tonga_a11[] =
 198{
 199	mmCB_HW_CONTROL, 0xfffdf3cf, 0x00007208,
 200	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
 201	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 202	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 203	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 204	mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x000000fc,
 205	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 206	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
 207	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 208	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 209	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 210	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 211	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000002fb,
 212	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x0000543b,
 213	mmTCP_CHAN_STEER_LO, 0xffffffff, 0xa9210876,
 214	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 215};
 216
 217static const u32 tonga_golden_common_all[] =
 218{
 219	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 220	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
 221	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
 222	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 223	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 224	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 225	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 226	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
 227};
 228
 229static const u32 tonga_mgcg_cgcg_init[] =
 230{
 231	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 232	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 233	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 234	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 235	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
 236	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
 237	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
 238	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 239	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 240	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 241	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 242	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 243	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 244	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 245	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 246	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 247	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 248	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 249	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 250	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 251	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 252	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 253	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
 254	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 255	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 256	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 257	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 258	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 259	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 260	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 261	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 262	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 263	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 264	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 265	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 266	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 267	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 268	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 269	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
 270	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 271	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 272	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 273	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 274	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
 275	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 276	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 277	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 278	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 279	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
 280	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 281	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 282	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 283	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 284	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 285	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 286	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 287	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 288	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 289	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
 290	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 291	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 292	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 293	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 294	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
 295	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 296	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 297	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 298	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 299	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
 300	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 301	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 302	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 303	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 304	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
 305	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 306};
 307
 308static const u32 golden_settings_vegam_a11[] =
 309{
 310	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
 311	mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000,
 312	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 313	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 314	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 315	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 316	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a,
 317	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e,
 318	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 319	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
 320	mmSQ_CONFIG, 0x07f80000, 0x01180000,
 321	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 322	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 323	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
 324	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 325	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054,
 326	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 327};
 328
 329static const u32 vegam_golden_common_all[] =
 330{
 331	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 332	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 333	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 334	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 335	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 336	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 337};
 338
 339static const u32 golden_settings_polaris11_a11[] =
 340{
 341	mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208,
 342	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
 343	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 344	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 345	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 346	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 347	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
 348	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
 349	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 350	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
 351	mmSQ_CONFIG, 0x07f80000, 0x01180000,
 352	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 353	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 354	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f3,
 355	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 356	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003210,
 357	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 358};
 359
 360static const u32 polaris11_golden_common_all[] =
 361{
 362	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 363	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011002,
 364	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 365	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 366	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 367	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 368};
 369
 370static const u32 golden_settings_polaris10_a11[] =
 371{
 372	mmATC_MISC_CG, 0x000c0fc0, 0x000c0200,
 373	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
 374	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
 375	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 376	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 377	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 378	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 379	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
 380	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002a,
 381	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 382	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
 383	mmSQ_CONFIG, 0x07f80000, 0x07180000,
 384	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 385	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 386	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
 387	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 388	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 389};
 390
 391static const u32 polaris10_golden_common_all[] =
 392{
 393	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 394	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
 395	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
 396	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 397	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 398	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 399	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 400	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 401};
 402
 403static const u32 fiji_golden_common_all[] =
 404{
 405	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 406	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
 407	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
 408	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 409	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 410	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 411	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 412	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 413	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 414	mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009,
 415};
 416
 417static const u32 golden_settings_fiji_a10[] =
 418{
 419	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 420	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 421	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 422	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 423	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 424	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 425	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 426	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 427	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 428	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
 429	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 430};
 431
 432static const u32 fiji_mgcg_cgcg_init[] =
 433{
 434	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 435	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 436	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 437	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 438	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
 439	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
 440	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
 441	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 442	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 443	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 444	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 445	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 446	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 447	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 448	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 449	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 450	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 451	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 452	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 453	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 454	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 455	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 456	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
 457	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 458	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 459	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 460	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 461	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 462	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 463	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 464	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 465	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 466	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 467	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
 468	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 469};
 470
 471static const u32 golden_settings_iceland_a11[] =
 472{
 473	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
 474	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 475	mmDB_DEBUG3, 0xc0000000, 0xc0000000,
 476	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 477	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 478	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 479	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x00000002,
 480	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
 481	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
 482	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 483	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 484	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 485	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 486	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f1,
 487	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 488	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010,
 489};
 490
 491static const u32 iceland_golden_common_all[] =
 492{
 493	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 494	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
 495	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
 496	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
 497	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 498	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 499	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 500	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
 501};
 502
 503static const u32 iceland_mgcg_cgcg_init[] =
 504{
 505	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 506	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 507	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 508	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 509	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0xc0000100,
 510	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0xc0000100,
 511	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0xc0000100,
 512	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 513	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 514	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 515	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 516	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 517	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 518	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 519	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 520	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 521	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 522	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 523	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 524	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 525	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 526	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 527	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0xff000100,
 528	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 529	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 530	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 531	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 532	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 533	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 534	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 535	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 536	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 537	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 538	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
 539	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 540	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 541	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 542	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 543	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
 544	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 545	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 546	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 547	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 548	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
 549	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 550	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 551	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 552	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 553	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
 554	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 555	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 556	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 557	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 558	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
 559	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 560	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 561	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 562	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 563	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
 564	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 565	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 566	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 567	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 568	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
 569};
 570
 571static const u32 cz_golden_settings_a11[] =
 572{
 573	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
 574	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 575	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 576	mmPA_SC_ENHANCE, 0xffffffff, 0x00000001,
 577	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 578	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
 579	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 580	mmTA_CNTL_AUX, 0x000f000f, 0x00010000,
 581	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 582	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 583	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f3,
 584	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00001302
 585};
 586
 587static const u32 cz_golden_common_all[] =
 588{
 589	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 590	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
 591	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
 592	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
 593	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 594	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 595	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 596	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
 597};
 598
 599static const u32 cz_mgcg_cgcg_init[] =
 600{
 601	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 602	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 603	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 604	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 605	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
 606	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
 607	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x00000100,
 608	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 609	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 610	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 611	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 612	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 613	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 614	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 615	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 616	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 617	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 618	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 619	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 620	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 621	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 622	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 623	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
 624	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 625	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 626	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 627	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 628	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 629	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 630	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 631	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 632	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 633	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 634	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 635	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 636	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 637	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 638	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 639	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
 640	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 641	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 642	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 643	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 644	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
 645	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 646	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 647	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 648	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 649	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
 650	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 651	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 652	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 653	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 654	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 655	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 656	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 657	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 658	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 659	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
 660	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 661	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 662	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 663	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 664	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
 665	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 666	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 667	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 668	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 669	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
 670	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 671	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 672	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 673	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 674	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
 675	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 676};
 677
 678static const u32 stoney_golden_settings_a11[] =
 679{
 680	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 681	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 682	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 683	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 684	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 685	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 686	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 687	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 688	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1,
 689	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010,
 690};
 691
 692static const u32 stoney_golden_common_all[] =
 693{
 694	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 695	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000,
 696	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
 697	mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001,
 698	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 699	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 700	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 701	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 702};
 703
 704static const u32 stoney_mgcg_cgcg_init[] =
 705{
 706	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 707	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
 708	mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
 709	mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
 710	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200,
 711};
 712
 713
 714static const char * const sq_edc_source_names[] = {
 715	"SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
 716	"SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
 717	"SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
 718	"SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
 719	"SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
 720	"SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
 721	"SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
 722};
 723
 724static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
 725static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
 726static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
 727static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev);
 728static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev);
 729static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev);
 730static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring);
 731static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring);
 732
 733#define CG_ACLK_CNTL__ACLK_DIVIDER_MASK                    0x0000007fL
 734#define CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT                  0x00000000L
 735
 736static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
 737{
 738	uint32_t data;
 739
 740	switch (adev->asic_type) {
 741	case CHIP_TOPAZ:
 742		amdgpu_device_program_register_sequence(adev,
 743							iceland_mgcg_cgcg_init,
 744							ARRAY_SIZE(iceland_mgcg_cgcg_init));
 745		amdgpu_device_program_register_sequence(adev,
 746							golden_settings_iceland_a11,
 747							ARRAY_SIZE(golden_settings_iceland_a11));
 748		amdgpu_device_program_register_sequence(adev,
 749							iceland_golden_common_all,
 750							ARRAY_SIZE(iceland_golden_common_all));
 751		break;
 752	case CHIP_FIJI:
 753		amdgpu_device_program_register_sequence(adev,
 754							fiji_mgcg_cgcg_init,
 755							ARRAY_SIZE(fiji_mgcg_cgcg_init));
 756		amdgpu_device_program_register_sequence(adev,
 757							golden_settings_fiji_a10,
 758							ARRAY_SIZE(golden_settings_fiji_a10));
 759		amdgpu_device_program_register_sequence(adev,
 760							fiji_golden_common_all,
 761							ARRAY_SIZE(fiji_golden_common_all));
 762		break;
 763
 764	case CHIP_TONGA:
 765		amdgpu_device_program_register_sequence(adev,
 766							tonga_mgcg_cgcg_init,
 767							ARRAY_SIZE(tonga_mgcg_cgcg_init));
 768		amdgpu_device_program_register_sequence(adev,
 769							golden_settings_tonga_a11,
 770							ARRAY_SIZE(golden_settings_tonga_a11));
 771		amdgpu_device_program_register_sequence(adev,
 772							tonga_golden_common_all,
 773							ARRAY_SIZE(tonga_golden_common_all));
 774		break;
 775	case CHIP_VEGAM:
 776		amdgpu_device_program_register_sequence(adev,
 777							golden_settings_vegam_a11,
 778							ARRAY_SIZE(golden_settings_vegam_a11));
 779		amdgpu_device_program_register_sequence(adev,
 780							vegam_golden_common_all,
 781							ARRAY_SIZE(vegam_golden_common_all));
 782		break;
 783	case CHIP_POLARIS11:
 784	case CHIP_POLARIS12:
 785		amdgpu_device_program_register_sequence(adev,
 786							golden_settings_polaris11_a11,
 787							ARRAY_SIZE(golden_settings_polaris11_a11));
 788		amdgpu_device_program_register_sequence(adev,
 789							polaris11_golden_common_all,
 790							ARRAY_SIZE(polaris11_golden_common_all));
 791		break;
 792	case CHIP_POLARIS10:
 793		amdgpu_device_program_register_sequence(adev,
 794							golden_settings_polaris10_a11,
 795							ARRAY_SIZE(golden_settings_polaris10_a11));
 796		amdgpu_device_program_register_sequence(adev,
 797							polaris10_golden_common_all,
 798							ARRAY_SIZE(polaris10_golden_common_all));
 799		data = RREG32_SMC(ixCG_ACLK_CNTL);
 800		data &= ~CG_ACLK_CNTL__ACLK_DIVIDER_MASK;
 801		data |= 0x18 << CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT;
 802		WREG32_SMC(ixCG_ACLK_CNTL, data);
 803		if ((adev->pdev->device == 0x67DF) && (adev->pdev->revision == 0xc7) &&
 804		    ((adev->pdev->subsystem_device == 0xb37 && adev->pdev->subsystem_vendor == 0x1002) ||
 805		     (adev->pdev->subsystem_device == 0x4a8 && adev->pdev->subsystem_vendor == 0x1043) ||
 806		     (adev->pdev->subsystem_device == 0x9480 && adev->pdev->subsystem_vendor == 0x1680))) {
 807			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1E, 0xDD);
 808			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1F, 0xD0);
 809		}
 810		break;
 811	case CHIP_CARRIZO:
 812		amdgpu_device_program_register_sequence(adev,
 813							cz_mgcg_cgcg_init,
 814							ARRAY_SIZE(cz_mgcg_cgcg_init));
 815		amdgpu_device_program_register_sequence(adev,
 816							cz_golden_settings_a11,
 817							ARRAY_SIZE(cz_golden_settings_a11));
 818		amdgpu_device_program_register_sequence(adev,
 819							cz_golden_common_all,
 820							ARRAY_SIZE(cz_golden_common_all));
 821		break;
 822	case CHIP_STONEY:
 823		amdgpu_device_program_register_sequence(adev,
 824							stoney_mgcg_cgcg_init,
 825							ARRAY_SIZE(stoney_mgcg_cgcg_init));
 826		amdgpu_device_program_register_sequence(adev,
 827							stoney_golden_settings_a11,
 828							ARRAY_SIZE(stoney_golden_settings_a11));
 829		amdgpu_device_program_register_sequence(adev,
 830							stoney_golden_common_all,
 831							ARRAY_SIZE(stoney_golden_common_all));
 832		break;
 833	default:
 834		break;
 835	}
 836}
 837
 
 
 
 
 
 
 
 838static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring)
 839{
 840	struct amdgpu_device *adev = ring->adev;
 
 841	uint32_t tmp = 0;
 842	unsigned i;
 843	int r;
 844
 845	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
 846	r = amdgpu_ring_alloc(ring, 3);
 847	if (r)
 848		return r;
 849
 
 
 
 
 
 850	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
 851	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
 852	amdgpu_ring_write(ring, 0xDEADBEEF);
 853	amdgpu_ring_commit(ring);
 854
 855	for (i = 0; i < adev->usec_timeout; i++) {
 856		tmp = RREG32(mmSCRATCH_REG0);
 857		if (tmp == 0xDEADBEEF)
 858			break;
 859		udelay(1);
 860	}
 861
 862	if (i >= adev->usec_timeout)
 863		r = -ETIMEDOUT;
 864
 
 
 865	return r;
 866}
 867
 868static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 869{
 870	struct amdgpu_device *adev = ring->adev;
 871	struct amdgpu_ib ib;
 872	struct dma_fence *f = NULL;
 873
 874	unsigned int index;
 875	uint64_t gpu_addr;
 876	uint32_t tmp;
 877	long r;
 878
 879	r = amdgpu_device_wb_get(adev, &index);
 880	if (r)
 881		return r;
 882
 883	gpu_addr = adev->wb.gpu_addr + (index * 4);
 884	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
 885	memset(&ib, 0, sizeof(ib));
 886
 887	r = amdgpu_ib_get(adev, NULL, 20, AMDGPU_IB_POOL_DIRECT, &ib);
 888	if (r)
 889		goto err1;
 890
 891	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
 892	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
 893	ib.ptr[2] = lower_32_bits(gpu_addr);
 894	ib.ptr[3] = upper_32_bits(gpu_addr);
 895	ib.ptr[4] = 0xDEADBEEF;
 896	ib.length_dw = 5;
 897
 898	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
 899	if (r)
 900		goto err2;
 901
 902	r = dma_fence_wait_timeout(f, false, timeout);
 903	if (r == 0) {
 904		r = -ETIMEDOUT;
 905		goto err2;
 906	} else if (r < 0) {
 907		goto err2;
 908	}
 909
 910	tmp = adev->wb.wb[index];
 911	if (tmp == 0xDEADBEEF)
 912		r = 0;
 913	else
 914		r = -EINVAL;
 915
 916err2:
 917	amdgpu_ib_free(adev, &ib, NULL);
 918	dma_fence_put(f);
 919err1:
 920	amdgpu_device_wb_free(adev, index);
 921	return r;
 922}
 923
 924
 925static void gfx_v8_0_free_microcode(struct amdgpu_device *adev)
 926{
 927	amdgpu_ucode_release(&adev->gfx.pfp_fw);
 928	amdgpu_ucode_release(&adev->gfx.me_fw);
 929	amdgpu_ucode_release(&adev->gfx.ce_fw);
 930	amdgpu_ucode_release(&adev->gfx.rlc_fw);
 931	amdgpu_ucode_release(&adev->gfx.mec_fw);
 
 
 
 
 
 932	if ((adev->asic_type != CHIP_STONEY) &&
 933	    (adev->asic_type != CHIP_TOPAZ))
 934		amdgpu_ucode_release(&adev->gfx.mec2_fw);
 
 935
 936	kfree(adev->gfx.rlc.register_list_format);
 937}
 938
 939static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
 940{
 941	const char *chip_name;
 
 942	int err;
 943	struct amdgpu_firmware_info *info = NULL;
 944	const struct common_firmware_header *header = NULL;
 945	const struct gfx_firmware_header_v1_0 *cp_hdr;
 946	const struct rlc_firmware_header_v2_0 *rlc_hdr;
 947	unsigned int *tmp = NULL, i;
 948
 949	DRM_DEBUG("\n");
 950
 951	switch (adev->asic_type) {
 952	case CHIP_TOPAZ:
 953		chip_name = "topaz";
 954		break;
 955	case CHIP_TONGA:
 956		chip_name = "tonga";
 957		break;
 958	case CHIP_CARRIZO:
 959		chip_name = "carrizo";
 960		break;
 961	case CHIP_FIJI:
 962		chip_name = "fiji";
 963		break;
 964	case CHIP_STONEY:
 965		chip_name = "stoney";
 966		break;
 967	case CHIP_POLARIS10:
 968		chip_name = "polaris10";
 969		break;
 970	case CHIP_POLARIS11:
 971		chip_name = "polaris11";
 972		break;
 973	case CHIP_POLARIS12:
 974		chip_name = "polaris12";
 975		break;
 976	case CHIP_VEGAM:
 977		chip_name = "vegam";
 978		break;
 979	default:
 980		BUG();
 981	}
 982
 983	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
 984		err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
 985					   "amdgpu/%s_pfp_2.bin", chip_name);
 986		if (err == -ENODEV) {
 987			err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
 988						   "amdgpu/%s_pfp.bin", chip_name);
 989		}
 990	} else {
 991		err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
 992					   "amdgpu/%s_pfp.bin", chip_name);
 993	}
 994	if (err)
 995		goto out;
 
 
 
 996	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
 997	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
 998	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
 999
1000	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1001		err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1002					   "amdgpu/%s_me_2.bin", chip_name);
1003		if (err == -ENODEV) {
1004			err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1005						   "amdgpu/%s_me.bin", chip_name);
1006		}
1007	} else {
1008		err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1009					   "amdgpu/%s_me.bin", chip_name);
1010	}
1011	if (err)
1012		goto out;
 
 
 
1013	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1014	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1015
1016	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1017
1018	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1019		err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1020					   "amdgpu/%s_ce_2.bin", chip_name);
1021		if (err == -ENODEV) {
1022			err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1023						   "amdgpu/%s_ce.bin", chip_name);
1024		}
1025	} else {
1026		err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1027					   "amdgpu/%s_ce.bin", chip_name);
1028	}
1029	if (err)
1030		goto out;
 
 
 
1031	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1032	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1033	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1034
1035	/*
1036	 * Support for MCBP/Virtualization in combination with chained IBs is
1037	 * formal released on feature version #46
1038	 */
1039	if (adev->gfx.ce_feature_version >= 46 &&
1040	    adev->gfx.pfp_feature_version >= 46) {
1041		adev->virt.chained_ib_support = true;
1042		DRM_INFO("Chained IB support enabled!\n");
1043	} else
1044		adev->virt.chained_ib_support = false;
1045
1046	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw,
1047				   "amdgpu/%s_rlc.bin", chip_name);
1048	if (err)
1049		goto out;
 
1050	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1051	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1052	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1053
1054	adev->gfx.rlc.save_and_restore_offset =
1055			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1056	adev->gfx.rlc.clear_state_descriptor_offset =
1057			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1058	adev->gfx.rlc.avail_scratch_ram_locations =
1059			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1060	adev->gfx.rlc.reg_restore_list_size =
1061			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1062	adev->gfx.rlc.reg_list_format_start =
1063			le32_to_cpu(rlc_hdr->reg_list_format_start);
1064	adev->gfx.rlc.reg_list_format_separate_start =
1065			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1066	adev->gfx.rlc.starting_offsets_start =
1067			le32_to_cpu(rlc_hdr->starting_offsets_start);
1068	adev->gfx.rlc.reg_list_format_size_bytes =
1069			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1070	adev->gfx.rlc.reg_list_size_bytes =
1071			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1072
1073	adev->gfx.rlc.register_list_format =
1074			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1075					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1076
1077	if (!adev->gfx.rlc.register_list_format) {
1078		err = -ENOMEM;
1079		goto out;
1080	}
1081
1082	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1083			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1084	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1085		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1086
1087	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1088
1089	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1090			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1091	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1092		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1093
1094	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1095		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1096					   "amdgpu/%s_mec_2.bin", chip_name);
1097		if (err == -ENODEV) {
1098			err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1099						   "amdgpu/%s_mec.bin", chip_name);
1100		}
1101	} else {
1102		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1103					   "amdgpu/%s_mec.bin", chip_name);
1104	}
1105	if (err)
1106		goto out;
 
 
 
1107	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1108	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1109	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1110
1111	if ((adev->asic_type != CHIP_STONEY) &&
1112	    (adev->asic_type != CHIP_TOPAZ)) {
1113		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1114			err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1115						   "amdgpu/%s_mec2_2.bin", chip_name);
1116			if (err == -ENODEV) {
1117				err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1118							   "amdgpu/%s_mec2.bin", chip_name);
1119			}
1120		} else {
1121			err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1122						   "amdgpu/%s_mec2.bin", chip_name);
1123		}
1124		if (!err) {
 
 
 
1125			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1126				adev->gfx.mec2_fw->data;
1127			adev->gfx.mec2_fw_version =
1128				le32_to_cpu(cp_hdr->header.ucode_version);
1129			adev->gfx.mec2_feature_version =
1130				le32_to_cpu(cp_hdr->ucode_feature_version);
1131		} else {
1132			err = 0;
1133			adev->gfx.mec2_fw = NULL;
1134		}
1135	}
1136
1137	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1138	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1139	info->fw = adev->gfx.pfp_fw;
1140	header = (const struct common_firmware_header *)info->fw->data;
1141	adev->firmware.fw_size +=
1142		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1143
1144	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1145	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1146	info->fw = adev->gfx.me_fw;
1147	header = (const struct common_firmware_header *)info->fw->data;
1148	adev->firmware.fw_size +=
1149		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1150
1151	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1152	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1153	info->fw = adev->gfx.ce_fw;
1154	header = (const struct common_firmware_header *)info->fw->data;
1155	adev->firmware.fw_size +=
1156		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1157
1158	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1159	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1160	info->fw = adev->gfx.rlc_fw;
1161	header = (const struct common_firmware_header *)info->fw->data;
1162	adev->firmware.fw_size +=
1163		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1164
1165	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1166	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1167	info->fw = adev->gfx.mec_fw;
1168	header = (const struct common_firmware_header *)info->fw->data;
1169	adev->firmware.fw_size +=
1170		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1171
1172	/* we need account JT in */
1173	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1174	adev->firmware.fw_size +=
1175		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1176
1177	if (amdgpu_sriov_vf(adev)) {
1178		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1179		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1180		info->fw = adev->gfx.mec_fw;
1181		adev->firmware.fw_size +=
1182			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1183	}
1184
1185	if (adev->gfx.mec2_fw) {
1186		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1187		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1188		info->fw = adev->gfx.mec2_fw;
1189		header = (const struct common_firmware_header *)info->fw->data;
1190		adev->firmware.fw_size +=
1191			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1192	}
1193
1194out:
1195	if (err) {
1196		dev_err(adev->dev, "gfx8: Failed to load firmware %s gfx firmware\n", chip_name);
1197		amdgpu_ucode_release(&adev->gfx.pfp_fw);
1198		amdgpu_ucode_release(&adev->gfx.me_fw);
1199		amdgpu_ucode_release(&adev->gfx.ce_fw);
1200		amdgpu_ucode_release(&adev->gfx.rlc_fw);
1201		amdgpu_ucode_release(&adev->gfx.mec_fw);
1202		amdgpu_ucode_release(&adev->gfx.mec2_fw);
 
 
 
 
 
 
 
 
1203	}
1204	return err;
1205}
1206
1207static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1208				    volatile u32 *buffer)
1209{
1210	u32 count = 0, i;
1211	const struct cs_section_def *sect = NULL;
1212	const struct cs_extent_def *ext = NULL;
1213
1214	if (adev->gfx.rlc.cs_data == NULL)
1215		return;
1216	if (buffer == NULL)
1217		return;
1218
1219	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1220	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1221
1222	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1223	buffer[count++] = cpu_to_le32(0x80000000);
1224	buffer[count++] = cpu_to_le32(0x80000000);
1225
1226	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1227		for (ext = sect->section; ext->extent != NULL; ++ext) {
1228			if (sect->id == SECT_CONTEXT) {
1229				buffer[count++] =
1230					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1231				buffer[count++] = cpu_to_le32(ext->reg_index -
1232						PACKET3_SET_CONTEXT_REG_START);
1233				for (i = 0; i < ext->reg_count; i++)
1234					buffer[count++] = cpu_to_le32(ext->extent[i]);
1235			} else {
1236				return;
1237			}
1238		}
1239	}
1240
1241	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1242	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1243			PACKET3_SET_CONTEXT_REG_START);
1244	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1245	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1246
1247	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1248	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1249
1250	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1251	buffer[count++] = cpu_to_le32(0);
1252}
1253
1254static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1255{
1256	if (adev->asic_type == CHIP_CARRIZO)
1257		return 5;
1258	else
1259		return 4;
1260}
1261
1262static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1263{
1264	const struct cs_section_def *cs_data;
1265	int r;
1266
1267	adev->gfx.rlc.cs_data = vi_cs_data;
1268
1269	cs_data = adev->gfx.rlc.cs_data;
1270
1271	if (cs_data) {
1272		/* init clear state block */
1273		r = amdgpu_gfx_rlc_init_csb(adev);
1274		if (r)
1275			return r;
1276	}
1277
1278	if ((adev->asic_type == CHIP_CARRIZO) ||
1279	    (adev->asic_type == CHIP_STONEY)) {
1280		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1281		r = amdgpu_gfx_rlc_init_cpt(adev);
1282		if (r)
1283			return r;
1284	}
1285
1286	/* init spm vmid with 0xf */
1287	if (adev->gfx.rlc.funcs->update_spm_vmid)
1288		adev->gfx.rlc.funcs->update_spm_vmid(adev, NULL, 0xf);
1289
1290	return 0;
1291}
1292
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1293static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1294{
1295	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1296}
1297
1298static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1299{
1300	int r;
1301	u32 *hpd;
1302	size_t mec_hpd_size;
1303
1304	bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1305
1306	/* take ownership of the relevant compute queues */
1307	amdgpu_gfx_compute_queue_acquire(adev);
1308
1309	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
1310	if (mec_hpd_size) {
1311		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1312					      AMDGPU_GEM_DOMAIN_VRAM |
1313					      AMDGPU_GEM_DOMAIN_GTT,
1314					      &adev->gfx.mec.hpd_eop_obj,
1315					      &adev->gfx.mec.hpd_eop_gpu_addr,
1316					      (void **)&hpd);
1317		if (r) {
1318			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1319			return r;
1320		}
1321
1322		memset(hpd, 0, mec_hpd_size);
1323
1324		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1325		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
 
 
 
 
 
 
1326	}
1327
 
 
 
 
 
1328	return 0;
1329}
1330
1331static const u32 vgpr_init_compute_shader[] =
1332{
1333	0x7e000209, 0x7e020208,
1334	0x7e040207, 0x7e060206,
1335	0x7e080205, 0x7e0a0204,
1336	0x7e0c0203, 0x7e0e0202,
1337	0x7e100201, 0x7e120200,
1338	0x7e140209, 0x7e160208,
1339	0x7e180207, 0x7e1a0206,
1340	0x7e1c0205, 0x7e1e0204,
1341	0x7e200203, 0x7e220202,
1342	0x7e240201, 0x7e260200,
1343	0x7e280209, 0x7e2a0208,
1344	0x7e2c0207, 0x7e2e0206,
1345	0x7e300205, 0x7e320204,
1346	0x7e340203, 0x7e360202,
1347	0x7e380201, 0x7e3a0200,
1348	0x7e3c0209, 0x7e3e0208,
1349	0x7e400207, 0x7e420206,
1350	0x7e440205, 0x7e460204,
1351	0x7e480203, 0x7e4a0202,
1352	0x7e4c0201, 0x7e4e0200,
1353	0x7e500209, 0x7e520208,
1354	0x7e540207, 0x7e560206,
1355	0x7e580205, 0x7e5a0204,
1356	0x7e5c0203, 0x7e5e0202,
1357	0x7e600201, 0x7e620200,
1358	0x7e640209, 0x7e660208,
1359	0x7e680207, 0x7e6a0206,
1360	0x7e6c0205, 0x7e6e0204,
1361	0x7e700203, 0x7e720202,
1362	0x7e740201, 0x7e760200,
1363	0x7e780209, 0x7e7a0208,
1364	0x7e7c0207, 0x7e7e0206,
1365	0xbf8a0000, 0xbf810000,
1366};
1367
1368static const u32 sgpr_init_compute_shader[] =
1369{
1370	0xbe8a0100, 0xbe8c0102,
1371	0xbe8e0104, 0xbe900106,
1372	0xbe920108, 0xbe940100,
1373	0xbe960102, 0xbe980104,
1374	0xbe9a0106, 0xbe9c0108,
1375	0xbe9e0100, 0xbea00102,
1376	0xbea20104, 0xbea40106,
1377	0xbea60108, 0xbea80100,
1378	0xbeaa0102, 0xbeac0104,
1379	0xbeae0106, 0xbeb00108,
1380	0xbeb20100, 0xbeb40102,
1381	0xbeb60104, 0xbeb80106,
1382	0xbeba0108, 0xbebc0100,
1383	0xbebe0102, 0xbec00104,
1384	0xbec20106, 0xbec40108,
1385	0xbec60100, 0xbec80102,
1386	0xbee60004, 0xbee70005,
1387	0xbeea0006, 0xbeeb0007,
1388	0xbee80008, 0xbee90009,
1389	0xbefc0000, 0xbf8a0000,
1390	0xbf810000, 0x00000000,
1391};
1392
1393static const u32 vgpr_init_regs[] =
1394{
1395	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1396	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1397	mmCOMPUTE_NUM_THREAD_X, 256*4,
1398	mmCOMPUTE_NUM_THREAD_Y, 1,
1399	mmCOMPUTE_NUM_THREAD_Z, 1,
1400	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1401	mmCOMPUTE_PGM_RSRC2, 20,
1402	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1403	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1404	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1405	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1406	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1407	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1408	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1409	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1410	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1411	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1412};
1413
1414static const u32 sgpr1_init_regs[] =
1415{
1416	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1417	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1418	mmCOMPUTE_NUM_THREAD_X, 256*5,
1419	mmCOMPUTE_NUM_THREAD_Y, 1,
1420	mmCOMPUTE_NUM_THREAD_Z, 1,
1421	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1422	mmCOMPUTE_PGM_RSRC2, 20,
1423	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1424	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1425	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1426	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1427	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1428	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1429	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1430	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1431	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1432	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1433};
1434
1435static const u32 sgpr2_init_regs[] =
1436{
1437	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1438	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1439	mmCOMPUTE_NUM_THREAD_X, 256*5,
1440	mmCOMPUTE_NUM_THREAD_Y, 1,
1441	mmCOMPUTE_NUM_THREAD_Z, 1,
1442	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1443	mmCOMPUTE_PGM_RSRC2, 20,
1444	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1445	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1446	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1447	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1448	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1449	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1450	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1451	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1452	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1453	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1454};
1455
1456static const u32 sec_ded_counter_registers[] =
1457{
1458	mmCPC_EDC_ATC_CNT,
1459	mmCPC_EDC_SCRATCH_CNT,
1460	mmCPC_EDC_UCODE_CNT,
1461	mmCPF_EDC_ATC_CNT,
1462	mmCPF_EDC_ROQ_CNT,
1463	mmCPF_EDC_TAG_CNT,
1464	mmCPG_EDC_ATC_CNT,
1465	mmCPG_EDC_DMA_CNT,
1466	mmCPG_EDC_TAG_CNT,
1467	mmDC_EDC_CSINVOC_CNT,
1468	mmDC_EDC_RESTORE_CNT,
1469	mmDC_EDC_STATE_CNT,
1470	mmGDS_EDC_CNT,
1471	mmGDS_EDC_GRBM_CNT,
1472	mmGDS_EDC_OA_DED,
1473	mmSPI_EDC_CNT,
1474	mmSQC_ATC_EDC_GATCL1_CNT,
1475	mmSQC_EDC_CNT,
1476	mmSQ_EDC_DED_CNT,
1477	mmSQ_EDC_INFO,
1478	mmSQ_EDC_SEC_CNT,
1479	mmTCC_EDC_CNT,
1480	mmTCP_ATC_EDC_GATCL1_CNT,
1481	mmTCP_EDC_CNT,
1482	mmTD_EDC_CNT
1483};
1484
1485static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1486{
1487	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1488	struct amdgpu_ib ib;
1489	struct dma_fence *f = NULL;
1490	int r, i;
1491	u32 tmp;
1492	unsigned total_size, vgpr_offset, sgpr_offset;
1493	u64 gpu_addr;
1494
1495	/* only supported on CZ */
1496	if (adev->asic_type != CHIP_CARRIZO)
1497		return 0;
1498
1499	/* bail if the compute ring is not ready */
1500	if (!ring->sched.ready)
1501		return 0;
1502
1503	tmp = RREG32(mmGB_EDC_MODE);
1504	WREG32(mmGB_EDC_MODE, 0);
1505
1506	total_size =
1507		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1508	total_size +=
1509		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1510	total_size +=
1511		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1512	total_size = ALIGN(total_size, 256);
1513	vgpr_offset = total_size;
1514	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1515	sgpr_offset = total_size;
1516	total_size += sizeof(sgpr_init_compute_shader);
1517
1518	/* allocate an indirect buffer to put the commands in */
1519	memset(&ib, 0, sizeof(ib));
1520	r = amdgpu_ib_get(adev, NULL, total_size,
1521					AMDGPU_IB_POOL_DIRECT, &ib);
1522	if (r) {
1523		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1524		return r;
1525	}
1526
1527	/* load the compute shaders */
1528	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1529		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1530
1531	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1532		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1533
1534	/* init the ib length to 0 */
1535	ib.length_dw = 0;
1536
1537	/* VGPR */
1538	/* write the register state for the compute dispatch */
1539	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1540		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1541		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1542		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1543	}
1544	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1545	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1546	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1547	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1548	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1549	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1550
1551	/* write dispatch packet */
1552	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1553	ib.ptr[ib.length_dw++] = 8; /* x */
1554	ib.ptr[ib.length_dw++] = 1; /* y */
1555	ib.ptr[ib.length_dw++] = 1; /* z */
1556	ib.ptr[ib.length_dw++] =
1557		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1558
1559	/* write CS partial flush packet */
1560	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1561	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1562
1563	/* SGPR1 */
1564	/* write the register state for the compute dispatch */
1565	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1566		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1567		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1568		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1569	}
1570	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1571	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1572	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1573	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1574	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1575	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1576
1577	/* write dispatch packet */
1578	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1579	ib.ptr[ib.length_dw++] = 8; /* x */
1580	ib.ptr[ib.length_dw++] = 1; /* y */
1581	ib.ptr[ib.length_dw++] = 1; /* z */
1582	ib.ptr[ib.length_dw++] =
1583		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1584
1585	/* write CS partial flush packet */
1586	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1587	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1588
1589	/* SGPR2 */
1590	/* write the register state for the compute dispatch */
1591	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1592		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1593		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1594		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1595	}
1596	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1597	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1598	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1599	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1600	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1601	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1602
1603	/* write dispatch packet */
1604	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1605	ib.ptr[ib.length_dw++] = 8; /* x */
1606	ib.ptr[ib.length_dw++] = 1; /* y */
1607	ib.ptr[ib.length_dw++] = 1; /* z */
1608	ib.ptr[ib.length_dw++] =
1609		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1610
1611	/* write CS partial flush packet */
1612	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1613	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1614
1615	/* shedule the ib on the ring */
1616	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1617	if (r) {
1618		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1619		goto fail;
1620	}
1621
1622	/* wait for the GPU to finish processing the IB */
1623	r = dma_fence_wait(f, false);
1624	if (r) {
1625		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1626		goto fail;
1627	}
1628
1629	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1630	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1631	WREG32(mmGB_EDC_MODE, tmp);
1632
1633	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1634	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1635	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1636
1637
1638	/* read back registers to clear the counters */
1639	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1640		RREG32(sec_ded_counter_registers[i]);
1641
1642fail:
1643	amdgpu_ib_free(adev, &ib, NULL);
1644	dma_fence_put(f);
1645
1646	return r;
1647}
1648
1649static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1650{
1651	u32 gb_addr_config;
1652	u32 mc_arb_ramcfg;
1653	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1654	u32 tmp;
1655	int ret;
1656
1657	switch (adev->asic_type) {
1658	case CHIP_TOPAZ:
1659		adev->gfx.config.max_shader_engines = 1;
1660		adev->gfx.config.max_tile_pipes = 2;
1661		adev->gfx.config.max_cu_per_sh = 6;
1662		adev->gfx.config.max_sh_per_se = 1;
1663		adev->gfx.config.max_backends_per_se = 2;
1664		adev->gfx.config.max_texture_channel_caches = 2;
1665		adev->gfx.config.max_gprs = 256;
1666		adev->gfx.config.max_gs_threads = 32;
1667		adev->gfx.config.max_hw_contexts = 8;
1668
1669		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1670		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1671		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1672		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1673		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1674		break;
1675	case CHIP_FIJI:
1676		adev->gfx.config.max_shader_engines = 4;
1677		adev->gfx.config.max_tile_pipes = 16;
1678		adev->gfx.config.max_cu_per_sh = 16;
1679		adev->gfx.config.max_sh_per_se = 1;
1680		adev->gfx.config.max_backends_per_se = 4;
1681		adev->gfx.config.max_texture_channel_caches = 16;
1682		adev->gfx.config.max_gprs = 256;
1683		adev->gfx.config.max_gs_threads = 32;
1684		adev->gfx.config.max_hw_contexts = 8;
1685
1686		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1687		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1688		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1689		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1690		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1691		break;
1692	case CHIP_POLARIS11:
1693	case CHIP_POLARIS12:
1694		ret = amdgpu_atombios_get_gfx_info(adev);
1695		if (ret)
1696			return ret;
1697		adev->gfx.config.max_gprs = 256;
1698		adev->gfx.config.max_gs_threads = 32;
1699		adev->gfx.config.max_hw_contexts = 8;
1700
1701		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1702		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1703		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1704		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1705		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1706		break;
1707	case CHIP_POLARIS10:
1708	case CHIP_VEGAM:
1709		ret = amdgpu_atombios_get_gfx_info(adev);
1710		if (ret)
1711			return ret;
1712		adev->gfx.config.max_gprs = 256;
1713		adev->gfx.config.max_gs_threads = 32;
1714		adev->gfx.config.max_hw_contexts = 8;
1715
1716		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1717		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1718		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1719		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1720		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1721		break;
1722	case CHIP_TONGA:
1723		adev->gfx.config.max_shader_engines = 4;
1724		adev->gfx.config.max_tile_pipes = 8;
1725		adev->gfx.config.max_cu_per_sh = 8;
1726		adev->gfx.config.max_sh_per_se = 1;
1727		adev->gfx.config.max_backends_per_se = 2;
1728		adev->gfx.config.max_texture_channel_caches = 8;
1729		adev->gfx.config.max_gprs = 256;
1730		adev->gfx.config.max_gs_threads = 32;
1731		adev->gfx.config.max_hw_contexts = 8;
1732
1733		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1734		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1735		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1736		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1737		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1738		break;
1739	case CHIP_CARRIZO:
1740		adev->gfx.config.max_shader_engines = 1;
1741		adev->gfx.config.max_tile_pipes = 2;
1742		adev->gfx.config.max_sh_per_se = 1;
1743		adev->gfx.config.max_backends_per_se = 2;
1744		adev->gfx.config.max_cu_per_sh = 8;
1745		adev->gfx.config.max_texture_channel_caches = 2;
1746		adev->gfx.config.max_gprs = 256;
1747		adev->gfx.config.max_gs_threads = 32;
1748		adev->gfx.config.max_hw_contexts = 8;
1749
1750		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1751		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1752		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1753		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1754		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1755		break;
1756	case CHIP_STONEY:
1757		adev->gfx.config.max_shader_engines = 1;
1758		adev->gfx.config.max_tile_pipes = 2;
1759		adev->gfx.config.max_sh_per_se = 1;
1760		adev->gfx.config.max_backends_per_se = 1;
1761		adev->gfx.config.max_cu_per_sh = 3;
1762		adev->gfx.config.max_texture_channel_caches = 2;
1763		adev->gfx.config.max_gprs = 256;
1764		adev->gfx.config.max_gs_threads = 16;
1765		adev->gfx.config.max_hw_contexts = 8;
1766
1767		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1768		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1769		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1770		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1771		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1772		break;
1773	default:
1774		adev->gfx.config.max_shader_engines = 2;
1775		adev->gfx.config.max_tile_pipes = 4;
1776		adev->gfx.config.max_cu_per_sh = 2;
1777		adev->gfx.config.max_sh_per_se = 1;
1778		adev->gfx.config.max_backends_per_se = 2;
1779		adev->gfx.config.max_texture_channel_caches = 4;
1780		adev->gfx.config.max_gprs = 256;
1781		adev->gfx.config.max_gs_threads = 32;
1782		adev->gfx.config.max_hw_contexts = 8;
1783
1784		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1785		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1786		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1787		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1788		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1789		break;
1790	}
1791
 
1792	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1793	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1794
1795	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
1796				MC_ARB_RAMCFG, NOOFBANK);
1797	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
1798				MC_ARB_RAMCFG, NOOFRANKS);
1799
1800	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1801	adev->gfx.config.mem_max_burst_length_bytes = 256;
1802	if (adev->flags & AMD_IS_APU) {
1803		/* Get memory bank mapping mode. */
1804		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1805		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1806		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1807
1808		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1809		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1810		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1811
1812		/* Validate settings in case only one DIMM installed. */
1813		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1814			dimm00_addr_map = 0;
1815		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1816			dimm01_addr_map = 0;
1817		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1818			dimm10_addr_map = 0;
1819		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1820			dimm11_addr_map = 0;
1821
1822		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1823		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1824		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1825			adev->gfx.config.mem_row_size_in_kb = 2;
1826		else
1827			adev->gfx.config.mem_row_size_in_kb = 1;
1828	} else {
1829		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1830		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1831		if (adev->gfx.config.mem_row_size_in_kb > 4)
1832			adev->gfx.config.mem_row_size_in_kb = 4;
1833	}
1834
1835	adev->gfx.config.shader_engine_tile_size = 32;
1836	adev->gfx.config.num_gpus = 1;
1837	adev->gfx.config.multi_gpu_tile_size = 64;
1838
1839	/* fix up row size */
1840	switch (adev->gfx.config.mem_row_size_in_kb) {
1841	case 1:
1842	default:
1843		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1844		break;
1845	case 2:
1846		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1847		break;
1848	case 4:
1849		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1850		break;
1851	}
1852	adev->gfx.config.gb_addr_config = gb_addr_config;
1853
1854	return 0;
1855}
1856
1857static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1858					int mec, int pipe, int queue)
1859{
1860	int r;
1861	unsigned irq_type;
1862	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1863	unsigned int hw_prio;
1864
1865	ring = &adev->gfx.compute_ring[ring_id];
1866
1867	/* mec0 is me1 */
1868	ring->me = mec + 1;
1869	ring->pipe = pipe;
1870	ring->queue = queue;
1871
1872	ring->ring_obj = NULL;
1873	ring->use_doorbell = true;
1874	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1875	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1876				+ (ring_id * GFX8_MEC_HPD_SIZE);
1877	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1878
1879	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1880		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1881		+ ring->pipe;
1882
1883	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1884			AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
1885	/* type-2 packets are deprecated on MEC, use type-3 instead */
1886	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1887			     hw_prio, NULL);
1888	if (r)
1889		return r;
1890
1891
1892	return 0;
1893}
1894
1895static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1896
1897static int gfx_v8_0_sw_init(struct amdgpu_ip_block *ip_block)
1898{
1899	int i, j, k, r, ring_id;
1900	int xcc_id = 0;
1901	struct amdgpu_ring *ring;
1902	struct amdgpu_device *adev = ip_block->adev;
 
