Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
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/firmware.h>
  25#include <linux/module.h>
  26#include <linux/pci.h>
  27
  28#include <drm/drm_cache.h>
  29#include "amdgpu.h"
  30#include "gmc_v6_0.h"
  31#include "amdgpu_ucode.h"
  32#include "amdgpu_gem.h"
  33
  34#include "bif/bif_3_0_d.h"
  35#include "bif/bif_3_0_sh_mask.h"
  36#include "oss/oss_1_0_d.h"
  37#include "oss/oss_1_0_sh_mask.h"
  38#include "gmc/gmc_6_0_d.h"
  39#include "gmc/gmc_6_0_sh_mask.h"
  40#include "dce/dce_6_0_d.h"
  41#include "dce/dce_6_0_sh_mask.h"
  42#include "si_enums.h"
  43
  44static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev);
  45static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
  46static int gmc_v6_0_wait_for_idle(void *handle);
  47
  48MODULE_FIRMWARE("amdgpu/tahiti_mc.bin");
  49MODULE_FIRMWARE("amdgpu/pitcairn_mc.bin");
  50MODULE_FIRMWARE("amdgpu/verde_mc.bin");
  51MODULE_FIRMWARE("amdgpu/oland_mc.bin");
  52MODULE_FIRMWARE("amdgpu/hainan_mc.bin");
  53MODULE_FIRMWARE("amdgpu/si58_mc.bin");
  54
  55#define MC_SEQ_MISC0__MT__MASK   0xf0000000
  56#define MC_SEQ_MISC0__MT__GDDR1  0x10000000
  57#define MC_SEQ_MISC0__MT__DDR2   0x20000000
  58#define MC_SEQ_MISC0__MT__GDDR3  0x30000000
  59#define MC_SEQ_MISC0__MT__GDDR4  0x40000000
  60#define MC_SEQ_MISC0__MT__GDDR5  0x50000000
  61#define MC_SEQ_MISC0__MT__HBM    0x60000000
  62#define MC_SEQ_MISC0__MT__DDR3   0xB0000000
  63
  64
  65static const u32 crtc_offsets[6] =
  66{
  67	SI_CRTC0_REGISTER_OFFSET,
  68	SI_CRTC1_REGISTER_OFFSET,
  69	SI_CRTC2_REGISTER_OFFSET,
  70	SI_CRTC3_REGISTER_OFFSET,
  71	SI_CRTC4_REGISTER_OFFSET,
  72	SI_CRTC5_REGISTER_OFFSET
  73};
  74
  75static void gmc_v6_0_mc_stop(struct amdgpu_device *adev)
 
  76{
  77	u32 blackout;
  78
 
 
 
  79	gmc_v6_0_wait_for_idle((void *)adev);
  80
  81	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
  82	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
  83		/* Block CPU access */
  84		WREG32(mmBIF_FB_EN, 0);
  85		/* blackout the MC */
  86		blackout = REG_SET_FIELD(blackout,
  87					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
  88		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
  89	}
  90	/* wait for the MC to settle */
  91	udelay(100);
  92
  93}
  94
  95static void gmc_v6_0_mc_resume(struct amdgpu_device *adev)
 
  96{
  97	u32 tmp;
  98
  99	/* unblackout the MC */
 100	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 101	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
 102	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
 103	/* allow CPU access */
 104	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
 105	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
 106	WREG32(mmBIF_FB_EN, tmp);
 
 
 