1903
1904	switch (adev->asic_type) {
1905	case CHIP_TONGA:
1906	case CHIP_CARRIZO:
1907	case CHIP_FIJI:
1908	case CHIP_POLARIS10:
1909	case CHIP_POLARIS11:
1910	case CHIP_POLARIS12:
1911	case CHIP_VEGAM:
1912		adev->gfx.mec.num_mec = 2;
1913		break;
1914	case CHIP_TOPAZ:
1915	case CHIP_STONEY:
1916	default:
1917		adev->gfx.mec.num_mec = 1;
1918		break;
1919	}
1920
1921	adev->gfx.mec.num_pipe_per_mec = 4;
1922	adev->gfx.mec.num_queue_per_pipe = 8;
1923
1924	/* EOP Event */
1925	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1926	if (r)
1927		return r;
1928
1929	/* Privileged reg */
1930	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1931			      &adev->gfx.priv_reg_irq);
1932	if (r)
1933		return r;
1934
1935	/* Privileged inst */
1936	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1937			      &adev->gfx.priv_inst_irq);
1938	if (r)
1939		return r;
1940
1941	/* Add CP EDC/ECC irq  */
1942	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1943			      &adev->gfx.cp_ecc_error_irq);
1944	if (r)
1945		return r;
1946
1947	/* SQ interrupts. */
1948	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
1949			      &adev->gfx.sq_irq);
1950	if (r) {
1951		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
1952		return r;
1953	}
1954
1955	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
1956
1957	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1958
 
 
1959	r = gfx_v8_0_init_microcode(adev);
1960	if (r) {
1961		DRM_ERROR("Failed to load gfx firmware!\n");
1962		return r;
1963	}
1964
1965	r = adev->gfx.rlc.funcs->init(adev);
1966	if (r) {
1967		DRM_ERROR("Failed to init rlc BOs!\n");
1968		return r;
1969	}
1970
1971	r = gfx_v8_0_mec_init(adev);
1972	if (r) {
1973		DRM_ERROR("Failed to init MEC BOs!\n");
1974		return r;
1975	}
1976
1977	/* set up the gfx ring */
1978	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
1979		ring = &adev->gfx.gfx_ring[i];
1980		ring->ring_obj = NULL;
1981		sprintf(ring->name, "gfx");
1982		/* no gfx doorbells on iceland */
1983		if (adev->asic_type != CHIP_TOPAZ) {
1984			ring->use_doorbell = true;
1985			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
1986		}
1987
1988		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
1989				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
1990				     AMDGPU_RING_PRIO_DEFAULT, NULL);
1991		if (r)
1992			return r;
1993	}
1994
1995
1996	/* set up the compute queues - allocate horizontally across pipes */
1997	ring_id = 0;
1998	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1999		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2000			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2001				if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
2002								     k, j))
2003					continue;
2004
2005				r = gfx_v8_0_compute_ring_init(adev,
2006								ring_id,
2007								i, k, j);
2008				if (r)
2009					return r;
2010
2011				ring_id++;
2012			}
2013		}
2014	}
2015
2016	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE, 0);
2017	if (r) {
2018		DRM_ERROR("Failed to init KIQ BOs!\n");
2019		return r;
2020	}
2021
2022	r = amdgpu_gfx_kiq_init_ring(adev, xcc_id);
 
2023	if (r)
2024		return r;
2025
2026	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2027	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation), 0);
2028	if (r)
2029		return r;
2030
2031	adev->gfx.ce_ram_size = 0x8000;
2032
2033	r = gfx_v8_0_gpu_early_init(adev);
2034	if (r)
2035		return r;
2036
2037	return 0;
2038}
2039
2040static int gfx_v8_0_sw_fini(struct amdgpu_ip_block *ip_block)
2041{
2042	struct amdgpu_device *adev = ip_block->adev;
2043	int i;
2044
2045	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2046		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2047	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2048		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2049
2050	amdgpu_gfx_mqd_sw_fini(adev, 0);
2051	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
2052	amdgpu_gfx_kiq_fini(adev, 0);
2053
2054	gfx_v8_0_mec_fini(adev);
2055	amdgpu_gfx_rlc_fini(adev);
2056	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2057				&adev->gfx.rlc.clear_state_gpu_addr,
2058				(void **)&adev->gfx.rlc.cs_ptr);
2059	if ((adev->asic_type == CHIP_CARRIZO) ||
2060	    (adev->asic_type == CHIP_STONEY)) {
2061		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2062				&adev->gfx.rlc.cp_table_gpu_addr,
2063				(void **)&adev->gfx.rlc.cp_table_ptr);
2064	}
2065	gfx_v8_0_free_microcode(adev);
2066
2067	return 0;
2068}
2069
2070static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2071{
2072	uint32_t *modearray, *mod2array;
2073	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2074	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2075	u32 reg_offset;
2076
2077	modearray = adev->gfx.config.tile_mode_array;
2078	mod2array = adev->gfx.config.macrotile_mode_array;
2079
2080	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2081		modearray[reg_offset] = 0;
2082
2083	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2084		mod2array[reg_offset] = 0;
2085
2086	switch (adev->asic_type) {
2087	case CHIP_TOPAZ:
2088		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2089				PIPE_CONFIG(ADDR_SURF_P2) |
2090				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2091				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2092		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2093				PIPE_CONFIG(ADDR_SURF_P2) |
2094				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2095				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2096		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2097				PIPE_CONFIG(ADDR_SURF_P2) |
2098				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2099				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2100		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2101				PIPE_CONFIG(ADDR_SURF_P2) |
2102				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2103				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2104		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2105				PIPE_CONFIG(ADDR_SURF_P2) |
2106				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2107				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2108		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2109				PIPE_CONFIG(ADDR_SURF_P2) |
2110				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2111				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2112		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2113				PIPE_CONFIG(ADDR_SURF_P2) |
2114				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2115				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2116		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2117				PIPE_CONFIG(ADDR_SURF_P2));
2118		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2119				PIPE_CONFIG(ADDR_SURF_P2) |
2120				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2121				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2122		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2123				 PIPE_CONFIG(ADDR_SURF_P2) |
2124				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2125				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2126		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2127				 PIPE_CONFIG(ADDR_SURF_P2) |
2128				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2129				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2130		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2131				 PIPE_CONFIG(ADDR_SURF_P2) |
2132				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2133				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2134		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2135				 PIPE_CONFIG(ADDR_SURF_P2) |
2136				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2137				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2138		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2139				 PIPE_CONFIG(ADDR_SURF_P2) |
2140				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2141				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2142		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2143				 PIPE_CONFIG(ADDR_SURF_P2) |
2144				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2145				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2146		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2147				 PIPE_CONFIG(ADDR_SURF_P2) |
2148				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2149				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2150		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2151				 PIPE_CONFIG(ADDR_SURF_P2) |
2152				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2153				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2154		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2155				 PIPE_CONFIG(ADDR_SURF_P2) |
2156				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2157				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2158		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2159				 PIPE_CONFIG(ADDR_SURF_P2) |
2160				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2161				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2162		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2163				 PIPE_CONFIG(ADDR_SURF_P2) |
2164				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2165				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2166		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2167				 PIPE_CONFIG(ADDR_SURF_P2) |
2168				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2169				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2170		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2171				 PIPE_CONFIG(ADDR_SURF_P2) |
2172				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2173				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2174		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2175				 PIPE_CONFIG(ADDR_SURF_P2) |
2176				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2177				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2178		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2179				 PIPE_CONFIG(ADDR_SURF_P2) |
2180				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2181				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2182		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2183				 PIPE_CONFIG(ADDR_SURF_P2) |
2184				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2185				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2186		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2187				 PIPE_CONFIG(ADDR_SURF_P2) |
2188				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2189				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2190
2191		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2192				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2193				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2194				NUM_BANKS(ADDR_SURF_8_BANK));
2195		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2196				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2197				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2198				NUM_BANKS(ADDR_SURF_8_BANK));
2199		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2200				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2201				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2202				NUM_BANKS(ADDR_SURF_8_BANK));
2203		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2204				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2205				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2206				NUM_BANKS(ADDR_SURF_8_BANK));
2207		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2208				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2209				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2210				NUM_BANKS(ADDR_SURF_8_BANK));
2211		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2212				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2213				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2214				NUM_BANKS(ADDR_SURF_8_BANK));
2215		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2216				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2217				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2218				NUM_BANKS(ADDR_SURF_8_BANK));
2219		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2220				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2221				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2222				NUM_BANKS(ADDR_SURF_16_BANK));
2223		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2224				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2225				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2226				NUM_BANKS(ADDR_SURF_16_BANK));
2227		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2228				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2229				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2230				 NUM_BANKS(ADDR_SURF_16_BANK));
2231		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2232				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2233				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2234				 NUM_BANKS(ADDR_SURF_16_BANK));
2235		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2236				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2237				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2238				 NUM_BANKS(ADDR_SURF_16_BANK));
2239		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2240				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2241				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2242				 NUM_BANKS(ADDR_SURF_16_BANK));
2243		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2244				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2245				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2246				 NUM_BANKS(ADDR_SURF_8_BANK));
2247
2248		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2249			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2250			    reg_offset != 23)
2251				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2252
2253		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2254			if (reg_offset != 7)
2255				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2256
2257		break;
2258	case CHIP_FIJI:
2259	case CHIP_VEGAM:
2260		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2261				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2262				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2263				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2264		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2265				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2266				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2267				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2268		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2269				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2270				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2271				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2272		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2273				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2274				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2275				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2276		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2277				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2278				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2279				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2280		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2281				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2282				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2283				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2284		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2285				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2286				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2287				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2288		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2289				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2290				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2291				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2292		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2293				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2294		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2295				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2296				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2297				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2298		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2299				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2300				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2301				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2302		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2303				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2304				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2305				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2306		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2307				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2308				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2309				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2310		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2311				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2312				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2313				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2314		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2315				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2316				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2317				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2318		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2319				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2320				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2321				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2322		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2323				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2324				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2325				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2326		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2327				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2328				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2329				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2330		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2331				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2332				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2333				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2334		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2335				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2336				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2337				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2338		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2339				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2340				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2341				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2342		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2343				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2344				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2345				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2346		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2347				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2348				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2349				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2350		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2351				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2352				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2353				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2354		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2355				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2356				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2357				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2358		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2359				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2360				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2361				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2362		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2363				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2364				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2365				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2366		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2367				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2368				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2369				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2370		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2371				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2372				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2373				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2374		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2375				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2376				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2377				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2378		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2379				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2380				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2381				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2382
2383		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2384				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2385				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2386				NUM_BANKS(ADDR_SURF_8_BANK));
2387		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2388				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2389				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2390				NUM_BANKS(ADDR_SURF_8_BANK));
2391		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2392				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2393				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2394				NUM_BANKS(ADDR_SURF_8_BANK));
2395		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2396				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2397				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2398				NUM_BANKS(ADDR_SURF_8_BANK));
2399		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2400				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2401				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2402				NUM_BANKS(ADDR_SURF_8_BANK));
2403		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2404				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2405				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2406				NUM_BANKS(ADDR_SURF_8_BANK));
2407		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2408				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2409				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2410				NUM_BANKS(ADDR_SURF_8_BANK));
2411		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2412				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2413				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2414				NUM_BANKS(ADDR_SURF_8_BANK));
2415		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2416				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2417				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2418				NUM_BANKS(ADDR_SURF_8_BANK));
2419		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2420				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2421				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2422				 NUM_BANKS(ADDR_SURF_8_BANK));
2423		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2424				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2425				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2426				 NUM_BANKS(ADDR_SURF_8_BANK));
2427		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2428				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2429				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2430				 NUM_BANKS(ADDR_SURF_8_BANK));
2431		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2432				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2433				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2434				 NUM_BANKS(ADDR_SURF_8_BANK));
2435		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2436				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2437				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2438				 NUM_BANKS(ADDR_SURF_4_BANK));
2439
2440		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2441			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2442
2443		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2444			if (reg_offset != 7)
2445				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2446
2447		break;
2448	case CHIP_TONGA:
2449		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2450				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2451				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2452				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2453		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2454				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2455				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2456				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2457		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2458				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2459				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2460				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2461		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2462				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2463				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2464				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2465		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2466				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2467				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2468				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2469		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2470				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2471				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2472				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2473		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2474				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2475				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2476				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2477		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2478				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2479				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2480				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2481		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2482				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2483		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2484				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2485				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2486				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2487		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2488				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2489				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2490				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2491		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2492				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2493				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2494				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2495		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2496				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2497				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2498				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2499		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2500				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2501				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2502				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2503		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2504				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2505				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2506				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2507		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2508				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2509				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2510				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2511		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2512				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2513				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2514				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2515		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2516				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2517				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2518				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2519		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2520				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2521				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2522				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2523		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2524				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2525				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2526				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2527		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2528				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2529				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2530				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2531		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2532				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2533				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2534				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2535		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2536				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2537				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2538				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2539		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2540				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2541				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2542				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2543		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2544				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2545				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2546				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2547		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2548				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2549				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2550				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2551		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2552				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2553				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2554				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2555		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2556				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2557				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2558				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2559		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2560				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2561				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2562				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2563		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2564				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2565				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2566				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2567		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2568				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2569				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2570				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2571
2572		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2573				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2574				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2575				NUM_BANKS(ADDR_SURF_16_BANK));
2576		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2577				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2578				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2579				NUM_BANKS(ADDR_SURF_16_BANK));
2580		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2581				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2582				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2583				NUM_BANKS(ADDR_SURF_16_BANK));
2584		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2585				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2586				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2587				NUM_BANKS(ADDR_SURF_16_BANK));
2588		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2589				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2590				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2591				NUM_BANKS(ADDR_SURF_16_BANK));
2592		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2593				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2594				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2595				NUM_BANKS(ADDR_SURF_16_BANK));
2596		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2597				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2598				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2599				NUM_BANKS(ADDR_SURF_16_BANK));
2600		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2601				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2602				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2603				NUM_BANKS(ADDR_SURF_16_BANK));
2604		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2605				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2606				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2607				NUM_BANKS(ADDR_SURF_16_BANK));
2608		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2609				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2610				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2611				 NUM_BANKS(ADDR_SURF_16_BANK));
2612		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2613				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2614				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2615				 NUM_BANKS(ADDR_SURF_16_BANK));
2616		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2617				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2618				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2619				 NUM_BANKS(ADDR_SURF_8_BANK));
2620		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2621				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2622				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2623				 NUM_BANKS(ADDR_SURF_4_BANK));
2624		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2625				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2626				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2627				 NUM_BANKS(ADDR_SURF_4_BANK));
2628
2629		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2630			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2631
2632		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2633			if (reg_offset != 7)
2634				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2635
2636		break;
2637	case CHIP_POLARIS11:
2638	case CHIP_POLARIS12:
2639		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2640				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2641				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2642				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2643		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2644				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2645				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2646				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2647		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2648				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2649				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2650				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2651		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2652				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2653				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2654				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2655		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2656				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2657				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2658				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2659		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2660				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2661				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2662				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2663		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2664				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2665				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2666				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2667		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2668				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2669				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2670				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2671		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2672				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2673		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2674				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2675				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2676				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2677		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2678				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2679				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2680				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2681		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2682				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2683				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2684				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2685		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2686				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2687				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2688				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2689		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2690				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2691				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2692				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2693		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2694				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2695				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2696				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2697		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2698				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2699				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2700				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2701		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2702				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2703				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2704				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2705		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2706				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2707				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2708				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2709		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2710				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2711				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2712				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2713		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2714				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2715				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2716				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2717		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2718				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2719				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2720				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2721		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2722				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2723				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2724				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2725		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2726				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2727				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2728				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2729		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2730				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2731				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2732				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2733		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2734				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2735				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2736				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2737		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2738				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2739				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2740				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2741		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2742				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2743				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2744				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2745		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2746				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2747				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2748				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2749		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2750				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2751				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2752				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2753		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2754				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2755				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2756				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2757		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2758				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2759				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2760				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2761
2762		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2763				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2764				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2765				NUM_BANKS(ADDR_SURF_16_BANK));
2766
2767		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2768				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2769				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2770				NUM_BANKS(ADDR_SURF_16_BANK));
2771
2772		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2773				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2774				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2775				NUM_BANKS(ADDR_SURF_16_BANK));
2776
2777		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2778				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2779				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2780				NUM_BANKS(ADDR_SURF_16_BANK));
2781
2782		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2783				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2784				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2785				NUM_BANKS(ADDR_SURF_16_BANK));
2786
2787		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2788				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2789				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2790				NUM_BANKS(ADDR_SURF_16_BANK));
2791
2792		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2793				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2794				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2795				NUM_BANKS(ADDR_SURF_16_BANK));
2796
2797		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2798				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2799				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2800				NUM_BANKS(ADDR_SURF_16_BANK));
2801
2802		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2803				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2804				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2805				NUM_BANKS(ADDR_SURF_16_BANK));
2806
2807		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2808				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2809				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2810				NUM_BANKS(ADDR_SURF_16_BANK));
2811
2812		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2813				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2814				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2815				NUM_BANKS(ADDR_SURF_16_BANK));
2816
2817		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2818				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2819				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2820				NUM_BANKS(ADDR_SURF_16_BANK));
2821
2822		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2823				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2824				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2825				NUM_BANKS(ADDR_SURF_8_BANK));
2826
2827		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2828				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2829				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2830				NUM_BANKS(ADDR_SURF_4_BANK));
2831
2832		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2833			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2834
2835		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2836			if (reg_offset != 7)
2837				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2838
2839		break;
2840	case CHIP_POLARIS10:
2841		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2842				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2843				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2844				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2845		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2846				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2847				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2848				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2849		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2850				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2851				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2852				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2853		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2854				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2855				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2856				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2857		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2858				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2859				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2860				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2861		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2862				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2863				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2864				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2865		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2866				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2867				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2868				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2869		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2870				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2871				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2872				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2873		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2874				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2875		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2876				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2877				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2878				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2879		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2880				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2881				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2882				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2883		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2884				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2885				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2886				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2887		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2888				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2889				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2890				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2891		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2892				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2893				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2894				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2895		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2896				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2897				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2898				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2899		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2900				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2901				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2902				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2903		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2904				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2905				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2906				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2907		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2908				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2909				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2910				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2911		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2912				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2913				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2914				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2915		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2916				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2917				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2918				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2919		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2920				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2921				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2922				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2923		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2924				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2925				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2926				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2927		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2928				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2929				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2930				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2931		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2932				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2933				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2934				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2935		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2936				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2937				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2938				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2939		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2940				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2941				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2942				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2943		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2944				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2945				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2946				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2947		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2948				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2949				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2950				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2951		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2952				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2953				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2954				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2955		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2956				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2957				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2958				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2959		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2960				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2961				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2962				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2963
2964		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2965				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2966				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2967				NUM_BANKS(ADDR_SURF_16_BANK));
2968
2969		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2970				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2971				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2972				NUM_BANKS(ADDR_SURF_16_BANK));
2973
2974		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2975				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2976				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2977				NUM_BANKS(ADDR_SURF_16_BANK));
2978
2979		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2980				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2981				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2982				NUM_BANKS(ADDR_SURF_16_BANK));
2983
2984		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2985				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2986				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2987				NUM_BANKS(ADDR_SURF_16_BANK));
2988
2989		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2990				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2991				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2992				NUM_BANKS(ADDR_SURF_16_BANK));
2993
2994		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2995				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2996				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2997				NUM_BANKS(ADDR_SURF_16_BANK));
2998
2999		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3000				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3001				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3002				NUM_BANKS(ADDR_SURF_16_BANK));
3003
3004		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3005				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3006				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3007				NUM_BANKS(ADDR_SURF_16_BANK));
3008
3009		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3010				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3011				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3012				NUM_BANKS(ADDR_SURF_16_BANK));
3013
3014		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3015				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3016				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3017				NUM_BANKS(ADDR_SURF_16_BANK));
3018
3019		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3020				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3021				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3022				NUM_BANKS(ADDR_SURF_8_BANK));
3023
3024		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3025				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3026				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3027				NUM_BANKS(ADDR_SURF_4_BANK));
3028
3029		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3030				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3031				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3032				NUM_BANKS(ADDR_SURF_4_BANK));
3033
3034		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3035			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3036
3037		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3038			if (reg_offset != 7)
3039				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3040
3041		break;
3042	case CHIP_STONEY:
3043		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3044				PIPE_CONFIG(ADDR_SURF_P2) |
3045				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3046				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3047		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3048				PIPE_CONFIG(ADDR_SURF_P2) |
3049				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3050				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3051		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3052				PIPE_CONFIG(ADDR_SURF_P2) |
3053				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3054				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3055		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3056				PIPE_CONFIG(ADDR_SURF_P2) |
3057				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3058				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3059		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3060				PIPE_CONFIG(ADDR_SURF_P2) |
3061				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3062				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3063		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3064				PIPE_CONFIG(ADDR_SURF_P2) |
3065				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3066				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3067		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3068				PIPE_CONFIG(ADDR_SURF_P2) |
3069				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3070				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3071		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3072				PIPE_CONFIG(ADDR_SURF_P2));
3073		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3074				PIPE_CONFIG(ADDR_SURF_P2) |
3075				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3076				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3077		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3078				 PIPE_CONFIG(ADDR_SURF_P2) |
3079				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3080				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3081		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3082				 PIPE_CONFIG(ADDR_SURF_P2) |
3083				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3084				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3085		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3086				 PIPE_CONFIG(ADDR_SURF_P2) |
3087				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3088				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3089		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3090				 PIPE_CONFIG(ADDR_SURF_P2) |
3091				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3092				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3093		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3094				 PIPE_CONFIG(ADDR_SURF_P2) |
3095				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3096				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3097		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3098				 PIPE_CONFIG(ADDR_SURF_P2) |
3099				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3100				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3101		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3102				 PIPE_CONFIG(ADDR_SURF_P2) |
3103				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3104				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3105		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3106				 PIPE_CONFIG(ADDR_SURF_P2) |
3107				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3108				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3109		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3110				 PIPE_CONFIG(ADDR_SURF_P2) |
3111				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3112				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3113		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3114				 PIPE_CONFIG(ADDR_SURF_P2) |
3115				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3116				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3117		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3118				 PIPE_CONFIG(ADDR_SURF_P2) |
3119				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3120				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3121		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3122				 PIPE_CONFIG(ADDR_SURF_P2) |
3123				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3124				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3125		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3126				 PIPE_CONFIG(ADDR_SURF_P2) |
3127				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3128				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3129		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3130				 PIPE_CONFIG(ADDR_SURF_P2) |
3131				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3132				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3133		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3134				 PIPE_CONFIG(ADDR_SURF_P2) |
3135				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3136				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3137		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3138				 PIPE_CONFIG(ADDR_SURF_P2) |
3139				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3140				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3141		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3142				 PIPE_CONFIG(ADDR_SURF_P2) |
3143				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3144				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3145
3146		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3147				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3148				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3149				NUM_BANKS(ADDR_SURF_8_BANK));
3150		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3151				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3152				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3153				NUM_BANKS(ADDR_SURF_8_BANK));
3154		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3155				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3156				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3157				NUM_BANKS(ADDR_SURF_8_BANK));
3158		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3159				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3160				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3161				NUM_BANKS(ADDR_SURF_8_BANK));
3162		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3163				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3164				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3165				NUM_BANKS(ADDR_SURF_8_BANK));
3166		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3167				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3168				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3169				NUM_BANKS(ADDR_SURF_8_BANK));
3170		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3171				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3172				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3173				NUM_BANKS(ADDR_SURF_8_BANK));
3174		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3175				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3176				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3177				NUM_BANKS(ADDR_SURF_16_BANK));
3178		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3179				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3180				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3181				NUM_BANKS(ADDR_SURF_16_BANK));
3182		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3183				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3184				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3185				 NUM_BANKS(ADDR_SURF_16_BANK));
3186		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3187				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3188				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3189				 NUM_BANKS(ADDR_SURF_16_BANK));
3190		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3191				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3192				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3193				 NUM_BANKS(ADDR_SURF_16_BANK));
3194		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3195				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3196				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3197				 NUM_BANKS(ADDR_SURF_16_BANK));
3198		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3199				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3200				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3201				 NUM_BANKS(ADDR_SURF_8_BANK));
3202
3203		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3204			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3205			    reg_offset != 23)
3206				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3207
3208		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3209			if (reg_offset != 7)
3210				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3211
3212		break;
3213	default:
3214		dev_warn(adev->dev,
3215			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3216			 adev->asic_type);
3217		fallthrough;
3218
3219	case CHIP_CARRIZO:
3220		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3221				PIPE_CONFIG(ADDR_SURF_P2) |
3222				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3223				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3224		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3225				PIPE_CONFIG(ADDR_SURF_P2) |
3226				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3227				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3228		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3229				PIPE_CONFIG(ADDR_SURF_P2) |
3230				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3231				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3232		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3233				PIPE_CONFIG(ADDR_SURF_P2) |
3234				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3235				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3236		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3237				PIPE_CONFIG(ADDR_SURF_P2) |
3238				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3239				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3240		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3241				PIPE_CONFIG(ADDR_SURF_P2) |
3242				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3243				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3244		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3245				PIPE_CONFIG(ADDR_SURF_P2) |
3246				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3247				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3248		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3249				PIPE_CONFIG(ADDR_SURF_P2));
3250		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3251				PIPE_CONFIG(ADDR_SURF_P2) |
3252				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3253				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3254		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3255				 PIPE_CONFIG(ADDR_SURF_P2) |
3256				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3257				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3258		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3259				 PIPE_CONFIG(ADDR_SURF_P2) |
3260				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3261				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3262		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3263				 PIPE_CONFIG(ADDR_SURF_P2) |
3264				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3265				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3266		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3267				 PIPE_CONFIG(ADDR_SURF_P2) |
3268				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3269				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3270		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3271				 PIPE_CONFIG(ADDR_SURF_P2) |
3272				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3273				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3274		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3275				 PIPE_CONFIG(ADDR_SURF_P2) |
3276				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3277				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3278		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3279				 PIPE_CONFIG(ADDR_SURF_P2) |
3280				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3281				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3282		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3283				 PIPE_CONFIG(ADDR_SURF_P2) |
3284				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3285				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3286		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3287				 PIPE_CONFIG(ADDR_SURF_P2) |
3288				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3289				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3290		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3291				 PIPE_CONFIG(ADDR_SURF_P2) |
3292				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3293				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3294		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3295				 PIPE_CONFIG(ADDR_SURF_P2) |
3296				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3297				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3298		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3299				 PIPE_CONFIG(ADDR_SURF_P2) |
3300				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3301				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3302		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3303				 PIPE_CONFIG(ADDR_SURF_P2) |
3304				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3305				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3306		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3307				 PIPE_CONFIG(ADDR_SURF_P2) |
3308				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3309				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3310		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3311				 PIPE_CONFIG(ADDR_SURF_P2) |
3312				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3313				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3314		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3315				 PIPE_CONFIG(ADDR_SURF_P2) |
3316				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3317				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3318		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3319				 PIPE_CONFIG(ADDR_SURF_P2) |
3320				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3321				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3322
3323		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3324				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3325				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3326				NUM_BANKS(ADDR_SURF_8_BANK));
3327		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3328				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3329				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3330				NUM_BANKS(ADDR_SURF_8_BANK));
3331		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3332				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3333				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3334				NUM_BANKS(ADDR_SURF_8_BANK));
3335		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3336				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3337				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3338				NUM_BANKS(ADDR_SURF_8_BANK));
3339		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3340				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3341				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3342				NUM_BANKS(ADDR_SURF_8_BANK));
3343		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3344				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3345				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3346				NUM_BANKS(ADDR_SURF_8_BANK));
3347		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3348				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3349				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3350				NUM_BANKS(ADDR_SURF_8_BANK));
3351		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3352				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3353				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3354				NUM_BANKS(ADDR_SURF_16_BANK));
3355		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3356				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3357				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3358				NUM_BANKS(ADDR_SURF_16_BANK));
3359		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3360				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3361				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3362				 NUM_BANKS(ADDR_SURF_16_BANK));
3363		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3364				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3365				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3366				 NUM_BANKS(ADDR_SURF_16_BANK));
3367		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3368				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3369				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3370				 NUM_BANKS(ADDR_SURF_16_BANK));
3371		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3372				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3373				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3374				 NUM_BANKS(ADDR_SURF_16_BANK));
3375		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3376				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3377				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3378				 NUM_BANKS(ADDR_SURF_8_BANK));
3379
3380		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3381			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3382			    reg_offset != 23)
3383				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3384
3385		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3386			if (reg_offset != 7)
3387				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3388
3389		break;
3390	}
3391}
3392
3393static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3394				  u32 se_num, u32 sh_num, u32 instance,
3395				  int xcc_id)
3396{
3397	u32 data;
3398
3399	if (instance == 0xffffffff)
3400		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3401	else
3402		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3403
3404	if (se_num == 0xffffffff)
3405		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3406	else
3407		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3408
3409	if (sh_num == 0xffffffff)
3410		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3411	else
3412		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3413
3414	WREG32(mmGRBM_GFX_INDEX, data);
3415}
3416
3417static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3418				  u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
3419{
3420	vi_srbm_select(adev, me, pipe, q, vm);
3421}
3422
3423static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3424{
3425	u32 data, mask;
3426
3427	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3428		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3429
3430	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3431
3432	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3433					 adev->gfx.config.max_sh_per_se);
3434
3435	return (~data) & mask;
3436}
3437
3438static void
3439gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3440{
3441	switch (adev->asic_type) {
3442	case CHIP_FIJI:
3443	case CHIP_VEGAM:
3444		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3445			  RB_XSEL2(1) | PKR_MAP(2) |
3446			  PKR_XSEL(1) | PKR_YSEL(1) |
3447			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3448		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3449			   SE_PAIR_YSEL(2);
3450		break;
3451	case CHIP_TONGA:
3452	case CHIP_POLARIS10:
3453		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3454			  SE_XSEL(1) | SE_YSEL(1);
3455		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3456			   SE_PAIR_YSEL(2);
3457		break;
3458	case CHIP_TOPAZ:
3459	case CHIP_CARRIZO:
3460		*rconf |= RB_MAP_PKR0(2);
3461		*rconf1 |= 0x0;
3462		break;
3463	case CHIP_POLARIS11:
3464	case CHIP_POLARIS12:
3465		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3466			  SE_XSEL(1) | SE_YSEL(1);
3467		*rconf1 |= 0x0;
3468		break;
3469	case CHIP_STONEY:
3470		*rconf |= 0x0;
3471		*rconf1 |= 0x0;
3472		break;
3473	default:
3474		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3475		break;
3476	}
3477}
3478
3479static void
3480gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3481					u32 raster_config, u32 raster_config_1,
3482					unsigned rb_mask, unsigned num_rb)
3483{
3484	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3485	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3486	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3487	unsigned rb_per_se = num_rb / num_se;
3488	unsigned se_mask[4];
3489	unsigned se;
3490
3491	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3492	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3493	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3494	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3495
3496	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3497	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3498	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3499
3500	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3501			     (!se_mask[2] && !se_mask[3]))) {
3502		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3503
3504		if (!se_mask[0] && !se_mask[1]) {
3505			raster_config_1 |=
3506				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3507		} else {
3508			raster_config_1 |=
3509				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3510		}
3511	}
3512
3513	for (se = 0; se < num_se; se++) {
3514		unsigned raster_config_se = raster_config;
3515		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3516		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3517		int idx = (se / 2) * 2;
3518
3519		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3520			raster_config_se &= ~SE_MAP_MASK;
3521
3522			if (!se_mask[idx]) {
3523				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3524			} else {
3525				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3526			}
3527		}
3528
3529		pkr0_mask &= rb_mask;
3530		pkr1_mask &= rb_mask;
3531		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3532			raster_config_se &= ~PKR_MAP_MASK;
3533
3534			if (!pkr0_mask) {
3535				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3536			} else {
3537				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3538			}
3539		}
3540
3541		if (rb_per_se >= 2) {
3542			unsigned rb0_mask = 1 << (se * rb_per_se);
3543			unsigned rb1_mask = rb0_mask << 1;
3544
3545			rb0_mask &= rb_mask;
3546			rb1_mask &= rb_mask;
3547			if (!rb0_mask || !rb1_mask) {
3548				raster_config_se &= ~RB_MAP_PKR0_MASK;
3549
3550				if (!rb0_mask) {
3551					raster_config_se |=
3552						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3553				} else {
3554					raster_config_se |=
3555						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3556				}
3557			}
3558
3559			if (rb_per_se > 2) {
3560				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3561				rb1_mask = rb0_mask << 1;
3562				rb0_mask &= rb_mask;
3563				rb1_mask &= rb_mask;
3564				if (!rb0_mask || !rb1_mask) {
3565					raster_config_se &= ~RB_MAP_PKR1_MASK;
3566
3567					if (!rb0_mask) {
3568						raster_config_se |=
3569							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3570					} else {
3571						raster_config_se |=
3572							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3573					}
3574				}
3575			}
3576		}
3577
3578		/* GRBM_GFX_INDEX has a different offset on VI */
3579		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff, 0);
3580		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3581		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3582	}
3583
3584	/* GRBM_GFX_INDEX has a different offset on VI */
3585	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3586}
3587
3588static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3589{
3590	int i, j;
3591	u32 data;
3592	u32 raster_config = 0, raster_config_1 = 0;
3593	u32 active_rbs = 0;
3594	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3595					adev->gfx.config.max_sh_per_se;
3596	unsigned num_rb_pipes;
3597
3598	mutex_lock(&adev->grbm_idx_mutex);
3599	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3600		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3601			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3602			data = gfx_v8_0_get_rb_active_bitmap(adev);
3603			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3604					       rb_bitmap_width_per_sh);
3605		}
3606	}
3607	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3608
3609	adev->gfx.config.backend_enable_mask = active_rbs;
3610	adev->gfx.config.num_rbs = hweight32(active_rbs);
3611
3612	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3613			     adev->gfx.config.max_shader_engines, 16);
3614
3615	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3616
3617	if (!adev->gfx.config.backend_enable_mask ||
3618			adev->gfx.config.num_rbs >= num_rb_pipes) {
3619		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3620		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3621	} else {
3622		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3623							adev->gfx.config.backend_enable_mask,
3624							num_rb_pipes);
3625	}
3626
3627	/* cache the values for userspace */
3628	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3629		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3630			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3631			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3632				RREG32(mmCC_RB_BACKEND_DISABLE);
3633			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3634				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3635			adev->gfx.config.rb_config[i][j].raster_config =
3636				RREG32(mmPA_SC_RASTER_CONFIG);
3637			adev->gfx.config.rb_config[i][j].raster_config_1 =
3638				RREG32(mmPA_SC_RASTER_CONFIG_1);
3639		}
3640	}
3641	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3642	mutex_unlock(&adev->grbm_idx_mutex);
3643}
3644
3645#define DEFAULT_SH_MEM_BASES	(0x6000)
3646/**
3647 * gfx_v8_0_init_compute_vmid - gart enable
3648 *
3649 * @adev: amdgpu_device pointer
3650 *
3651 * Initialize compute vmid sh_mem registers
3652 *
3653 */
 
 
 
3654static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3655{
3656	int i;
3657	uint32_t sh_mem_config;
3658	uint32_t sh_mem_bases;
3659
3660	/*
3661	 * Configure apertures:
3662	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3663	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3664	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3665	 */
3666	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3667
3668	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3669			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3670			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3671			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3672			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3673			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3674
3675	mutex_lock(&adev->srbm_mutex);
3676	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3677		vi_srbm_select(adev, 0, 0, 0, i);
3678		/* CP and shaders */
3679		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3680		WREG32(mmSH_MEM_APE1_BASE, 1);
3681		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3682		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3683	}
3684	vi_srbm_select(adev, 0, 0, 0, 0);
3685	mutex_unlock(&adev->srbm_mutex);
3686
3687	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3688	   access. These should be enabled by FW for target VMIDs. */
3689	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3690		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3691		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3692		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3693		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3694	}
3695}
3696
3697static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3698{
3699	int vmid;
3700
3701	/*
3702	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3703	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3704	 * the driver can enable them for graphics. VMID0 should maintain
3705	 * access so that HWS firmware can save/restore entries.
3706	 */
3707	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
3708		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3709		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3710		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3711		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3712	}
3713}
3714
3715static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3716{
3717	switch (adev->asic_type) {
3718	default:
3719		adev->gfx.config.double_offchip_lds_buf = 1;
3720		break;
3721	case CHIP_CARRIZO:
3722	case CHIP_STONEY:
3723		adev->gfx.config.double_offchip_lds_buf = 0;
3724		break;
3725	}
3726}
3727
3728static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3729{
3730	u32 tmp, sh_static_mem_cfg;
3731	int i;
3732
3733	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3734	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3735	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3736	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3737
3738	gfx_v8_0_tiling_mode_table_init(adev);
3739	gfx_v8_0_setup_rb(adev);
3740	gfx_v8_0_get_cu_info(adev);
3741	gfx_v8_0_config_init(adev);
3742
3743	/* XXX SH_MEM regs */
3744	/* where to put LDS, scratch, GPUVM in FSA64 space */
3745	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3746				   SWIZZLE_ENABLE, 1);
3747	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3748				   ELEMENT_SIZE, 1);
3749	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3750				   INDEX_STRIDE, 3);
3751	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3752
3753	mutex_lock(&adev->srbm_mutex);
3754	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3755		vi_srbm_select(adev, 0, 0, 0, i);
3756		/* CP and shaders */
3757		if (i == 0) {
3758			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3759			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3760			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3761					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3762			WREG32(mmSH_MEM_CONFIG, tmp);
3763			WREG32(mmSH_MEM_BASES, 0);
3764		} else {
3765			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3766			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3767			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3768					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3769			WREG32(mmSH_MEM_CONFIG, tmp);
3770			tmp = adev->gmc.shared_aperture_start >> 48;
3771			WREG32(mmSH_MEM_BASES, tmp);
3772		}
3773
3774		WREG32(mmSH_MEM_APE1_BASE, 1);
3775		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3776	}
3777	vi_srbm_select(adev, 0, 0, 0, 0);
3778	mutex_unlock(&adev->srbm_mutex);
3779
3780	gfx_v8_0_init_compute_vmid(adev);
3781	gfx_v8_0_init_gds_vmid(adev);
3782
3783	mutex_lock(&adev->grbm_idx_mutex);
3784	/*
3785	 * making sure that the following register writes will be broadcasted
3786	 * to all the shaders
3787	 */
3788	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3789
3790	WREG32(mmPA_SC_FIFO_SIZE,
3791		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3792			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3793		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3794			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3795		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3796			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3797		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3798			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3799
3800	tmp = RREG32(mmSPI_ARB_PRIORITY);
3801	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3802	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3803	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3804	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3805	WREG32(mmSPI_ARB_PRIORITY, tmp);
3806
3807	mutex_unlock(&adev->grbm_idx_mutex);
3808
3809}
3810
3811static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3812{
3813	u32 i, j, k;
3814	u32 mask;
3815
3816	mutex_lock(&adev->grbm_idx_mutex);
3817	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3818		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3819			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3820			for (k = 0; k < adev->usec_timeout; k++) {
3821				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3822					break;
3823				udelay(1);
3824			}
3825			if (k == adev->usec_timeout) {
3826				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3827						      0xffffffff, 0xffffffff, 0);
3828				mutex_unlock(&adev->grbm_idx_mutex);
3829				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3830					 i, j);
3831				return;
3832			}
3833		}
3834	}
3835	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3836	mutex_unlock(&adev->grbm_idx_mutex);
3837
3838	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3839		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3840		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3841		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3842	for (k = 0; k < adev->usec_timeout; k++) {
3843		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3844			break;
3845		udelay(1);
3846	}
3847}
3848
3849static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3850					       bool enable)
3851{
3852	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3853
3854	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3855	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3856	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3857	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3858
3859	WREG32(mmCP_INT_CNTL_RING0, tmp);
3860}
3861
3862static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3863{
3864	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
3865	/* csib */
3866	WREG32(mmRLC_CSIB_ADDR_HI,
3867			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3868	WREG32(mmRLC_CSIB_ADDR_LO,
3869			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3870	WREG32(mmRLC_CSIB_LENGTH,
3871			adev->gfx.rlc.clear_state_size);
3872}
3873
3874static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3875				int ind_offset,
3876				int list_size,
3877				int *unique_indices,
3878				int *indices_count,
3879				int max_indices,
3880				int *ind_start_offsets,
3881				int *offset_count,
3882				int max_offset)
3883{
3884	int indices;
3885	bool new_entry = true;
3886
3887	for (; ind_offset < list_size; ind_offset++) {
3888
3889		if (new_entry) {
3890			new_entry = false;
3891			ind_start_offsets[*offset_count] = ind_offset;
3892			*offset_count = *offset_count + 1;
3893			BUG_ON(*offset_count >= max_offset);
3894		}
3895
3896		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3897			new_entry = true;
3898			continue;
3899		}
3900
3901		ind_offset += 2;
3902
3903		/* look for the matching indice */
3904		for (indices = 0;
3905			indices < *indices_count;
3906			indices++) {
3907			if (unique_indices[indices] ==
3908				register_list_format[ind_offset])
3909				break;
3910		}
3911
3912		if (indices >= *indices_count) {
3913			unique_indices[*indices_count] =
3914				register_list_format[ind_offset];
3915			indices = *indices_count;
3916			*indices_count = *indices_count + 1;
3917			BUG_ON(*indices_count >= max_indices);
3918		}
3919
3920		register_list_format[ind_offset] = indices;
3921	}
3922}
3923
3924static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3925{
3926	int i, temp, data;
3927	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3928	int indices_count = 0;
3929	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3930	int offset_count = 0;
3931
3932	int list_size;
3933	unsigned int *register_list_format =
3934		kmemdup(adev->gfx.rlc.register_list_format,
3935			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3936	if (!register_list_format)
3937		return -ENOMEM;
3938
3939	gfx_v8_0_parse_ind_reg_list(register_list_format,
3940				RLC_FormatDirectRegListLength,
3941				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3942				unique_indices,
3943				&indices_count,
3944				ARRAY_SIZE(unique_indices),
3945				indirect_start_offsets,
3946				&offset_count,
3947				ARRAY_SIZE(indirect_start_offsets));
3948
3949	/* save and restore list */
3950	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
3951
3952	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
3953	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
3954		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
3955
3956	/* indirect list */
3957	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
3958	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
3959		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
3960
3961	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
3962	list_size = list_size >> 1;
3963	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
3964	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
3965
3966	/* starting offsets starts */
3967	WREG32(mmRLC_GPM_SCRATCH_ADDR,
3968		adev->gfx.rlc.starting_offsets_start);
3969	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
3970		WREG32(mmRLC_GPM_SCRATCH_DATA,
3971				indirect_start_offsets[i]);
3972
3973	/* unique indices */
3974	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
3975	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
3976	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
3977		if (unique_indices[i] != 0) {
3978			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
3979			WREG32(data + i, unique_indices[i] >> 20);
3980		}
3981	}
3982	kfree(register_list_format);
3983
3984	return 0;
3985}
3986
3987static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
3988{
3989	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
3990}
3991
3992static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
3993{
3994	uint32_t data;
3995
3996	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
3997
3998	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
3999	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4000	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4001	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4002	WREG32(mmRLC_PG_DELAY, data);
4003
4004	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4005	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4006
4007}
4008
4009static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4010						bool enable)
4011{
4012	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4013}
4014
4015static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4016						  bool enable)
4017{
4018	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4019}
4020
4021static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4022{
4023	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4024}
4025
4026static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4027{
4028	if ((adev->asic_type == CHIP_CARRIZO) ||
4029	    (adev->asic_type == CHIP_STONEY)) {
4030		gfx_v8_0_init_csb(adev);
4031		gfx_v8_0_init_save_restore_list(adev);
4032		gfx_v8_0_enable_save_restore_machine(adev);
4033		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4034		gfx_v8_0_init_power_gating(adev);
4035		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4036	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4037		   (adev->asic_type == CHIP_POLARIS12) ||
4038		   (adev->asic_type == CHIP_VEGAM)) {
4039		gfx_v8_0_init_csb(adev);
4040		gfx_v8_0_init_save_restore_list(adev);
4041		gfx_v8_0_enable_save_restore_machine(adev);
4042		gfx_v8_0_init_power_gating(adev);
4043	}
4044
4045}
4046
4047static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4048{
4049	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4050
4051	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4052	gfx_v8_0_wait_for_rlc_serdes(adev);
4053}
4054
4055static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4056{
4057	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4058	udelay(50);
4059
4060	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4061	udelay(50);
4062}
4063
4064static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4065{
4066	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4067
4068	/* carrizo do enable cp interrupt after cp inited */
4069	if (!(adev->flags & AMD_IS_APU))
4070		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4071
4072	udelay(50);
4073}
4074
4075static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4076{
4077	if (amdgpu_sriov_vf(adev)) {
4078		gfx_v8_0_init_csb(adev);
4079		return 0;
4080	}
4081
4082	adev->gfx.rlc.funcs->stop(adev);
4083	adev->gfx.rlc.funcs->reset(adev);
4084	gfx_v8_0_init_pg(adev);
4085	adev->gfx.rlc.funcs->start(adev);
4086
4087	return 0;
4088}
4089
4090static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4091{
 