 
 107}
 108
 109static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
 110{
 111	const char *chip_name;
 112	char fw_name[30];
 113	int err;
 114	bool is_58_fw = false;
 115
 116	DRM_DEBUG("\n");
 117
 118	switch (adev->asic_type) {
 119	case CHIP_TAHITI:
 120		chip_name = "tahiti";
 121		break;
 122	case CHIP_PITCAIRN:
 123		chip_name = "pitcairn";
 124		break;
 125	case CHIP_VERDE:
 126		chip_name = "verde";
 127		break;
 128	case CHIP_OLAND:
 129		chip_name = "oland";
 130		break;
 131	case CHIP_HAINAN:
 132		chip_name = "hainan";
 133		break;
 134	default: BUG();
 135	}
 136
 137	/* this memory configuration requires special firmware */
 138	if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
 139		is_58_fw = true;
 140
 141	if (is_58_fw)
 142		snprintf(fw_name, sizeof(fw_name), "amdgpu/si58_mc.bin");
 143	else
 144		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
 145	err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
 146	if (err)
 147		goto out;
 148
 149	err = amdgpu_ucode_validate(adev->gmc.fw);
 150
 151out:
 152	if (err) {
 153		dev_err(adev->dev,
 154		       "si_mc: Failed to load firmware \"%s\"\n",
 155		       fw_name);
 156		release_firmware(adev->gmc.fw);
 157		adev->gmc.fw = NULL;
 158	}
 159	return err;
 160}
 161
 162static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
 163{
 164	const __le32 *new_fw_data = NULL;
 165	u32 running;
 166	const __le32 *new_io_mc_regs = NULL;
 167	int i, regs_size, ucode_size;
 168	const struct mc_firmware_header_v1_0 *hdr;
 169
 170	if (!adev->gmc.fw)
 171		return -EINVAL;
 172
 173	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 174
 175	amdgpu_ucode_print_mc_hdr(&hdr->header);
 176
 177	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 178	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 179	new_io_mc_regs = (const __le32 *)
 180		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 181	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 182	new_fw_data = (const __le32 *)
 183		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 184
 185	running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
 186
 187	if (running == 0) {
 188
 189		/* reset the engine and set to writable */
 190		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 191		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 192
 193		/* load mc io regs */
 194		for (i = 0; i < regs_size; i++) {
 195			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
 196			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
 197		}
 198		/* load the MC ucode */
 199		for (i = 0; i < ucode_size; i++) {
 200			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
 201		}
 202
 203		/* put the engine back into the active state */
 204		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 205		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 206		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 207
 208		/* wait for training to complete */
 209		for (i = 0; i < adev->usec_timeout; i++) {
 210			if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
 211				break;
 212			udelay(1);
 213		}
 214		for (i = 0; i < adev->usec_timeout; i++) {
 215			if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
 216				break;
 217			udelay(1);
 218		}
 219
 220	}
 221
 222	return 0;
 223}
 224
 225static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
 226				       struct amdgpu_gmc *mc)
 227{
 228	u64 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
 229	base <<= 24;
 230
 231	amdgpu_gmc_vram_location(adev, mc, base);
 232	amdgpu_gmc_gart_location(adev, mc);
 
 
 
 233}
 234
 235static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
 236{
 
 
 237	int i, j;
 238
 239	/* Initialize HDP */
 240	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
 241		WREG32((0xb05 + j), 0x00000000);
 242		WREG32((0xb06 + j), 0x00000000);
 243		WREG32((0xb07 + j), 0x00000000);
 244		WREG32((0xb08 + j), 0x00000000);
 245		WREG32((0xb09 + j), 0x00000000);
 246	}
 247	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
 248
 
 
 
 
 
 249	if (gmc_v6_0_wait_for_idle((void *)adev)) {
 250		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 251	}
 252
 253	if (adev->mode_info.num_crtc) {
 254		u32 tmp;
 255
 256		/* Lockout access through VGA aperture*/
 257		tmp = RREG32(mmVGA_HDP_CONTROL);
 258		tmp |= VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK;
 259		WREG32(mmVGA_HDP_CONTROL, tmp);
 260
 261		/* disable VGA render */
 262		tmp = RREG32(mmVGA_RENDER_CONTROL);
 263		tmp &= ~VGA_VSTATUS_CNTL;
 264		WREG32(mmVGA_RENDER_CONTROL, tmp);
 265	}
 266	/* Update configuration */
 267	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
 268	       adev->gmc.vram_start >> 12);
 269	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 270	       adev->gmc.vram_end >> 12);
 271	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
 272	       adev->vram_scratch.gpu_addr >> 12);
 
 
 
 
 
 
 
 273	WREG32(mmMC_VM_AGP_BASE, 0);
 274	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
 275	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
 276
 277	if (gmc_v6_0_wait_for_idle((void *)adev)) {
 278		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 279	}
 