4092	u32 tmp = RREG32(mmCP_ME_CNTL);
4093
4094	if (enable) {
4095		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4096		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4097		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4098	} else {
4099		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4100		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4101		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
 
 
4102	}
4103	WREG32(mmCP_ME_CNTL, tmp);
4104	udelay(50);
4105}
4106
4107static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4108{
4109	u32 count = 0;
4110	const struct cs_section_def *sect = NULL;
4111	const struct cs_extent_def *ext = NULL;
4112
4113	/* begin clear state */
4114	count += 2;
4115	/* context control state */
4116	count += 3;
4117
4118	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4119		for (ext = sect->section; ext->extent != NULL; ++ext) {
4120			if (sect->id == SECT_CONTEXT)
4121				count += 2 + ext->reg_count;
4122			else
4123				return 0;
4124		}
4125	}
4126	/* pa_sc_raster_config/pa_sc_raster_config1 */
4127	count += 4;
4128	/* end clear state */
4129	count += 2;
4130	/* clear state */
4131	count += 2;
4132
4133	return count;
4134}
4135
4136static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4137{
4138	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4139	const struct cs_section_def *sect = NULL;
4140	const struct cs_extent_def *ext = NULL;
4141	int r, i;
4142
4143	/* init the CP */
4144	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4145	WREG32(mmCP_ENDIAN_SWAP, 0);
4146	WREG32(mmCP_DEVICE_ID, 1);
4147
4148	gfx_v8_0_cp_gfx_enable(adev, true);
4149
4150	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4151	if (r) {
4152		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4153		return r;
4154	}
4155
4156	/* clear state buffer */
4157	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4158	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4159
4160	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4161	amdgpu_ring_write(ring, 0x80000000);
4162	amdgpu_ring_write(ring, 0x80000000);
4163
4164	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4165		for (ext = sect->section; ext->extent != NULL; ++ext) {
4166			if (sect->id == SECT_CONTEXT) {
4167				amdgpu_ring_write(ring,
4168				       PACKET3(PACKET3_SET_CONTEXT_REG,
4169					       ext->reg_count));
4170				amdgpu_ring_write(ring,
4171				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4172				for (i = 0; i < ext->reg_count; i++)
4173					amdgpu_ring_write(ring, ext->extent[i]);
4174			}
4175		}
4176	}
4177
4178	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4179	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4180	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4181	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4182
4183	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4184	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4185
4186	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4187	amdgpu_ring_write(ring, 0);
4188
4189	/* init the CE partitions */
4190	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4191	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4192	amdgpu_ring_write(ring, 0x8000);
4193	amdgpu_ring_write(ring, 0x8000);
4194
4195	amdgpu_ring_commit(ring);
4196
4197	return 0;
4198}
4199static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4200{
4201	u32 tmp;
4202	/* no gfx doorbells on iceland */
4203	if (adev->asic_type == CHIP_TOPAZ)
4204		return;
4205
4206	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4207
4208	if (ring->use_doorbell) {
4209		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4210				DOORBELL_OFFSET, ring->doorbell_index);
4211		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4212						DOORBELL_HIT, 0);
4213		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4214					    DOORBELL_EN, 1);
4215	} else {
4216		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4217	}
4218
4219	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4220
4221	if (adev->flags & AMD_IS_APU)
4222		return;
4223
4224	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4225					DOORBELL_RANGE_LOWER,
4226					adev->doorbell_index.gfx_ring0);
4227	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4228
4229	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4230		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4231}
4232
4233static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4234{
4235	struct amdgpu_ring *ring;
4236	u32 tmp;
4237	u32 rb_bufsz;
4238	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4239
4240	/* Set the write pointer delay */
4241	WREG32(mmCP_RB_WPTR_DELAY, 0);
4242
4243	/* set the RB to use vmid 0 */
4244	WREG32(mmCP_RB_VMID, 0);
4245
4246	/* Set ring buffer size */
4247	ring = &adev->gfx.gfx_ring[0];
4248	rb_bufsz = order_base_2(ring->ring_size / 8);
4249	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4250	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4251	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4252	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4253#ifdef __BIG_ENDIAN
4254	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4255#endif
4256	WREG32(mmCP_RB0_CNTL, tmp);
4257
4258	/* Initialize the ring buffer's read and write pointers */
4259	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4260	ring->wptr = 0;
4261	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4262
4263	/* set the wb address whether it's enabled or not */
4264	rptr_addr = ring->rptr_gpu_addr;
4265	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4266	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4267
4268	wptr_gpu_addr = ring->wptr_gpu_addr;
4269	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4270	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4271	mdelay(1);
4272	WREG32(mmCP_RB0_CNTL, tmp);
4273
4274	rb_addr = ring->gpu_addr >> 8;
4275	WREG32(mmCP_RB0_BASE, rb_addr);
4276	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4277
4278	gfx_v8_0_set_cpg_door_bell(adev, ring);
4279	/* start the ring */
4280	amdgpu_ring_clear_ring(ring);
4281	gfx_v8_0_cp_gfx_start(adev);
 
4282
4283	return 0;
4284}
4285
4286static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4287{
 
 
4288	if (enable) {
4289		WREG32(mmCP_MEC_CNTL, 0);
4290	} else {
4291		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4292		adev->gfx.kiq[0].ring.sched.ready = false;
 
 
4293	}
4294	udelay(50);
4295}
4296
4297/* KIQ functions */
4298static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4299{
4300	uint32_t tmp;
4301	struct amdgpu_device *adev = ring->adev;
4302
4303	/* tell RLC which is KIQ queue */
4304	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4305	tmp &= 0xffffff00;
4306	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4307	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4308	tmp |= 0x80;
4309	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4310}
4311
4312static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4313{
4314	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
4315	uint64_t queue_mask = 0;
4316	int r, i;
4317
4318	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4319		if (!test_bit(i, adev->gfx.mec_bitmap[0].queue_bitmap))
4320			continue;
4321
4322		/* This situation may be hit in the future if a new HW
4323		 * generation exposes more than 64 queues. If so, the
4324		 * definition of queue_mask needs updating */
4325		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4326			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4327			break;
4328		}
4329
4330		queue_mask |= (1ull << i);
4331	}
4332
4333	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4334	if (r) {
4335		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4336		return r;
4337	}
4338	/* set resources */
4339	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4340	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4341	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4342	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4343	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4344	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4345	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4346	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4347	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4348		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4349		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4350		uint64_t wptr_addr = ring->wptr_gpu_addr;
4351
4352		/* map queues */
4353		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4354		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4355		amdgpu_ring_write(kiq_ring,
4356				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4357		amdgpu_ring_write(kiq_ring,
4358				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4359				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4360				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4361				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4362		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4363		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4364		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4365		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4366	}
4367
4368	amdgpu_ring_commit(kiq_ring);
4369
4370	return 0;
4371}
4372
4373static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4374{
4375	int i, r = 0;
4376
4377	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4378		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4379		for (i = 0; i < adev->usec_timeout; i++) {
4380			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4381				break;
4382			udelay(1);
4383		}
4384		if (i == adev->usec_timeout)
4385			r = -ETIMEDOUT;
4386	}
4387	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4388	WREG32(mmCP_HQD_PQ_RPTR, 0);
4389	WREG32(mmCP_HQD_PQ_WPTR, 0);
4390
4391	return r;
4392}
4393
4394static void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd *mqd)
4395{
4396	struct amdgpu_device *adev = ring->adev;
4397
4398	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4399		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
4400			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
4401			mqd->cp_hqd_queue_priority =
4402				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
4403		}
4404	}
4405}
4406
4407static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4408{
4409	struct amdgpu_device *adev = ring->adev;
4410	struct vi_mqd *mqd = ring->mqd_ptr;
4411	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4412	uint32_t tmp;
4413
4414	mqd->header = 0xC0310800;
4415	mqd->compute_pipelinestat_enable = 0x00000001;
4416	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4417	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4418	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4419	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4420	mqd->compute_misc_reserved = 0x00000003;
4421	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4422						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4423	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4424						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4425	eop_base_addr = ring->eop_gpu_addr >> 8;
4426	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4427	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4428
4429	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4430	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4431	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4432			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4433
4434	mqd->cp_hqd_eop_control = tmp;
4435
4436	/* enable doorbell? */
4437	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4438			    CP_HQD_PQ_DOORBELL_CONTROL,
4439			    DOORBELL_EN,
4440			    ring->use_doorbell ? 1 : 0);
4441
4442	mqd->cp_hqd_pq_doorbell_control = tmp;
4443
4444	/* set the pointer to the MQD */
4445	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4446	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4447
4448	/* set MQD vmid to 0 */
4449	tmp = RREG32(mmCP_MQD_CONTROL);
4450	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4451	mqd->cp_mqd_control = tmp;
4452
4453	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4454	hqd_gpu_addr = ring->gpu_addr >> 8;
4455	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4456	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4457
4458	/* set up the HQD, this is similar to CP_RB0_CNTL */
4459	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4460	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4461			    (order_base_2(ring->ring_size / 4) - 1));
4462	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4463			(order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
4464#ifdef __BIG_ENDIAN
4465	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4466#endif
4467	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4468	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4469	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4470	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4471	mqd->cp_hqd_pq_control = tmp;
4472
4473	/* set the wb address whether it's enabled or not */
4474	wb_gpu_addr = ring->rptr_gpu_addr;
4475	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4476	mqd->cp_hqd_pq_rptr_report_addr_hi =
4477		upper_32_bits(wb_gpu_addr) & 0xffff;
4478
4479	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4480	wb_gpu_addr = ring->wptr_gpu_addr;
4481	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4482	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4483
4484	tmp = 0;
4485	/* enable the doorbell if requested */
4486	if (ring->use_doorbell) {
4487		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4488		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4489				DOORBELL_OFFSET, ring->doorbell_index);
4490
4491		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4492					 DOORBELL_EN, 1);
4493		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4494					 DOORBELL_SOURCE, 0);
4495		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4496					 DOORBELL_HIT, 0);
4497	}
4498
4499	mqd->cp_hqd_pq_doorbell_control = tmp;
4500
4501	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4502	ring->wptr = 0;
4503	mqd->cp_hqd_pq_wptr = ring->wptr;
4504	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4505
4506	/* set the vmid for the queue */
4507	mqd->cp_hqd_vmid = 0;
4508
4509	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4510	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4511	mqd->cp_hqd_persistent_state = tmp;
4512
4513	/* set MTYPE */
4514	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4515	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4516	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4517	mqd->cp_hqd_ib_control = tmp;
4518
4519	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4520	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4521	mqd->cp_hqd_iq_timer = tmp;
4522
4523	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4524	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4525	mqd->cp_hqd_ctx_save_control = tmp;
4526
4527	/* defaults */
4528	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4529	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
 
 
 
4530	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4531	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4532	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4533	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4534	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4535	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4536	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4537	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4538	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4539	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4540
4541	/* set static priority for a queue/ring */
4542	gfx_v8_0_mqd_set_priority(ring, mqd);
4543	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4544
4545	/* map_queues packet doesn't need activate the queue,
4546	 * so only kiq need set this field.
4547	 */
4548	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
4549		mqd->cp_hqd_active = 1;
4550
4551	return 0;
4552}
4553
4554static int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4555			struct vi_mqd *mqd)
4556{
4557	uint32_t mqd_reg;
4558	uint32_t *mqd_data;
4559
4560	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4561	mqd_data = &mqd->cp_mqd_base_addr_lo;
4562
4563	/* disable wptr polling */
4564	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4565
4566	/* program all HQD registers */
4567	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4568		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4569
4570	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4571	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4572	 * on ASICs that do not support context-save.
4573	 * EOP writes/reads can start anywhere in the ring.
4574	 */
4575	if (adev->asic_type != CHIP_TONGA) {
4576		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4577		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4578		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4579	}
4580
4581	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4582		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4583
4584	/* activate the HQD */
4585	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4586		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4587
4588	return 0;
4589}
4590
4591static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4592{
4593	struct amdgpu_device *adev = ring->adev;
4594	struct vi_mqd *mqd = ring->mqd_ptr;
 
4595
4596	gfx_v8_0_kiq_setting(ring);
4597
4598	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4599		/* reset MQD to a clean status */
4600		if (adev->gfx.kiq[0].mqd_backup)
4601			memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(struct vi_mqd_allocation));
4602
4603		/* reset ring buffer */
4604		ring->wptr = 0;
4605		amdgpu_ring_clear_ring(ring);
4606		mutex_lock(&adev->srbm_mutex);
4607		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4608		gfx_v8_0_mqd_commit(adev, mqd);
4609		vi_srbm_select(adev, 0, 0, 0, 0);
4610		mutex_unlock(&adev->srbm_mutex);
4611	} else {
4612		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4613		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4614		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4615		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
4616			amdgpu_ring_clear_ring(ring);
4617		mutex_lock(&adev->srbm_mutex);
4618		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4619		gfx_v8_0_mqd_init(ring);
4620		gfx_v8_0_mqd_commit(adev, mqd);
4621		vi_srbm_select(adev, 0, 0, 0, 0);
4622		mutex_unlock(&adev->srbm_mutex);
4623
4624		if (adev->gfx.kiq[0].mqd_backup)
4625			memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(struct vi_mqd_allocation));
4626	}
4627
4628	return 0;
4629}
4630
4631static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4632{
4633	struct amdgpu_device *adev = ring->adev;
4634	struct vi_mqd *mqd = ring->mqd_ptr;
4635	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4636
4637	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4638		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4639		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4640		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4641		mutex_lock(&adev->srbm_mutex);
4642		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4643		gfx_v8_0_mqd_init(ring);
4644		vi_srbm_select(adev, 0, 0, 0, 0);
4645		mutex_unlock(&adev->srbm_mutex);
4646
4647		if (adev->gfx.mec.mqd_backup[mqd_idx])
4648			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4649	} else {
4650		/* restore MQD to a clean status */
4651		if (adev->gfx.mec.mqd_backup[mqd_idx])
4652			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4653		/* reset ring buffer */
4654		ring->wptr = 0;
4655		amdgpu_ring_clear_ring(ring);
 
 
4656	}
4657	return 0;
4658}
4659
4660static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4661{
4662	if (adev->asic_type > CHIP_TONGA) {
4663		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4664		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4665	}
4666	/* enable doorbells */
4667	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4668}
4669
4670static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4671{
4672	struct amdgpu_ring *ring;
4673	int r;
4674
4675	ring = &adev->gfx.kiq[0].ring;
4676
4677	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4678	if (unlikely(r != 0))
4679		return r;
4680
4681	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4682	if (unlikely(r != 0)) {
4683		amdgpu_bo_unreserve(ring->mqd_obj);
4684		return r;
4685	}
4686
4687	gfx_v8_0_kiq_init_queue(ring);
4688	amdgpu_bo_kunmap(ring->mqd_obj);
4689	ring->mqd_ptr = NULL;
4690	amdgpu_bo_unreserve(ring->mqd_obj);
 
4691	return 0;
4692}
4693
4694static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4695{
4696	struct amdgpu_ring *ring = NULL;
4697	int r = 0, i;
4698
4699	gfx_v8_0_cp_compute_enable(adev, true);
4700
4701	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4702		ring = &adev->gfx.compute_ring[i];
4703
4704		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4705		if (unlikely(r != 0))
4706			goto done;
4707		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4708		if (!r) {
4709			r = gfx_v8_0_kcq_init_queue(ring);
4710			amdgpu_bo_kunmap(ring->mqd_obj);
4711			ring->mqd_ptr = NULL;
4712		}
4713		amdgpu_bo_unreserve(ring->mqd_obj);
4714		if (r)
4715			goto done;
4716	}
4717
4718	gfx_v8_0_set_mec_doorbell_range(adev);
4719
4720	r = gfx_v8_0_kiq_kcq_enable(adev);
4721	if (r)
4722		goto done;
4723
4724done:
4725	return r;
4726}
4727
4728static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4729{
4730	int r, i;
4731	struct amdgpu_ring *ring;
4732
4733	/* collect all the ring_tests here, gfx, kiq, compute */
4734	ring = &adev->gfx.gfx_ring[0];
4735	r = amdgpu_ring_test_helper(ring);
4736	if (r)
4737		return r;
4738
4739	ring = &adev->gfx.kiq[0].ring;
4740	r = amdgpu_ring_test_helper(ring);
4741	if (r)
4742		return r;
4743
4744	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4745		ring = &adev->gfx.compute_ring[i];
4746		amdgpu_ring_test_helper(ring);
4747	}
4748
4749	return 0;
4750}
4751
4752static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4753{
4754	int r;
4755
4756	if (!(adev->flags & AMD_IS_APU))
4757		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4758
4759	r = gfx_v8_0_kiq_resume(adev);
4760	if (r)
4761		return r;
4762
4763	r = gfx_v8_0_cp_gfx_resume(adev);
4764	if (r)
4765		return r;
4766
4767	r = gfx_v8_0_kcq_resume(adev);
4768	if (r)
4769		return r;
4770
4771	r = gfx_v8_0_cp_test_all_rings(adev);
4772	if (r)
4773		return r;
4774
4775	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4776
4777	return 0;
4778}
4779
4780static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4781{
4782	gfx_v8_0_cp_gfx_enable(adev, enable);
4783	gfx_v8_0_cp_compute_enable(adev, enable);
4784}
4785
4786static int gfx_v8_0_hw_init(struct amdgpu_ip_block *ip_block)
4787{
4788	int r;
4789	struct amdgpu_device *adev = ip_block->adev;
4790
4791	gfx_v8_0_init_golden_registers(adev);
4792	gfx_v8_0_constants_init(adev);
4793
 
 
 
 
4794	r = adev->gfx.rlc.funcs->resume(adev);
4795	if (r)
4796		return r;
4797
4798	r = gfx_v8_0_cp_resume(adev);
4799
4800	return r;
4801}
4802
4803static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4804{
4805	int r, i;
4806	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
4807
4808	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4809	if (r)
4810		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4811
4812	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4813		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4814
4815		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4816		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4817						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4818						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4819						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4820						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4821		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4822		amdgpu_ring_write(kiq_ring, 0);
4823		amdgpu_ring_write(kiq_ring, 0);
4824		amdgpu_ring_write(kiq_ring, 0);
4825	}
4826	/* Submit unmap queue packet */
4827	amdgpu_ring_commit(kiq_ring);
4828	/*
4829	 * Ring test will do a basic scratch register change check. Just run
4830	 * this to ensure that unmap queues that is submitted before got
4831	 * processed successfully before returning.
4832	 */
4833	r = amdgpu_ring_test_helper(kiq_ring);
4834	if (r)
4835		DRM_ERROR("KCQ disable failed\n");
4836
4837	return r;
4838}
4839
4840static bool gfx_v8_0_is_idle(void *handle)
4841{
4842	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4843
4844	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4845		|| RREG32(mmGRBM_STATUS2) != 0x8)
4846		return false;
4847	else
4848		return true;
4849}
4850
4851static bool gfx_v8_0_rlc_is_idle(void *handle)
4852{
4853	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4854
4855	if (RREG32(mmGRBM_STATUS2) != 0x8)
4856		return false;
4857	else
4858		return true;
4859}
4860
4861static int gfx_v8_0_wait_for_rlc_idle(void *handle)
4862{
4863	unsigned int i;
4864	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4865
4866	for (i = 0; i < adev->usec_timeout; i++) {
4867		if (gfx_v8_0_rlc_is_idle(handle))
4868			return 0;
4869
4870		udelay(1);
4871	}
4872	return -ETIMEDOUT;
4873}
4874
4875static int gfx_v8_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
4876{
4877	unsigned int i;
4878	struct amdgpu_device *adev = ip_block->adev;
4879
4880	for (i = 0; i < adev->usec_timeout; i++) {
4881		if (gfx_v8_0_is_idle(adev))
4882			return 0;
4883
4884		udelay(1);
4885	}
4886	return -ETIMEDOUT;
4887}
4888
4889static int gfx_v8_0_hw_fini(struct amdgpu_ip_block *ip_block)
4890{
4891	struct amdgpu_device *adev = ip_block->adev;
4892
4893	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4894	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4895
4896	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4897
4898	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4899
4900	/* disable KCQ to avoid CPC touch memory not valid anymore */
4901	gfx_v8_0_kcq_disable(adev);
4902
4903	if (amdgpu_sriov_vf(adev)) {
4904		pr_debug("For SRIOV client, shouldn't do anything.\n");
4905		return 0;
4906	}
4907
4908	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4909	if (!gfx_v8_0_wait_for_idle(ip_block))
4910		gfx_v8_0_cp_enable(adev, false);
4911	else
4912		pr_err("cp is busy, skip halt cp\n");
4913	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4914		adev->gfx.rlc.funcs->stop(adev);
4915	else
4916		pr_err("rlc is busy, skip halt rlc\n");
4917	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
 
 
4918
4919	return 0;
4920}
4921
4922static int gfx_v8_0_suspend(struct amdgpu_ip_block *ip_block)
4923{
4924	return gfx_v8_0_hw_fini(ip_block);
4925}
4926
4927static int gfx_v8_0_resume(struct amdgpu_ip_block *ip_block)
4928{
4929	return gfx_v8_0_hw_init(ip_block);
4930}
4931
4932static bool gfx_v8_0_check_soft_reset(struct amdgpu_ip_block *ip_block)
4933{
4934	struct amdgpu_device *adev = ip_block->adev;
4935	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4936	u32 tmp;
4937
4938	/* GRBM_STATUS */
4939	tmp = RREG32(mmGRBM_STATUS);
4940	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4941		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4942		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4943		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4944		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4945		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4946		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4947		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4948						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4949		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4950						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4951		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4952						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4953	}
4954
4955	/* GRBM_STATUS2 */
4956	tmp = RREG32(mmGRBM_STATUS2);
4957	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4958		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4959						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4960
4961	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
4962	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
4963	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
4964		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4965						SOFT_RESET_CPF, 1);
4966		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4967						SOFT_RESET_CPC, 1);
4968		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4969						SOFT_RESET_CPG, 1);
4970		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
4971						SOFT_RESET_GRBM, 1);
4972	}
4973
4974	/* SRBM_STATUS */
4975	tmp = RREG32(mmSRBM_STATUS);
4976	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
4977		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4978						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4979	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
4980		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4981						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
4982
4983	if (grbm_soft_reset || srbm_soft_reset) {
4984		adev->gfx.grbm_soft_reset = grbm_soft_reset;
4985		adev->gfx.srbm_soft_reset = srbm_soft_reset;
4986		return true;
4987	} else {
4988		adev->gfx.grbm_soft_reset = 0;
4989		adev->gfx.srbm_soft_reset = 0;
4990		return false;
4991	}
4992}
4993
4994static int gfx_v8_0_pre_soft_reset(struct amdgpu_ip_block *ip_block)
4995{
4996	struct amdgpu_device *adev = ip_block->adev;
4997	u32 grbm_soft_reset = 0;
4998
4999	if ((!adev->gfx.grbm_soft_reset) &&
5000	    (!adev->gfx.srbm_soft_reset))
5001		return 0;
5002
5003	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5004
5005	/* stop the rlc */
5006	adev->gfx.rlc.funcs->stop(adev);
5007
5008	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5009	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5010		/* Disable GFX parsing/prefetching */
5011		gfx_v8_0_cp_gfx_enable(adev, false);
5012
5013	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5014	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5015	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5016	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5017		int i;
5018
5019		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5020			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5021
5022			mutex_lock(&adev->srbm_mutex);
5023			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5024			gfx_v8_0_deactivate_hqd(adev, 2);
5025			vi_srbm_select(adev, 0, 0, 0, 0);
5026			mutex_unlock(&adev->srbm_mutex);
5027		}
5028		/* Disable MEC parsing/prefetching */
5029		gfx_v8_0_cp_compute_enable(adev, false);
5030	}
5031
5032	return 0;
5033}
5034
5035static int gfx_v8_0_soft_reset(struct amdgpu_ip_block *ip_block)
5036{
5037	struct amdgpu_device *adev = ip_block->adev;
5038	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5039	u32 tmp;
5040
5041	if ((!adev->gfx.grbm_soft_reset) &&
5042	    (!adev->gfx.srbm_soft_reset))
5043		return 0;
5044
5045	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5046	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5047
5048	if (grbm_soft_reset || srbm_soft_reset) {
5049		tmp = RREG32(mmGMCON_DEBUG);
5050		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5051		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5052		WREG32(mmGMCON_DEBUG, tmp);
5053		udelay(50);
5054	}
5055
5056	if (grbm_soft_reset) {
5057		tmp = RREG32(mmGRBM_SOFT_RESET);
5058		tmp |= grbm_soft_reset;
5059		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5060		WREG32(mmGRBM_SOFT_RESET, tmp);
5061		tmp = RREG32(mmGRBM_SOFT_RESET);
5062
5063		udelay(50);
5064
5065		tmp &= ~grbm_soft_reset;
5066		WREG32(mmGRBM_SOFT_RESET, tmp);
5067		tmp = RREG32(mmGRBM_SOFT_RESET);
5068	}
5069
5070	if (srbm_soft_reset) {
5071		tmp = RREG32(mmSRBM_SOFT_RESET);
5072		tmp |= srbm_soft_reset;
5073		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5074		WREG32(mmSRBM_SOFT_RESET, tmp);
5075		tmp = RREG32(mmSRBM_SOFT_RESET);
5076
5077		udelay(50);
5078
5079		tmp &= ~srbm_soft_reset;
5080		WREG32(mmSRBM_SOFT_RESET, tmp);
5081		tmp = RREG32(mmSRBM_SOFT_RESET);
5082	}
5083
5084	if (grbm_soft_reset || srbm_soft_reset) {
5085		tmp = RREG32(mmGMCON_DEBUG);
5086		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5087		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5088		WREG32(mmGMCON_DEBUG, tmp);
5089	}
5090
5091	/* Wait a little for things to settle down */
5092	udelay(50);
5093
5094	return 0;
5095}
5096
5097static int gfx_v8_0_post_soft_reset(struct amdgpu_ip_block *ip_block)
5098{
5099	struct amdgpu_device *adev = ip_block->adev;
5100	u32 grbm_soft_reset = 0;
5101
5102	if ((!adev->gfx.grbm_soft_reset) &&
5103	    (!adev->gfx.srbm_soft_reset))
5104		return 0;
5105
5106	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5107
5108	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5109	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5110	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5111	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5112		int i;
5113
5114		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5115			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5116
5117			mutex_lock(&adev->srbm_mutex);
5118			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5119			gfx_v8_0_deactivate_hqd(adev, 2);
5120			vi_srbm_select(adev, 0, 0, 0, 0);
5121			mutex_unlock(&adev->srbm_mutex);
5122		}
5123		gfx_v8_0_kiq_resume(adev);
5124		gfx_v8_0_kcq_resume(adev);
5125	}
5126
5127	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5128	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5129		gfx_v8_0_cp_gfx_resume(adev);
5130
5131	gfx_v8_0_cp_test_all_rings(adev);
5132
5133	adev->gfx.rlc.funcs->start(adev);
5134
5135	return 0;
5136}
5137
5138/**
5139 * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5140 *
5141 * @adev: amdgpu_device pointer
5142 *
5143 * Fetches a GPU clock counter snapshot.
5144 * Returns the 64 bit clock counter snapshot.
5145 */
5146static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5147{
5148	uint64_t clock;
5149
5150	mutex_lock(&adev->gfx.gpu_clock_mutex);
5151	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5152	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5153		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5154	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5155	return clock;
5156}
5157
5158static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5159					  uint32_t vmid,
5160					  uint32_t gds_base, uint32_t gds_size,
5161					  uint32_t gws_base, uint32_t gws_size,
5162					  uint32_t oa_base, uint32_t oa_size)
5163{
5164	/* GDS Base */
5165	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5166	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5167				WRITE_DATA_DST_SEL(0)));
5168	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5169	amdgpu_ring_write(ring, 0);
5170	amdgpu_ring_write(ring, gds_base);
5171
5172	/* GDS Size */
5173	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5174	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5175				WRITE_DATA_DST_SEL(0)));
5176	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
5177	amdgpu_ring_write(ring, 0);
5178	amdgpu_ring_write(ring, gds_size);
5179
5180	/* GWS */
5181	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5182	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5183				WRITE_DATA_DST_SEL(0)));
5184	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
5185	amdgpu_ring_write(ring, 0);
5186	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5187
5188	/* OA */
5189	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5190	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5191				WRITE_DATA_DST_SEL(0)));
5192	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
5193	amdgpu_ring_write(ring, 0);
5194	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5195}
5196
5197static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5198{
5199	WREG32(mmSQ_IND_INDEX,
5200		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5201		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5202		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5203		(SQ_IND_INDEX__FORCE_READ_MASK));
5204	return RREG32(mmSQ_IND_DATA);
5205}
5206
5207static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5208			   uint32_t wave, uint32_t thread,
5209			   uint32_t regno, uint32_t num, uint32_t *out)
5210{
5211	WREG32(mmSQ_IND_INDEX,
5212		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5213		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5214		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5215		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5216		(SQ_IND_INDEX__FORCE_READ_MASK) |
5217		(SQ_IND_INDEX__AUTO_INCR_MASK));
5218	while (num--)
5219		*(out++) = RREG32(mmSQ_IND_DATA);
5220}
5221
5222static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5223{
5224	/* type 0 wave data */
5225	dst[(*no_fields)++] = 0;
5226	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5227	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5228	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5229	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5230	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5231	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5232	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5233	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5234	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5235	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5236	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5237	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5238	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5239	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5240	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5241	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5242	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5243	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
5244	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
5245}
5246
5247static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
5248				     uint32_t wave, uint32_t start,
5249				     uint32_t size, uint32_t *dst)
5250{
5251	wave_read_regs(
5252		adev, simd, wave, 0,
5253		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5254}
5255
5256
5257static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5258	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5259	.select_se_sh = &gfx_v8_0_select_se_sh,
5260	.read_wave_data = &gfx_v8_0_read_wave_data,
5261	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5262	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5263};
5264
5265static int gfx_v8_0_early_init(struct amdgpu_ip_block *ip_block)
5266{
5267	struct amdgpu_device *adev = ip_block->adev;
5268
5269	adev->gfx.xcc_mask = 1;
5270	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5271	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
5272					  AMDGPU_MAX_COMPUTE_RINGS);
5273	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5274	gfx_v8_0_set_ring_funcs(adev);
5275	gfx_v8_0_set_irq_funcs(adev);
5276	gfx_v8_0_set_gds_init(adev);
5277	gfx_v8_0_set_rlc_funcs(adev);
5278
5279	return 0;
5280}
5281
5282static int gfx_v8_0_late_init(struct amdgpu_ip_block *ip_block)
5283{
5284	struct amdgpu_device *adev = ip_block->adev;
5285	int r;
5286
5287	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5288	if (r)
5289		return r;
5290
5291	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5292	if (r)
5293		return r;
5294
5295	/* requires IBs so do in late init after IB pool is initialized */
5296	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5297	if (r)
5298		return r;
5299
5300	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5301	if (r) {
5302		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5303		return r;
5304	}
5305
5306	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5307	if (r) {
5308		DRM_ERROR(
5309			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5310			r);
5311		return r;
5312	}
5313
5314	return 0;
5315}
5316
5317static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5318						       bool enable)
5319{
5320	if ((adev->asic_type == CHIP_POLARIS11) ||
5321	    (adev->asic_type == CHIP_POLARIS12) ||
5322	    (adev->asic_type == CHIP_VEGAM))
 
5323		/* Send msg to SMU via Powerplay */
5324		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5325
5326	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5327}
5328
5329static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5330							bool enable)
5331{
5332	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5333}
5334
5335static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5336		bool enable)
5337{
5338	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5339}
5340
5341static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5342					  bool enable)
5343{
5344	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5345}
5346
5347static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5348						bool enable)
5349{
5350	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5351
5352	/* Read any GFX register to wake up GFX. */
5353	if (!enable)
5354		RREG32(mmDB_RENDER_CONTROL);
5355}
5356
5357static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5358					  bool enable)
5359{
5360	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5361		cz_enable_gfx_cg_power_gating(adev, true);
5362		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5363			cz_enable_gfx_pipeline_power_gating(adev, true);
5364	} else {
5365		cz_enable_gfx_cg_power_gating(adev, false);
5366		cz_enable_gfx_pipeline_power_gating(adev, false);
5367	}
5368}
5369
5370static int gfx_v8_0_set_powergating_state(void *handle,
5371					  enum amd_powergating_state state)
5372{
5373	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5374	bool enable = (state == AMD_PG_STATE_GATE);
5375
5376	if (amdgpu_sriov_vf(adev))
5377		return 0;
5378
5379	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5380				AMD_PG_SUPPORT_RLC_SMU_HS |
5381				AMD_PG_SUPPORT_CP |
5382				AMD_PG_SUPPORT_GFX_DMG))
5383		amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5384	switch (adev->asic_type) {
5385	case CHIP_CARRIZO:
5386	case CHIP_STONEY:
5387
5388		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5389			cz_enable_sck_slow_down_on_power_up(adev, true);
5390			cz_enable_sck_slow_down_on_power_down(adev, true);
5391		} else {
5392			cz_enable_sck_slow_down_on_power_up(adev, false);
5393			cz_enable_sck_slow_down_on_power_down(adev, false);
5394		}
5395		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5396			cz_enable_cp_power_gating(adev, true);
5397		else
5398			cz_enable_cp_power_gating(adev, false);
5399
5400		cz_update_gfx_cg_power_gating(adev, enable);
5401
5402		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5403			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5404		else
5405			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5406
5407		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5408			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5409		else
5410			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5411		break;
5412	case CHIP_POLARIS11:
5413	case CHIP_POLARIS12:
5414	case CHIP_VEGAM:
5415		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5416			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5417		else
5418			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5419
5420		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5421			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5422		else
5423			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5424
5425		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5426			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5427		else
5428			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5429		break;
5430	default:
5431		break;
5432	}
5433	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5434				AMD_PG_SUPPORT_RLC_SMU_HS |
5435				AMD_PG_SUPPORT_CP |
5436				AMD_PG_SUPPORT_GFX_DMG))
5437		amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5438	return 0;
5439}
5440
5441static void gfx_v8_0_get_clockgating_state(void *handle, u64 *flags)
5442{
5443	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5444	int data;
5445
5446	if (amdgpu_sriov_vf(adev))
5447		*flags = 0;
5448
5449	/* AMD_CG_SUPPORT_GFX_MGCG */
5450	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5451	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5452		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5453
5454	/* AMD_CG_SUPPORT_GFX_CGLG */
5455	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5456	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5457		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5458
5459	/* AMD_CG_SUPPORT_GFX_CGLS */
5460	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5461		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5462
5463	/* AMD_CG_SUPPORT_GFX_CGTS */
5464	data = RREG32(mmCGTS_SM_CTRL_REG);
5465	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5466		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5467
5468	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5469	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5470		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5471
5472	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5473	data = RREG32(mmRLC_MEM_SLP_CNTL);
5474	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5475		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5476
5477	/* AMD_CG_SUPPORT_GFX_CP_LS */
5478	data = RREG32(mmCP_MEM_SLP_CNTL);
5479	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5480		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5481}
5482
5483static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5484				     uint32_t reg_addr, uint32_t cmd)
5485{
5486	uint32_t data;
5487
5488	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
5489
5490	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5491	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5492
5493	data = RREG32(mmRLC_SERDES_WR_CTRL);
5494	if (adev->asic_type == CHIP_STONEY)
5495		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5496			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5497			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5498			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5499			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5500			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5501			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5502			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5503			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5504	else
5505		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5506			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5507			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5508			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5509			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5510			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5511			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5512			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5513			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5514			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5515			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5516	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5517		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5518		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5519		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5520
5521	WREG32(mmRLC_SERDES_WR_CTRL, data);
5522}
5523
5524#define MSG_ENTER_RLC_SAFE_MODE     1
5525#define MSG_EXIT_RLC_SAFE_MODE      0
5526#define RLC_GPR_REG2__REQ_MASK 0x00000001
5527#define RLC_GPR_REG2__REQ__SHIFT 0
5528#define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5529#define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5530
5531static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5532{
5533	uint32_t rlc_setting;
5534
5535	rlc_setting = RREG32(mmRLC_CNTL);
5536	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5537		return false;
5538
5539	return true;
5540}
5541
5542static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
5543{
5544	uint32_t data;
5545	unsigned i;
5546	data = RREG32(mmRLC_CNTL);
5547	data |= RLC_SAFE_MODE__CMD_MASK;
5548	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5549	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5550	WREG32(mmRLC_SAFE_MODE, data);
5551
5552	/* wait for RLC_SAFE_MODE */
5553	for (i = 0; i < adev->usec_timeout; i++) {
5554		if ((RREG32(mmRLC_GPM_STAT) &
5555		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5556		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5557		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5558		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5559			break;
5560		udelay(1);
5561	}
5562	for (i = 0; i < adev->usec_timeout; i++) {
5563		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5564			break;
5565		udelay(1);
5566	}
5567}
5568
5569static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
5570{
5571	uint32_t data;
5572	unsigned i;
5573
5574	data = RREG32(mmRLC_CNTL);
5575	data |= RLC_SAFE_MODE__CMD_MASK;
5576	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5577	WREG32(mmRLC_SAFE_MODE, data);
5578
5579	for (i = 0; i < adev->usec_timeout; i++) {
5580		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5581			break;
5582		udelay(1);
5583	}
5584}
5585
5586static void gfx_v8_0_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, unsigned vmid)
5587{
5588	u32 data;
5589
5590	amdgpu_gfx_off_ctrl(adev, false);
5591
5592	if (amdgpu_sriov_is_pp_one_vf(adev))
5593		data = RREG32_NO_KIQ(mmRLC_SPM_VMID);
5594	else
5595		data = RREG32(mmRLC_SPM_VMID);
5596
5597	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
5598	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
5599
5600	if (amdgpu_sriov_is_pp_one_vf(adev))
5601		WREG32_NO_KIQ(mmRLC_SPM_VMID, data);
5602	else
5603		WREG32(mmRLC_SPM_VMID, data);
5604
5605	amdgpu_gfx_off_ctrl(adev, true);
5606}
5607
5608static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5609	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5610	.set_safe_mode = gfx_v8_0_set_safe_mode,
5611	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5612	.init = gfx_v8_0_rlc_init,
5613	.get_csb_size = gfx_v8_0_get_csb_size,
5614	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5615	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5616	.resume = gfx_v8_0_rlc_resume,
5617	.stop = gfx_v8_0_rlc_stop,
5618	.reset = gfx_v8_0_rlc_reset,
5619	.start = gfx_v8_0_rlc_start,
5620	.update_spm_vmid = gfx_v8_0_update_spm_vmid
5621};
5622
5623static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5624						      bool enable)
5625{
5626	uint32_t temp, data;
5627
5628	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5629
5630	/* It is disabled by HW by default */
5631	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5632		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5633			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5634				/* 1 - RLC memory Light sleep */
5635				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5636
5637			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5638				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5639		}
5640
5641		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5642		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5643		if (adev->flags & AMD_IS_APU)
5644			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5645				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5646				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5647		else
5648			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5649				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5650				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5651				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5652
5653		if (temp != data)
5654			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5655
5656		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5657		gfx_v8_0_wait_for_rlc_serdes(adev);
5658
5659		/* 5 - clear mgcg override */
5660		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5661
5662		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5663			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5664			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5665			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5666			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5667			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5668			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5669			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5670			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5671				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5672			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5673			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5674			if (temp != data)
5675				WREG32(mmCGTS_SM_CTRL_REG, data);
5676		}
5677		udelay(50);
5678
5679		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5680		gfx_v8_0_wait_for_rlc_serdes(adev);
5681	} else {
5682		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5683		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5684		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5685				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5686				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5687				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5688		if (temp != data)
5689			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5690
5691		/* 2 - disable MGLS in RLC */
5692		data = RREG32(mmRLC_MEM_SLP_CNTL);
5693		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5694			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5695			WREG32(mmRLC_MEM_SLP_CNTL, data);
5696		}
5697
5698		/* 3 - disable MGLS in CP */
5699		data = RREG32(mmCP_MEM_SLP_CNTL);
5700		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5701			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5702			WREG32(mmCP_MEM_SLP_CNTL, data);
5703		}
5704
5705		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5706		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5707		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5708				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5709		if (temp != data)
5710			WREG32(mmCGTS_SM_CTRL_REG, data);
5711
5712		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5713		gfx_v8_0_wait_for_rlc_serdes(adev);
5714
5715		/* 6 - set mgcg override */
5716		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5717
5718		udelay(50);
5719
5720		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5721		gfx_v8_0_wait_for_rlc_serdes(adev);
5722	}
5723
5724	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5725}
5726
5727static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5728						      bool enable)
5729{
5730	uint32_t temp, temp1, data, data1;
5731
5732	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5733
5734	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5735
5736	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5737		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5738		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5739		if (temp1 != data1)
5740			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5741
5742		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5743		gfx_v8_0_wait_for_rlc_serdes(adev);
5744
5745		/* 2 - clear cgcg override */
5746		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5747
5748		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5749		gfx_v8_0_wait_for_rlc_serdes(adev);
5750
5751		/* 3 - write cmd to set CGLS */
5752		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5753
5754		/* 4 - enable cgcg */
5755		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5756
5757		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5758			/* enable cgls*/
5759			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5760
5761			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5762			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5763
5764			if (temp1 != data1)
5765				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5766		} else {
5767			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5768		}
5769
5770		if (temp != data)
5771			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5772
5773		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5774		 * Cmp_busy/GFX_Idle interrupts
5775		 */
5776		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5777	} else {
5778		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5779		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5780
5781		/* TEST CGCG */
5782		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5783		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5784				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5785		if (temp1 != data1)
5786			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5787
5788		/* read gfx register to wake up cgcg */
5789		RREG32(mmCB_CGTT_SCLK_CTRL);
5790		RREG32(mmCB_CGTT_SCLK_CTRL);
5791		RREG32(mmCB_CGTT_SCLK_CTRL);
5792		RREG32(mmCB_CGTT_SCLK_CTRL);
5793
5794		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5795		gfx_v8_0_wait_for_rlc_serdes(adev);
5796
5797		/* write cmd to Set CGCG Override */
5798		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5799
5800		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5801		gfx_v8_0_wait_for_rlc_serdes(adev);
5802
5803		/* write cmd to Clear CGLS */
5804		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5805
5806		/* disable cgcg, cgls should be disabled too. */
5807		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5808			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5809		if (temp != data)
5810			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5811		/* enable interrupts again for PG */
5812		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5813	}
5814
5815	gfx_v8_0_wait_for_rlc_serdes(adev);
5816
5817	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5818}
5819static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5820					    bool enable)
5821{
5822	if (enable) {
5823		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5824		 * ===  MGCG + MGLS + TS(CG/LS) ===
5825		 */
5826		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5827		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5828	} else {
5829		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5830		 * ===  CGCG + CGLS ===
5831		 */
5832		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5833		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5834	}
5835	return 0;
5836}
5837
5838static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5839					  enum amd_clockgating_state state)
5840{
5841	uint32_t msg_id, pp_state = 0;
5842	uint32_t pp_support_state = 0;
5843
5844	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5845		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5846			pp_support_state = PP_STATE_SUPPORT_LS;
5847			pp_state = PP_STATE_LS;
5848		}
5849		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5850			pp_support_state |= PP_STATE_SUPPORT_CG;
5851			pp_state |= PP_STATE_CG;
5852		}
5853		if (state == AMD_CG_STATE_UNGATE)
5854			pp_state = 0;
5855
5856		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5857				PP_BLOCK_GFX_CG,
5858				pp_support_state,
5859				pp_state);
5860		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5861	}
5862
5863	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5864		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5865			pp_support_state = PP_STATE_SUPPORT_LS;
5866			pp_state = PP_STATE_LS;
5867		}
5868
5869		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5870			pp_support_state |= PP_STATE_SUPPORT_CG;
5871			pp_state |= PP_STATE_CG;
5872		}
5873
5874		if (state == AMD_CG_STATE_UNGATE)
5875			pp_state = 0;
5876
5877		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5878				PP_BLOCK_GFX_MG,
5879				pp_support_state,
5880				pp_state);
5881		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5882	}
5883
5884	return 0;
5885}
5886
5887static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5888					  enum amd_clockgating_state state)
5889{
5890
5891	uint32_t msg_id, pp_state = 0;
5892	uint32_t pp_support_state = 0;
5893
5894	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5895		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5896			pp_support_state = PP_STATE_SUPPORT_LS;
5897			pp_state = PP_STATE_LS;
5898		}
5899		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5900			pp_support_state |= PP_STATE_SUPPORT_CG;
5901			pp_state |= PP_STATE_CG;
5902		}
5903		if (state == AMD_CG_STATE_UNGATE)
5904			pp_state = 0;
5905
5906		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5907				PP_BLOCK_GFX_CG,
5908				pp_support_state,
5909				pp_state);
5910		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5911	}
5912
5913	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5914		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5915			pp_support_state = PP_STATE_SUPPORT_LS;
5916			pp_state = PP_STATE_LS;
5917		}
5918		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5919			pp_support_state |= PP_STATE_SUPPORT_CG;
5920			pp_state |= PP_STATE_CG;
5921		}
5922		if (state == AMD_CG_STATE_UNGATE)
5923			pp_state = 0;
5924
5925		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5926				PP_BLOCK_GFX_3D,
5927				pp_support_state,
5928				pp_state);
5929		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5930	}
5931
5932	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5933		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5934			pp_support_state = PP_STATE_SUPPORT_LS;
5935			pp_state = PP_STATE_LS;
5936		}
5937
5938		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5939			pp_support_state |= PP_STATE_SUPPORT_CG;
5940			pp_state |= PP_STATE_CG;
5941		}
5942
5943		if (state == AMD_CG_STATE_UNGATE)
5944			pp_state = 0;
5945
5946		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5947				PP_BLOCK_GFX_MG,
5948				pp_support_state,
5949				pp_state);
5950		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5951	}
5952
5953	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
5954		pp_support_state = PP_STATE_SUPPORT_LS;
5955
5956		if (state == AMD_CG_STATE_UNGATE)
5957			pp_state = 0;
5958		else
5959			pp_state = PP_STATE_LS;
5960
5961		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5962				PP_BLOCK_GFX_RLC,
5963				pp_support_state,
5964				pp_state);
5965		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5966	}
5967
5968	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
5969		pp_support_state = PP_STATE_SUPPORT_LS;
5970
5971		if (state == AMD_CG_STATE_UNGATE)
5972			pp_state = 0;
5973		else
5974			pp_state = PP_STATE_LS;
5975		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5976			PP_BLOCK_GFX_CP,
5977			pp_support_state,
5978			pp_state);
5979		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
 
5980	}
5981
5982	return 0;
5983}
5984
5985static int gfx_v8_0_set_clockgating_state(void *handle,
5986					  enum amd_clockgating_state state)
5987{
5988	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5989
5990	if (amdgpu_sriov_vf(adev))
5991		return 0;
5992
5993	switch (adev->asic_type) {
5994	case CHIP_FIJI:
5995	case CHIP_CARRIZO:
5996	case CHIP_STONEY:
5997		gfx_v8_0_update_gfx_clock_gating(adev,
5998						 state == AMD_CG_STATE_GATE);
5999		break;
6000	case CHIP_TONGA:
6001		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
6002		break;
6003	case CHIP_POLARIS10:
6004	case CHIP_POLARIS11:
6005	case CHIP_POLARIS12:
6006	case CHIP_VEGAM:
6007		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
6008		break;
6009	default:
6010		break;
6011	}
6012	return 0;
6013}
6014
6015static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6016{
6017	return *ring->rptr_cpu_addr;
6018}
6019
6020static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6021{
6022	struct amdgpu_device *adev = ring->adev;
6023
6024	if (ring->use_doorbell)
6025		/* XXX check if swapping is necessary on BE */
6026		return *ring->wptr_cpu_addr;
6027	else
6028		return RREG32(mmCP_RB0_WPTR);
6029}
6030
6031static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6032{
6033	struct amdgpu_device *adev = ring->adev;
6034
6035	if (ring->use_doorbell) {
6036		/* XXX check if swapping is necessary on BE */
6037		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6038		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6039	} else {
6040		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6041		(void)RREG32(mmCP_RB0_WPTR);
6042	}
6043}
6044
6045static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6046{
6047	u32 ref_and_mask, reg_mem_engine;
6048
6049	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6050	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6051		switch (ring->me) {
6052		case 1:
6053			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6054			break;
6055		case 2:
6056			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6057			break;
6058		default:
6059			return;
6060		}
6061		reg_mem_engine = 0;
6062	} else {
6063		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6064		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6065	}
6066
6067	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6068	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6069				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6070				 reg_mem_engine));
6071	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6072	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6073	amdgpu_ring_write(ring, ref_and_mask);
6074	amdgpu_ring_write(ring, ref_and_mask);
6075	amdgpu_ring_write(ring, 0x20); /* poll interval */
6076}
6077
6078static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6079{
6080	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6081	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6082		EVENT_INDEX(4));
6083
6084	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6085	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6086		EVENT_INDEX(0));
6087}
6088
6089static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6090					struct amdgpu_job *job,
6091					struct amdgpu_ib *ib,
6092					uint32_t flags)
6093{
6094	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6095	u32 header, control = 0;
6096
6097	if (ib->flags & AMDGPU_IB_FLAG_CE)
6098		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6099	else
6100		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6101
6102	control |= ib->length_dw | (vmid << 24);
6103
6104	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6105		control |= INDIRECT_BUFFER_PRE_ENB(1);
6106
6107		if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
6108			gfx_v8_0_ring_emit_de_meta(ring);
6109	}
6110
6111	amdgpu_ring_write(ring, header);
6112	amdgpu_ring_write(ring,
6113#ifdef __BIG_ENDIAN
6114			  (2 << 0) |
6115#endif
6116			  (ib->gpu_addr & 0xFFFFFFFC));
6117	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6118	amdgpu_ring_write(ring, control);
6119}
6120
6121static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6122					  struct amdgpu_job *job,
6123					  struct amdgpu_ib *ib,
6124					  uint32_t flags)
6125{
6126	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6127	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6128
6129	/* Currently, there is a high possibility to get wave ID mismatch
6130	 * between ME and GDS, leading to a hw deadlock, because ME generates
6131	 * different wave IDs than the GDS expects. This situation happens
6132	 * randomly when at least 5 compute pipes use GDS ordered append.
6133	 * The wave IDs generated by ME are also wrong after suspend/resume.
6134	 * Those are probably bugs somewhere else in the kernel driver.
6135	 *
6136	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6137	 * GDS to 0 for this ring (me/pipe).
6138	 */
6139	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6140		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6141		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6142		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6143	}
6144
6145	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6146	amdgpu_ring_write(ring,
6147#ifdef __BIG_ENDIAN
6148				(2 << 0) |
6149#endif
6150				(ib->gpu_addr & 0xFFFFFFFC));
6151	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6152	amdgpu_ring_write(ring, control);
6153}
6154
6155static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6156					 u64 seq, unsigned flags)
6157{
6158	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6159	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6160	bool exec = flags & AMDGPU_FENCE_FLAG_EXEC;
6161
6162	/* Workaround for cache flush problems. First send a dummy EOP
6163	 * event down the pipe with seq one below.
6164	 */
6165	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6166	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6167				 EOP_TC_ACTION_EN |
6168				 EOP_TC_WB_ACTION_EN |
6169				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6170				 EVENT_INDEX(5)));
6171	amdgpu_ring_write(ring, addr & 0xfffffffc);
6172	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6173				DATA_SEL(1) | INT_SEL(0));
6174	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
6175	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
6176
6177	/* Then send the real EOP event down the pipe:
6178	 * EVENT_WRITE_EOP - flush caches, send int */
6179	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6180	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6181				 EOP_TC_ACTION_EN |
6182				 EOP_TC_WB_ACTION_EN |
6183				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6184				 EVENT_INDEX(5) |
6185				 (exec ? EOP_EXEC : 0)));
6186	amdgpu_ring_write(ring, addr & 0xfffffffc);
6187	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6188			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6189	amdgpu_ring_write(ring, lower_32_bits(seq));
6190	amdgpu_ring_write(ring, upper_32_bits(seq));
6191
6192}
6193
6194static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6195{
6196	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6197	uint32_t seq = ring->fence_drv.sync_seq;
6198	uint64_t addr = ring->fence_drv.gpu_addr;
6199
6200	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6201	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6202				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6203				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6204	amdgpu_ring_write(ring, addr & 0xfffffffc);
6205	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6206	amdgpu_ring_write(ring, seq);
6207	amdgpu_ring_write(ring, 0xffffffff);
6208	amdgpu_ring_write(ring, 4); /* poll interval */
6209}
6210
6211static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6212					unsigned vmid, uint64_t pd_addr)
6213{
6214	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6215
6216	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6217
6218	/* wait for the invalidate to complete */
6219	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6220	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6221				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6222				 WAIT_REG_MEM_ENGINE(0))); /* me */
6223	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6224	amdgpu_ring_write(ring, 0);
6225	amdgpu_ring_write(ring, 0); /* ref */
6226	amdgpu_ring_write(ring, 0); /* mask */
6227	amdgpu_ring_write(ring, 0x20); /* poll interval */
6228
6229	/* compute doesn't have PFP */
6230	if (usepfp) {
6231		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6232		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6233		amdgpu_ring_write(ring, 0x0);
6234	}
6235}
6236
6237static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6238{
6239	return *ring->wptr_cpu_addr;
6240}
6241
6242static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6243{
6244	struct amdgpu_device *adev = ring->adev;
6245
6246	/* XXX check if swapping is necessary on BE */
6247	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6248	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6249}
6250
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6251static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6252					     u64 addr, u64 seq,
6253					     unsigned flags)
6254{
6255	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6256	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6257
6258	/* RELEASE_MEM - flush caches, send int */
6259	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6260	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6261				 EOP_TC_ACTION_EN |
6262				 EOP_TC_WB_ACTION_EN |
6263				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6264				 EVENT_INDEX(5)));
6265	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6266	amdgpu_ring_write(ring, addr & 0xfffffffc);
6267	amdgpu_ring_write(ring, upper_32_bits(addr));
6268	amdgpu_ring_write(ring, lower_32_bits(seq));
6269	amdgpu_ring_write(ring, upper_32_bits(seq));
6270}
6271
6272static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6273					 u64 seq, unsigned int flags)
6274{
6275	/* we only allocate 32bit for each seq wb address */
6276	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6277
6278	/* write fence seq to the "addr" */
6279	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6280	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6281				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
6282	amdgpu_ring_write(ring, lower_32_bits(addr));
6283	amdgpu_ring_write(ring, upper_32_bits(addr));
6284	amdgpu_ring_write(ring, lower_32_bits(seq));
6285
6286	if (flags & AMDGPU_FENCE_FLAG_INT) {
6287		/* set register to trigger INT */
6288		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6289		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6290					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6291		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6292		amdgpu_ring_write(ring, 0);
6293		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6294	}
6295}
6296
6297static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6298{
6299	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6300	amdgpu_ring_write(ring, 0);
6301}
6302
6303static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6304{
6305	uint32_t dw2 = 0;
6306
6307	if (amdgpu_sriov_vf(ring->adev))
6308		gfx_v8_0_ring_emit_ce_meta(ring);
6309
6310	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6311	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6312		gfx_v8_0_ring_emit_vgt_flush(ring);
6313		/* set load_global_config & load_global_uconfig */
6314		dw2 |= 0x8001;
6315		/* set load_cs_sh_regs */
6316		dw2 |= 0x01000000;
6317		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6318		dw2 |= 0x10002;
6319
6320		/* set load_ce_ram if preamble presented */
6321		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6322			dw2 |= 0x10000000;
6323	} else {
6324		/* still load_ce_ram if this is the first time preamble presented
6325		 * although there is no context switch happens.
6326		 */
6327		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6328			dw2 |= 0x10000000;
6329	}
6330
6331	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6332	amdgpu_ring_write(ring, dw2);
6333	amdgpu_ring_write(ring, 0);
6334}
6335
6336static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
6337						  uint64_t addr)
6338{
6339	unsigned ret;
6340
6341	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6342	amdgpu_ring_write(ring, lower_32_bits(addr));
6343	amdgpu_ring_write(ring, upper_32_bits(addr));
6344	/* discard following DWs if *cond_exec_gpu_addr==0 */
6345	amdgpu_ring_write(ring, 0);
6346	ret = ring->wptr & ring->buf_mask;
6347	/* patch dummy value later */
6348	amdgpu_ring_write(ring, 0);
6349	return ret;
6350}
6351
6352static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
6353				    uint32_t reg_val_offs)
 
 
 
 
 
 
 
 
 
 
 
 
 
6354{
6355	struct amdgpu_device *adev = ring->adev;
6356
6357	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6358	amdgpu_ring_write(ring, 0 |	/* src: register*/
6359				(5 << 8) |	/* dst: memory */
6360				(1 << 20));	/* write confirm */
6361	amdgpu_ring_write(ring, reg);
6362	amdgpu_ring_write(ring, 0);
6363	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6364				reg_val_offs * 4));
6365	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6366				reg_val_offs * 4));
6367}
6368
6369static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6370				  uint32_t val)
6371{
6372	uint32_t cmd;
6373
6374	switch (ring->funcs->type) {
6375	case AMDGPU_RING_TYPE_GFX:
6376		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6377		break;
6378	case AMDGPU_RING_TYPE_KIQ:
6379		cmd = 1 << 16; /* no inc addr */
6380		break;
6381	default:
6382		cmd = WR_CONFIRM;
6383		break;
6384	}
6385
6386	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6387	amdgpu_ring_write(ring, cmd);
6388	amdgpu_ring_write(ring, reg);
6389	amdgpu_ring_write(ring, 0);
6390	amdgpu_ring_write(ring, val);
6391}
6392
6393static void gfx_v8_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
6394				  int mem_space, int opt, uint32_t addr0,
6395				  uint32_t addr1, uint32_t ref, uint32_t mask,
6396				  uint32_t inv)
6397{
6398	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6399	amdgpu_ring_write(ring,
6400			  /* memory (1) or register (0) */
6401			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
6402			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
6403			   WAIT_REG_MEM_FUNCTION(3) |  /* equal */
6404			   WAIT_REG_MEM_ENGINE(eng_sel)));
6405
6406	if (mem_space)
6407		BUG_ON(addr0 & 0x3); /* Dword align */
6408	amdgpu_ring_write(ring, addr0);
6409	amdgpu_ring_write(ring, addr1);
6410	amdgpu_ring_write(ring, ref);
6411	amdgpu_ring_write(ring, mask);
6412	amdgpu_ring_write(ring, inv); /* poll interval */
6413}
6414
6415static void gfx_v8_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
6416					uint32_t val, uint32_t mask)
6417{
6418	gfx_v8_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
6419}
6420
6421static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6422{
6423	struct amdgpu_device *adev = ring->adev;
6424	uint32_t value = 0;
6425
6426	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6427	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6428	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6429	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6430	WREG32(mmSQ_CMD, value);
6431}
6432
6433static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6434						 enum amdgpu_interrupt_state state)
6435{
6436	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6437		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6438}
6439
6440static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6441						     int me, int pipe,
6442						     enum amdgpu_interrupt_state state)
6443{
6444	u32 mec_int_cntl, mec_int_cntl_reg;
6445
6446	/*
6447	 * amdgpu controls only the first MEC. That's why this function only
6448	 * handles the setting of interrupts for this specific MEC. All other
6449	 * pipes' interrupts are set by amdkfd.
6450	 */
6451
6452	if (me == 1) {
6453		switch (pipe) {
6454		case 0:
6455			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6456			break;
6457		case 1:
6458			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6459			break;
6460		case 2:
6461			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6462			break;
6463		case 3:
6464			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6465			break;
6466		default:
6467			DRM_DEBUG("invalid pipe %d\n", pipe);
6468			return;
6469		}
6470	} else {
6471		DRM_DEBUG("invalid me %d\n", me);
6472		return;
6473	}
6474
6475	switch (state) {
6476	case AMDGPU_IRQ_STATE_DISABLE:
6477		mec_int_cntl = RREG32(mec_int_cntl_reg);
6478		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6479		WREG32(mec_int_cntl_reg, mec_int_cntl);
6480		break;
6481	case AMDGPU_IRQ_STATE_ENABLE:
6482		mec_int_cntl = RREG32(mec_int_cntl_reg);
6483		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6484		WREG32(mec_int_cntl_reg, mec_int_cntl);
6485		break;
6486	default:
6487		break;
6488	}
6489}
6490
6491static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6492					     struct amdgpu_irq_src *source,
6493					     unsigned type,
6494					     enum amdgpu_interrupt_state state)
6495{
6496	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6497		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6498
6499	return 0;
6500}
6501
6502static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6503					      struct amdgpu_irq_src *source,
6504					      unsigned type,
6505					      enum amdgpu_interrupt_state state)
6506{
6507	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6508		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6509
6510	return 0;
6511}
6512
6513static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6514					    struct amdgpu_irq_src *src,
6515					    unsigned type,
6516					    enum amdgpu_interrupt_state state)
6517{
6518	switch (type) {
6519	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6520		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6521		break;
6522	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6523		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6524		break;
6525	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6526		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6527		break;
6528	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6529		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6530		break;
6531	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6532		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6533		break;
6534	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6535		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6536		break;
6537	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6538		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6539		break;
6540	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6541		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6542		break;
6543	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6544		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6545		break;
6546	default:
6547		break;
6548	}
6549	return 0;
6550}
6551
6552static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6553					 struct amdgpu_irq_src *source,
6554					 unsigned int type,
6555					 enum amdgpu_interrupt_state state)
6556{
6557	int enable_flag;
6558
6559	switch (state) {
6560	case AMDGPU_IRQ_STATE_DISABLE:
6561		enable_flag = 0;
6562		break;
6563
6564	case AMDGPU_IRQ_STATE_ENABLE:
6565		enable_flag = 1;
6566		break;
6567
6568	default:
6569		return -EINVAL;
6570	}
6571
6572	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6573	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6574	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6575	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6576	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6577	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6578		     enable_flag);
6579	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6580		     enable_flag);
6581	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6582		     enable_flag);
6583	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6584		     enable_flag);
6585	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6586		     enable_flag);
6587	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6588		     enable_flag);
6589	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6590		     enable_flag);
6591	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6592		     enable_flag);
6593
6594	return 0;
6595}
6596
6597static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6598				     struct amdgpu_irq_src *source,
6599				     unsigned int type,
6600				     enum amdgpu_interrupt_state state)
6601{
6602	int enable_flag;
6603
6604	switch (state) {
6605	case AMDGPU_IRQ_STATE_DISABLE:
6606		enable_flag = 1;
6607		break;
6608
6609	case AMDGPU_IRQ_STATE_ENABLE:
6610		enable_flag = 0;
6611		break;
6612
6613	default:
6614		return -EINVAL;
6615	}
6616
6617	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6618		     enable_flag);
6619
6620	return 0;
6621}
6622
6623static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6624			    struct amdgpu_irq_src *source,
6625			    struct amdgpu_iv_entry *entry)
6626{
6627	int i;
6628	u8 me_id, pipe_id, queue_id;
6629	struct amdgpu_ring *ring;
6630
6631	DRM_DEBUG("IH: CP EOP\n");
6632	me_id = (entry->ring_id & 0x0c) >> 2;
6633	pipe_id = (entry->ring_id & 0x03) >> 0;
6634	queue_id = (entry->ring_id & 0x70) >> 4;
6635
6636	switch (me_id) {
6637	case 0:
6638		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6639		break;
6640	case 1:
6641	case 2:
6642		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6643			ring = &adev->gfx.compute_ring[i];
6644			/* Per-queue interrupt is supported for MEC starting from VI.
6645			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6646			  */
6647			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6648				amdgpu_fence_process(ring);
6649		}
6650		break;
6651	}
6652	return 0;
6653}
6654
6655static void gfx_v8_0_fault(struct amdgpu_device *adev,
6656			   struct amdgpu_iv_entry *entry)
6657{
6658	u8 me_id, pipe_id, queue_id;
6659	struct amdgpu_ring *ring;
6660	int i;
6661
6662	me_id = (entry->ring_id & 0x0c) >> 2;
6663	pipe_id = (entry->ring_id & 0x03) >> 0;
6664	queue_id = (entry->ring_id & 0x70) >> 4;
6665
6666	switch (me_id) {
6667	case 0:
6668		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6669		break;
6670	case 1:
6671	case 2:
6672		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6673			ring = &adev->gfx.compute_ring[i];
6674			if (ring->me == me_id && ring->pipe == pipe_id &&
6675			    ring->queue == queue_id)
6676				drm_sched_fault(&ring->sched);
6677		}
6678		break;
6679	}
6680}
6681
6682static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6683				 struct amdgpu_irq_src *source,
6684				 struct amdgpu_iv_entry *entry)
6685{
6686	DRM_ERROR("Illegal register access in command stream\n");
6687	gfx_v8_0_fault(adev, entry);
6688	return 0;
6689}
6690
6691static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6692				  struct amdgpu_irq_src *source,
6693				  struct amdgpu_iv_entry *entry)
6694{
6695	DRM_ERROR("Illegal instruction in command stream\n");
6696	gfx_v8_0_fault(adev, entry);
6697	return 0;
6698}
6699
6700static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6701				     struct amdgpu_irq_src *source,
6702				     struct amdgpu_iv_entry *entry)
6703{
6704	DRM_ERROR("CP EDC/ECC error detected.");
6705	return 0;
6706}
6707
6708static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data,
6709				  bool from_wq)
6710{
6711	u32 enc, se_id, sh_id, cu_id;
6712	char type[20];
6713	int sq_edc_source = -1;
6714
6715	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6716	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6717
6718	switch (enc) {
6719		case 0:
6720			DRM_INFO("SQ general purpose intr detected:"
6721					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6722					"host_cmd_overflow %d, cmd_timestamp %d,"
6723					"reg_timestamp %d, thread_trace_buff_full %d,"
6724					"wlt %d, thread_trace %d.\n",
6725					se_id,
6726					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6727					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6728					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6729					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6730					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6731					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6732					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6733					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6734					);
6735			break;
6736		case 1:
6737		case 2:
6738
6739			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6740			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6741
6742			/*
6743			 * This function can be called either directly from ISR
6744			 * or from BH in which case we can access SQ_EDC_INFO
6745			 * instance
6746			 */
6747			if (from_wq) {
6748				mutex_lock(&adev->grbm_idx_mutex);
6749				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id, 0);
6750
6751				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6752
6753				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
6754				mutex_unlock(&adev->grbm_idx_mutex);
6755			}
6756
6757			if (enc == 1)
6758				sprintf(type, "instruction intr");
6759			else
6760				sprintf(type, "EDC/ECC error");
6761
6762			DRM_INFO(
6763				"SQ %s detected: "
6764					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6765					"trap %s, sq_ed_info.source %s.\n",
6766					type, se_id, sh_id, cu_id,
6767					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6768					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6769					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6770					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6771					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6772				);
6773			break;
6774		default:
6775			DRM_ERROR("SQ invalid encoding type\n.");
6776	}
6777}
6778
6779static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6780{
6781
6782	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6783	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6784
6785	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data, true);
6786}
6787
6788static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6789			   struct amdgpu_irq_src *source,
6790			   struct amdgpu_iv_entry *entry)
6791{
6792	unsigned ih_data = entry->src_data[0];
6793
6794	/*
6795	 * Try to submit work so SQ_EDC_INFO can be accessed from
6796	 * BH. If previous work submission hasn't finished yet
6797	 * just print whatever info is possible directly from the ISR.
6798	 */
6799	if (work_pending(&adev->gfx.sq_work.work)) {
6800		gfx_v8_0_parse_sq_irq(adev, ih_data, false);
6801	} else {
6802		adev->gfx.sq_work.ih_data = ih_data;
6803		schedule_work(&adev->gfx.sq_work.work);
6804	}
6805
6806	return 0;
6807}
6808
6809static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring)
6810{
6811	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
6812	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6813			  PACKET3_TC_ACTION_ENA |
6814			  PACKET3_SH_KCACHE_ACTION_ENA |
6815			  PACKET3_SH_ICACHE_ACTION_ENA |
6816			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6817	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6818	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
6819	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
6820}
6821
6822static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
6823{
6824	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6825	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6826			  PACKET3_TC_ACTION_ENA |
6827			  PACKET3_SH_KCACHE_ACTION_ENA |
6828			  PACKET3_SH_ICACHE_ACTION_ENA |
6829			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6830	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
6831	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
6832	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
6833	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
6834	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
6835}
6836
6837
6838/* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6839#define mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT	0x0000007f
6840static void gfx_v8_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6841					uint32_t pipe, bool enable)
6842{
6843	uint32_t val;
6844	uint32_t wcl_cs_reg;
6845
6846	val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT;
6847
6848	switch (pipe) {
6849	case 0:
6850		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS0;
6851		break;
6852	case 1:
6853		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS1;
6854		break;
6855	case 2:
6856		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS2;
6857		break;
6858	case 3:
6859		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS3;
6860		break;
6861	default:
6862		DRM_DEBUG("invalid pipe %d\n", pipe);
6863		return;
6864	}
6865
6866	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6867
6868}
6869
6870#define mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT	0x07ffffff
6871static void gfx_v8_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6872{
6873	struct amdgpu_device *adev = ring->adev;
6874	uint32_t val;
6875	int i;
6876
6877	/* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6878	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6879	 * around 25% of gpu resources.
6880	 */
6881	val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6882	amdgpu_ring_emit_wreg(ring, mmSPI_WCL_PIPE_PERCENT_GFX, val);
6883
6884	/* Restrict waves for normal/low priority compute queues as well
6885	 * to get best QoS for high priority compute jobs.
6886	 *
6887	 * amdgpu controls only 1st ME(0-3 CS pipes).
6888	 */
6889	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6890		if (i != ring->pipe)
6891			gfx_v8_0_emit_wave_limit_cs(ring, i, enable);
6892
6893	}
6894
6895}
6896
6897static int gfx_v8_0_reset_kgq(struct amdgpu_ring *ring, unsigned int vmid)
6898{
6899	struct amdgpu_device *adev = ring->adev;
6900	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
6901	struct amdgpu_ring *kiq_ring = &kiq->ring;
6902	unsigned long flags;
6903	u32 tmp;
6904	int r;
6905
6906	if (amdgpu_sriov_vf(adev))
6907		return -EINVAL;
6908
6909	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
6910		return -EINVAL;
6911
6912	spin_lock_irqsave(&kiq->ring_lock, flags);
6913
6914	if (amdgpu_ring_alloc(kiq_ring, 5)) {
6915		spin_unlock_irqrestore(&kiq->ring_lock, flags);
6916		return -ENOMEM;
6917	}
6918
6919	tmp = REG_SET_FIELD(0, CP_VMID_RESET, RESET_REQUEST, 1 << vmid);
6920	gfx_v8_0_ring_emit_wreg(kiq_ring, mmCP_VMID_RESET, tmp);
6921	amdgpu_ring_commit(kiq_ring);
6922
6923	spin_unlock_irqrestore(&kiq->ring_lock, flags);
6924
6925	r = amdgpu_ring_test_ring(kiq_ring);
6926	if (r)
6927		return r;
6928
6929	if (amdgpu_ring_alloc(ring, 7 + 12 + 5))
6930		return -ENOMEM;
6931	gfx_v8_0_ring_emit_fence_gfx(ring, ring->fence_drv.gpu_addr,
6932				     ring->fence_drv.sync_seq, AMDGPU_FENCE_FLAG_EXEC);
6933	gfx_v8_0_ring_emit_reg_wait(ring, mmCP_VMID_RESET, 0, 0xffff);
6934	gfx_v8_0_ring_emit_wreg(ring, mmCP_VMID_RESET, 0);
6935
6936	return amdgpu_ring_test_ring(ring);
6937}
6938
6939static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6940	.name = "gfx_v8_0",
6941	.early_init = gfx_v8_0_early_init,
6942	.late_init = gfx_v8_0_late_init,
6943	.sw_init = gfx_v8_0_sw_init,
6944	.sw_fini = gfx_v8_0_sw_fini,
6945	.hw_init = gfx_v8_0_hw_init,
6946	.hw_fini = gfx_v8_0_hw_fini,
6947	.suspend = gfx_v8_0_suspend,
6948	.resume = gfx_v8_0_resume,
6949	.is_idle = gfx_v8_0_is_idle,
6950	.wait_for_idle = gfx_v8_0_wait_for_idle,
6951	.check_soft_reset = gfx_v8_0_check_soft_reset,
6952	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6953	.soft_reset = gfx_v8_0_soft_reset,
6954	.post_soft_reset = gfx_v8_0_post_soft_reset,
6955	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6956	.set_powergating_state = gfx_v8_0_set_powergating_state,
6957	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6958};
6959
6960static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6961	.type = AMDGPU_RING_TYPE_GFX,
6962	.align_mask = 0xff,
6963	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6964	.support_64bit_ptrs = false,
6965	.get_rptr = gfx_v8_0_ring_get_rptr,
6966	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6967	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6968	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6969		5 +  /* COND_EXEC */
6970		7 +  /* PIPELINE_SYNC */
6971		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6972		12 +  /* FENCE for VM_FLUSH */
6973		20 + /* GDS switch */
6974		4 + /* double SWITCH_BUFFER,
6975		       the first COND_EXEC jump to the place just
6976			   prior to this double SWITCH_BUFFER  */
6977		5 + /* COND_EXEC */
6978		7 +	 /*	HDP_flush */
6979		4 +	 /*	VGT_flush */
6980		14 + /*	CE_META */
6981		31 + /*	DE_META */
6982		3 + /* CNTX_CTRL */
6983		5 + /* HDP_INVL */
6984		12 + 12 + /* FENCE x2 */
6985		2 + /* SWITCH_BUFFER */
6986		5, /* SURFACE_SYNC */
6987	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6988	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6989	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6990	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6991	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6992	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6993	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6994	.test_ring = gfx_v8_0_ring_test_ring,
6995	.test_ib = gfx_v8_0_ring_test_ib,
6996	.insert_nop = amdgpu_ring_insert_nop,
6997	.pad_ib = amdgpu_ring_generic_pad_ib,
6998	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6999	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
7000	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
 
7001	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7002	.soft_recovery = gfx_v8_0_ring_soft_recovery,
7003	.emit_mem_sync = gfx_v8_0_emit_mem_sync,
7004	.reset = gfx_v8_0_reset_kgq,
7005};
7006
7007static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
7008	.type = AMDGPU_RING_TYPE_COMPUTE,
7009	.align_mask = 0xff,
7010	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
7011	.support_64bit_ptrs = false,
7012	.get_rptr = gfx_v8_0_ring_get_rptr,
7013	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
7014	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7015	.emit_frame_size =
7016		20 + /* gfx_v8_0_ring_emit_gds_switch */
7017		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7018		5 + /* hdp_invalidate */
7019		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7020		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
7021		7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
7022		7 + /* gfx_v8_0_emit_mem_sync_compute */
7023		5 + /* gfx_v8_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
7024		15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
7025	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7026	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
7027	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
7028	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
7029	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
7030	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
7031	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
7032	.test_ring = gfx_v8_0_ring_test_ring,
7033	.test_ib = gfx_v8_0_ring_test_ib,
7034	.insert_nop = amdgpu_ring_insert_nop,
7035	.pad_ib = amdgpu_ring_generic_pad_ib,
 
7036	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7037	.soft_recovery = gfx_v8_0_ring_soft_recovery,
7038	.emit_mem_sync = gfx_v8_0_emit_mem_sync_compute,
7039	.emit_wave_limit = gfx_v8_0_emit_wave_limit,
7040};
7041
7042static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
7043	.type = AMDGPU_RING_TYPE_KIQ,
7044	.align_mask = 0xff,
7045	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
7046	.support_64bit_ptrs = false,
7047	.get_rptr = gfx_v8_0_ring_get_rptr,
7048	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
7049	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7050	.emit_frame_size =
7051		20 + /* gfx_v8_0_ring_emit_gds_switch */
7052		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7053		5 + /* hdp_invalidate */
7054		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7055		17 + /* gfx_v8_0_ring_emit_vm_flush */
7056		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7057	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7058	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
7059	.test_ring = gfx_v8_0_ring_test_ring,
7060	.insert_nop = amdgpu_ring_insert_nop,
7061	.pad_ib = amdgpu_ring_generic_pad_ib,
7062	.emit_rreg = gfx_v8_0_ring_emit_rreg,
7063	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7064};
7065
7066static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
7067{
7068	int i;
7069
7070	adev->gfx.kiq[0].ring.funcs = &gfx_v8_0_ring_funcs_kiq;
7071
7072	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7073		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7074
7075	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7076		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7077}
7078
7079static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7080	.set = gfx_v8_0_set_eop_interrupt_state,
7081	.process = gfx_v8_0_eop_irq,
7082};
7083
7084static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7085	.set = gfx_v8_0_set_priv_reg_fault_state,
7086	.process = gfx_v8_0_priv_reg_irq,
7087};
7088
7089static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7090	.set = gfx_v8_0_set_priv_inst_fault_state,
7091	.process = gfx_v8_0_priv_inst_irq,
7092};
7093
7094static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7095	.set = gfx_v8_0_set_cp_ecc_int_state,
7096	.process = gfx_v8_0_cp_ecc_error_irq,
7097};
7098
7099static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7100	.set = gfx_v8_0_set_sq_int_state,
7101	.process = gfx_v8_0_sq_irq,
7102};
7103
7104static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7105{
7106	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7107	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7108
7109	adev->gfx.priv_reg_irq.num_types = 1;
7110	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7111
7112	adev->gfx.priv_inst_irq.num_types = 1;
7113	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7114
7115	adev->gfx.cp_ecc_error_irq.num_types = 1;
7116	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7117
7118	adev->gfx.sq_irq.num_types = 1;
7119	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7120}
7121
7122static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7123{
7124	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7125}
7126
7127static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7128{
7129	/* init asci gds info */
7130	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7131	adev->gds.gws_size = 64;
7132	adev->gds.oa_size = 16;
7133	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7134}
7135
7136static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7137						 u32 bitmap)
7138{
7139	u32 data;
7140
7141	if (!bitmap)
7142		return;
7143
7144	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7145	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7146
7147	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7148}
7149
7150static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7151{
7152	u32 data, mask;
7153
7154	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7155		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7156
7157	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7158
7159	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7160}
7161
7162static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7163{
7164	int i, j, k, counter, active_cu_number = 0;
7165	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7166	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7167	unsigned disable_masks[4 * 2];
7168	u32 ao_cu_num;
7169
7170	memset(cu_info, 0, sizeof(*cu_info));
7171
7172	if (adev->flags & AMD_IS_APU)
7173		ao_cu_num = 2;
7174	else
7175		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7176
7177	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7178
7179	mutex_lock(&adev->grbm_idx_mutex);
7180	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7181		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7182			mask = 1;
7183			ao_bitmap = 0;
7184			counter = 0;
7185			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
7186			if (i < 4 && j < 2)
7187				gfx_v8_0_set_user_cu_inactive_bitmap(
7188					adev, disable_masks[i * 2 + j]);
7189			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7190			cu_info->bitmap[0][i][j] = bitmap;
7191
7192			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7193				if (bitmap & mask) {
7194					if (counter < ao_cu_num)
7195						ao_bitmap |= mask;
7196					counter ++;
7197				}
7198				mask <<= 1;
7199			}
7200			active_cu_number += counter;
7201			if (i < 2 && j < 2)
7202				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7203			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7204		}
7205	}
7206	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
7207	mutex_unlock(&adev->grbm_idx_mutex);
7208
7209	cu_info->number = active_cu_number;
7210	cu_info->ao_cu_mask = ao_cu_mask;
7211	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7212	cu_info->max_waves_per_simd = 10;
7213	cu_info->max_scratch_slots_per_cu = 32;
7214	cu_info->wave_front_size = 64;
7215	cu_info->lds_size = 64;
7216}
7217
7218const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7219{
7220	.type = AMD_IP_BLOCK_TYPE_GFX,
7221	.major = 8,
7222	.minor = 0,
7223	.rev = 0,
7224	.funcs = &gfx_v8_0_ip_funcs,
7225};
7226
7227const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7228{
7229	.type = AMD_IP_BLOCK_TYPE_GFX,
7230	.major = 8,
7231	.minor = 1,
7232	.rev = 0,
7233	.funcs = &gfx_v8_0_ip_funcs,
7234};
7235
7236static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7237{
7238	uint64_t ce_payload_addr;
7239	int cnt_ce;
7240	union {
7241		struct vi_ce_ib_state regular;
7242		struct vi_ce_ib_state_chained_ib chained;
7243	} ce_payload = {};
7244
7245	if (ring->adev->virt.chained_ib_support) {
7246		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7247			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7248		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7249	} else {
7250		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7251			offsetof(struct vi_gfx_meta_data, ce_payload);
7252		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7253	}
7254
7255	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7256	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7257				WRITE_DATA_DST_SEL(8) |
7258				WR_CONFIRM) |
7259				WRITE_DATA_CACHE_POLICY(0));
7260	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7261	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7262	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7263}
7264
7265static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7266{
7267	uint64_t de_payload_addr, gds_addr, csa_addr;
7268	int cnt_de;
7269	union {
7270		struct vi_de_ib_state regular;
7271		struct vi_de_ib_state_chained_ib chained;
7272	} de_payload = {};
7273
7274	csa_addr = amdgpu_csa_vaddr(ring->adev);
7275	gds_addr = csa_addr + 4096;
7276	if (ring->adev->virt.chained_ib_support) {
7277		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7278		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7279		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7280		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7281	} else {
7282		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7283		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7284		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7285		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7286	}
7287
7288	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7289	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7290				WRITE_DATA_DST_SEL(8) |
7291				WR_CONFIRM) |
7292				WRITE_DATA_CACHE_POLICY(0));
7293	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7294	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7295	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7296}
v5.4
   1/*
   2 * Copyright 2014 Advanced Micro Devices, Inc.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20 * OTHER DEALINGS IN THE SOFTWARE.
  21 *
  22 */
  23
  24#include <linux/delay.h>
  25#include <linux/kernel.h>
  26#include <linux/firmware.h>
  27#include <linux/module.h>
  28#include <linux/pci.h>
  29
  30#include "amdgpu.h"
  31#include "amdgpu_gfx.h"
 
  32#include "vi.h"
  33#include "vi_structs.h"
  34#include "vid.h"
  35#include "amdgpu_ucode.h"
  36#include "amdgpu_atombios.h"
  37#include "atombios_i2c.h"
  38#include "clearstate_vi.h"
  39
  40#include "gmc/gmc_8_2_d.h"
  41#include "gmc/gmc_8_2_sh_mask.h"
  42
  43#include "oss/oss_3_0_d.h"
  44#include "oss/oss_3_0_sh_mask.h"
  45
  46#include "bif/bif_5_0_d.h"
  47#include "bif/bif_5_0_sh_mask.h"
  48#include "gca/gfx_8_0_d.h"
  49#include "gca/gfx_8_0_enum.h"
  50#include "gca/gfx_8_0_sh_mask.h"
  51
  52#include "dce/dce_10_0_d.h"
  53#include "dce/dce_10_0_sh_mask.h"
  54
  55#include "smu/smu_7_1_3_d.h"
  56
  57#include "ivsrcid/ivsrcid_vislands30.h"
  58
  59#define GFX8_NUM_GFX_RINGS     1
  60#define GFX8_MEC_HPD_SIZE 4096
  61
  62#define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001
  63#define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001
  64#define POLARIS11_GB_ADDR_CONFIG_GOLDEN 0x22011002
  65#define TONGA_GB_ADDR_CONFIG_GOLDEN 0x22011003
  66
  67#define ARRAY_MODE(x)					((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT)
  68#define PIPE_CONFIG(x)					((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT)
  69#define TILE_SPLIT(x)					((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT)
  70#define MICRO_TILE_MODE_NEW(x)				((x) << GB_TILE_MODE0__MICRO_TILE_MODE_NEW__SHIFT)
  71#define SAMPLE_SPLIT(x)					((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT)
  72#define BANK_WIDTH(x)					((x) << GB_MACROTILE_MODE0__BANK_WIDTH__SHIFT)
  73#define BANK_HEIGHT(x)					((x) << GB_MACROTILE_MODE0__BANK_HEIGHT__SHIFT)
  74#define MACRO_TILE_ASPECT(x)				((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT)
  75#define NUM_BANKS(x)					((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT)
  76
  77#define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK            0x00000001L
  78#define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK            0x00000002L
  79#define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK           0x00000004L
  80#define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK           0x00000008L
  81#define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK           0x00000010L
  82#define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK           0x00000020L
  83
  84/* BPM SERDES CMD */
  85#define SET_BPM_SERDES_CMD    1
  86#define CLE_BPM_SERDES_CMD    0
  87
  88/* BPM Register Address*/
  89enum {
  90	BPM_REG_CGLS_EN = 0,        /* Enable/Disable CGLS */
  91	BPM_REG_CGLS_ON,            /* ON/OFF CGLS: shall be controlled by RLC FW */
  92	BPM_REG_CGCG_OVERRIDE,      /* Set/Clear CGCG Override */
  93	BPM_REG_MGCG_OVERRIDE,      /* Set/Clear MGCG Override */
  94	BPM_REG_FGCG_OVERRIDE,      /* Set/Clear FGCG Override */
  95	BPM_REG_FGCG_MAX
  96};
  97
  98#define RLC_FormatDirectRegListLength        14
  99
 100MODULE_FIRMWARE("amdgpu/carrizo_ce.bin");
 101MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin");
 102MODULE_FIRMWARE("amdgpu/carrizo_me.bin");
 103MODULE_FIRMWARE("amdgpu/carrizo_mec.bin");
 104MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin");
 105MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin");
 106
 107MODULE_FIRMWARE("amdgpu/stoney_ce.bin");
 108MODULE_FIRMWARE("amdgpu/stoney_pfp.bin");
 109MODULE_FIRMWARE("amdgpu/stoney_me.bin");
 110MODULE_FIRMWARE("amdgpu/stoney_mec.bin");
 111MODULE_FIRMWARE("amdgpu/stoney_rlc.bin");
 112
 113MODULE_FIRMWARE("amdgpu/tonga_ce.bin");
 114MODULE_FIRMWARE("amdgpu/tonga_pfp.bin");
 115MODULE_FIRMWARE("amdgpu/tonga_me.bin");
 116MODULE_FIRMWARE("amdgpu/tonga_mec.bin");
 117MODULE_FIRMWARE("amdgpu/tonga_mec2.bin");
 118MODULE_FIRMWARE("amdgpu/tonga_rlc.bin");
 119
 120MODULE_FIRMWARE("amdgpu/topaz_ce.bin");
 121MODULE_FIRMWARE("amdgpu/topaz_pfp.bin");
 122MODULE_FIRMWARE("amdgpu/topaz_me.bin");
 123MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
 124MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
 125
 126MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
 127MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
 128MODULE_FIRMWARE("amdgpu/fiji_me.bin");
 129MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
 130MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
 131MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
 132
 133MODULE_FIRMWARE("amdgpu/polaris10_ce.bin");
 134MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin");
 135MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin");
 136MODULE_FIRMWARE("amdgpu/polaris10_pfp_2.bin");
 137MODULE_FIRMWARE("amdgpu/polaris10_me.bin");
 138MODULE_FIRMWARE("amdgpu/polaris10_me_2.bin");
 139MODULE_FIRMWARE("amdgpu/polaris10_mec.bin");
 140MODULE_FIRMWARE("amdgpu/polaris10_mec_2.bin");
 141MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin");
 142MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin");
 143MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin");
 144
 145MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
 146MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
 147MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
 148MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
 149MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
 150MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
 151MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
 152MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
 153MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
 154MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
 155MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
 156
 157MODULE_FIRMWARE("amdgpu/polaris12_ce.bin");
 158MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin");
 159MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin");
 160MODULE_FIRMWARE("amdgpu/polaris12_pfp_2.bin");
 161MODULE_FIRMWARE("amdgpu/polaris12_me.bin");
 162MODULE_FIRMWARE("amdgpu/polaris12_me_2.bin");
 163MODULE_FIRMWARE("amdgpu/polaris12_mec.bin");
 164MODULE_FIRMWARE("amdgpu/polaris12_mec_2.bin");
 165MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin");
 166MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin");
 167MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin");
 168
 169MODULE_FIRMWARE("amdgpu/vegam_ce.bin");
 170MODULE_FIRMWARE("amdgpu/vegam_pfp.bin");
 171MODULE_FIRMWARE("amdgpu/vegam_me.bin");
 172MODULE_FIRMWARE("amdgpu/vegam_mec.bin");
 173MODULE_FIRMWARE("amdgpu/vegam_mec2.bin");
 174MODULE_FIRMWARE("amdgpu/vegam_rlc.bin");
 175
 176static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
 177{
 178	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
 179	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
 180	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
 181	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
 182	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
 183	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
 184	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
 185	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
 186	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
 187	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
 188	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
 189	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
 190	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
 191	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
 192	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
 193	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
 194};
 195
 196static const u32 golden_settings_tonga_a11[] =
 197{
 198	mmCB_HW_CONTROL, 0xfffdf3cf, 0x00007208,
 199	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
 200	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 201	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 202	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 203	mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x000000fc,
 204	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 205	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
 206	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 207	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 208	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 209	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 210	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000002fb,
 211	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x0000543b,
 212	mmTCP_CHAN_STEER_LO, 0xffffffff, 0xa9210876,
 213	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 214};
 215
 216static const u32 tonga_golden_common_all[] =
 217{
 218	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 219	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
 220	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
 221	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 222	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 223	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 224	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 225	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
 226};
 227
 228static const u32 tonga_mgcg_cgcg_init[] =
 229{
 230	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 231	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 232	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 233	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 234	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
 235	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
 236	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
 237	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 238	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 239	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 240	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 241	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 242	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 243	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 244	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 245	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 246	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 247	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 248	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 249	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 250	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 251	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 252	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
 253	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 254	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 255	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 256	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 257	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 258	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 259	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 260	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 261	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 262	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 263	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 264	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 265	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 266	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 267	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 268	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
 269	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 270	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 271	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 272	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 273	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
 274	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 275	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 276	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 277	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 278	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
 279	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 280	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 281	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 282	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 283	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 284	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 285	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 286	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 287	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 288	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
 289	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 290	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 291	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 292	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 293	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
 294	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 295	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 296	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 297	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 298	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
 299	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 300	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 301	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 302	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 303	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
 304	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 305};
 306
 307static const u32 golden_settings_vegam_a11[] =
 308{
 309	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
 310	mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000,
 311	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 312	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 313	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 314	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 315	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a,
 316	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e,
 317	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 318	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
 319	mmSQ_CONFIG, 0x07f80000, 0x01180000,
 320	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 321	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 322	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
 323	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 324	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054,
 325	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 326};
 327
 328static const u32 vegam_golden_common_all[] =
 329{
 330	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 331	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 332	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 333	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 334	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 335	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 336};
 337
 338static const u32 golden_settings_polaris11_a11[] =
 339{
 340	mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208,
 341	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
 342	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 343	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 344	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 345	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 346	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
 347	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
 348	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 349	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
 350	mmSQ_CONFIG, 0x07f80000, 0x01180000,
 351	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 352	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 353	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f3,
 354	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 355	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003210,
 356	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 357};
 358
 359static const u32 polaris11_golden_common_all[] =
 360{
 361	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 362	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011002,
 363	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 364	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 365	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 366	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 367};
 368
 369static const u32 golden_settings_polaris10_a11[] =
 370{
 371	mmATC_MISC_CG, 0x000c0fc0, 0x000c0200,
 372	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
 373	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
 374	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 375	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 376	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 377	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 378	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
 379	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002a,
 380	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 381	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
 382	mmSQ_CONFIG, 0x07f80000, 0x07180000,
 383	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 384	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 385	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
 386	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 387	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 388};
 389
 390static const u32 polaris10_golden_common_all[] =
 391{
 392	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 393	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
 394	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
 395	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 396	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 397	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 398	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 399	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 400};
 401
 402static const u32 fiji_golden_common_all[] =
 403{
 404	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 405	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
 406	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
 407	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
 408	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 409	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 410	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 411	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 412	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 413	mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009,
 414};
 415
 416static const u32 golden_settings_fiji_a10[] =
 417{
 418	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
 419	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 420	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 421	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 422	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 423	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 424	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 425	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 426	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 427	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
 428	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
 429};
 430
 431static const u32 fiji_mgcg_cgcg_init[] =
 432{
 433	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 434	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 435	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 436	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 437	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
 438	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
 439	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
 440	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 441	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 442	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 443	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 444	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 445	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 446	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 447	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 448	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 449	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 450	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 451	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 452	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 453	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 454	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 455	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
 456	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 457	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 458	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 459	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 460	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 461	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 462	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 463	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 464	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 465	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 466	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
 467	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 468};
 469
 470static const u32 golden_settings_iceland_a11[] =
 471{
 472	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
 473	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 474	mmDB_DEBUG3, 0xc0000000, 0xc0000000,
 475	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 476	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 477	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 478	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x00000002,
 479	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
 480	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
 481	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 482	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 483	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 484	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 485	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f1,
 486	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
 487	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010,
 488};
 489
 490static const u32 iceland_golden_common_all[] =
 491{
 492	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 493	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
 494	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
 495	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
 496	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 497	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 498	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 499	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
 500};
 501
 502static const u32 iceland_mgcg_cgcg_init[] =
 503{
 504	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 505	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 506	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 507	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 508	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0xc0000100,
 509	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0xc0000100,
 510	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0xc0000100,
 511	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 512	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 513	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 514	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 515	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 516	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 517	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 518	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 519	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 520	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 521	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 522	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 523	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 524	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 525	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 526	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0xff000100,
 527	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 528	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 529	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 530	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 531	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 532	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 533	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 534	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 535	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 536	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 537	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
 538	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 539	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 540	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 541	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 542	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
 543	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 544	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 545	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 546	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 547	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
 548	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 549	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 550	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 551	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 552	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
 553	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 554	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 555	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 556	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 557	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
 558	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 559	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 560	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 561	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 562	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
 563	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 564	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 565	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 566	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 567	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
 568};
 569
 570static const u32 cz_golden_settings_a11[] =
 571{
 572	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
 573	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 574	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 575	mmPA_SC_ENHANCE, 0xffffffff, 0x00000001,
 576	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 577	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
 578	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
 579	mmTA_CNTL_AUX, 0x000f000f, 0x00010000,
 580	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 581	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 582	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f3,
 583	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00001302
 584};
 585
 586static const u32 cz_golden_common_all[] =
 587{
 588	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 589	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
 590	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
 591	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
 592	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 593	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 594	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 595	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
 596};
 597
 598static const u32 cz_mgcg_cgcg_init[] =
 599{
 600	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
 601	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 602	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 603	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
 604	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
 605	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
 606	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x00000100,
 607	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
 608	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
 609	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
 610	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
 611	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
 612	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
 613	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
 614	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
 615	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
 616	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
 617	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
 618	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
 619	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
 620	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
 621	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
 622	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
 623	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
 624	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
 625	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
 626	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
 627	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 628	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
 629	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
 630	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 631	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 632	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 633	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 634	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 635	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 636	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 637	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 638	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
 639	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 640	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 641	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 642	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 643	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
 644	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 645	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 646	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 647	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 648	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
 649	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 650	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 651	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 652	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 653	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
 654	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 655	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 656	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 657	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 658	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
 659	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 660	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 661	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 662	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 663	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
 664	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 665	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 666	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
 667	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
 668	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
 669	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
 670	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
 671	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
 672	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
 673	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
 674	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
 675};
 676
 677static const u32 stoney_golden_settings_a11[] =
 678{
 679	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
 680	mmGB_GPU_ID, 0x0000000f, 0x00000000,
 681	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
 682	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
 683	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
 684	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
 685	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
 686	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
 687	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1,
 688	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010,
 689};
 690
 691static const u32 stoney_golden_common_all[] =
 692{
 693	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 694	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000,
 695	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
 696	mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001,
 697	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
 698	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
 699	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
 700	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
 701};
 702
 703static const u32 stoney_mgcg_cgcg_init[] =
 704{
 705	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
 706	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
 707	mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
 708	mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
 709	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200,
 710};
 711
 712
 713static const char * const sq_edc_source_names[] = {
 714	"SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
 715	"SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
 716	"SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
 717	"SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
 718	"SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
 719	"SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
 720	"SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
 721};
 722
 723static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
 724static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
 725static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
 726static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev);
 727static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev);
 728static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev);
 729static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring);
 730static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring);
 731
 
 
 
 732static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
 733{
 
 
 734	switch (adev->asic_type) {
 735	case CHIP_TOPAZ:
 736		amdgpu_device_program_register_sequence(adev,
 737							iceland_mgcg_cgcg_init,
 738							ARRAY_SIZE(iceland_mgcg_cgcg_init));
 739		amdgpu_device_program_register_sequence(adev,
 740							golden_settings_iceland_a11,
 741							ARRAY_SIZE(golden_settings_iceland_a11));
 742		amdgpu_device_program_register_sequence(adev,
 743							iceland_golden_common_all,
 744							ARRAY_SIZE(iceland_golden_common_all));
 745		break;
 746	case CHIP_FIJI:
 747		amdgpu_device_program_register_sequence(adev,
 748							fiji_mgcg_cgcg_init,
 749							ARRAY_SIZE(fiji_mgcg_cgcg_init));
 750		amdgpu_device_program_register_sequence(adev,
 751							golden_settings_fiji_a10,
 752							ARRAY_SIZE(golden_settings_fiji_a10));
 753		amdgpu_device_program_register_sequence(adev,
 754							fiji_golden_common_all,
 755							ARRAY_SIZE(fiji_golden_common_all));
 756		break;
 757
 758	case CHIP_TONGA:
 759		amdgpu_device_program_register_sequence(adev,
 760							tonga_mgcg_cgcg_init,
 761							ARRAY_SIZE(tonga_mgcg_cgcg_init));
 762		amdgpu_device_program_register_sequence(adev,
 763							golden_settings_tonga_a11,
 764							ARRAY_SIZE(golden_settings_tonga_a11));
 765		amdgpu_device_program_register_sequence(adev,
 766							tonga_golden_common_all,
 767							ARRAY_SIZE(tonga_golden_common_all));
 768		break;
 769	case CHIP_VEGAM:
 770		amdgpu_device_program_register_sequence(adev,
 771							golden_settings_vegam_a11,
 772							ARRAY_SIZE(golden_settings_vegam_a11));
 773		amdgpu_device_program_register_sequence(adev,
 774							vegam_golden_common_all,
 775							ARRAY_SIZE(vegam_golden_common_all));
 776		break;
 777	case CHIP_POLARIS11:
 778	case CHIP_POLARIS12:
 779		amdgpu_device_program_register_sequence(adev,
 780							golden_settings_polaris11_a11,
 781							ARRAY_SIZE(golden_settings_polaris11_a11));
 782		amdgpu_device_program_register_sequence(adev,
 783							polaris11_golden_common_all,
 784							ARRAY_SIZE(polaris11_golden_common_all));
 785		break;
 786	case CHIP_POLARIS10:
 787		amdgpu_device_program_register_sequence(adev,
 788							golden_settings_polaris10_a11,
 789							ARRAY_SIZE(golden_settings_polaris10_a11));
 790		amdgpu_device_program_register_sequence(adev,
 791							polaris10_golden_common_all,
 792							ARRAY_SIZE(polaris10_golden_common_all));
 793		WREG32_SMC(ixCG_ACLK_CNTL, 0x0000001C);
 794		if (adev->pdev->revision == 0xc7 &&
 
 
 
 795		    ((adev->pdev->subsystem_device == 0xb37 && adev->pdev->subsystem_vendor == 0x1002) ||
 796		     (adev->pdev->subsystem_device == 0x4a8 && adev->pdev->subsystem_vendor == 0x1043) ||
 797		     (adev->pdev->subsystem_device == 0x9480 && adev->pdev->subsystem_vendor == 0x1682))) {
 798			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1E, 0xDD);
 799			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1F, 0xD0);
 800		}
 801		break;
 802	case CHIP_CARRIZO:
 803		amdgpu_device_program_register_sequence(adev,
 804							cz_mgcg_cgcg_init,
 805							ARRAY_SIZE(cz_mgcg_cgcg_init));
 806		amdgpu_device_program_register_sequence(adev,
 807							cz_golden_settings_a11,
 808							ARRAY_SIZE(cz_golden_settings_a11));
 809		amdgpu_device_program_register_sequence(adev,
 810							cz_golden_common_all,
 811							ARRAY_SIZE(cz_golden_common_all));
 812		break;
 813	case CHIP_STONEY:
 814		amdgpu_device_program_register_sequence(adev,
 815							stoney_mgcg_cgcg_init,
 816							ARRAY_SIZE(stoney_mgcg_cgcg_init));
 817		amdgpu_device_program_register_sequence(adev,
 818							stoney_golden_settings_a11,
 819							ARRAY_SIZE(stoney_golden_settings_a11));
 820		amdgpu_device_program_register_sequence(adev,
 821							stoney_golden_common_all,
 822							ARRAY_SIZE(stoney_golden_common_all));
 823		break;
 824	default:
 825		break;
 826	}
 827}
 828
 829static void gfx_v8_0_scratch_init(struct amdgpu_device *adev)
 830{
 831	adev->gfx.scratch.num_reg = 8;
 832	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
 833	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
 834}
 835
 836static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring)
 837{
 838	struct amdgpu_device *adev = ring->adev;
 839	uint32_t scratch;
 840	uint32_t tmp = 0;
 841	unsigned i;
 842	int r;
 843
 844	r = amdgpu_gfx_scratch_get(adev, &scratch);
 
 845	if (r)
 846		return r;
 847
 848	WREG32(scratch, 0xCAFEDEAD);
 849	r = amdgpu_ring_alloc(ring, 3);
 850	if (r)
 851		goto error_free_scratch;
 852
 853	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
 854	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
 855	amdgpu_ring_write(ring, 0xDEADBEEF);
 856	amdgpu_ring_commit(ring);
 857
 858	for (i = 0; i < adev->usec_timeout; i++) {
 859		tmp = RREG32(scratch);
 860		if (tmp == 0xDEADBEEF)
 861			break;
 862		udelay(1);
 863	}
 864
 865	if (i >= adev->usec_timeout)
 866		r = -ETIMEDOUT;
 867
 868error_free_scratch:
 869	amdgpu_gfx_scratch_free(adev, scratch);
 870	return r;
 871}
 872
 873static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 874{
 875	struct amdgpu_device *adev = ring->adev;
 876	struct amdgpu_ib ib;
 877	struct dma_fence *f = NULL;
 878
 879	unsigned int index;
 880	uint64_t gpu_addr;
 881	uint32_t tmp;
 882	long r;
 883
 884	r = amdgpu_device_wb_get(adev, &index);
 885	if (r)
 886		return r;
 887
 888	gpu_addr = adev->wb.gpu_addr + (index * 4);
 889	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
 890	memset(&ib, 0, sizeof(ib));
 891	r = amdgpu_ib_get(adev, NULL, 16, &ib);
 
 892	if (r)
 893		goto err1;
 894
 895	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
 896	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
 897	ib.ptr[2] = lower_32_bits(gpu_addr);
 898	ib.ptr[3] = upper_32_bits(gpu_addr);
 899	ib.ptr[4] = 0xDEADBEEF;
 900	ib.length_dw = 5;
 901
 902	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
 903	if (r)
 904		goto err2;
 905
 906	r = dma_fence_wait_timeout(f, false, timeout);
 907	if (r == 0) {
 908		r = -ETIMEDOUT;
 909		goto err2;
 910	} else if (r < 0) {
 911		goto err2;
 912	}
 913
 914	tmp = adev->wb.wb[index];
 915	if (tmp == 0xDEADBEEF)
 916		r = 0;
 917	else
 918		r = -EINVAL;
 919
 920err2:
 921	amdgpu_ib_free(adev, &ib, NULL);
 922	dma_fence_put(f);
 923err1:
 924	amdgpu_device_wb_free(adev, index);
 925	return r;
 926}
 927
 928
 929static void gfx_v8_0_free_microcode(struct amdgpu_device *adev)
 930{
 931	release_firmware(adev->gfx.pfp_fw);
 932	adev->gfx.pfp_fw = NULL;
 933	release_firmware(adev->gfx.me_fw);
 934	adev->gfx.me_fw = NULL;
 935	release_firmware(adev->gfx.ce_fw);
 936	adev->gfx.ce_fw = NULL;
 937	release_firmware(adev->gfx.rlc_fw);
 938	adev->gfx.rlc_fw = NULL;
 939	release_firmware(adev->gfx.mec_fw);
 940	adev->gfx.mec_fw = NULL;
 941	if ((adev->asic_type != CHIP_STONEY) &&
 942	    (adev->asic_type != CHIP_TOPAZ))
 943		release_firmware(adev->gfx.mec2_fw);
 944	adev->gfx.mec2_fw = NULL;
 945
 946	kfree(adev->gfx.rlc.register_list_format);
 947}
 948
 949static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
 950{
 951	const char *chip_name;
 952	char fw_name[30];
 953	int err;
 954	struct amdgpu_firmware_info *info = NULL;
 955	const struct common_firmware_header *header = NULL;
 956	const struct gfx_firmware_header_v1_0 *cp_hdr;
 957	const struct rlc_firmware_header_v2_0 *rlc_hdr;
 958	unsigned int *tmp = NULL, i;
 959
 960	DRM_DEBUG("\n");
 961
 962	switch (adev->asic_type) {
 963	case CHIP_TOPAZ:
 964		chip_name = "topaz";
 965		break;
 966	case CHIP_TONGA:
 967		chip_name = "tonga";
 968		break;
 969	case CHIP_CARRIZO:
 970		chip_name = "carrizo";
 971		break;
 972	case CHIP_FIJI:
 973		chip_name = "fiji";
 974		break;
 975	case CHIP_STONEY:
 976		chip_name = "stoney";
 977		break;
 978	case CHIP_POLARIS10:
 979		chip_name = "polaris10";
 980		break;
 981	case CHIP_POLARIS11:
 982		chip_name = "polaris11";
 983		break;
 984	case CHIP_POLARIS12:
 985		chip_name = "polaris12";
 986		break;
 987	case CHIP_VEGAM:
 988		chip_name = "vegam";
 989		break;
 990	default:
 991		BUG();
 992	}
 993
 994	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
 995		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp_2.bin", chip_name);
 996		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
 997		if (err == -ENOENT) {
 998			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
 999			err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1000		}
1001	} else {
1002		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1003		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1004	}
1005	if (err)
1006		goto out;
1007	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1008	if (err)
1009		goto out;
1010	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1011	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1012	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1013
1014	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1015		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me_2.bin", chip_name);
1016		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1017		if (err == -ENOENT) {
1018			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1019			err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1020		}
1021	} else {
1022		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1023		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1024	}
1025	if (err)
1026		goto out;
1027	err = amdgpu_ucode_validate(adev->gfx.me_fw);
1028	if (err)
1029		goto out;
1030	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1031	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1032
1033	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1034
1035	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1036		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce_2.bin", chip_name);
1037		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1038		if (err == -ENOENT) {
1039			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1040			err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1041		}
1042	} else {
1043		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1044		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1045	}
1046	if (err)
1047		goto out;
1048	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1049	if (err)
1050		goto out;
1051	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1052	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1053	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1054
1055	/*
1056	 * Support for MCBP/Virtualization in combination with chained IBs is
1057	 * formal released on feature version #46
1058	 */
1059	if (adev->gfx.ce_feature_version >= 46 &&
1060	    adev->gfx.pfp_feature_version >= 46) {
1061		adev->virt.chained_ib_support = true;
1062		DRM_INFO("Chained IB support enabled!\n");
1063	} else
1064		adev->virt.chained_ib_support = false;
1065
1066	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1067	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1068	if (err)
1069		goto out;
1070	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1071	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1072	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1073	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1074
1075	adev->gfx.rlc.save_and_restore_offset =
1076			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1077	adev->gfx.rlc.clear_state_descriptor_offset =
1078			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1079	adev->gfx.rlc.avail_scratch_ram_locations =
1080			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1081	adev->gfx.rlc.reg_restore_list_size =
1082			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1083	adev->gfx.rlc.reg_list_format_start =
1084			le32_to_cpu(rlc_hdr->reg_list_format_start);
1085	adev->gfx.rlc.reg_list_format_separate_start =
1086			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1087	adev->gfx.rlc.starting_offsets_start =
1088			le32_to_cpu(rlc_hdr->starting_offsets_start);
1089	adev->gfx.rlc.reg_list_format_size_bytes =
1090			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1091	adev->gfx.rlc.reg_list_size_bytes =
1092			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1093
1094	adev->gfx.rlc.register_list_format =
1095			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1096					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1097
1098	if (!adev->gfx.rlc.register_list_format) {
1099		err = -ENOMEM;
1100		goto out;
1101	}
1102
1103	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1104			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1105	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1106		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1107
1108	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1109
1110	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1111			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1112	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1113		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1114
1115	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1116		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec_2.bin", chip_name);
1117		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1118		if (err == -ENOENT) {
1119			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1120			err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1121		}
1122	} else {
1123		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1124		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1125	}
1126	if (err)
1127		goto out;
1128	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1129	if (err)
1130		goto out;
1131	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1132	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1133	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1134
1135	if ((adev->asic_type != CHIP_STONEY) &&
1136	    (adev->asic_type != CHIP_TOPAZ)) {
1137		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1138			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2_2.bin", chip_name);
1139			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1140			if (err == -ENOENT) {
1141				snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1142				err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1143			}
1144		} else {
1145			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1146			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1147		}
1148		if (!err) {
1149			err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1150			if (err)
1151				goto out;
1152			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1153				adev->gfx.mec2_fw->data;
1154			adev->gfx.mec2_fw_version =
1155				le32_to_cpu(cp_hdr->header.ucode_version);
1156			adev->gfx.mec2_feature_version =
1157				le32_to_cpu(cp_hdr->ucode_feature_version);
1158		} else {
1159			err = 0;
1160			adev->gfx.mec2_fw = NULL;
1161		}
1162	}
1163
1164	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1165	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1166	info->fw = adev->gfx.pfp_fw;
1167	header = (const struct common_firmware_header *)info->fw->data;
1168	adev->firmware.fw_size +=
1169		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1170
1171	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1172	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1173	info->fw = adev->gfx.me_fw;
1174	header = (const struct common_firmware_header *)info->fw->data;
1175	adev->firmware.fw_size +=
1176		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1177
1178	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1179	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1180	info->fw = adev->gfx.ce_fw;
1181	header = (const struct common_firmware_header *)info->fw->data;
1182	adev->firmware.fw_size +=
1183		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1184
1185	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1186	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1187	info->fw = adev->gfx.rlc_fw;
1188	header = (const struct common_firmware_header *)info->fw->data;
1189	adev->firmware.fw_size +=
1190		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1191
1192	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1193	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1194	info->fw = adev->gfx.mec_fw;
1195	header = (const struct common_firmware_header *)info->fw->data;
1196	adev->firmware.fw_size +=
1197		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1198
1199	/* we need account JT in */
1200	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1201	adev->firmware.fw_size +=
1202		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1203
1204	if (amdgpu_sriov_vf(adev)) {
1205		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1206		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1207		info->fw = adev->gfx.mec_fw;
1208		adev->firmware.fw_size +=
1209			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1210	}
1211
1212	if (adev->gfx.mec2_fw) {
1213		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1214		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1215		info->fw = adev->gfx.mec2_fw;
1216		header = (const struct common_firmware_header *)info->fw->data;
1217		adev->firmware.fw_size +=
1218			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1219	}
1220
1221out:
1222	if (err) {
1223		dev_err(adev->dev,
1224			"gfx8: Failed to load firmware \"%s\"\n",
1225			fw_name);
1226		release_firmware(adev->gfx.pfp_fw);
1227		adev->gfx.pfp_fw = NULL;
1228		release_firmware(adev->gfx.me_fw);
1229		adev->gfx.me_fw = NULL;
1230		release_firmware(adev->gfx.ce_fw);
1231		adev->gfx.ce_fw = NULL;
1232		release_firmware(adev->gfx.rlc_fw);
1233		adev->gfx.rlc_fw = NULL;
1234		release_firmware(adev->gfx.mec_fw);
1235		adev->gfx.mec_fw = NULL;
1236		release_firmware(adev->gfx.mec2_fw);
1237		adev->gfx.mec2_fw = NULL;
1238	}
1239	return err;
1240}
1241
1242static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1243				    volatile u32 *buffer)
1244{
1245	u32 count = 0, i;
1246	const struct cs_section_def *sect = NULL;
1247	const struct cs_extent_def *ext = NULL;
1248
1249	if (adev->gfx.rlc.cs_data == NULL)
1250		return;
1251	if (buffer == NULL)
1252		return;
1253
1254	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1255	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1256
1257	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1258	buffer[count++] = cpu_to_le32(0x80000000);
1259	buffer[count++] = cpu_to_le32(0x80000000);
1260
1261	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1262		for (ext = sect->section; ext->extent != NULL; ++ext) {
1263			if (sect->id == SECT_CONTEXT) {
1264				buffer[count++] =
1265					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1266				buffer[count++] = cpu_to_le32(ext->reg_index -
1267						PACKET3_SET_CONTEXT_REG_START);
1268				for (i = 0; i < ext->reg_count; i++)
1269					buffer[count++] = cpu_to_le32(ext->extent[i]);
1270			} else {
1271				return;
1272			}
1273		}
1274	}
1275
1276	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1277	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1278			PACKET3_SET_CONTEXT_REG_START);
1279	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1280	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1281
1282	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1283	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1284
1285	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1286	buffer[count++] = cpu_to_le32(0);
1287}
1288
1289static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1290{
1291	if (adev->asic_type == CHIP_CARRIZO)
1292		return 5;
1293	else
1294		return 4;
1295}
1296
1297static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1298{
1299	const struct cs_section_def *cs_data;
1300	int r;
1301
1302	adev->gfx.rlc.cs_data = vi_cs_data;
1303
1304	cs_data = adev->gfx.rlc.cs_data;
1305
1306	if (cs_data) {
1307		/* init clear state block */
1308		r = amdgpu_gfx_rlc_init_csb(adev);
1309		if (r)
1310			return r;
1311	}
1312
1313	if ((adev->asic_type == CHIP_CARRIZO) ||
1314	    (adev->asic_type == CHIP_STONEY)) {
1315		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1316		r = amdgpu_gfx_rlc_init_cpt(adev);
1317		if (r)
1318			return r;
1319	}
1320
 
 
 
 
1321	return 0;
1322}
1323
1324static int gfx_v8_0_csb_vram_pin(struct amdgpu_device *adev)
1325{
1326	int r;
1327
1328	r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, false);
1329	if (unlikely(r != 0))
1330		return r;
1331
1332	r = amdgpu_bo_pin(adev->gfx.rlc.clear_state_obj,
1333			AMDGPU_GEM_DOMAIN_VRAM);
1334	if (!r)
1335		adev->gfx.rlc.clear_state_gpu_addr =
1336			amdgpu_bo_gpu_offset(adev->gfx.rlc.clear_state_obj);
1337
1338	amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
1339
1340	return r;
1341}
1342
1343static void gfx_v8_0_csb_vram_unpin(struct amdgpu_device *adev)
1344{
1345	int r;
1346
1347	if (!adev->gfx.rlc.clear_state_obj)
1348		return;
1349
1350	r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, true);
1351	if (likely(r == 0)) {
1352		amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj);
1353		amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
1354	}
1355}
1356
1357static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1358{
1359	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1360}
1361
1362static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1363{
1364	int r;
1365	u32 *hpd;
1366	size_t mec_hpd_size;
1367
1368	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1369
1370	/* take ownership of the relevant compute queues */
1371	amdgpu_gfx_compute_queue_acquire(adev);
1372
1373	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
 
 
 
 
 
 
 
 
 
 
 
 
 
1374
1375	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1376				      AMDGPU_GEM_DOMAIN_VRAM,
1377				      &adev->gfx.mec.hpd_eop_obj,
1378				      &adev->gfx.mec.hpd_eop_gpu_addr,
1379				      (void **)&hpd);
1380	if (r) {
1381		dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1382		return r;
1383	}
1384
1385	memset(hpd, 0, mec_hpd_size);
1386
1387	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1388	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1389
1390	return 0;
1391}
1392
1393static const u32 vgpr_init_compute_shader[] =
1394{
1395	0x7e000209, 0x7e020208,
1396	0x7e040207, 0x7e060206,
1397	0x7e080205, 0x7e0a0204,
1398	0x7e0c0203, 0x7e0e0202,
1399	0x7e100201, 0x7e120200,
1400	0x7e140209, 0x7e160208,
1401	0x7e180207, 0x7e1a0206,
1402	0x7e1c0205, 0x7e1e0204,
1403	0x7e200203, 0x7e220202,
1404	0x7e240201, 0x7e260200,
1405	0x7e280209, 0x7e2a0208,
1406	0x7e2c0207, 0x7e2e0206,
1407	0x7e300205, 0x7e320204,
1408	0x7e340203, 0x7e360202,
1409	0x7e380201, 0x7e3a0200,
1410	0x7e3c0209, 0x7e3e0208,
1411	0x7e400207, 0x7e420206,
1412	0x7e440205, 0x7e460204,
1413	0x7e480203, 0x7e4a0202,
1414	0x7e4c0201, 0x7e4e0200,
1415	0x7e500209, 0x7e520208,
1416	0x7e540207, 0x7e560206,
1417	0x7e580205, 0x7e5a0204,
1418	0x7e5c0203, 0x7e5e0202,
1419	0x7e600201, 0x7e620200,
1420	0x7e640209, 0x7e660208,
1421	0x7e680207, 0x7e6a0206,
1422	0x7e6c0205, 0x7e6e0204,
1423	0x7e700203, 0x7e720202,
1424	0x7e740201, 0x7e760200,
1425	0x7e780209, 0x7e7a0208,
1426	0x7e7c0207, 0x7e7e0206,
1427	0xbf8a0000, 0xbf810000,
1428};
1429
1430static const u32 sgpr_init_compute_shader[] =
1431{
1432	0xbe8a0100, 0xbe8c0102,
1433	0xbe8e0104, 0xbe900106,
1434	0xbe920108, 0xbe940100,
1435	0xbe960102, 0xbe980104,
1436	0xbe9a0106, 0xbe9c0108,
1437	0xbe9e0100, 0xbea00102,
1438	0xbea20104, 0xbea40106,
1439	0xbea60108, 0xbea80100,
1440	0xbeaa0102, 0xbeac0104,
1441	0xbeae0106, 0xbeb00108,
1442	0xbeb20100, 0xbeb40102,
1443	0xbeb60104, 0xbeb80106,
1444	0xbeba0108, 0xbebc0100,
1445	0xbebe0102, 0xbec00104,
1446	0xbec20106, 0xbec40108,
1447	0xbec60100, 0xbec80102,
1448	0xbee60004, 0xbee70005,
1449	0xbeea0006, 0xbeeb0007,
1450	0xbee80008, 0xbee90009,
1451	0xbefc0000, 0xbf8a0000,
1452	0xbf810000, 0x00000000,
1453};
1454
1455static const u32 vgpr_init_regs[] =
1456{
1457	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1458	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1459	mmCOMPUTE_NUM_THREAD_X, 256*4,
1460	mmCOMPUTE_NUM_THREAD_Y, 1,
1461	mmCOMPUTE_NUM_THREAD_Z, 1,
1462	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1463	mmCOMPUTE_PGM_RSRC2, 20,
1464	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1465	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1466	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1467	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1468	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1469	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1470	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1471	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1472	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1473	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1474};
1475
1476static const u32 sgpr1_init_regs[] =
1477{
1478	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1479	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1480	mmCOMPUTE_NUM_THREAD_X, 256*5,
1481	mmCOMPUTE_NUM_THREAD_Y, 1,
1482	mmCOMPUTE_NUM_THREAD_Z, 1,
1483	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1484	mmCOMPUTE_PGM_RSRC2, 20,
1485	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1486	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1487	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1488	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1489	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1490	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1491	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1492	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1493	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1494	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1495};
1496
1497static const u32 sgpr2_init_regs[] =
1498{
1499	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1500	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1501	mmCOMPUTE_NUM_THREAD_X, 256*5,
1502	mmCOMPUTE_NUM_THREAD_Y, 1,
1503	mmCOMPUTE_NUM_THREAD_Z, 1,
1504	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1505	mmCOMPUTE_PGM_RSRC2, 20,
1506	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1507	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1508	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1509	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1510	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1511	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1512	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1513	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1514	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1515	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1516};
1517
1518static const u32 sec_ded_counter_registers[] =
1519{
1520	mmCPC_EDC_ATC_CNT,
1521	mmCPC_EDC_SCRATCH_CNT,
1522	mmCPC_EDC_UCODE_CNT,
1523	mmCPF_EDC_ATC_CNT,
1524	mmCPF_EDC_ROQ_CNT,
1525	mmCPF_EDC_TAG_CNT,
1526	mmCPG_EDC_ATC_CNT,
1527	mmCPG_EDC_DMA_CNT,
1528	mmCPG_EDC_TAG_CNT,
1529	mmDC_EDC_CSINVOC_CNT,
1530	mmDC_EDC_RESTORE_CNT,
1531	mmDC_EDC_STATE_CNT,
1532	mmGDS_EDC_CNT,
1533	mmGDS_EDC_GRBM_CNT,
1534	mmGDS_EDC_OA_DED,
1535	mmSPI_EDC_CNT,
1536	mmSQC_ATC_EDC_GATCL1_CNT,
1537	mmSQC_EDC_CNT,
1538	mmSQ_EDC_DED_CNT,
1539	mmSQ_EDC_INFO,
1540	mmSQ_EDC_SEC_CNT,
1541	mmTCC_EDC_CNT,
1542	mmTCP_ATC_EDC_GATCL1_CNT,
1543	mmTCP_EDC_CNT,
1544	mmTD_EDC_CNT
1545};
1546
1547static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1548{
1549	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1550	struct amdgpu_ib ib;
1551	struct dma_fence *f = NULL;
1552	int r, i;
1553	u32 tmp;
1554	unsigned total_size, vgpr_offset, sgpr_offset;
1555	u64 gpu_addr;
1556
1557	/* only supported on CZ */
1558	if (adev->asic_type != CHIP_CARRIZO)
1559		return 0;
1560
1561	/* bail if the compute ring is not ready */
1562	if (!ring->sched.ready)
1563		return 0;
1564
1565	tmp = RREG32(mmGB_EDC_MODE);
1566	WREG32(mmGB_EDC_MODE, 0);
1567
1568	total_size =
1569		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1570	total_size +=
1571		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1572	total_size +=
1573		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1574	total_size = ALIGN(total_size, 256);
1575	vgpr_offset = total_size;
1576	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1577	sgpr_offset = total_size;
1578	total_size += sizeof(sgpr_init_compute_shader);
1579
1580	/* allocate an indirect buffer to put the commands in */
1581	memset(&ib, 0, sizeof(ib));
1582	r = amdgpu_ib_get(adev, NULL, total_size, &ib);
 
1583	if (r) {
1584		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1585		return r;
1586	}
1587
1588	/* load the compute shaders */
1589	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1590		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1591
1592	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1593		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1594
1595	/* init the ib length to 0 */
1596	ib.length_dw = 0;
1597
1598	/* VGPR */
1599	/* write the register state for the compute dispatch */
1600	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1601		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1602		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1603		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1604	}
1605	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1606	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1607	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1608	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1609	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1610	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1611
1612	/* write dispatch packet */
1613	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1614	ib.ptr[ib.length_dw++] = 8; /* x */
1615	ib.ptr[ib.length_dw++] = 1; /* y */
1616	ib.ptr[ib.length_dw++] = 1; /* z */
1617	ib.ptr[ib.length_dw++] =
1618		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1619
1620	/* write CS partial flush packet */
1621	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1622	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1623
1624	/* SGPR1 */
1625	/* write the register state for the compute dispatch */
1626	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1627		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1628		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1629		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1630	}
1631	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1632	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1633	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1634	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1635	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1636	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1637
1638	/* write dispatch packet */
1639	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1640	ib.ptr[ib.length_dw++] = 8; /* x */
1641	ib.ptr[ib.length_dw++] = 1; /* y */
1642	ib.ptr[ib.length_dw++] = 1; /* z */
1643	ib.ptr[ib.length_dw++] =
1644		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1645
1646	/* write CS partial flush packet */
1647	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1648	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1649
1650	/* SGPR2 */
1651	/* write the register state for the compute dispatch */
1652	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1653		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1654		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1655		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1656	}
1657	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1658	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1659	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1660	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1661	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1662	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1663
1664	/* write dispatch packet */
1665	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1666	ib.ptr[ib.length_dw++] = 8; /* x */
1667	ib.ptr[ib.length_dw++] = 1; /* y */
1668	ib.ptr[ib.length_dw++] = 1; /* z */
1669	ib.ptr[ib.length_dw++] =
1670		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1671
1672	/* write CS partial flush packet */
1673	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1674	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1675
1676	/* shedule the ib on the ring */
1677	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1678	if (r) {
1679		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1680		goto fail;
1681	}
1682
1683	/* wait for the GPU to finish processing the IB */
1684	r = dma_fence_wait(f, false);
1685	if (r) {
1686		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1687		goto fail;
1688	}
1689
1690	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1691	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1692	WREG32(mmGB_EDC_MODE, tmp);
1693
1694	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1695	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1696	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1697
1698
1699	/* read back registers to clear the counters */
1700	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1701		RREG32(sec_ded_counter_registers[i]);
1702
1703fail:
1704	amdgpu_ib_free(adev, &ib, NULL);
1705	dma_fence_put(f);
1706
1707	return r;
1708}
1709
1710static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1711{
1712	u32 gb_addr_config;
1713	u32 mc_shared_chmap, mc_arb_ramcfg;
1714	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1715	u32 tmp;
1716	int ret;
1717
1718	switch (adev->asic_type) {
1719	case CHIP_TOPAZ:
1720		adev->gfx.config.max_shader_engines = 1;
1721		adev->gfx.config.max_tile_pipes = 2;
1722		adev->gfx.config.max_cu_per_sh = 6;
1723		adev->gfx.config.max_sh_per_se = 1;
1724		adev->gfx.config.max_backends_per_se = 2;
1725		adev->gfx.config.max_texture_channel_caches = 2;
1726		adev->gfx.config.max_gprs = 256;
1727		adev->gfx.config.max_gs_threads = 32;
1728		adev->gfx.config.max_hw_contexts = 8;
1729
1730		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1731		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1732		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1733		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1734		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1735		break;
1736	case CHIP_FIJI:
1737		adev->gfx.config.max_shader_engines = 4;
1738		adev->gfx.config.max_tile_pipes = 16;
1739		adev->gfx.config.max_cu_per_sh = 16;
1740		adev->gfx.config.max_sh_per_se = 1;
1741		adev->gfx.config.max_backends_per_se = 4;
1742		adev->gfx.config.max_texture_channel_caches = 16;
1743		adev->gfx.config.max_gprs = 256;
1744		adev->gfx.config.max_gs_threads = 32;
1745		adev->gfx.config.max_hw_contexts = 8;
1746
1747		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1748		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1749		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1750		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1751		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1752		break;
1753	case CHIP_POLARIS11:
1754	case CHIP_POLARIS12:
1755		ret = amdgpu_atombios_get_gfx_info(adev);
1756		if (ret)
1757			return ret;
1758		adev->gfx.config.max_gprs = 256;
1759		adev->gfx.config.max_gs_threads = 32;
1760		adev->gfx.config.max_hw_contexts = 8;
1761
1762		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1763		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1764		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1765		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1766		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1767		break;
1768	case CHIP_POLARIS10:
1769	case CHIP_VEGAM:
1770		ret = amdgpu_atombios_get_gfx_info(adev);
1771		if (ret)
1772			return ret;
1773		adev->gfx.config.max_gprs = 256;
1774		adev->gfx.config.max_gs_threads = 32;
1775		adev->gfx.config.max_hw_contexts = 8;
1776
1777		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1778		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1779		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1780		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1781		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1782		break;
1783	case CHIP_TONGA:
1784		adev->gfx.config.max_shader_engines = 4;
1785		adev->gfx.config.max_tile_pipes = 8;
1786		adev->gfx.config.max_cu_per_sh = 8;
1787		adev->gfx.config.max_sh_per_se = 1;
1788		adev->gfx.config.max_backends_per_se = 2;
1789		adev->gfx.config.max_texture_channel_caches = 8;
1790		adev->gfx.config.max_gprs = 256;
1791		adev->gfx.config.max_gs_threads = 32;
1792		adev->gfx.config.max_hw_contexts = 8;
1793
1794		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1795		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1796		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1797		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1798		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1799		break;
1800	case CHIP_CARRIZO:
1801		adev->gfx.config.max_shader_engines = 1;
1802		adev->gfx.config.max_tile_pipes = 2;
1803		adev->gfx.config.max_sh_per_se = 1;
1804		adev->gfx.config.max_backends_per_se = 2;
1805		adev->gfx.config.max_cu_per_sh = 8;
1806		adev->gfx.config.max_texture_channel_caches = 2;
1807		adev->gfx.config.max_gprs = 256;
1808		adev->gfx.config.max_gs_threads = 32;
1809		adev->gfx.config.max_hw_contexts = 8;
1810
1811		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1812		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1813		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1814		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1815		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1816		break;
1817	case CHIP_STONEY:
1818		adev->gfx.config.max_shader_engines = 1;
1819		adev->gfx.config.max_tile_pipes = 2;
1820		adev->gfx.config.max_sh_per_se = 1;
1821		adev->gfx.config.max_backends_per_se = 1;
1822		adev->gfx.config.max_cu_per_sh = 3;
1823		adev->gfx.config.max_texture_channel_caches = 2;
1824		adev->gfx.config.max_gprs = 256;
1825		adev->gfx.config.max_gs_threads = 16;
1826		adev->gfx.config.max_hw_contexts = 8;
1827
1828		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1829		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1830		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1831		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1832		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1833		break;
1834	default:
1835		adev->gfx.config.max_shader_engines = 2;
1836		adev->gfx.config.max_tile_pipes = 4;
1837		adev->gfx.config.max_cu_per_sh = 2;
1838		adev->gfx.config.max_sh_per_se = 1;
1839		adev->gfx.config.max_backends_per_se = 2;
1840		adev->gfx.config.max_texture_channel_caches = 4;
1841		adev->gfx.config.max_gprs = 256;
1842		adev->gfx.config.max_gs_threads = 32;
1843		adev->gfx.config.max_hw_contexts = 8;
1844
1845		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1846		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1847		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1848		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1849		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1850		break;
1851	}
1852
1853	mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP);
1854	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1855	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1856
 
 
 
 
 
1857	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1858	adev->gfx.config.mem_max_burst_length_bytes = 256;
1859	if (adev->flags & AMD_IS_APU) {
1860		/* Get memory bank mapping mode. */
1861		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1862		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1863		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1864
1865		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1866		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1867		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1868
1869		/* Validate settings in case only one DIMM installed. */
1870		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1871			dimm00_addr_map = 0;
1872		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1873			dimm01_addr_map = 0;
1874		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1875			dimm10_addr_map = 0;
1876		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1877			dimm11_addr_map = 0;
1878
1879		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1880		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1881		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1882			adev->gfx.config.mem_row_size_in_kb = 2;
1883		else
1884			adev->gfx.config.mem_row_size_in_kb = 1;
1885	} else {
1886		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1887		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1888		if (adev->gfx.config.mem_row_size_in_kb > 4)
1889			adev->gfx.config.mem_row_size_in_kb = 4;
1890	}
1891
1892	adev->gfx.config.shader_engine_tile_size = 32;
1893	adev->gfx.config.num_gpus = 1;
1894	adev->gfx.config.multi_gpu_tile_size = 64;
1895
1896	/* fix up row size */
1897	switch (adev->gfx.config.mem_row_size_in_kb) {
1898	case 1:
1899	default:
1900		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1901		break;
1902	case 2:
1903		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1904		break;
1905	case 4:
1906		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1907		break;
1908	}
1909	adev->gfx.config.gb_addr_config = gb_addr_config;
1910
1911	return 0;
1912}
1913
1914static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1915					int mec, int pipe, int queue)
1916{
1917	int r;
1918	unsigned irq_type;
1919	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
 
1920
1921	ring = &adev->gfx.compute_ring[ring_id];
1922
1923	/* mec0 is me1 */
1924	ring->me = mec + 1;
1925	ring->pipe = pipe;
1926	ring->queue = queue;
1927
1928	ring->ring_obj = NULL;
1929	ring->use_doorbell = true;
1930	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1931	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1932				+ (ring_id * GFX8_MEC_HPD_SIZE);
1933	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1934
1935	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1936		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1937		+ ring->pipe;
1938
 
 
1939	/* type-2 packets are deprecated on MEC, use type-3 instead */
1940	r = amdgpu_ring_init(adev, ring, 1024,
1941			&adev->gfx.eop_irq, irq_type);
1942	if (r)
1943		return r;
1944
1945
1946	return 0;
1947}
1948
1949static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1950
1951static int gfx_v8_0_sw_init(void *handle)
1952{
1953	int i, j, k, r, ring_id;
 
1954	struct amdgpu_ring *ring;
1955	struct amdgpu_kiq *kiq;
1956	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1957
1958	switch (adev->asic_type) {
1959	case CHIP_TONGA:
1960	case CHIP_CARRIZO:
1961	case CHIP_FIJI:
1962	case CHIP_POLARIS10:
1963	case CHIP_POLARIS11:
1964	case CHIP_POLARIS12:
1965	case CHIP_VEGAM:
1966		adev->gfx.mec.num_mec = 2;
1967		break;
1968	case CHIP_TOPAZ:
1969	case CHIP_STONEY:
1970	default:
1971		adev->gfx.mec.num_mec = 1;
1972		break;
1973	}
1974
1975	adev->gfx.mec.num_pipe_per_mec = 4;
1976	adev->gfx.mec.num_queue_per_pipe = 8;
1977
1978	/* EOP Event */
1979	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1980	if (r)
1981		return r;
1982
1983	/* Privileged reg */
1984	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1985			      &adev->gfx.priv_reg_irq);
1986	if (r)
1987		return r;
1988
1989	/* Privileged inst */
1990	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1991			      &adev->gfx.priv_inst_irq);
1992	if (r)
1993		return r;
1994
1995	/* Add CP EDC/ECC irq  */
1996	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1997			      &adev->gfx.cp_ecc_error_irq);
1998	if (r)
1999		return r;
2000
2001	/* SQ interrupts. */
2002	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
2003			      &adev->gfx.sq_irq);
2004	if (r) {
2005		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
2006		return r;
2007	}
2008
2009	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
2010
2011	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2012
2013	gfx_v8_0_scratch_init(adev);
2014
2015	r = gfx_v8_0_init_microcode(adev);
2016	if (r) {
2017		DRM_ERROR("Failed to load gfx firmware!\n");
2018		return r;
2019	}
2020
2021	r = adev->gfx.rlc.funcs->init(adev);
2022	if (r) {
2023		DRM_ERROR("Failed to init rlc BOs!\n");
2024		return r;
2025	}
2026
2027	r = gfx_v8_0_mec_init(adev);
2028	if (r) {
2029		DRM_ERROR("Failed to init MEC BOs!\n");
2030		return r;
2031	}
2032
2033	/* set up the gfx ring */
2034	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2035		ring = &adev->gfx.gfx_ring[i];
2036		ring->ring_obj = NULL;
2037		sprintf(ring->name, "gfx");
2038		/* no gfx doorbells on iceland */
2039		if (adev->asic_type != CHIP_TOPAZ) {
2040			ring->use_doorbell = true;
2041			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
2042		}
2043
2044		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2045				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP);
 
2046		if (r)
2047			return r;
2048	}
2049
2050
2051	/* set up the compute queues - allocate horizontally across pipes */
2052	ring_id = 0;
2053	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2054		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2055			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2056				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
 
2057					continue;
2058
2059				r = gfx_v8_0_compute_ring_init(adev,
2060								ring_id,
2061								i, k, j);
2062				if (r)
2063					return r;
2064
2065				ring_id++;
2066			}
2067		}
2068	}
2069
2070	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE);
2071	if (r) {
2072		DRM_ERROR("Failed to init KIQ BOs!\n");
2073		return r;
2074	}
2075
2076	kiq = &adev->gfx.kiq;
2077	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2078	if (r)
2079		return r;
2080
2081	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2082	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation));
2083	if (r)
2084		return r;
2085
2086	adev->gfx.ce_ram_size = 0x8000;
2087
2088	r = gfx_v8_0_gpu_early_init(adev);
2089	if (r)
2090		return r;
2091
2092	return 0;
2093}
2094
2095static int gfx_v8_0_sw_fini(void *handle)
2096{
2097	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2098	int i;
2099
2100	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2101		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2102	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2103		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2104
2105	amdgpu_gfx_mqd_sw_fini(adev);
2106	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring, &adev->gfx.kiq.irq);
2107	amdgpu_gfx_kiq_fini(adev);
2108
2109	gfx_v8_0_mec_fini(adev);
2110	amdgpu_gfx_rlc_fini(adev);
2111	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2112				&adev->gfx.rlc.clear_state_gpu_addr,
2113				(void **)&adev->gfx.rlc.cs_ptr);
2114	if ((adev->asic_type == CHIP_CARRIZO) ||
2115	    (adev->asic_type == CHIP_STONEY)) {
2116		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2117				&adev->gfx.rlc.cp_table_gpu_addr,
2118				(void **)&adev->gfx.rlc.cp_table_ptr);
2119	}
2120	gfx_v8_0_free_microcode(adev);
2121
2122	return 0;
2123}
2124
2125static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2126{
2127	uint32_t *modearray, *mod2array;
2128	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2129	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2130	u32 reg_offset;
2131
2132	modearray = adev->gfx.config.tile_mode_array;
2133	mod2array = adev->gfx.config.macrotile_mode_array;
2134
2135	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2136		modearray[reg_offset] = 0;
2137
2138	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2139		mod2array[reg_offset] = 0;
2140
2141	switch (adev->asic_type) {
2142	case CHIP_TOPAZ:
2143		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2144				PIPE_CONFIG(ADDR_SURF_P2) |
2145				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2146				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2147		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2148				PIPE_CONFIG(ADDR_SURF_P2) |
2149				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2150				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2151		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2152				PIPE_CONFIG(ADDR_SURF_P2) |
2153				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2154				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2155		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2156				PIPE_CONFIG(ADDR_SURF_P2) |
2157				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2158				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2159		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2160				PIPE_CONFIG(ADDR_SURF_P2) |
2161				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2162				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2163		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2164				PIPE_CONFIG(ADDR_SURF_P2) |
2165				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2166				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2167		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2168				PIPE_CONFIG(ADDR_SURF_P2) |
2169				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2170				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2171		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2172				PIPE_CONFIG(ADDR_SURF_P2));
2173		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2174				PIPE_CONFIG(ADDR_SURF_P2) |
2175				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2176				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2177		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2178				 PIPE_CONFIG(ADDR_SURF_P2) |
2179				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2180				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2181		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2182				 PIPE_CONFIG(ADDR_SURF_P2) |
2183				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2184				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2185		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2186				 PIPE_CONFIG(ADDR_SURF_P2) |
2187				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2188				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2189		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2190				 PIPE_CONFIG(ADDR_SURF_P2) |
2191				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2192				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2193		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2194				 PIPE_CONFIG(ADDR_SURF_P2) |
2195				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2196				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2197		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2198				 PIPE_CONFIG(ADDR_SURF_P2) |
2199				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2200				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2201		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2202				 PIPE_CONFIG(ADDR_SURF_P2) |
2203				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2204				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2205		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2206				 PIPE_CONFIG(ADDR_SURF_P2) |
2207				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2208				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2209		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2210				 PIPE_CONFIG(ADDR_SURF_P2) |
2211				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2212				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2213		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2214				 PIPE_CONFIG(ADDR_SURF_P2) |
2215				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2216				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2217		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2218				 PIPE_CONFIG(ADDR_SURF_P2) |
2219				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2220				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2221		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2222				 PIPE_CONFIG(ADDR_SURF_P2) |
2223				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2224				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2225		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2226				 PIPE_CONFIG(ADDR_SURF_P2) |
2227				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2228				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2229		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2230				 PIPE_CONFIG(ADDR_SURF_P2) |
2231				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2232				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2233		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2234				 PIPE_CONFIG(ADDR_SURF_P2) |
2235				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2236				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2237		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2238				 PIPE_CONFIG(ADDR_SURF_P2) |
2239				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2240				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2241		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2242				 PIPE_CONFIG(ADDR_SURF_P2) |
2243				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2244				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2245
2246		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2247				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2248				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2249				NUM_BANKS(ADDR_SURF_8_BANK));
2250		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2251				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2252				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2253				NUM_BANKS(ADDR_SURF_8_BANK));
2254		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2255				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2256				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2257				NUM_BANKS(ADDR_SURF_8_BANK));
2258		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2259				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2260				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2261				NUM_BANKS(ADDR_SURF_8_BANK));
2262		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2263				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2264				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2265				NUM_BANKS(ADDR_SURF_8_BANK));
2266		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2267				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2268				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2269				NUM_BANKS(ADDR_SURF_8_BANK));
2270		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2271				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2272				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2273				NUM_BANKS(ADDR_SURF_8_BANK));
2274		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2275				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2276				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2277				NUM_BANKS(ADDR_SURF_16_BANK));
2278		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2279				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2280				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2281				NUM_BANKS(ADDR_SURF_16_BANK));
2282		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2283				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2284				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2285				 NUM_BANKS(ADDR_SURF_16_BANK));
2286		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2287				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2288				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2289				 NUM_BANKS(ADDR_SURF_16_BANK));
2290		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2291				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2292				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2293				 NUM_BANKS(ADDR_SURF_16_BANK));
2294		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2295				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2296				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2297				 NUM_BANKS(ADDR_SURF_16_BANK));
2298		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2299				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2300				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2301				 NUM_BANKS(ADDR_SURF_8_BANK));
2302
2303		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2304			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2305			    reg_offset != 23)
2306				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2307
2308		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2309			if (reg_offset != 7)
2310				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2311
2312		break;
2313	case CHIP_FIJI:
2314	case CHIP_VEGAM:
2315		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2316				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2317				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2318				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2319		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2320				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2321				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2322				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2323		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2324				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2325				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2326				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2327		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2328				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2329				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2330				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2331		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2332				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2333				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2334				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2335		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2336				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2337				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2338				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2339		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2340				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2341				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2342				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2343		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2344				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2345				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2346				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2347		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2348				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2349		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2350				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2351				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2352				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2353		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2354				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2355				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2356				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2357		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2358				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2359				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2360				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2361		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2362				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2363				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2364				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2365		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2366				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2367				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2368				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2369		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2370				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2371				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2372				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2373		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2374				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2375				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2376				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2377		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2378				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2379				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2380				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2381		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2382				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2383				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2384				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2385		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2386				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2387				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2388				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2389		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2390				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2391				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2392				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2393		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2394				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2395				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2396				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2397		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2398				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2399				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2400				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2401		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2402				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2403				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2404				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2405		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2406				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2407				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2408				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2409		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2410				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2411				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2412				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2413		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2414				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2415				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2416				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2417		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2418				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2419				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2420				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2421		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2422				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2423				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2424				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2425		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2426				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2427				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2428				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2429		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2430				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2431				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2432				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2433		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2434				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2435				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2436				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2437
2438		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2439				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2440				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2441				NUM_BANKS(ADDR_SURF_8_BANK));
2442		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2443				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2444				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2445				NUM_BANKS(ADDR_SURF_8_BANK));
2446		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2447				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2448				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2449				NUM_BANKS(ADDR_SURF_8_BANK));
2450		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2451				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2452				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2453				NUM_BANKS(ADDR_SURF_8_BANK));
2454		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2455				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2456				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2457				NUM_BANKS(ADDR_SURF_8_BANK));
2458		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2459				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2460				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2461				NUM_BANKS(ADDR_SURF_8_BANK));
2462		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2463				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2464				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2465				NUM_BANKS(ADDR_SURF_8_BANK));
2466		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2467				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2468				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2469				NUM_BANKS(ADDR_SURF_8_BANK));
2470		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2471				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2472				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2473				NUM_BANKS(ADDR_SURF_8_BANK));
2474		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2475				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2476				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2477				 NUM_BANKS(ADDR_SURF_8_BANK));
2478		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2479				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2480				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2481				 NUM_BANKS(ADDR_SURF_8_BANK));
2482		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2483				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2484				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2485				 NUM_BANKS(ADDR_SURF_8_BANK));
2486		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2487				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2488				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2489				 NUM_BANKS(ADDR_SURF_8_BANK));
2490		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2491				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2492				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2493				 NUM_BANKS(ADDR_SURF_4_BANK));
2494
2495		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2496			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2497
2498		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2499			if (reg_offset != 7)
2500				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2501
2502		break;
2503	case CHIP_TONGA:
2504		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2505				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2506				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2507				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2508		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2509				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2510				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2511				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2512		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2513				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2514				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2515				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2516		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2517				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2518				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2519				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2520		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2521				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2522				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2523				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2524		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2525				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2526				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2527				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2528		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2529				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2530				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2531				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2532		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2533				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2534				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2535				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2536		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2537				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2538		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2539				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2540				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2541				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2542		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2543				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2544				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2545				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2546		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2547				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2548				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2549				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2550		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2551				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2552				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2553				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2554		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2555				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2556				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2557				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2558		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2559				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2560				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2561				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2562		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2563				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2564				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2565				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2566		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2567				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2568				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2569				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2570		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2571				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2572				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2573				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2574		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2575				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2576				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2577				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2578		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2579				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2580				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2581				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2582		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2583				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2584				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2585				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2586		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2587				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2588				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2589				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2590		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2591				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2592				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2593				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2594		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2595				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2596				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2597				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2598		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2599				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2600				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2601				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2602		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2603				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2604				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2605				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2606		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2607				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2608				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2609				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2610		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2611				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2612				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2613				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2614		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2615				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2616				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2617				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2618		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2619				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2620				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2621				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2622		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2623				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2624				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2625				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2626
2627		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2628				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2629				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2630				NUM_BANKS(ADDR_SURF_16_BANK));
2631		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2632				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2633				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2634				NUM_BANKS(ADDR_SURF_16_BANK));
2635		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2636				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2637				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2638				NUM_BANKS(ADDR_SURF_16_BANK));
2639		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2640				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2641				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2642				NUM_BANKS(ADDR_SURF_16_BANK));
2643		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2644				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2645				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2646				NUM_BANKS(ADDR_SURF_16_BANK));
2647		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2648				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2649				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2650				NUM_BANKS(ADDR_SURF_16_BANK));
2651		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2652				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2653				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2654				NUM_BANKS(ADDR_SURF_16_BANK));
2655		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2656				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2657				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2658				NUM_BANKS(ADDR_SURF_16_BANK));
2659		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2660				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2661				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2662				NUM_BANKS(ADDR_SURF_16_BANK));
2663		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2664				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2665				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2666				 NUM_BANKS(ADDR_SURF_16_BANK));
2667		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2668				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2669				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2670				 NUM_BANKS(ADDR_SURF_16_BANK));
2671		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2672				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2673				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2674				 NUM_BANKS(ADDR_SURF_8_BANK));
2675		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2676				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2677				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2678				 NUM_BANKS(ADDR_SURF_4_BANK));
2679		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2680				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2681				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2682				 NUM_BANKS(ADDR_SURF_4_BANK));
2683
2684		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2685			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2686
2687		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2688			if (reg_offset != 7)
2689				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2690
2691		break;
2692	case CHIP_POLARIS11:
2693	case CHIP_POLARIS12:
2694		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2695				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2696				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2697				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2698		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2699				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2700				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2701				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2702		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2703				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2704				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2705				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2706		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2707				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2708				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2709				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2710		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2711				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2712				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2713				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2714		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2715				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2716				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2717				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2718		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2719				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2720				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2721				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2722		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2723				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2724				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2725				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2726		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2727				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2728		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2729				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2730				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2731				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2732		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2733				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2734				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2735				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2736		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2737				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2738				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2739				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2740		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2741				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2742				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2743				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2744		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2745				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2746				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2747				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2748		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2749				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2750				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2751				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2752		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2753				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2754				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2755				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2756		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2757				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2758				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2759				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2760		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2761				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2762				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2763				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2764		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2765				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2766				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2767				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2768		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2769				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2770				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2771				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2772		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2773				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2774				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2775				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2776		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2777				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2778				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2779				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2780		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2781				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2782				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2783				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2784		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2785				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2786				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2787				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2788		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2789				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2790				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2791				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2792		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2793				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2794				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2795				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2796		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2797				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2798				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2799				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2800		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2801				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2802				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2803				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2804		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2805				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2806				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2807				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2808		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2809				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2810				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2811				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2812		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2813				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2814				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2815				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2816
2817		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2818				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2819				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2820				NUM_BANKS(ADDR_SURF_16_BANK));
2821
2822		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2823				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2824				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2825				NUM_BANKS(ADDR_SURF_16_BANK));
2826
2827		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2828				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2829				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2830				NUM_BANKS(ADDR_SURF_16_BANK));
2831
2832		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2833				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2834				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2835				NUM_BANKS(ADDR_SURF_16_BANK));
2836
2837		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2838				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2839				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2840				NUM_BANKS(ADDR_SURF_16_BANK));
2841
2842		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2843				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2844				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2845				NUM_BANKS(ADDR_SURF_16_BANK));
2846
2847		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2848				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2849				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2850				NUM_BANKS(ADDR_SURF_16_BANK));
2851
2852		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2853				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2854				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2855				NUM_BANKS(ADDR_SURF_16_BANK));
2856
2857		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2858				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2859				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2860				NUM_BANKS(ADDR_SURF_16_BANK));
2861
2862		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2863				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2864				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2865				NUM_BANKS(ADDR_SURF_16_BANK));
2866
2867		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2868				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2869				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2870				NUM_BANKS(ADDR_SURF_16_BANK));
2871
2872		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2873				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2874				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2875				NUM_BANKS(ADDR_SURF_16_BANK));
2876
2877		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2878				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2879				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2880				NUM_BANKS(ADDR_SURF_8_BANK));
2881
2882		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2883				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2884				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2885				NUM_BANKS(ADDR_SURF_4_BANK));
2886
2887		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2888			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2889
2890		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2891			if (reg_offset != 7)
2892				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2893
2894		break;
2895	case CHIP_POLARIS10:
2896		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2897				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2898				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2899				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2900		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2901				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2902				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2903				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2904		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2905				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2906				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2907				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2908		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2909				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2910				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2911				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2912		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2913				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2914				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2915				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2916		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2917				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2918				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2919				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2920		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2921				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2922				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2923				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2924		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2925				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2926				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2927				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2928		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2929				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2930		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2931				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2932				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2933				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2934		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2935				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2936				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2937				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2938		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2939				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2940				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2941				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2942		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2943				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2944				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2945				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2946		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2947				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2948				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2949				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2950		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2951				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2952				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2953				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2954		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2955				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2956				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2957				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2958		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2959				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2960				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2961				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2962		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2963				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2964				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2965				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2966		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2967				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2968				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2969				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2970		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2971				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2972				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2973				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2974		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2975				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2976				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2977				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2978		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2979				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2980				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2981				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2982		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2983				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2984				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2985				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2986		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2987				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2988				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2989				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2990		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2991				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2992				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2993				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2994		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2995				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2996				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2997				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2998		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2999				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
3000				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3001				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3002		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3003				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
3004				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3005				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3006		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3007				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
3008				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3009				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3010		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3011				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
3012				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3013				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3014		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3015				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
3016				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3017				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3018
3019		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3020				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3021				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3022				NUM_BANKS(ADDR_SURF_16_BANK));
3023
3024		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3025				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3026				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3027				NUM_BANKS(ADDR_SURF_16_BANK));
3028
3029		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3030				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3031				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3032				NUM_BANKS(ADDR_SURF_16_BANK));
3033
3034		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3035				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3036				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3037				NUM_BANKS(ADDR_SURF_16_BANK));
3038
3039		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3040				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3041				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3042				NUM_BANKS(ADDR_SURF_16_BANK));
3043
3044		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3045				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3046				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3047				NUM_BANKS(ADDR_SURF_16_BANK));
3048
3049		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3050				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3051				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3052				NUM_BANKS(ADDR_SURF_16_BANK));
3053
3054		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3055				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3056				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3057				NUM_BANKS(ADDR_SURF_16_BANK));
3058
3059		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3060				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3061				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3062				NUM_BANKS(ADDR_SURF_16_BANK));
3063
3064		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3065				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3066				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3067				NUM_BANKS(ADDR_SURF_16_BANK));
3068
3069		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3070				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3071				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3072				NUM_BANKS(ADDR_SURF_16_BANK));
3073
3074		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3075				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3076				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3077				NUM_BANKS(ADDR_SURF_8_BANK));
3078
3079		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3080				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3081				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3082				NUM_BANKS(ADDR_SURF_4_BANK));
3083
3084		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3085				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3086				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3087				NUM_BANKS(ADDR_SURF_4_BANK));
3088
3089		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3090			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3091
3092		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3093			if (reg_offset != 7)
3094				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3095
3096		break;
3097	case CHIP_STONEY:
3098		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3099				PIPE_CONFIG(ADDR_SURF_P2) |
3100				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3101				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3102		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3103				PIPE_CONFIG(ADDR_SURF_P2) |
3104				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3105				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3106		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3107				PIPE_CONFIG(ADDR_SURF_P2) |
3108				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3109				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3110		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3111				PIPE_CONFIG(ADDR_SURF_P2) |
3112				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3113				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3114		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3115				PIPE_CONFIG(ADDR_SURF_P2) |
3116				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3117				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3118		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3119				PIPE_CONFIG(ADDR_SURF_P2) |
3120				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3121				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3122		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3123				PIPE_CONFIG(ADDR_SURF_P2) |
3124				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3125				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3126		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3127				PIPE_CONFIG(ADDR_SURF_P2));
3128		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3129				PIPE_CONFIG(ADDR_SURF_P2) |
3130				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3131				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3132		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3133				 PIPE_CONFIG(ADDR_SURF_P2) |
3134				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3135				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3136		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3137				 PIPE_CONFIG(ADDR_SURF_P2) |
3138				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3139				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3140		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3141				 PIPE_CONFIG(ADDR_SURF_P2) |
3142				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3143				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3144		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3145				 PIPE_CONFIG(ADDR_SURF_P2) |
3146				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3147				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3148		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3149				 PIPE_CONFIG(ADDR_SURF_P2) |
3150				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3151				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3152		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3153				 PIPE_CONFIG(ADDR_SURF_P2) |
3154				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3155				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3156		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3157				 PIPE_CONFIG(ADDR_SURF_P2) |
3158				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3159				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3160		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3161				 PIPE_CONFIG(ADDR_SURF_P2) |
3162				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3163				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3164		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3165				 PIPE_CONFIG(ADDR_SURF_P2) |
3166				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3167				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3168		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3169				 PIPE_CONFIG(ADDR_SURF_P2) |
3170				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3171				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3172		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3173				 PIPE_CONFIG(ADDR_SURF_P2) |
3174				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3175				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3176		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3177				 PIPE_CONFIG(ADDR_SURF_P2) |
3178				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3179				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3180		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3181				 PIPE_CONFIG(ADDR_SURF_P2) |
3182				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3183				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3184		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3185				 PIPE_CONFIG(ADDR_SURF_P2) |
3186				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3187				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3188		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3189				 PIPE_CONFIG(ADDR_SURF_P2) |
3190				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3191				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3192		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3193				 PIPE_CONFIG(ADDR_SURF_P2) |
3194				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3195				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3196		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3197				 PIPE_CONFIG(ADDR_SURF_P2) |
3198				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3199				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3200
3201		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3202				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3203				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3204				NUM_BANKS(ADDR_SURF_8_BANK));
3205		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3206				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3207				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3208				NUM_BANKS(ADDR_SURF_8_BANK));
3209		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3210				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3211				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3212				NUM_BANKS(ADDR_SURF_8_BANK));
3213		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3214				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3215				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3216				NUM_BANKS(ADDR_SURF_8_BANK));
3217		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3218				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3219				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3220				NUM_BANKS(ADDR_SURF_8_BANK));
3221		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3222				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3223				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3224				NUM_BANKS(ADDR_SURF_8_BANK));
3225		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3226				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3227				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3228				NUM_BANKS(ADDR_SURF_8_BANK));
3229		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3230				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3231				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3232				NUM_BANKS(ADDR_SURF_16_BANK));
3233		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3234				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3235				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3236				NUM_BANKS(ADDR_SURF_16_BANK));
3237		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3238				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3239				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3240				 NUM_BANKS(ADDR_SURF_16_BANK));
3241		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3242				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3243				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3244				 NUM_BANKS(ADDR_SURF_16_BANK));
3245		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3246				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3247				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3248				 NUM_BANKS(ADDR_SURF_16_BANK));
3249		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3250				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3251				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3252				 NUM_BANKS(ADDR_SURF_16_BANK));
3253		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3254				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3255				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3256				 NUM_BANKS(ADDR_SURF_8_BANK));
3257
3258		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3259			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3260			    reg_offset != 23)
3261				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3262
3263		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3264			if (reg_offset != 7)
3265				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3266
3267		break;
3268	default:
3269		dev_warn(adev->dev,
3270			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3271			 adev->asic_type);
3272		/* fall through */
3273
3274	case CHIP_CARRIZO:
3275		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3276				PIPE_CONFIG(ADDR_SURF_P2) |
3277				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3278				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3279		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3280				PIPE_CONFIG(ADDR_SURF_P2) |
3281				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3282				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3283		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3284				PIPE_CONFIG(ADDR_SURF_P2) |
3285				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3286				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3287		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3288				PIPE_CONFIG(ADDR_SURF_P2) |
3289				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3290				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3291		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3292				PIPE_CONFIG(ADDR_SURF_P2) |
3293				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3294				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3295		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3296				PIPE_CONFIG(ADDR_SURF_P2) |
3297				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3298				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3299		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3300				PIPE_CONFIG(ADDR_SURF_P2) |
3301				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3302				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3303		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3304				PIPE_CONFIG(ADDR_SURF_P2));
3305		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3306				PIPE_CONFIG(ADDR_SURF_P2) |
3307				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3308				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3309		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3310				 PIPE_CONFIG(ADDR_SURF_P2) |
3311				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3312				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3313		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3314				 PIPE_CONFIG(ADDR_SURF_P2) |
3315				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3316				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3317		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3318				 PIPE_CONFIG(ADDR_SURF_P2) |
3319				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3320				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3321		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3322				 PIPE_CONFIG(ADDR_SURF_P2) |
3323				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3324				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3325		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3326				 PIPE_CONFIG(ADDR_SURF_P2) |
3327				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3328				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3329		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3330				 PIPE_CONFIG(ADDR_SURF_P2) |
3331				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3332				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3333		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3334				 PIPE_CONFIG(ADDR_SURF_P2) |
3335				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3336				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3337		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3338				 PIPE_CONFIG(ADDR_SURF_P2) |
3339				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3340				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3341		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3342				 PIPE_CONFIG(ADDR_SURF_P2) |
3343				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3344				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3345		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3346				 PIPE_CONFIG(ADDR_SURF_P2) |
3347				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3348				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3349		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3350				 PIPE_CONFIG(ADDR_SURF_P2) |
3351				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3352				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3353		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3354				 PIPE_CONFIG(ADDR_SURF_P2) |
3355				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3356				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3357		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3358				 PIPE_CONFIG(ADDR_SURF_P2) |
3359				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3360				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3361		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3362				 PIPE_CONFIG(ADDR_SURF_P2) |
3363				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3364				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3365		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3366				 PIPE_CONFIG(ADDR_SURF_P2) |
3367				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3368				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3369		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3370				 PIPE_CONFIG(ADDR_SURF_P2) |
3371				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3372				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3373		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3374				 PIPE_CONFIG(ADDR_SURF_P2) |
3375				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3376				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3377
3378		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3379				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3380				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3381				NUM_BANKS(ADDR_SURF_8_BANK));
3382		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3383				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3384				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3385				NUM_BANKS(ADDR_SURF_8_BANK));
3386		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3387				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3388				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3389				NUM_BANKS(ADDR_SURF_8_BANK));
3390		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3391				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3392				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3393				NUM_BANKS(ADDR_SURF_8_BANK));
3394		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3395				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3396				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3397				NUM_BANKS(ADDR_SURF_8_BANK));
3398		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3399				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3400				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3401				NUM_BANKS(ADDR_SURF_8_BANK));
3402		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3403				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3404				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3405				NUM_BANKS(ADDR_SURF_8_BANK));
3406		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3407				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3408				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3409				NUM_BANKS(ADDR_SURF_16_BANK));
3410		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3411				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3412				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3413				NUM_BANKS(ADDR_SURF_16_BANK));
3414		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3415				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3416				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3417				 NUM_BANKS(ADDR_SURF_16_BANK));
3418		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3419				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3420				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3421				 NUM_BANKS(ADDR_SURF_16_BANK));
3422		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3423				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3424				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3425				 NUM_BANKS(ADDR_SURF_16_BANK));
3426		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3427				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3428				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3429				 NUM_BANKS(ADDR_SURF_16_BANK));
3430		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3431				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3432				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3433				 NUM_BANKS(ADDR_SURF_8_BANK));
3434
3435		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3436			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3437			    reg_offset != 23)
3438				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3439
3440		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3441			if (reg_offset != 7)
3442				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3443
3444		break;
3445	}
3446}
3447
3448static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3449				  u32 se_num, u32 sh_num, u32 instance)
 
3450{
3451	u32 data;
3452
3453	if (instance == 0xffffffff)
3454		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3455	else
3456		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3457
3458	if (se_num == 0xffffffff)
3459		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3460	else
3461		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3462
3463	if (sh_num == 0xffffffff)
3464		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3465	else
3466		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3467
3468	WREG32(mmGRBM_GFX_INDEX, data);
3469}
3470
3471static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3472				  u32 me, u32 pipe, u32 q, u32 vm)
3473{
3474	vi_srbm_select(adev, me, pipe, q, vm);
3475}
3476
3477static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3478{
3479	u32 data, mask;
3480
3481	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3482		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3483
3484	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3485
3486	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3487					 adev->gfx.config.max_sh_per_se);
3488
3489	return (~data) & mask;
3490}
3491
3492static void
3493gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3494{
3495	switch (adev->asic_type) {
3496	case CHIP_FIJI:
3497	case CHIP_VEGAM:
3498		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3499			  RB_XSEL2(1) | PKR_MAP(2) |
3500			  PKR_XSEL(1) | PKR_YSEL(1) |
3501			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3502		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3503			   SE_PAIR_YSEL(2);
3504		break;
3505	case CHIP_TONGA:
3506	case CHIP_POLARIS10:
3507		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3508			  SE_XSEL(1) | SE_YSEL(1);
3509		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3510			   SE_PAIR_YSEL(2);
3511		break;
3512	case CHIP_TOPAZ:
3513	case CHIP_CARRIZO:
3514		*rconf |= RB_MAP_PKR0(2);
3515		*rconf1 |= 0x0;
3516		break;
3517	case CHIP_POLARIS11:
3518	case CHIP_POLARIS12:
3519		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3520			  SE_XSEL(1) | SE_YSEL(1);
3521		*rconf1 |= 0x0;
3522		break;
3523	case CHIP_STONEY:
3524		*rconf |= 0x0;
3525		*rconf1 |= 0x0;
3526		break;
3527	default:
3528		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3529		break;
3530	}
3531}
3532
3533static void
3534gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3535					u32 raster_config, u32 raster_config_1,
3536					unsigned rb_mask, unsigned num_rb)
3537{
3538	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3539	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3540	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3541	unsigned rb_per_se = num_rb / num_se;
3542	unsigned se_mask[4];
3543	unsigned se;
3544
3545	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3546	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3547	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3548	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3549
3550	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3551	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3552	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3553
3554	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3555			     (!se_mask[2] && !se_mask[3]))) {
3556		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3557
3558		if (!se_mask[0] && !se_mask[1]) {
3559			raster_config_1 |=
3560				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3561		} else {
3562			raster_config_1 |=
3563				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3564		}
3565	}
3566
3567	for (se = 0; se < num_se; se++) {
3568		unsigned raster_config_se = raster_config;
3569		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3570		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3571		int idx = (se / 2) * 2;
3572
3573		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3574			raster_config_se &= ~SE_MAP_MASK;
3575
3576			if (!se_mask[idx]) {
3577				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3578			} else {
3579				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3580			}
3581		}
3582
3583		pkr0_mask &= rb_mask;
3584		pkr1_mask &= rb_mask;
3585		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3586			raster_config_se &= ~PKR_MAP_MASK;
3587
3588			if (!pkr0_mask) {
3589				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3590			} else {
3591				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3592			}
3593		}
3594
3595		if (rb_per_se >= 2) {
3596			unsigned rb0_mask = 1 << (se * rb_per_se);
3597			unsigned rb1_mask = rb0_mask << 1;
3598
3599			rb0_mask &= rb_mask;
3600			rb1_mask &= rb_mask;
3601			if (!rb0_mask || !rb1_mask) {
3602				raster_config_se &= ~RB_MAP_PKR0_MASK;
3603
3604				if (!rb0_mask) {
3605					raster_config_se |=
3606						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3607				} else {
3608					raster_config_se |=
3609						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3610				}
3611			}
3612
3613			if (rb_per_se > 2) {
3614				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3615				rb1_mask = rb0_mask << 1;
3616				rb0_mask &= rb_mask;
3617				rb1_mask &= rb_mask;
3618				if (!rb0_mask || !rb1_mask) {
3619					raster_config_se &= ~RB_MAP_PKR1_MASK;
3620
3621					if (!rb0_mask) {
3622						raster_config_se |=
3623							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3624					} else {
3625						raster_config_se |=
3626							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3627					}
3628				}
3629			}
3630		}
3631
3632		/* GRBM_GFX_INDEX has a different offset on VI */
3633		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
3634		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3635		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3636	}
3637
3638	/* GRBM_GFX_INDEX has a different offset on VI */
3639	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3640}
3641
3642static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3643{
3644	int i, j;
3645	u32 data;
3646	u32 raster_config = 0, raster_config_1 = 0;
3647	u32 active_rbs = 0;
3648	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3649					adev->gfx.config.max_sh_per_se;
3650	unsigned num_rb_pipes;
3651
3652	mutex_lock(&adev->grbm_idx_mutex);
3653	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3654		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3655			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3656			data = gfx_v8_0_get_rb_active_bitmap(adev);
3657			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3658					       rb_bitmap_width_per_sh);
3659		}
3660	}
3661	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3662
3663	adev->gfx.config.backend_enable_mask = active_rbs;
3664	adev->gfx.config.num_rbs = hweight32(active_rbs);
3665
3666	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3667			     adev->gfx.config.max_shader_engines, 16);
3668
3669	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3670
3671	if (!adev->gfx.config.backend_enable_mask ||
3672			adev->gfx.config.num_rbs >= num_rb_pipes) {
3673		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3674		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3675	} else {
3676		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3677							adev->gfx.config.backend_enable_mask,
3678							num_rb_pipes);
3679	}
3680
3681	/* cache the values for userspace */
3682	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3683		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3684			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3685			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3686				RREG32(mmCC_RB_BACKEND_DISABLE);
3687			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3688				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3689			adev->gfx.config.rb_config[i][j].raster_config =
3690				RREG32(mmPA_SC_RASTER_CONFIG);
3691			adev->gfx.config.rb_config[i][j].raster_config_1 =
3692				RREG32(mmPA_SC_RASTER_CONFIG_1);
3693		}
3694	}
3695	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3696	mutex_unlock(&adev->grbm_idx_mutex);
3697}
3698
 
3699/**
3700 * gfx_v8_0_init_compute_vmid - gart enable
3701 *
3702 * @adev: amdgpu_device pointer
3703 *
3704 * Initialize compute vmid sh_mem registers
3705 *
3706 */
3707#define DEFAULT_SH_MEM_BASES	(0x6000)
3708#define FIRST_COMPUTE_VMID	(8)
3709#define LAST_COMPUTE_VMID	(16)
3710static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3711{
3712	int i;
3713	uint32_t sh_mem_config;
3714	uint32_t sh_mem_bases;
3715
3716	/*
3717	 * Configure apertures:
3718	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3719	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3720	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3721	 */
3722	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3723
3724	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3725			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3726			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3727			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3728			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3729			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3730
3731	mutex_lock(&adev->srbm_mutex);
3732	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
3733		vi_srbm_select(adev, 0, 0, 0, i);
3734		/* CP and shaders */
3735		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3736		WREG32(mmSH_MEM_APE1_BASE, 1);
3737		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3738		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3739	}
3740	vi_srbm_select(adev, 0, 0, 0, 0);
3741	mutex_unlock(&adev->srbm_mutex);
3742
3743	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3744	   acccess. These should be enabled by FW for target VMIDs. */
3745	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
3746		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3747		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3748		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3749		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3750	}
3751}
3752
3753static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3754{
3755	int vmid;
3756
3757	/*
3758	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3759	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3760	 * the driver can enable them for graphics. VMID0 should maintain
3761	 * access so that HWS firmware can save/restore entries.
3762	 */
3763	for (vmid = 1; vmid < 16; vmid++) {
3764		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3765		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3766		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3767		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3768	}
3769}
3770
3771static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3772{
3773	switch (adev->asic_type) {
3774	default:
3775		adev->gfx.config.double_offchip_lds_buf = 1;
3776		break;
3777	case CHIP_CARRIZO:
3778	case CHIP_STONEY:
3779		adev->gfx.config.double_offchip_lds_buf = 0;
3780		break;
3781	}
3782}
3783
3784static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3785{
3786	u32 tmp, sh_static_mem_cfg;
3787	int i;
3788
3789	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3790	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3791	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3792	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3793
3794	gfx_v8_0_tiling_mode_table_init(adev);
3795	gfx_v8_0_setup_rb(adev);
3796	gfx_v8_0_get_cu_info(adev);
3797	gfx_v8_0_config_init(adev);
3798
3799	/* XXX SH_MEM regs */
3800	/* where to put LDS, scratch, GPUVM in FSA64 space */
3801	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3802				   SWIZZLE_ENABLE, 1);
3803	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3804				   ELEMENT_SIZE, 1);
3805	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3806				   INDEX_STRIDE, 3);
3807	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3808
3809	mutex_lock(&adev->srbm_mutex);
3810	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3811		vi_srbm_select(adev, 0, 0, 0, i);
3812		/* CP and shaders */
3813		if (i == 0) {
3814			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3815			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3816			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3817					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3818			WREG32(mmSH_MEM_CONFIG, tmp);
3819			WREG32(mmSH_MEM_BASES, 0);
3820		} else {
3821			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3822			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3823			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3824					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3825			WREG32(mmSH_MEM_CONFIG, tmp);
3826			tmp = adev->gmc.shared_aperture_start >> 48;
3827			WREG32(mmSH_MEM_BASES, tmp);
3828		}
3829
3830		WREG32(mmSH_MEM_APE1_BASE, 1);
3831		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3832	}
3833	vi_srbm_select(adev, 0, 0, 0, 0);
3834	mutex_unlock(&adev->srbm_mutex);
3835
3836	gfx_v8_0_init_compute_vmid(adev);
3837	gfx_v8_0_init_gds_vmid(adev);
3838
3839	mutex_lock(&adev->grbm_idx_mutex);
3840	/*
3841	 * making sure that the following register writes will be broadcasted
3842	 * to all the shaders
3843	 */
3844	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3845
3846	WREG32(mmPA_SC_FIFO_SIZE,
3847		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3848			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3849		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3850			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3851		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3852			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3853		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3854			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3855
3856	tmp = RREG32(mmSPI_ARB_PRIORITY);
3857	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3858	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3859	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3860	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3861	WREG32(mmSPI_ARB_PRIORITY, tmp);
3862
3863	mutex_unlock(&adev->grbm_idx_mutex);
3864
3865}
3866
3867static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3868{
3869	u32 i, j, k;
3870	u32 mask;
3871
3872	mutex_lock(&adev->grbm_idx_mutex);
3873	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3874		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3875			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3876			for (k = 0; k < adev->usec_timeout; k++) {
3877				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3878					break;
3879				udelay(1);
3880			}
3881			if (k == adev->usec_timeout) {
3882				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3883						      0xffffffff, 0xffffffff);
3884				mutex_unlock(&adev->grbm_idx_mutex);
3885				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3886					 i, j);
3887				return;
3888			}
3889		}
3890	}
3891	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3892	mutex_unlock(&adev->grbm_idx_mutex);
3893
3894	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3895		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3896		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3897		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3898	for (k = 0; k < adev->usec_timeout; k++) {
3899		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3900			break;
3901		udelay(1);
3902	}
3903}
3904
3905static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3906					       bool enable)
3907{
3908	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3909
3910	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3911	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3912	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3913	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3914
3915	WREG32(mmCP_INT_CNTL_RING0, tmp);
3916}
3917
3918static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3919{
 
3920	/* csib */
3921	WREG32(mmRLC_CSIB_ADDR_HI,
3922			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3923	WREG32(mmRLC_CSIB_ADDR_LO,
3924			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3925	WREG32(mmRLC_CSIB_LENGTH,
3926			adev->gfx.rlc.clear_state_size);
3927}
3928
3929static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3930				int ind_offset,
3931				int list_size,
3932				int *unique_indices,
3933				int *indices_count,
3934				int max_indices,
3935				int *ind_start_offsets,
3936				int *offset_count,
3937				int max_offset)
3938{
3939	int indices;
3940	bool new_entry = true;
3941
3942	for (; ind_offset < list_size; ind_offset++) {
3943
3944		if (new_entry) {
3945			new_entry = false;
3946			ind_start_offsets[*offset_count] = ind_offset;
3947			*offset_count = *offset_count + 1;
3948			BUG_ON(*offset_count >= max_offset);
3949		}
3950
3951		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3952			new_entry = true;
3953			continue;
3954		}
3955
3956		ind_offset += 2;
3957
3958		/* look for the matching indice */
3959		for (indices = 0;
3960			indices < *indices_count;
3961			indices++) {
3962			if (unique_indices[indices] ==
3963				register_list_format[ind_offset])
3964				break;
3965		}
3966
3967		if (indices >= *indices_count) {
3968			unique_indices[*indices_count] =
3969				register_list_format[ind_offset];
3970			indices = *indices_count;
3971			*indices_count = *indices_count + 1;
3972			BUG_ON(*indices_count >= max_indices);
3973		}
3974
3975		register_list_format[ind_offset] = indices;
3976	}
3977}
3978
3979static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3980{
3981	int i, temp, data;
3982	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3983	int indices_count = 0;
3984	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3985	int offset_count = 0;
3986
3987	int list_size;
3988	unsigned int *register_list_format =
3989		kmemdup(adev->gfx.rlc.register_list_format,
3990			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3991	if (!register_list_format)
3992		return -ENOMEM;
3993
3994	gfx_v8_0_parse_ind_reg_list(register_list_format,
3995				RLC_FormatDirectRegListLength,
3996				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3997				unique_indices,
3998				&indices_count,
3999				ARRAY_SIZE(unique_indices),
4000				indirect_start_offsets,
4001				&offset_count,
4002				ARRAY_SIZE(indirect_start_offsets));
4003
4004	/* save and restore list */
4005	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
4006
4007	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
4008	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
4009		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
4010
4011	/* indirect list */
4012	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
4013	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
4014		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
4015
4016	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
4017	list_size = list_size >> 1;
4018	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
4019	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
4020
4021	/* starting offsets starts */
4022	WREG32(mmRLC_GPM_SCRATCH_ADDR,
4023		adev->gfx.rlc.starting_offsets_start);
4024	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
4025		WREG32(mmRLC_GPM_SCRATCH_DATA,
4026				indirect_start_offsets[i]);
4027
4028	/* unique indices */
4029	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
4030	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
4031	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
4032		if (unique_indices[i] != 0) {
4033			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
4034			WREG32(data + i, unique_indices[i] >> 20);
4035		}
4036	}
4037	kfree(register_list_format);
4038
4039	return 0;
4040}
4041
4042static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
4043{
4044	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
4045}
4046
4047static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
4048{
4049	uint32_t data;
4050
4051	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
4052
4053	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
4054	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4055	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4056	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4057	WREG32(mmRLC_PG_DELAY, data);
4058
4059	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4060	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4061
4062}
4063
4064static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4065						bool enable)
4066{
4067	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4068}
4069
4070static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4071						  bool enable)
4072{
4073	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4074}
4075
4076static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4077{
4078	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4079}
4080
4081static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4082{
4083	if ((adev->asic_type == CHIP_CARRIZO) ||
4084	    (adev->asic_type == CHIP_STONEY)) {
4085		gfx_v8_0_init_csb(adev);
4086		gfx_v8_0_init_save_restore_list(adev);
4087		gfx_v8_0_enable_save_restore_machine(adev);
4088		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4089		gfx_v8_0_init_power_gating(adev);
4090		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4091	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4092		   (adev->asic_type == CHIP_POLARIS12) ||
4093		   (adev->asic_type == CHIP_VEGAM)) {
4094		gfx_v8_0_init_csb(adev);
4095		gfx_v8_0_init_save_restore_list(adev);
4096		gfx_v8_0_enable_save_restore_machine(adev);
4097		gfx_v8_0_init_power_gating(adev);
4098	}
4099
4100}
4101
4102static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4103{
4104	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4105
4106	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4107	gfx_v8_0_wait_for_rlc_serdes(adev);
4108}
4109
4110static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4111{
4112	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4113	udelay(50);
4114
4115	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4116	udelay(50);
4117}
4118
4119static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4120{
4121	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4122
4123	/* carrizo do enable cp interrupt after cp inited */
4124	if (!(adev->flags & AMD_IS_APU))
4125		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4126
4127	udelay(50);
4128}
4129
4130static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4131{
4132	if (amdgpu_sriov_vf(adev)) {
4133		gfx_v8_0_init_csb(adev);
4134		return 0;
4135	}
4136
4137	adev->gfx.rlc.funcs->stop(adev);
4138	adev->gfx.rlc.funcs->reset(adev);
4139	gfx_v8_0_init_pg(adev);
4140	adev->gfx.rlc.funcs->start(adev);
4141
4142	return 0;
4143}
4144
4145static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4146{
4147	int i;
4148	u32 tmp = RREG32(mmCP_ME_CNTL);
4149
4150	if (enable) {
4151		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4152		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4153		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4154	} else {
4155		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4156		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4157		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
4158		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4159			adev->gfx.gfx_ring[i].sched.ready = false;
4160	}
4161	WREG32(mmCP_ME_CNTL, tmp);
4162	udelay(50);
4163}
4164
4165static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4166{
4167	u32 count = 0;
4168	const struct cs_section_def *sect = NULL;
4169	const struct cs_extent_def *ext = NULL;
4170
4171	/* begin clear state */
4172	count += 2;
4173	/* context control state */
4174	count += 3;
4175
4176	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4177		for (ext = sect->section; ext->extent != NULL; ++ext) {
4178			if (sect->id == SECT_CONTEXT)
4179				count += 2 + ext->reg_count;
4180			else
4181				return 0;
4182		}
4183	}
4184	/* pa_sc_raster_config/pa_sc_raster_config1 */
4185	count += 4;
4186	/* end clear state */
4187	count += 2;
4188	/* clear state */
4189	count += 2;
4190
4191	return count;
4192}
4193
4194static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4195{
4196	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4197	const struct cs_section_def *sect = NULL;
4198	const struct cs_extent_def *ext = NULL;
4199	int r, i;
4200
4201	/* init the CP */
4202	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4203	WREG32(mmCP_ENDIAN_SWAP, 0);
4204	WREG32(mmCP_DEVICE_ID, 1);
4205
4206	gfx_v8_0_cp_gfx_enable(adev, true);
4207
4208	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4209	if (r) {
4210		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4211		return r;
4212	}
4213
4214	/* clear state buffer */
4215	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4216	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4217
4218	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4219	amdgpu_ring_write(ring, 0x80000000);
4220	amdgpu_ring_write(ring, 0x80000000);
4221
4222	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4223		for (ext = sect->section; ext->extent != NULL; ++ext) {
4224			if (sect->id == SECT_CONTEXT) {
4225				amdgpu_ring_write(ring,
4226				       PACKET3(PACKET3_SET_CONTEXT_REG,
4227					       ext->reg_count));
4228				amdgpu_ring_write(ring,
4229				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4230				for (i = 0; i < ext->reg_count; i++)
4231					amdgpu_ring_write(ring, ext->extent[i]);
4232			}
4233		}
4234	}
4235
4236	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4237	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4238	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4239	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4240
4241	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4242	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4243
4244	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4245	amdgpu_ring_write(ring, 0);
4246
4247	/* init the CE partitions */
4248	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4249	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4250	amdgpu_ring_write(ring, 0x8000);
4251	amdgpu_ring_write(ring, 0x8000);
4252
4253	amdgpu_ring_commit(ring);
4254
4255	return 0;
4256}
4257static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4258{
4259	u32 tmp;
4260	/* no gfx doorbells on iceland */
4261	if (adev->asic_type == CHIP_TOPAZ)
4262		return;
4263
4264	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4265
4266	if (ring->use_doorbell) {
4267		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4268				DOORBELL_OFFSET, ring->doorbell_index);
4269		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4270						DOORBELL_HIT, 0);
4271		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4272					    DOORBELL_EN, 1);
4273	} else {
4274		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4275	}
4276
4277	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4278
4279	if (adev->flags & AMD_IS_APU)
4280		return;
4281
4282	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4283					DOORBELL_RANGE_LOWER,
4284					adev->doorbell_index.gfx_ring0);
4285	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4286
4287	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4288		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4289}
4290
4291static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4292{
4293	struct amdgpu_ring *ring;
4294	u32 tmp;
4295	u32 rb_bufsz;
4296	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4297
4298	/* Set the write pointer delay */
4299	WREG32(mmCP_RB_WPTR_DELAY, 0);
4300
4301	/* set the RB to use vmid 0 */
4302	WREG32(mmCP_RB_VMID, 0);
4303
4304	/* Set ring buffer size */
4305	ring = &adev->gfx.gfx_ring[0];
4306	rb_bufsz = order_base_2(ring->ring_size / 8);
4307	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4308	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4309	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4310	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4311#ifdef __BIG_ENDIAN
4312	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4313#endif
4314	WREG32(mmCP_RB0_CNTL, tmp);
4315
4316	/* Initialize the ring buffer's read and write pointers */
4317	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4318	ring->wptr = 0;
4319	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4320
4321	/* set the wb address wether it's enabled or not */
4322	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
4323	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4324	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4325
4326	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4327	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4328	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4329	mdelay(1);
4330	WREG32(mmCP_RB0_CNTL, tmp);
4331
4332	rb_addr = ring->gpu_addr >> 8;
4333	WREG32(mmCP_RB0_BASE, rb_addr);
4334	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4335
4336	gfx_v8_0_set_cpg_door_bell(adev, ring);
4337	/* start the ring */
4338	amdgpu_ring_clear_ring(ring);
4339	gfx_v8_0_cp_gfx_start(adev);
4340	ring->sched.ready = true;
4341
4342	return 0;
4343}
4344
4345static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4346{
4347	int i;
4348
4349	if (enable) {
4350		WREG32(mmCP_MEC_CNTL, 0);
4351	} else {
4352		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4353		for (i = 0; i < adev->gfx.num_compute_rings; i++)
4354			adev->gfx.compute_ring[i].sched.ready = false;
4355		adev->gfx.kiq.ring.sched.ready = false;
4356	}
4357	udelay(50);
4358}
4359
4360/* KIQ functions */
4361static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4362{
4363	uint32_t tmp;
4364	struct amdgpu_device *adev = ring->adev;
4365
4366	/* tell RLC which is KIQ queue */
4367	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4368	tmp &= 0xffffff00;
4369	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4370	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4371	tmp |= 0x80;
4372	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4373}
4374
4375static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4376{
4377	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4378	uint64_t queue_mask = 0;
4379	int r, i;
4380
4381	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4382		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
4383			continue;
4384
4385		/* This situation may be hit in the future if a new HW
4386		 * generation exposes more than 64 queues. If so, the
4387		 * definition of queue_mask needs updating */
4388		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4389			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4390			break;
4391		}
4392
4393		queue_mask |= (1ull << i);
4394	}
4395
4396	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4397	if (r) {
4398		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4399		return r;
4400	}
4401	/* set resources */
4402	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4403	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4404	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4405	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4406	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4407	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4408	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4409	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4410	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4411		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4412		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4413		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4414
4415		/* map queues */
4416		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4417		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4418		amdgpu_ring_write(kiq_ring,
4419				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4420		amdgpu_ring_write(kiq_ring,
4421				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4422				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4423				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4424				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4425		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4426		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4427		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4428		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4429	}
4430
4431	amdgpu_ring_commit(kiq_ring);
4432
4433	return 0;
4434}
4435
4436static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4437{
4438	int i, r = 0;
4439
4440	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4441		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4442		for (i = 0; i < adev->usec_timeout; i++) {
4443			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4444				break;
4445			udelay(1);
4446		}
4447		if (i == adev->usec_timeout)
4448			r = -ETIMEDOUT;
4449	}
4450	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4451	WREG32(mmCP_HQD_PQ_RPTR, 0);
4452	WREG32(mmCP_HQD_PQ_WPTR, 0);
4453
4454	return r;
4455}
4456
 
 
 
 
 
 
 
 
 
 
 
 
 
4457static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4458{
4459	struct amdgpu_device *adev = ring->adev;
4460	struct vi_mqd *mqd = ring->mqd_ptr;
4461	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4462	uint32_t tmp;
4463
4464	mqd->header = 0xC0310800;
4465	mqd->compute_pipelinestat_enable = 0x00000001;
4466	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4467	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4468	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4469	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4470	mqd->compute_misc_reserved = 0x00000003;
4471	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4472						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4473	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4474						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4475	eop_base_addr = ring->eop_gpu_addr >> 8;
4476	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4477	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4478
4479	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4480	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4481	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4482			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4483
4484	mqd->cp_hqd_eop_control = tmp;
4485
4486	/* enable doorbell? */
4487	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4488			    CP_HQD_PQ_DOORBELL_CONTROL,
4489			    DOORBELL_EN,
4490			    ring->use_doorbell ? 1 : 0);
4491
4492	mqd->cp_hqd_pq_doorbell_control = tmp;
4493
4494	/* set the pointer to the MQD */
4495	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4496	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4497
4498	/* set MQD vmid to 0 */
4499	tmp = RREG32(mmCP_MQD_CONTROL);
4500	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4501	mqd->cp_mqd_control = tmp;
4502
4503	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4504	hqd_gpu_addr = ring->gpu_addr >> 8;
4505	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4506	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4507
4508	/* set up the HQD, this is similar to CP_RB0_CNTL */
4509	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4510	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4511			    (order_base_2(ring->ring_size / 4) - 1));
4512	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4513			((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
4514#ifdef __BIG_ENDIAN
4515	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4516#endif
4517	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4518	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4519	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4520	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4521	mqd->cp_hqd_pq_control = tmp;
4522
4523	/* set the wb address whether it's enabled or not */
4524	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
4525	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4526	mqd->cp_hqd_pq_rptr_report_addr_hi =
4527		upper_32_bits(wb_gpu_addr) & 0xffff;
4528
4529	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4530	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4531	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4532	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4533
4534	tmp = 0;
4535	/* enable the doorbell if requested */
4536	if (ring->use_doorbell) {
4537		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4538		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4539				DOORBELL_OFFSET, ring->doorbell_index);
4540
4541		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4542					 DOORBELL_EN, 1);
4543		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4544					 DOORBELL_SOURCE, 0);
4545		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4546					 DOORBELL_HIT, 0);
4547	}
4548
4549	mqd->cp_hqd_pq_doorbell_control = tmp;
4550
4551	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4552	ring->wptr = 0;
4553	mqd->cp_hqd_pq_wptr = ring->wptr;
4554	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4555
4556	/* set the vmid for the queue */
4557	mqd->cp_hqd_vmid = 0;
4558
4559	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4560	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4561	mqd->cp_hqd_persistent_state = tmp;
4562
4563	/* set MTYPE */
4564	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4565	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4566	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4567	mqd->cp_hqd_ib_control = tmp;
4568
4569	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4570	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4571	mqd->cp_hqd_iq_timer = tmp;
4572
4573	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4574	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4575	mqd->cp_hqd_ctx_save_control = tmp;
4576
4577	/* defaults */
4578	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4579	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
4580	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
4581	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
4582	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4583	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4584	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4585	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4586	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4587	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4588	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4589	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4590	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4591	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4592	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4593
4594	/* activate the queue */
4595	mqd->cp_hqd_active = 1;
 
 
 
 
 
 
 
4596
4597	return 0;
4598}
4599
4600int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4601			struct vi_mqd *mqd)
4602{
4603	uint32_t mqd_reg;
4604	uint32_t *mqd_data;
4605
4606	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4607	mqd_data = &mqd->cp_mqd_base_addr_lo;
4608
4609	/* disable wptr polling */
4610	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4611
4612	/* program all HQD registers */
4613	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4614		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4615
4616	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4617	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4618	 * on ASICs that do not support context-save.
4619	 * EOP writes/reads can start anywhere in the ring.
4620	 */
4621	if (adev->asic_type != CHIP_TONGA) {
4622		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4623		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4624		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4625	}
4626
4627	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4628		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4629
4630	/* activate the HQD */
4631	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4632		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4633
4634	return 0;
4635}
4636
4637static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4638{
4639	struct amdgpu_device *adev = ring->adev;
4640	struct vi_mqd *mqd = ring->mqd_ptr;
4641	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
4642
4643	gfx_v8_0_kiq_setting(ring);
4644
4645	if (adev->in_gpu_reset) { /* for GPU_RESET case */
4646		/* reset MQD to a clean status */
4647		if (adev->gfx.mec.mqd_backup[mqd_idx])
4648			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4649
4650		/* reset ring buffer */
4651		ring->wptr = 0;
4652		amdgpu_ring_clear_ring(ring);
4653		mutex_lock(&adev->srbm_mutex);
4654		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4655		gfx_v8_0_mqd_commit(adev, mqd);
4656		vi_srbm_select(adev, 0, 0, 0, 0);
4657		mutex_unlock(&adev->srbm_mutex);
4658	} else {
4659		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4660		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4661		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
 
 
4662		mutex_lock(&adev->srbm_mutex);
4663		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4664		gfx_v8_0_mqd_init(ring);
4665		gfx_v8_0_mqd_commit(adev, mqd);
4666		vi_srbm_select(adev, 0, 0, 0, 0);
4667		mutex_unlock(&adev->srbm_mutex);
4668
4669		if (adev->gfx.mec.mqd_backup[mqd_idx])
4670			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4671	}
4672
4673	return 0;
4674}
4675
4676static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4677{
4678	struct amdgpu_device *adev = ring->adev;
4679	struct vi_mqd *mqd = ring->mqd_ptr;
4680	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4681
4682	if (!adev->in_gpu_reset && !adev->in_suspend) {
4683		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4684		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4685		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4686		mutex_lock(&adev->srbm_mutex);
4687		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4688		gfx_v8_0_mqd_init(ring);
4689		vi_srbm_select(adev, 0, 0, 0, 0);
4690		mutex_unlock(&adev->srbm_mutex);
4691
4692		if (adev->gfx.mec.mqd_backup[mqd_idx])
4693			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4694	} else if (adev->in_gpu_reset) { /* for GPU_RESET case */
4695		/* reset MQD to a clean status */
4696		if (adev->gfx.mec.mqd_backup[mqd_idx])
4697			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4698		/* reset ring buffer */
4699		ring->wptr = 0;
4700		amdgpu_ring_clear_ring(ring);
4701	} else {
4702		amdgpu_ring_clear_ring(ring);
4703	}
4704	return 0;
4705}
4706
4707static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4708{
4709	if (adev->asic_type > CHIP_TONGA) {
4710		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4711		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4712	}
4713	/* enable doorbells */
4714	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4715}
4716
4717static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4718{
4719	struct amdgpu_ring *ring;
4720	int r;
4721
4722	ring = &adev->gfx.kiq.ring;
4723
4724	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4725	if (unlikely(r != 0))
4726		return r;
4727
4728	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4729	if (unlikely(r != 0))
 
4730		return r;
 
4731
4732	gfx_v8_0_kiq_init_queue(ring);
4733	amdgpu_bo_kunmap(ring->mqd_obj);
4734	ring->mqd_ptr = NULL;
4735	amdgpu_bo_unreserve(ring->mqd_obj);
4736	ring->sched.ready = true;
4737	return 0;
4738}
4739
4740static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4741{
4742	struct amdgpu_ring *ring = NULL;
4743	int r = 0, i;
4744
4745	gfx_v8_0_cp_compute_enable(adev, true);
4746
4747	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4748		ring = &adev->gfx.compute_ring[i];
4749
4750		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4751		if (unlikely(r != 0))
4752			goto done;
4753		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4754		if (!r) {
4755			r = gfx_v8_0_kcq_init_queue(ring);
4756			amdgpu_bo_kunmap(ring->mqd_obj);
4757			ring->mqd_ptr = NULL;
4758		}
4759		amdgpu_bo_unreserve(ring->mqd_obj);
4760		if (r)
4761			goto done;
4762	}
4763
4764	gfx_v8_0_set_mec_doorbell_range(adev);
4765
4766	r = gfx_v8_0_kiq_kcq_enable(adev);
4767	if (r)
4768		goto done;
4769
4770done:
4771	return r;
4772}
4773
4774static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4775{
4776	int r, i;
4777	struct amdgpu_ring *ring;
4778
4779	/* collect all the ring_tests here, gfx, kiq, compute */
4780	ring = &adev->gfx.gfx_ring[0];
4781	r = amdgpu_ring_test_helper(ring);
4782	if (r)
4783		return r;
4784
4785	ring = &adev->gfx.kiq.ring;
4786	r = amdgpu_ring_test_helper(ring);
4787	if (r)
4788		return r;
4789
4790	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4791		ring = &adev->gfx.compute_ring[i];
4792		amdgpu_ring_test_helper(ring);
4793	}
4794
4795	return 0;
4796}
4797
4798static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4799{
4800	int r;
4801
4802	if (!(adev->flags & AMD_IS_APU))
4803		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4804
4805	r = gfx_v8_0_kiq_resume(adev);
4806	if (r)
4807		return r;
4808
4809	r = gfx_v8_0_cp_gfx_resume(adev);
4810	if (r)
4811		return r;
4812
4813	r = gfx_v8_0_kcq_resume(adev);
4814	if (r)
4815		return r;
4816
4817	r = gfx_v8_0_cp_test_all_rings(adev);
4818	if (r)
4819		return r;
4820
4821	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4822
4823	return 0;
4824}
4825
4826static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4827{
4828	gfx_v8_0_cp_gfx_enable(adev, enable);
4829	gfx_v8_0_cp_compute_enable(adev, enable);
4830}
4831
4832static int gfx_v8_0_hw_init(void *handle)
4833{
4834	int r;
4835	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4836
4837	gfx_v8_0_init_golden_registers(adev);
4838	gfx_v8_0_constants_init(adev);
4839
4840	r = gfx_v8_0_csb_vram_pin(adev);
4841	if (r)
4842		return r;
4843
4844	r = adev->gfx.rlc.funcs->resume(adev);
4845	if (r)
4846		return r;
4847
4848	r = gfx_v8_0_cp_resume(adev);
4849
4850	return r;
4851}
4852
4853static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4854{
4855	int r, i;
4856	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4857
4858	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4859	if (r)
4860		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4861
4862	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4863		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4864
4865		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4866		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4867						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4868						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4869						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4870						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4871		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4872		amdgpu_ring_write(kiq_ring, 0);
4873		amdgpu_ring_write(kiq_ring, 0);
4874		amdgpu_ring_write(kiq_ring, 0);
4875	}
 
 
 
 
 
 
 
4876	r = amdgpu_ring_test_helper(kiq_ring);
4877	if (r)
4878		DRM_ERROR("KCQ disable failed\n");
4879
4880	return r;
4881}
4882
4883static bool gfx_v8_0_is_idle(void *handle)
4884{
4885	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4886
4887	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4888		|| RREG32(mmGRBM_STATUS2) != 0x8)
4889		return false;
4890	else
4891		return true;
4892}
4893
4894static bool gfx_v8_0_rlc_is_idle(void *handle)
4895{
4896	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4897
4898	if (RREG32(mmGRBM_STATUS2) != 0x8)
4899		return false;
4900	else
4901		return true;
4902}
4903
4904static int gfx_v8_0_wait_for_rlc_idle(void *handle)
4905{
4906	unsigned int i;
4907	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4908
4909	for (i = 0; i < adev->usec_timeout; i++) {
4910		if (gfx_v8_0_rlc_is_idle(handle))
4911			return 0;
4912
4913		udelay(1);
4914	}
4915	return -ETIMEDOUT;
4916}
4917
4918static int gfx_v8_0_wait_for_idle(void *handle)
4919{
4920	unsigned int i;
4921	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4922
4923	for (i = 0; i < adev->usec_timeout; i++) {
4924		if (gfx_v8_0_is_idle(handle))
4925			return 0;
4926
4927		udelay(1);
4928	}
4929	return -ETIMEDOUT;
4930}
4931
4932static int gfx_v8_0_hw_fini(void *handle)
4933{
4934	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4935
4936	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4937	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4938
4939	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4940
4941	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4942
4943	/* disable KCQ to avoid CPC touch memory not valid anymore */
4944	gfx_v8_0_kcq_disable(adev);
4945
4946	if (amdgpu_sriov_vf(adev)) {
4947		pr_debug("For SRIOV client, shouldn't do anything.\n");
4948		return 0;
4949	}
4950	amdgpu_gfx_rlc_enter_safe_mode(adev);
4951	if (!gfx_v8_0_wait_for_idle(adev))
 
4952		gfx_v8_0_cp_enable(adev, false);
4953	else
4954		pr_err("cp is busy, skip halt cp\n");
4955	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4956		adev->gfx.rlc.funcs->stop(adev);
4957	else
4958		pr_err("rlc is busy, skip halt rlc\n");
4959	amdgpu_gfx_rlc_exit_safe_mode(adev);
4960
4961	gfx_v8_0_csb_vram_unpin(adev);
4962
4963	return 0;
4964}
4965
4966static int gfx_v8_0_suspend(void *handle)
4967{
4968	return gfx_v8_0_hw_fini(handle);
4969}
4970
4971static int gfx_v8_0_resume(void *handle)
4972{
4973	return gfx_v8_0_hw_init(handle);
4974}
4975
4976static bool gfx_v8_0_check_soft_reset(void *handle)
4977{
4978	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4979	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4980	u32 tmp;
4981
4982	/* GRBM_STATUS */
4983	tmp = RREG32(mmGRBM_STATUS);
4984	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4985		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4986		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4987		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4988		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4989		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4990		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4991		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4992						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4993		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4994						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4995		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4996						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4997	}
4998
4999	/* GRBM_STATUS2 */
5000	tmp = RREG32(mmGRBM_STATUS2);
5001	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
5002		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
5003						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
5004
5005	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
5006	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
5007	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
5008		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
5009						SOFT_RESET_CPF, 1);
5010		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
5011						SOFT_RESET_CPC, 1);
5012		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
5013						SOFT_RESET_CPG, 1);
5014		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
5015						SOFT_RESET_GRBM, 1);
5016	}
5017
5018	/* SRBM_STATUS */
5019	tmp = RREG32(mmSRBM_STATUS);
5020	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
5021		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
5022						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
5023	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
5024		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
5025						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
5026
5027	if (grbm_soft_reset || srbm_soft_reset) {
5028		adev->gfx.grbm_soft_reset = grbm_soft_reset;
5029		adev->gfx.srbm_soft_reset = srbm_soft_reset;
5030		return true;
5031	} else {
5032		adev->gfx.grbm_soft_reset = 0;
5033		adev->gfx.srbm_soft_reset = 0;
5034		return false;
5035	}
5036}
5037
5038static int gfx_v8_0_pre_soft_reset(void *handle)
5039{
5040	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5041	u32 grbm_soft_reset = 0;
5042
5043	if ((!adev->gfx.grbm_soft_reset) &&
5044	    (!adev->gfx.srbm_soft_reset))
5045		return 0;
5046
5047	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5048
5049	/* stop the rlc */
5050	adev->gfx.rlc.funcs->stop(adev);
5051
5052	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5053	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5054		/* Disable GFX parsing/prefetching */
5055		gfx_v8_0_cp_gfx_enable(adev, false);
5056
5057	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5058	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5059	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5060	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5061		int i;
5062
5063		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5064			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5065
5066			mutex_lock(&adev->srbm_mutex);
5067			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5068			gfx_v8_0_deactivate_hqd(adev, 2);
5069			vi_srbm_select(adev, 0, 0, 0, 0);
5070			mutex_unlock(&adev->srbm_mutex);
5071		}
5072		/* Disable MEC parsing/prefetching */
5073		gfx_v8_0_cp_compute_enable(adev, false);
5074	}
5075
5076       return 0;
5077}
5078
5079static int gfx_v8_0_soft_reset(void *handle)
5080{
5081	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5082	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5083	u32 tmp;
5084
5085	if ((!adev->gfx.grbm_soft_reset) &&
5086	    (!adev->gfx.srbm_soft_reset))
5087		return 0;
5088
5089	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5090	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5091
5092	if (grbm_soft_reset || srbm_soft_reset) {
5093		tmp = RREG32(mmGMCON_DEBUG);
5094		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5095		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5096		WREG32(mmGMCON_DEBUG, tmp);
5097		udelay(50);
5098	}
5099
5100	if (grbm_soft_reset) {
5101		tmp = RREG32(mmGRBM_SOFT_RESET);
5102		tmp |= grbm_soft_reset;
5103		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5104		WREG32(mmGRBM_SOFT_RESET, tmp);
5105		tmp = RREG32(mmGRBM_SOFT_RESET);
5106
5107		udelay(50);
5108
5109		tmp &= ~grbm_soft_reset;
5110		WREG32(mmGRBM_SOFT_RESET, tmp);
5111		tmp = RREG32(mmGRBM_SOFT_RESET);
5112	}
5113
5114	if (srbm_soft_reset) {
5115		tmp = RREG32(mmSRBM_SOFT_RESET);
5116		tmp |= srbm_soft_reset;
5117		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5118		WREG32(mmSRBM_SOFT_RESET, tmp);
5119		tmp = RREG32(mmSRBM_SOFT_RESET);
5120
5121		udelay(50);
5122
5123		tmp &= ~srbm_soft_reset;
5124		WREG32(mmSRBM_SOFT_RESET, tmp);
5125		tmp = RREG32(mmSRBM_SOFT_RESET);
5126	}
5127
5128	if (grbm_soft_reset || srbm_soft_reset) {
5129		tmp = RREG32(mmGMCON_DEBUG);
5130		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5131		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5132		WREG32(mmGMCON_DEBUG, tmp);
5133	}
5134
5135	/* Wait a little for things to settle down */
5136	udelay(50);
5137
5138	return 0;
5139}
5140
5141static int gfx_v8_0_post_soft_reset(void *handle)
5142{
5143	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5144	u32 grbm_soft_reset = 0;
5145
5146	if ((!adev->gfx.grbm_soft_reset) &&
5147	    (!adev->gfx.srbm_soft_reset))
5148		return 0;
5149
5150	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5151
5152	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5153	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5154	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5155	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5156		int i;
5157
5158		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5159			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5160
5161			mutex_lock(&adev->srbm_mutex);
5162			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5163			gfx_v8_0_deactivate_hqd(adev, 2);
5164			vi_srbm_select(adev, 0, 0, 0, 0);
5165			mutex_unlock(&adev->srbm_mutex);
5166		}
5167		gfx_v8_0_kiq_resume(adev);
5168		gfx_v8_0_kcq_resume(adev);
5169	}
5170
5171	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5172	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5173		gfx_v8_0_cp_gfx_resume(adev);
5174
5175	gfx_v8_0_cp_test_all_rings(adev);
5176
5177	adev->gfx.rlc.funcs->start(adev);
5178
5179	return 0;
5180}
5181
5182/**
5183 * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5184 *
5185 * @adev: amdgpu_device pointer
5186 *
5187 * Fetches a GPU clock counter snapshot.
5188 * Returns the 64 bit clock counter snapshot.
5189 */
5190static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5191{
5192	uint64_t clock;
5193
5194	mutex_lock(&adev->gfx.gpu_clock_mutex);
5195	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5196	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5197		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5198	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5199	return clock;
5200}
5201
5202static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5203					  uint32_t vmid,
5204					  uint32_t gds_base, uint32_t gds_size,
5205					  uint32_t gws_base, uint32_t gws_size,
5206					  uint32_t oa_base, uint32_t oa_size)
5207{
5208	/* GDS Base */
5209	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5210	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5211				WRITE_DATA_DST_SEL(0)));
5212	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5213	amdgpu_ring_write(ring, 0);
5214	amdgpu_ring_write(ring, gds_base);
5215
5216	/* GDS Size */
5217	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5218	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5219				WRITE_DATA_DST_SEL(0)));
5220	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
5221	amdgpu_ring_write(ring, 0);
5222	amdgpu_ring_write(ring, gds_size);
5223
5224	/* GWS */
5225	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5226	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5227				WRITE_DATA_DST_SEL(0)));
5228	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
5229	amdgpu_ring_write(ring, 0);
5230	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5231
5232	/* OA */
5233	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5234	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5235				WRITE_DATA_DST_SEL(0)));
5236	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
5237	amdgpu_ring_write(ring, 0);
5238	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5239}
5240
5241static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5242{
5243	WREG32(mmSQ_IND_INDEX,
5244		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5245		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5246		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5247		(SQ_IND_INDEX__FORCE_READ_MASK));
5248	return RREG32(mmSQ_IND_DATA);
5249}
5250
5251static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5252			   uint32_t wave, uint32_t thread,
5253			   uint32_t regno, uint32_t num, uint32_t *out)
5254{
5255	WREG32(mmSQ_IND_INDEX,
5256		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5257		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5258		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5259		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5260		(SQ_IND_INDEX__FORCE_READ_MASK) |
5261		(SQ_IND_INDEX__AUTO_INCR_MASK));
5262	while (num--)
5263		*(out++) = RREG32(mmSQ_IND_DATA);
5264}
5265
5266static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5267{
5268	/* type 0 wave data */
5269	dst[(*no_fields)++] = 0;
5270	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5271	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5272	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5273	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5274	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5275	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5276	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5277	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5278	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5279	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5280	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5281	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5282	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5283	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5284	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5285	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5286	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5287	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
 
5288}
5289
5290static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
5291				     uint32_t wave, uint32_t start,
5292				     uint32_t size, uint32_t *dst)
5293{
5294	wave_read_regs(
5295		adev, simd, wave, 0,
5296		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5297}
5298
5299
5300static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5301	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5302	.select_se_sh = &gfx_v8_0_select_se_sh,
5303	.read_wave_data = &gfx_v8_0_read_wave_data,
5304	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5305	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5306};
5307
5308static int gfx_v8_0_early_init(void *handle)
5309{
5310	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5311
 
5312	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5313	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
 
5314	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5315	gfx_v8_0_set_ring_funcs(adev);
5316	gfx_v8_0_set_irq_funcs(adev);
5317	gfx_v8_0_set_gds_init(adev);
5318	gfx_v8_0_set_rlc_funcs(adev);
5319
5320	return 0;
5321}
5322
5323static int gfx_v8_0_late_init(void *handle)
5324{
5325	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5326	int r;
5327
5328	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5329	if (r)
5330		return r;
5331
5332	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5333	if (r)
5334		return r;
5335
5336	/* requires IBs so do in late init after IB pool is initialized */
5337	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5338	if (r)
5339		return r;
5340
5341	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5342	if (r) {
5343		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5344		return r;
5345	}
5346
5347	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5348	if (r) {
5349		DRM_ERROR(
5350			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5351			r);
5352		return r;
5353	}
5354
5355	return 0;
5356}
5357
5358static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5359						       bool enable)
5360{
5361	if (((adev->asic_type == CHIP_POLARIS11) ||
5362	    (adev->asic_type == CHIP_POLARIS12) ||
5363	    (adev->asic_type == CHIP_VEGAM)) &&
5364	    adev->powerplay.pp_funcs->set_powergating_by_smu)
5365		/* Send msg to SMU via Powerplay */
5366		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5367
5368	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5369}
5370
5371static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5372							bool enable)
5373{
5374	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5375}
5376
5377static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5378		bool enable)
5379{
5380	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5381}
5382
5383static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5384					  bool enable)
5385{
5386	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5387}
5388
5389static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5390						bool enable)
5391{
5392	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5393
5394	/* Read any GFX register to wake up GFX. */
5395	if (!enable)
5396		RREG32(mmDB_RENDER_CONTROL);
5397}
5398
5399static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5400					  bool enable)
5401{
5402	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5403		cz_enable_gfx_cg_power_gating(adev, true);
5404		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5405			cz_enable_gfx_pipeline_power_gating(adev, true);
5406	} else {
5407		cz_enable_gfx_cg_power_gating(adev, false);
5408		cz_enable_gfx_pipeline_power_gating(adev, false);
5409	}
5410}
5411
5412static int gfx_v8_0_set_powergating_state(void *handle,
5413					  enum amd_powergating_state state)
5414{
5415	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5416	bool enable = (state == AMD_PG_STATE_GATE);
5417
5418	if (amdgpu_sriov_vf(adev))
5419		return 0;
5420
5421	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5422				AMD_PG_SUPPORT_RLC_SMU_HS |
5423				AMD_PG_SUPPORT_CP |
5424				AMD_PG_SUPPORT_GFX_DMG))
5425		amdgpu_gfx_rlc_enter_safe_mode(adev);
5426	switch (adev->asic_type) {
5427	case CHIP_CARRIZO:
5428	case CHIP_STONEY:
5429
5430		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5431			cz_enable_sck_slow_down_on_power_up(adev, true);
5432			cz_enable_sck_slow_down_on_power_down(adev, true);
5433		} else {
5434			cz_enable_sck_slow_down_on_power_up(adev, false);
5435			cz_enable_sck_slow_down_on_power_down(adev, false);
5436		}
5437		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5438			cz_enable_cp_power_gating(adev, true);
5439		else
5440			cz_enable_cp_power_gating(adev, false);
5441
5442		cz_update_gfx_cg_power_gating(adev, enable);
5443
5444		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5445			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5446		else
5447			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5448
5449		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5450			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5451		else
5452			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5453		break;
5454	case CHIP_POLARIS11:
5455	case CHIP_POLARIS12:
5456	case CHIP_VEGAM:
5457		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5458			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5459		else
5460			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5461
5462		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5463			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5464		else
5465			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5466
5467		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5468			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5469		else
5470			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5471		break;
5472	default:
5473		break;
5474	}
5475	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5476				AMD_PG_SUPPORT_RLC_SMU_HS |
5477				AMD_PG_SUPPORT_CP |
5478				AMD_PG_SUPPORT_GFX_DMG))
5479		amdgpu_gfx_rlc_exit_safe_mode(adev);
5480	return 0;
5481}
5482
5483static void gfx_v8_0_get_clockgating_state(void *handle, u32 *flags)
5484{
5485	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5486	int data;
5487
5488	if (amdgpu_sriov_vf(adev))
5489		*flags = 0;
5490
5491	/* AMD_CG_SUPPORT_GFX_MGCG */
5492	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5493	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5494		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5495
5496	/* AMD_CG_SUPPORT_GFX_CGLG */
5497	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5498	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5499		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5500
5501	/* AMD_CG_SUPPORT_GFX_CGLS */
5502	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5503		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5504
5505	/* AMD_CG_SUPPORT_GFX_CGTS */
5506	data = RREG32(mmCGTS_SM_CTRL_REG);
5507	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5508		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5509
5510	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5511	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5512		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5513
5514	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5515	data = RREG32(mmRLC_MEM_SLP_CNTL);
5516	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5517		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5518
5519	/* AMD_CG_SUPPORT_GFX_CP_LS */
5520	data = RREG32(mmCP_MEM_SLP_CNTL);
5521	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5522		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5523}
5524
5525static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5526				     uint32_t reg_addr, uint32_t cmd)
5527{
5528	uint32_t data;
5529
5530	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5531
5532	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5533	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5534
5535	data = RREG32(mmRLC_SERDES_WR_CTRL);
5536	if (adev->asic_type == CHIP_STONEY)
5537		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5538			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5539			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5540			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5541			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5542			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5543			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5544			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5545			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5546	else
5547		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5548			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5549			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5550			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5551			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5552			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5553			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5554			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5555			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5556			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5557			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5558	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5559		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5560		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5561		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5562
5563	WREG32(mmRLC_SERDES_WR_CTRL, data);
5564}
5565
5566#define MSG_ENTER_RLC_SAFE_MODE     1
5567#define MSG_EXIT_RLC_SAFE_MODE      0
5568#define RLC_GPR_REG2__REQ_MASK 0x00000001
5569#define RLC_GPR_REG2__REQ__SHIFT 0
5570#define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5571#define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5572
5573static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5574{
5575	uint32_t rlc_setting;
5576
5577	rlc_setting = RREG32(mmRLC_CNTL);
5578	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5579		return false;
5580
5581	return true;
5582}
5583
5584static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev)
5585{
5586	uint32_t data;
5587	unsigned i;
5588	data = RREG32(mmRLC_CNTL);
5589	data |= RLC_SAFE_MODE__CMD_MASK;
5590	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5591	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5592	WREG32(mmRLC_SAFE_MODE, data);
5593
5594	/* wait for RLC_SAFE_MODE */
5595	for (i = 0; i < adev->usec_timeout; i++) {
5596		if ((RREG32(mmRLC_GPM_STAT) &
5597		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5598		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5599		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5600		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5601			break;
5602		udelay(1);
5603	}
5604	for (i = 0; i < adev->usec_timeout; i++) {
5605		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5606			break;
5607		udelay(1);
5608	}
5609}
5610
5611static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev)
5612{
5613	uint32_t data;
5614	unsigned i;
5615
5616	data = RREG32(mmRLC_CNTL);
5617	data |= RLC_SAFE_MODE__CMD_MASK;
5618	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5619	WREG32(mmRLC_SAFE_MODE, data);
5620
5621	for (i = 0; i < adev->usec_timeout; i++) {
5622		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5623			break;
5624		udelay(1);
5625	}
5626}
5627
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5628static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5629	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5630	.set_safe_mode = gfx_v8_0_set_safe_mode,
5631	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5632	.init = gfx_v8_0_rlc_init,
5633	.get_csb_size = gfx_v8_0_get_csb_size,
5634	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5635	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5636	.resume = gfx_v8_0_rlc_resume,
5637	.stop = gfx_v8_0_rlc_stop,
5638	.reset = gfx_v8_0_rlc_reset,
5639	.start = gfx_v8_0_rlc_start
 
5640};
5641
5642static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5643						      bool enable)
5644{
5645	uint32_t temp, data;
5646
5647	amdgpu_gfx_rlc_enter_safe_mode(adev);
5648
5649	/* It is disabled by HW by default */
5650	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5651		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5652			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5653				/* 1 - RLC memory Light sleep */
5654				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5655
5656			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5657				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5658		}
5659
5660		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5661		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5662		if (adev->flags & AMD_IS_APU)
5663			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5664				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5665				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5666		else
5667			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5668				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5669				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5670				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5671
5672		if (temp != data)
5673			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5674
5675		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5676		gfx_v8_0_wait_for_rlc_serdes(adev);
5677
5678		/* 5 - clear mgcg override */
5679		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5680
5681		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5682			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5683			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5684			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5685			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5686			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5687			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5688			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5689			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5690				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5691			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5692			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5693			if (temp != data)
5694				WREG32(mmCGTS_SM_CTRL_REG, data);
5695		}
5696		udelay(50);
5697
5698		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5699		gfx_v8_0_wait_for_rlc_serdes(adev);
5700	} else {
5701		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5702		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5703		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5704				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5705				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5706				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5707		if (temp != data)
5708			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5709
5710		/* 2 - disable MGLS in RLC */
5711		data = RREG32(mmRLC_MEM_SLP_CNTL);
5712		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5713			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5714			WREG32(mmRLC_MEM_SLP_CNTL, data);
5715		}
5716
5717		/* 3 - disable MGLS in CP */
5718		data = RREG32(mmCP_MEM_SLP_CNTL);
5719		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5720			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5721			WREG32(mmCP_MEM_SLP_CNTL, data);
5722		}
5723
5724		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5725		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5726		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5727				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5728		if (temp != data)
5729			WREG32(mmCGTS_SM_CTRL_REG, data);
5730
5731		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5732		gfx_v8_0_wait_for_rlc_serdes(adev);
5733
5734		/* 6 - set mgcg override */
5735		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5736
5737		udelay(50);
5738
5739		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5740		gfx_v8_0_wait_for_rlc_serdes(adev);
5741	}
5742
5743	amdgpu_gfx_rlc_exit_safe_mode(adev);
5744}
5745
5746static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5747						      bool enable)
5748{
5749	uint32_t temp, temp1, data, data1;
5750
5751	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5752
5753	amdgpu_gfx_rlc_enter_safe_mode(adev);
5754
5755	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5756		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5757		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5758		if (temp1 != data1)
5759			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5760
5761		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5762		gfx_v8_0_wait_for_rlc_serdes(adev);
5763
5764		/* 2 - clear cgcg override */
5765		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5766
5767		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5768		gfx_v8_0_wait_for_rlc_serdes(adev);
5769
5770		/* 3 - write cmd to set CGLS */
5771		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5772
5773		/* 4 - enable cgcg */
5774		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5775
5776		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5777			/* enable cgls*/
5778			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5779
5780			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5781			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5782
5783			if (temp1 != data1)
5784				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5785		} else {
5786			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5787		}
5788
5789		if (temp != data)
5790			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5791
5792		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5793		 * Cmp_busy/GFX_Idle interrupts
5794		 */
5795		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5796	} else {
5797		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5798		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5799
5800		/* TEST CGCG */
5801		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5802		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5803				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5804		if (temp1 != data1)
5805			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5806
5807		/* read gfx register to wake up cgcg */
5808		RREG32(mmCB_CGTT_SCLK_CTRL);
5809		RREG32(mmCB_CGTT_SCLK_CTRL);
5810		RREG32(mmCB_CGTT_SCLK_CTRL);
5811		RREG32(mmCB_CGTT_SCLK_CTRL);
5812
5813		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5814		gfx_v8_0_wait_for_rlc_serdes(adev);
5815
5816		/* write cmd to Set CGCG Overrride */
5817		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5818
5819		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5820		gfx_v8_0_wait_for_rlc_serdes(adev);
5821
5822		/* write cmd to Clear CGLS */
5823		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5824
5825		/* disable cgcg, cgls should be disabled too. */
5826		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5827			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5828		if (temp != data)
5829			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5830		/* enable interrupts again for PG */
5831		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5832	}
5833
5834	gfx_v8_0_wait_for_rlc_serdes(adev);
5835
5836	amdgpu_gfx_rlc_exit_safe_mode(adev);
5837}
5838static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5839					    bool enable)
5840{
5841	if (enable) {
5842		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5843		 * ===  MGCG + MGLS + TS(CG/LS) ===
5844		 */
5845		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5846		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5847	} else {
5848		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5849		 * ===  CGCG + CGLS ===
5850		 */
5851		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5852		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5853	}
5854	return 0;
5855}
5856
5857static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5858					  enum amd_clockgating_state state)
5859{
5860	uint32_t msg_id, pp_state = 0;
5861	uint32_t pp_support_state = 0;
5862
5863	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5864		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5865			pp_support_state = PP_STATE_SUPPORT_LS;
5866			pp_state = PP_STATE_LS;
5867		}
5868		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5869			pp_support_state |= PP_STATE_SUPPORT_CG;
5870			pp_state |= PP_STATE_CG;
5871		}
5872		if (state == AMD_CG_STATE_UNGATE)
5873			pp_state = 0;
5874
5875		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5876				PP_BLOCK_GFX_CG,
5877				pp_support_state,
5878				pp_state);
5879		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5880			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5881	}
5882
5883	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5884		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5885			pp_support_state = PP_STATE_SUPPORT_LS;
5886			pp_state = PP_STATE_LS;
5887		}
5888
5889		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5890			pp_support_state |= PP_STATE_SUPPORT_CG;
5891			pp_state |= PP_STATE_CG;
5892		}
5893
5894		if (state == AMD_CG_STATE_UNGATE)
5895			pp_state = 0;
5896
5897		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5898				PP_BLOCK_GFX_MG,
5899				pp_support_state,
5900				pp_state);
5901		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5902			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5903	}
5904
5905	return 0;
5906}
5907
5908static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5909					  enum amd_clockgating_state state)
5910{
5911
5912	uint32_t msg_id, pp_state = 0;
5913	uint32_t pp_support_state = 0;
5914
5915	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5916		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5917			pp_support_state = PP_STATE_SUPPORT_LS;
5918			pp_state = PP_STATE_LS;
5919		}
5920		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5921			pp_support_state |= PP_STATE_SUPPORT_CG;
5922			pp_state |= PP_STATE_CG;
5923		}
5924		if (state == AMD_CG_STATE_UNGATE)
5925			pp_state = 0;
5926
5927		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5928				PP_BLOCK_GFX_CG,
5929				pp_support_state,
5930				pp_state);
5931		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5932			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5933	}
5934
5935	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5936		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5937			pp_support_state = PP_STATE_SUPPORT_LS;
5938			pp_state = PP_STATE_LS;
5939		}
5940		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5941			pp_support_state |= PP_STATE_SUPPORT_CG;
5942			pp_state |= PP_STATE_CG;
5943		}
5944		if (state == AMD_CG_STATE_UNGATE)
5945			pp_state = 0;
5946
5947		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5948				PP_BLOCK_GFX_3D,
5949				pp_support_state,
5950				pp_state);
5951		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5952			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5953	}
5954
5955	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5956		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5957			pp_support_state = PP_STATE_SUPPORT_LS;
5958			pp_state = PP_STATE_LS;
5959		}
5960
5961		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5962			pp_support_state |= PP_STATE_SUPPORT_CG;
5963			pp_state |= PP_STATE_CG;
5964		}
5965
5966		if (state == AMD_CG_STATE_UNGATE)
5967			pp_state = 0;
5968
5969		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5970				PP_BLOCK_GFX_MG,
5971				pp_support_state,
5972				pp_state);
5973		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5974			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5975	}
5976
5977	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
5978		pp_support_state = PP_STATE_SUPPORT_LS;
5979
5980		if (state == AMD_CG_STATE_UNGATE)
5981			pp_state = 0;
5982		else
5983			pp_state = PP_STATE_LS;
5984
5985		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5986				PP_BLOCK_GFX_RLC,
5987				pp_support_state,
5988				pp_state);
5989		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5990			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5991	}
5992
5993	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
5994		pp_support_state = PP_STATE_SUPPORT_LS;
5995
5996		if (state == AMD_CG_STATE_UNGATE)
5997			pp_state = 0;
5998		else
5999			pp_state = PP_STATE_LS;
6000		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
6001			PP_BLOCK_GFX_CP,
6002			pp_support_state,
6003			pp_state);
6004		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
6005			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
6006	}
6007
6008	return 0;
6009}
6010
6011static int gfx_v8_0_set_clockgating_state(void *handle,
6012					  enum amd_clockgating_state state)
6013{
6014	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
6015
6016	if (amdgpu_sriov_vf(adev))
6017		return 0;
6018
6019	switch (adev->asic_type) {
6020	case CHIP_FIJI:
6021	case CHIP_CARRIZO:
6022	case CHIP_STONEY:
6023		gfx_v8_0_update_gfx_clock_gating(adev,
6024						 state == AMD_CG_STATE_GATE);
6025		break;
6026	case CHIP_TONGA:
6027		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
6028		break;
6029	case CHIP_POLARIS10:
6030	case CHIP_POLARIS11:
6031	case CHIP_POLARIS12:
6032	case CHIP_VEGAM:
6033		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
6034		break;
6035	default:
6036		break;
6037	}
6038	return 0;
6039}
6040
6041static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6042{
6043	return ring->adev->wb.wb[ring->rptr_offs];
6044}
6045
6046static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6047{
6048	struct amdgpu_device *adev = ring->adev;
6049
6050	if (ring->use_doorbell)
6051		/* XXX check if swapping is necessary on BE */
6052		return ring->adev->wb.wb[ring->wptr_offs];
6053	else
6054		return RREG32(mmCP_RB0_WPTR);
6055}
6056
6057static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6058{
6059	struct amdgpu_device *adev = ring->adev;
6060
6061	if (ring->use_doorbell) {
6062		/* XXX check if swapping is necessary on BE */
6063		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
6064		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6065	} else {
6066		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6067		(void)RREG32(mmCP_RB0_WPTR);
6068	}
6069}
6070
6071static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6072{
6073	u32 ref_and_mask, reg_mem_engine;
6074
6075	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6076	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6077		switch (ring->me) {
6078		case 1:
6079			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6080			break;
6081		case 2:
6082			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6083			break;
6084		default:
6085			return;
6086		}
6087		reg_mem_engine = 0;
6088	} else {
6089		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6090		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6091	}
6092
6093	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6094	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6095				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6096				 reg_mem_engine));
6097	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6098	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6099	amdgpu_ring_write(ring, ref_and_mask);
6100	amdgpu_ring_write(ring, ref_and_mask);
6101	amdgpu_ring_write(ring, 0x20); /* poll interval */
6102}
6103
6104static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6105{
6106	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6107	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6108		EVENT_INDEX(4));
6109
6110	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6111	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6112		EVENT_INDEX(0));
6113}
6114
6115static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6116					struct amdgpu_job *job,
6117					struct amdgpu_ib *ib,
6118					uint32_t flags)
6119{
6120	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6121	u32 header, control = 0;
6122
6123	if (ib->flags & AMDGPU_IB_FLAG_CE)
6124		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6125	else
6126		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6127
6128	control |= ib->length_dw | (vmid << 24);
6129
6130	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6131		control |= INDIRECT_BUFFER_PRE_ENB(1);
6132
6133		if (!(ib->flags & AMDGPU_IB_FLAG_CE))
6134			gfx_v8_0_ring_emit_de_meta(ring);
6135	}
6136
6137	amdgpu_ring_write(ring, header);
6138	amdgpu_ring_write(ring,
6139#ifdef __BIG_ENDIAN
6140			  (2 << 0) |
6141#endif
6142			  (ib->gpu_addr & 0xFFFFFFFC));
6143	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6144	amdgpu_ring_write(ring, control);
6145}
6146
6147static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6148					  struct amdgpu_job *job,
6149					  struct amdgpu_ib *ib,
6150					  uint32_t flags)
6151{
6152	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6153	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6154
6155	/* Currently, there is a high possibility to get wave ID mismatch
6156	 * between ME and GDS, leading to a hw deadlock, because ME generates
6157	 * different wave IDs than the GDS expects. This situation happens
6158	 * randomly when at least 5 compute pipes use GDS ordered append.
6159	 * The wave IDs generated by ME are also wrong after suspend/resume.
6160	 * Those are probably bugs somewhere else in the kernel driver.
6161	 *
6162	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6163	 * GDS to 0 for this ring (me/pipe).
6164	 */
6165	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6166		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6167		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6168		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6169	}
6170
6171	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6172	amdgpu_ring_write(ring,
6173#ifdef __BIG_ENDIAN
6174				(2 << 0) |
6175#endif
6176				(ib->gpu_addr & 0xFFFFFFFC));
6177	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6178	amdgpu_ring_write(ring, control);
6179}
6180
6181static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6182					 u64 seq, unsigned flags)
6183{
6184	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6185	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
 
6186
6187	/* EVENT_WRITE_EOP - flush caches, send int */
 
 
6188	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6189	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6190				 EOP_TC_ACTION_EN |
6191				 EOP_TC_WB_ACTION_EN |
6192				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6193				 EVENT_INDEX(5)));
6194	amdgpu_ring_write(ring, addr & 0xfffffffc);
6195	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6196			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6197	amdgpu_ring_write(ring, lower_32_bits(seq));
6198	amdgpu_ring_write(ring, upper_32_bits(seq));
6199
6200}
6201
6202static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6203{
6204	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6205	uint32_t seq = ring->fence_drv.sync_seq;
6206	uint64_t addr = ring->fence_drv.gpu_addr;
6207
6208	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6209	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6210				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6211				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6212	amdgpu_ring_write(ring, addr & 0xfffffffc);
6213	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6214	amdgpu_ring_write(ring, seq);
6215	amdgpu_ring_write(ring, 0xffffffff);
6216	amdgpu_ring_write(ring, 4); /* poll interval */
6217}
6218
6219static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6220					unsigned vmid, uint64_t pd_addr)
6221{
6222	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6223
6224	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6225
6226	/* wait for the invalidate to complete */
6227	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6228	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6229				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6230				 WAIT_REG_MEM_ENGINE(0))); /* me */
6231	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6232	amdgpu_ring_write(ring, 0);
6233	amdgpu_ring_write(ring, 0); /* ref */
6234	amdgpu_ring_write(ring, 0); /* mask */
6235	amdgpu_ring_write(ring, 0x20); /* poll interval */
6236
6237	/* compute doesn't have PFP */
6238	if (usepfp) {
6239		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6240		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6241		amdgpu_ring_write(ring, 0x0);
6242	}
6243}
6244
6245static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6246{
6247	return ring->adev->wb.wb[ring->wptr_offs];
6248}
6249
6250static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6251{
6252	struct amdgpu_device *adev = ring->adev;
6253
6254	/* XXX check if swapping is necessary on BE */
6255	adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
6256	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6257}
6258
6259static void gfx_v8_0_ring_set_pipe_percent(struct amdgpu_ring *ring,
6260					   bool acquire)
6261{
6262	struct amdgpu_device *adev = ring->adev;
6263	int pipe_num, tmp, reg;
6264	int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1;
6265
6266	pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe;
6267
6268	/* first me only has 2 entries, GFX and HP3D */
6269	if (ring->me > 0)
6270		pipe_num -= 2;
6271
6272	reg = mmSPI_WCL_PIPE_PERCENT_GFX + pipe_num;
6273	tmp = RREG32(reg);
6274	tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent);
6275	WREG32(reg, tmp);
6276}
6277
6278static void gfx_v8_0_pipe_reserve_resources(struct amdgpu_device *adev,
6279					    struct amdgpu_ring *ring,
6280					    bool acquire)
6281{
6282	int i, pipe;
6283	bool reserve;
6284	struct amdgpu_ring *iring;
6285
6286	mutex_lock(&adev->gfx.pipe_reserve_mutex);
6287	pipe = amdgpu_gfx_mec_queue_to_bit(adev, ring->me, ring->pipe, 0);
6288	if (acquire)
6289		set_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6290	else
6291		clear_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6292
6293	if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) {
6294		/* Clear all reservations - everyone reacquires all resources */
6295		for (i = 0; i < adev->gfx.num_gfx_rings; ++i)
6296			gfx_v8_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i],
6297						       true);
6298
6299		for (i = 0; i < adev->gfx.num_compute_rings; ++i)
6300			gfx_v8_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i],
6301						       true);
6302	} else {
6303		/* Lower all pipes without a current reservation */
6304		for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
6305			iring = &adev->gfx.gfx_ring[i];
6306			pipe = amdgpu_gfx_mec_queue_to_bit(adev,
6307							   iring->me,
6308							   iring->pipe,
6309							   0);
6310			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6311			gfx_v8_0_ring_set_pipe_percent(iring, reserve);
6312		}
6313
6314		for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
6315			iring = &adev->gfx.compute_ring[i];
6316			pipe = amdgpu_gfx_mec_queue_to_bit(adev,
6317							   iring->me,
6318							   iring->pipe,
6319							   0);
6320			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6321			gfx_v8_0_ring_set_pipe_percent(iring, reserve);
6322		}
6323	}
6324
6325	mutex_unlock(&adev->gfx.pipe_reserve_mutex);
6326}
6327
6328static void gfx_v8_0_hqd_set_priority(struct amdgpu_device *adev,
6329				      struct amdgpu_ring *ring,
6330				      bool acquire)
6331{
6332	uint32_t pipe_priority = acquire ? 0x2 : 0x0;
6333	uint32_t queue_priority = acquire ? 0xf : 0x0;
6334
6335	mutex_lock(&adev->srbm_mutex);
6336	vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
6337
6338	WREG32(mmCP_HQD_PIPE_PRIORITY, pipe_priority);
6339	WREG32(mmCP_HQD_QUEUE_PRIORITY, queue_priority);
6340
6341	vi_srbm_select(adev, 0, 0, 0, 0);
6342	mutex_unlock(&adev->srbm_mutex);
6343}
6344static void gfx_v8_0_ring_set_priority_compute(struct amdgpu_ring *ring,
6345					       enum drm_sched_priority priority)
6346{
6347	struct amdgpu_device *adev = ring->adev;
6348	bool acquire = priority == DRM_SCHED_PRIORITY_HIGH_HW;
6349
6350	if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
6351		return;
6352
6353	gfx_v8_0_hqd_set_priority(adev, ring, acquire);
6354	gfx_v8_0_pipe_reserve_resources(adev, ring, acquire);
6355}
6356
6357static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6358					     u64 addr, u64 seq,
6359					     unsigned flags)
6360{
6361	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6362	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6363
6364	/* RELEASE_MEM - flush caches, send int */
6365	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6366	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6367				 EOP_TC_ACTION_EN |
6368				 EOP_TC_WB_ACTION_EN |
6369				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6370				 EVENT_INDEX(5)));
6371	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6372	amdgpu_ring_write(ring, addr & 0xfffffffc);
6373	amdgpu_ring_write(ring, upper_32_bits(addr));
6374	amdgpu_ring_write(ring, lower_32_bits(seq));
6375	amdgpu_ring_write(ring, upper_32_bits(seq));
6376}
6377
6378static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6379					 u64 seq, unsigned int flags)
6380{
6381	/* we only allocate 32bit for each seq wb address */
6382	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6383
6384	/* write fence seq to the "addr" */
6385	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6386	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6387				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
6388	amdgpu_ring_write(ring, lower_32_bits(addr));
6389	amdgpu_ring_write(ring, upper_32_bits(addr));
6390	amdgpu_ring_write(ring, lower_32_bits(seq));
6391
6392	if (flags & AMDGPU_FENCE_FLAG_INT) {
6393		/* set register to trigger INT */
6394		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6395		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6396					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6397		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6398		amdgpu_ring_write(ring, 0);
6399		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6400	}
6401}
6402
6403static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6404{
6405	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6406	amdgpu_ring_write(ring, 0);
6407}
6408
6409static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6410{
6411	uint32_t dw2 = 0;
6412
6413	if (amdgpu_sriov_vf(ring->adev))
6414		gfx_v8_0_ring_emit_ce_meta(ring);
6415
6416	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6417	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6418		gfx_v8_0_ring_emit_vgt_flush(ring);
6419		/* set load_global_config & load_global_uconfig */
6420		dw2 |= 0x8001;
6421		/* set load_cs_sh_regs */
6422		dw2 |= 0x01000000;
6423		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6424		dw2 |= 0x10002;
6425
6426		/* set load_ce_ram if preamble presented */
6427		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6428			dw2 |= 0x10000000;
6429	} else {
6430		/* still load_ce_ram if this is the first time preamble presented
6431		 * although there is no context switch happens.
6432		 */
6433		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6434			dw2 |= 0x10000000;
6435	}
6436
6437	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6438	amdgpu_ring_write(ring, dw2);
6439	amdgpu_ring_write(ring, 0);
6440}
6441
6442static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
 
6443{
6444	unsigned ret;
6445
6446	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6447	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
6448	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
6449	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
 
6450	ret = ring->wptr & ring->buf_mask;
6451	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
 
6452	return ret;
6453}
6454
6455static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
6456{
6457	unsigned cur;
6458
6459	BUG_ON(offset > ring->buf_mask);
6460	BUG_ON(ring->ring[offset] != 0x55aa55aa);
6461
6462	cur = (ring->wptr & ring->buf_mask) - 1;
6463	if (likely(cur > offset))
6464		ring->ring[offset] = cur - offset;
6465	else
6466		ring->ring[offset] = (ring->ring_size >> 2) - offset + cur;
6467}
6468
6469static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg)
6470{
6471	struct amdgpu_device *adev = ring->adev;
6472
6473	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6474	amdgpu_ring_write(ring, 0 |	/* src: register*/
6475				(5 << 8) |	/* dst: memory */
6476				(1 << 20));	/* write confirm */
6477	amdgpu_ring_write(ring, reg);
6478	amdgpu_ring_write(ring, 0);
6479	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6480				adev->virt.reg_val_offs * 4));
6481	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6482				adev->virt.reg_val_offs * 4));
6483}
6484
6485static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6486				  uint32_t val)
6487{
6488	uint32_t cmd;
6489
6490	switch (ring->funcs->type) {
6491	case AMDGPU_RING_TYPE_GFX:
6492		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6493		break;
6494	case AMDGPU_RING_TYPE_KIQ:
6495		cmd = 1 << 16; /* no inc addr */
6496		break;
6497	default:
6498		cmd = WR_CONFIRM;
6499		break;
6500	}
6501
6502	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6503	amdgpu_ring_write(ring, cmd);
6504	amdgpu_ring_write(ring, reg);
6505	amdgpu_ring_write(ring, 0);
6506	amdgpu_ring_write(ring, val);
6507}
6508
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6509static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6510{
6511	struct amdgpu_device *adev = ring->adev;
6512	uint32_t value = 0;
6513
6514	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6515	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6516	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6517	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6518	WREG32(mmSQ_CMD, value);
6519}
6520
6521static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6522						 enum amdgpu_interrupt_state state)
6523{
6524	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6525		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6526}
6527
6528static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6529						     int me, int pipe,
6530						     enum amdgpu_interrupt_state state)
6531{
6532	u32 mec_int_cntl, mec_int_cntl_reg;
6533
6534	/*
6535	 * amdgpu controls only the first MEC. That's why this function only
6536	 * handles the setting of interrupts for this specific MEC. All other
6537	 * pipes' interrupts are set by amdkfd.
6538	 */
6539
6540	if (me == 1) {
6541		switch (pipe) {
6542		case 0:
6543			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6544			break;
6545		case 1:
6546			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6547			break;
6548		case 2:
6549			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6550			break;
6551		case 3:
6552			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6553			break;
6554		default:
6555			DRM_DEBUG("invalid pipe %d\n", pipe);
6556			return;
6557		}
6558	} else {
6559		DRM_DEBUG("invalid me %d\n", me);
6560		return;
6561	}
6562
6563	switch (state) {
6564	case AMDGPU_IRQ_STATE_DISABLE:
6565		mec_int_cntl = RREG32(mec_int_cntl_reg);
6566		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6567		WREG32(mec_int_cntl_reg, mec_int_cntl);
6568		break;
6569	case AMDGPU_IRQ_STATE_ENABLE:
6570		mec_int_cntl = RREG32(mec_int_cntl_reg);
6571		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6572		WREG32(mec_int_cntl_reg, mec_int_cntl);
6573		break;
6574	default:
6575		break;
6576	}
6577}
6578
6579static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6580					     struct amdgpu_irq_src *source,
6581					     unsigned type,
6582					     enum amdgpu_interrupt_state state)
6583{
6584	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6585		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6586
6587	return 0;
6588}
6589
6590static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6591					      struct amdgpu_irq_src *source,
6592					      unsigned type,
6593					      enum amdgpu_interrupt_state state)
6594{
6595	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6596		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6597
6598	return 0;
6599}
6600
6601static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6602					    struct amdgpu_irq_src *src,
6603					    unsigned type,
6604					    enum amdgpu_interrupt_state state)
6605{
6606	switch (type) {
6607	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6608		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6609		break;
6610	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6611		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6612		break;
6613	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6614		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6615		break;
6616	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6617		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6618		break;
6619	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6620		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6621		break;
6622	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6623		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6624		break;
6625	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6626		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6627		break;
6628	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6629		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6630		break;
6631	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6632		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6633		break;
6634	default:
6635		break;
6636	}
6637	return 0;
6638}
6639
6640static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6641					 struct amdgpu_irq_src *source,
6642					 unsigned int type,
6643					 enum amdgpu_interrupt_state state)
6644{
6645	int enable_flag;
6646
6647	switch (state) {
6648	case AMDGPU_IRQ_STATE_DISABLE:
6649		enable_flag = 0;
6650		break;
6651
6652	case AMDGPU_IRQ_STATE_ENABLE:
6653		enable_flag = 1;
6654		break;
6655
6656	default:
6657		return -EINVAL;
6658	}
6659
6660	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6661	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6662	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6663	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6664	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6665	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6666		     enable_flag);
6667	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6668		     enable_flag);
6669	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6670		     enable_flag);
6671	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6672		     enable_flag);
6673	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6674		     enable_flag);
6675	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6676		     enable_flag);
6677	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6678		     enable_flag);
6679	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6680		     enable_flag);
6681
6682	return 0;
6683}
6684
6685static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6686				     struct amdgpu_irq_src *source,
6687				     unsigned int type,
6688				     enum amdgpu_interrupt_state state)
6689{
6690	int enable_flag;
6691
6692	switch (state) {
6693	case AMDGPU_IRQ_STATE_DISABLE:
6694		enable_flag = 1;
6695		break;
6696
6697	case AMDGPU_IRQ_STATE_ENABLE:
6698		enable_flag = 0;
6699		break;
6700
6701	default:
6702		return -EINVAL;
6703	}
6704
6705	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6706		     enable_flag);
6707
6708	return 0;
6709}
6710
6711static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6712			    struct amdgpu_irq_src *source,
6713			    struct amdgpu_iv_entry *entry)
6714{
6715	int i;
6716	u8 me_id, pipe_id, queue_id;
6717	struct amdgpu_ring *ring;
6718
6719	DRM_DEBUG("IH: CP EOP\n");
6720	me_id = (entry->ring_id & 0x0c) >> 2;
6721	pipe_id = (entry->ring_id & 0x03) >> 0;
6722	queue_id = (entry->ring_id & 0x70) >> 4;
6723
6724	switch (me_id) {
6725	case 0:
6726		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6727		break;
6728	case 1:
6729	case 2:
6730		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6731			ring = &adev->gfx.compute_ring[i];
6732			/* Per-queue interrupt is supported for MEC starting from VI.
6733			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6734			  */
6735			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6736				amdgpu_fence_process(ring);
6737		}
6738		break;
6739	}
6740	return 0;
6741}
6742
6743static void gfx_v8_0_fault(struct amdgpu_device *adev,
6744			   struct amdgpu_iv_entry *entry)
6745{
6746	u8 me_id, pipe_id, queue_id;
6747	struct amdgpu_ring *ring;
6748	int i;
6749
6750	me_id = (entry->ring_id & 0x0c) >> 2;
6751	pipe_id = (entry->ring_id & 0x03) >> 0;
6752	queue_id = (entry->ring_id & 0x70) >> 4;
6753
6754	switch (me_id) {
6755	case 0:
6756		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6757		break;
6758	case 1:
6759	case 2:
6760		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6761			ring = &adev->gfx.compute_ring[i];
6762			if (ring->me == me_id && ring->pipe == pipe_id &&
6763			    ring->queue == queue_id)
6764				drm_sched_fault(&ring->sched);
6765		}
6766		break;
6767	}
6768}
6769
6770static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6771				 struct amdgpu_irq_src *source,
6772				 struct amdgpu_iv_entry *entry)
6773{
6774	DRM_ERROR("Illegal register access in command stream\n");
6775	gfx_v8_0_fault(adev, entry);
6776	return 0;
6777}
6778
6779static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6780				  struct amdgpu_irq_src *source,
6781				  struct amdgpu_iv_entry *entry)
6782{
6783	DRM_ERROR("Illegal instruction in command stream\n");
6784	gfx_v8_0_fault(adev, entry);
6785	return 0;
6786}
6787
6788static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6789				     struct amdgpu_irq_src *source,
6790				     struct amdgpu_iv_entry *entry)
6791{
6792	DRM_ERROR("CP EDC/ECC error detected.");
6793	return 0;
6794}
6795
6796static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data)
 
6797{
6798	u32 enc, se_id, sh_id, cu_id;
6799	char type[20];
6800	int sq_edc_source = -1;
6801
6802	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6803	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6804
6805	switch (enc) {
6806		case 0:
6807			DRM_INFO("SQ general purpose intr detected:"
6808					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6809					"host_cmd_overflow %d, cmd_timestamp %d,"
6810					"reg_timestamp %d, thread_trace_buff_full %d,"
6811					"wlt %d, thread_trace %d.\n",
6812					se_id,
6813					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6814					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6815					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6816					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6817					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6818					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6819					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6820					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6821					);
6822			break;
6823		case 1:
6824		case 2:
6825
6826			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6827			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6828
6829			/*
6830			 * This function can be called either directly from ISR
6831			 * or from BH in which case we can access SQ_EDC_INFO
6832			 * instance
6833			 */
6834			if (in_task()) {
6835				mutex_lock(&adev->grbm_idx_mutex);
6836				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id);
6837
6838				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6839
6840				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6841				mutex_unlock(&adev->grbm_idx_mutex);
6842			}
6843
6844			if (enc == 1)
6845				sprintf(type, "instruction intr");
6846			else
6847				sprintf(type, "EDC/ECC error");
6848
6849			DRM_INFO(
6850				"SQ %s detected: "
6851					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6852					"trap %s, sq_ed_info.source %s.\n",
6853					type, se_id, sh_id, cu_id,
6854					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6855					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6856					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6857					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6858					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6859				);
6860			break;
6861		default:
6862			DRM_ERROR("SQ invalid encoding type\n.");
6863	}
6864}
6865
6866static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6867{
6868
6869	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6870	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6871
6872	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data);
6873}
6874
6875static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6876			   struct amdgpu_irq_src *source,
6877			   struct amdgpu_iv_entry *entry)
6878{
6879	unsigned ih_data = entry->src_data[0];
6880
6881	/*
6882	 * Try to submit work so SQ_EDC_INFO can be accessed from
6883	 * BH. If previous work submission hasn't finished yet
6884	 * just print whatever info is possible directly from the ISR.
6885	 */
6886	if (work_pending(&adev->gfx.sq_work.work)) {
6887		gfx_v8_0_parse_sq_irq(adev, ih_data);
6888	} else {
6889		adev->gfx.sq_work.ih_data = ih_data;
6890		schedule_work(&adev->gfx.sq_work.work);
6891	}
6892
6893	return 0;
6894}
6895
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6896static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6897	.name = "gfx_v8_0",
6898	.early_init = gfx_v8_0_early_init,
6899	.late_init = gfx_v8_0_late_init,
6900	.sw_init = gfx_v8_0_sw_init,
6901	.sw_fini = gfx_v8_0_sw_fini,
6902	.hw_init = gfx_v8_0_hw_init,
6903	.hw_fini = gfx_v8_0_hw_fini,
6904	.suspend = gfx_v8_0_suspend,
6905	.resume = gfx_v8_0_resume,
6906	.is_idle = gfx_v8_0_is_idle,
6907	.wait_for_idle = gfx_v8_0_wait_for_idle,
6908	.check_soft_reset = gfx_v8_0_check_soft_reset,
6909	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6910	.soft_reset = gfx_v8_0_soft_reset,
6911	.post_soft_reset = gfx_v8_0_post_soft_reset,
6912	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6913	.set_powergating_state = gfx_v8_0_set_powergating_state,
6914	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6915};
6916
6917static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6918	.type = AMDGPU_RING_TYPE_GFX,
6919	.align_mask = 0xff,
6920	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6921	.support_64bit_ptrs = false,
6922	.get_rptr = gfx_v8_0_ring_get_rptr,
6923	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6924	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6925	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6926		5 +  /* COND_EXEC */
6927		7 +  /* PIPELINE_SYNC */
6928		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6929		8 +  /* FENCE for VM_FLUSH */
6930		20 + /* GDS switch */
6931		4 + /* double SWITCH_BUFFER,
6932		       the first COND_EXEC jump to the place just
6933			   prior to this double SWITCH_BUFFER  */
6934		5 + /* COND_EXEC */
6935		7 +	 /*	HDP_flush */
6936		4 +	 /*	VGT_flush */
6937		14 + /*	CE_META */
6938		31 + /*	DE_META */
6939		3 + /* CNTX_CTRL */
6940		5 + /* HDP_INVL */
6941		8 + 8 + /* FENCE x2 */
6942		2, /* SWITCH_BUFFER */
 
6943	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6944	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6945	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6946	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6947	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6948	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6949	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6950	.test_ring = gfx_v8_0_ring_test_ring,
6951	.test_ib = gfx_v8_0_ring_test_ib,
6952	.insert_nop = amdgpu_ring_insert_nop,
6953	.pad_ib = amdgpu_ring_generic_pad_ib,
6954	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6955	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
6956	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
6957	.patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec,
6958	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6959	.soft_recovery = gfx_v8_0_ring_soft_recovery,
 
 
6960};
6961
6962static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
6963	.type = AMDGPU_RING_TYPE_COMPUTE,
6964	.align_mask = 0xff,
6965	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6966	.support_64bit_ptrs = false,
6967	.get_rptr = gfx_v8_0_ring_get_rptr,
6968	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6969	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6970	.emit_frame_size =
6971		20 + /* gfx_v8_0_ring_emit_gds_switch */
6972		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6973		5 + /* hdp_invalidate */
6974		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6975		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
6976		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
 
 
 
6977	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6978	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
6979	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
6980	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6981	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6982	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6983	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6984	.test_ring = gfx_v8_0_ring_test_ring,
6985	.test_ib = gfx_v8_0_ring_test_ib,
6986	.insert_nop = amdgpu_ring_insert_nop,
6987	.pad_ib = amdgpu_ring_generic_pad_ib,
6988	.set_priority = gfx_v8_0_ring_set_priority_compute,
6989	.emit_wreg = gfx_v8_0_ring_emit_wreg,
 
 
 
6990};
6991
6992static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
6993	.type = AMDGPU_RING_TYPE_KIQ,
6994	.align_mask = 0xff,
6995	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6996	.support_64bit_ptrs = false,
6997	.get_rptr = gfx_v8_0_ring_get_rptr,
6998	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6999	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7000	.emit_frame_size =
7001		20 + /* gfx_v8_0_ring_emit_gds_switch */
7002		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7003		5 + /* hdp_invalidate */
7004		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7005		17 + /* gfx_v8_0_ring_emit_vm_flush */
7006		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7007	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7008	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
7009	.test_ring = gfx_v8_0_ring_test_ring,
7010	.insert_nop = amdgpu_ring_insert_nop,
7011	.pad_ib = amdgpu_ring_generic_pad_ib,
7012	.emit_rreg = gfx_v8_0_ring_emit_rreg,
7013	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7014};
7015
7016static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
7017{
7018	int i;
7019
7020	adev->gfx.kiq.ring.funcs = &gfx_v8_0_ring_funcs_kiq;
7021
7022	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7023		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7024
7025	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7026		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7027}
7028
7029static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7030	.set = gfx_v8_0_set_eop_interrupt_state,
7031	.process = gfx_v8_0_eop_irq,
7032};
7033
7034static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7035	.set = gfx_v8_0_set_priv_reg_fault_state,
7036	.process = gfx_v8_0_priv_reg_irq,
7037};
7038
7039static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7040	.set = gfx_v8_0_set_priv_inst_fault_state,
7041	.process = gfx_v8_0_priv_inst_irq,
7042};
7043
7044static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7045	.set = gfx_v8_0_set_cp_ecc_int_state,
7046	.process = gfx_v8_0_cp_ecc_error_irq,
7047};
7048
7049static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7050	.set = gfx_v8_0_set_sq_int_state,
7051	.process = gfx_v8_0_sq_irq,
7052};
7053
7054static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7055{
7056	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7057	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7058
7059	adev->gfx.priv_reg_irq.num_types = 1;
7060	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7061
7062	adev->gfx.priv_inst_irq.num_types = 1;
7063	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7064
7065	adev->gfx.cp_ecc_error_irq.num_types = 1;
7066	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7067
7068	adev->gfx.sq_irq.num_types = 1;
7069	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7070}
7071
7072static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7073{
7074	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7075}
7076
7077static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7078{
7079	/* init asci gds info */
7080	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7081	adev->gds.gws_size = 64;
7082	adev->gds.oa_size = 16;
7083	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7084}
7085
7086static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7087						 u32 bitmap)
7088{
7089	u32 data;
7090
7091	if (!bitmap)
7092		return;
7093
7094	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7095	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7096
7097	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7098}
7099
7100static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7101{
7102	u32 data, mask;
7103
7104	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7105		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7106
7107	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7108
7109	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7110}
7111
7112static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7113{
7114	int i, j, k, counter, active_cu_number = 0;
7115	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7116	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7117	unsigned disable_masks[4 * 2];
7118	u32 ao_cu_num;
7119
7120	memset(cu_info, 0, sizeof(*cu_info));
7121
7122	if (adev->flags & AMD_IS_APU)
7123		ao_cu_num = 2;
7124	else
7125		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7126
7127	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7128
7129	mutex_lock(&adev->grbm_idx_mutex);
7130	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7131		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7132			mask = 1;
7133			ao_bitmap = 0;
7134			counter = 0;
7135			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
7136			if (i < 4 && j < 2)
7137				gfx_v8_0_set_user_cu_inactive_bitmap(
7138					adev, disable_masks[i * 2 + j]);
7139			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7140			cu_info->bitmap[i][j] = bitmap;
7141
7142			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7143				if (bitmap & mask) {
7144					if (counter < ao_cu_num)
7145						ao_bitmap |= mask;
7146					counter ++;
7147				}
7148				mask <<= 1;
7149			}
7150			active_cu_number += counter;
7151			if (i < 2 && j < 2)
7152				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7153			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7154		}
7155	}
7156	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
7157	mutex_unlock(&adev->grbm_idx_mutex);
7158
7159	cu_info->number = active_cu_number;
7160	cu_info->ao_cu_mask = ao_cu_mask;
7161	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7162	cu_info->max_waves_per_simd = 10;
7163	cu_info->max_scratch_slots_per_cu = 32;
7164	cu_info->wave_front_size = 64;
7165	cu_info->lds_size = 64;
7166}
7167
7168const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7169{
7170	.type = AMD_IP_BLOCK_TYPE_GFX,
7171	.major = 8,
7172	.minor = 0,
7173	.rev = 0,
7174	.funcs = &gfx_v8_0_ip_funcs,
7175};
7176
7177const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7178{
7179	.type = AMD_IP_BLOCK_TYPE_GFX,
7180	.major = 8,
7181	.minor = 1,
7182	.rev = 0,
7183	.funcs = &gfx_v8_0_ip_funcs,
7184};
7185
7186static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7187{
7188	uint64_t ce_payload_addr;
7189	int cnt_ce;
7190	union {
7191		struct vi_ce_ib_state regular;
7192		struct vi_ce_ib_state_chained_ib chained;
7193	} ce_payload = {};
7194
7195	if (ring->adev->virt.chained_ib_support) {
7196		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7197			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7198		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7199	} else {
7200		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7201			offsetof(struct vi_gfx_meta_data, ce_payload);
7202		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7203	}
7204
7205	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7206	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7207				WRITE_DATA_DST_SEL(8) |
7208				WR_CONFIRM) |
7209				WRITE_DATA_CACHE_POLICY(0));
7210	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7211	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7212	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7213}
7214
7215static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7216{
7217	uint64_t de_payload_addr, gds_addr, csa_addr;
7218	int cnt_de;
7219	union {
7220		struct vi_de_ib_state regular;
7221		struct vi_de_ib_state_chained_ib chained;
7222	} de_payload = {};
7223
7224	csa_addr = amdgpu_csa_vaddr(ring->adev);
7225	gds_addr = csa_addr + 4096;
7226	if (ring->adev->virt.chained_ib_support) {
7227		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7228		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7229		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7230		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7231	} else {
7232		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7233		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7234		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7235		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7236	}
7237
7238	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7239	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7240				WRITE_DATA_DST_SEL(8) |
7241				WR_CONFIRM) |
7242				WRITE_DATA_CACHE_POLICY(0));
7243	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7244	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7245	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7246}