 280}
 281
 282static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
 283{
 284
 285	u32 tmp;
 286	int chansize, numchan;
 287	int r;
 288
 289	tmp = RREG32(mmMC_ARB_RAMCFG);
 290	if (tmp & (1 << 11)) {
 291		chansize = 16;
 292	} else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
 293		chansize = 64;
 294	} else {
 295		chansize = 32;
 296	}
 297	tmp = RREG32(mmMC_SHARED_CHMAP);
 298	switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
 299	case 0:
 300	default:
 301		numchan = 1;
 302		break;
 303	case 1:
 304		numchan = 2;
 305		break;
 306	case 2:
 307		numchan = 4;
 308		break;
 309	case 3:
 310		numchan = 8;
 311		break;
 312	case 4:
 313		numchan = 3;
 314		break;
 315	case 5:
 316		numchan = 6;
 317		break;
 318	case 6:
 319		numchan = 10;
 320		break;
 321	case 7:
 322		numchan = 12;
 323		break;
 324	case 8:
 325		numchan = 16;
 326		break;
 327	}
 328	adev->gmc.vram_width = numchan * chansize;
 
 
 
 329	/* size in MB on si */
 330	adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 331	adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 
 332
 333	if (!(adev->flags & AMD_IS_APU)) {
 334		r = amdgpu_device_resize_fb_bar(adev);
 335		if (r)
 336			return r;
 337	}
 338	adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
 339	adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
 340	adev->gmc.visible_vram_size = adev->gmc.aper_size;
 341
 342	/* set the gart size */
 343	if (amdgpu_gart_size == -1) {
 344		switch (adev->asic_type) {
 345		case CHIP_HAINAN:    /* no MM engines */
 346		default:
 347			adev->gmc.gart_size = 256ULL << 20;
 348			break;
 349		case CHIP_VERDE:    /* UVD, VCE do not support GPUVM */
 350		case CHIP_TAHITI:   /* UVD, VCE do not support GPUVM */
 351		case CHIP_PITCAIRN: /* UVD, VCE do not support GPUVM */
 352		case CHIP_OLAND:    /* UVD, VCE do not support GPUVM */
 353			adev->gmc.gart_size = 1024ULL << 20;
 354			break;
 355		}
 356	} else {
 357		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
 358	}
 359
 360	gmc_v6_0_vram_gtt_location(adev, &adev->gmc);
 361
 362	return 0;
 363}
 364
 365static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 366					uint32_t vmhub, uint32_t flush_type)
 367{
 368	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 369}
 370
 371static uint64_t gmc_v6_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 372					    unsigned vmid, uint64_t pd_addr)
 373{
 374	uint32_t reg;
 375
 376	/* write new base address */
 377	if (vmid < 8)
 378		reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
 379	else
 380		reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (vmid - 8);
 381	amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
 382
 383	/* bits 0-15 are the VM contexts0-15 */
 384	amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
 385
 386	return pd_addr;
 387}
 388
 389static uint64_t gmc_v6_0_get_vm_pte_flags(struct amdgpu_device *adev,
 390					  uint32_t flags)
 391{
 392	uint64_t pte_flag = 0;
 393
 394	if (flags & AMDGPU_VM_PAGE_READABLE)
 395		pte_flag |= AMDGPU_PTE_READABLE;
 396	if (flags & AMDGPU_VM_PAGE_WRITEABLE)
 397		pte_flag |= AMDGPU_PTE_WRITEABLE;
 398	if (flags & AMDGPU_VM_PAGE_PRT)
 399		pte_flag |= AMDGPU_PTE_PRT;
 400
 401	return pte_flag;
 402}
 403
 404static void gmc_v6_0_get_vm_pde(struct amdgpu_device *adev, int level,
 405				uint64_t *addr, uint64_t *flags)
 406{
 407	BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
 408}
 409
 410static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
 411					      bool value)
 412{
 413	u32 tmp;
 414
 415	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 416	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 417			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 418	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 419			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 420	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 421			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 422	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 423			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 424	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 425			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 426	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 427			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 428	WREG32(mmVM_CONTEXT1_CNTL, tmp);
 429}
 430
 431 /**
 432   + * gmc_v8_0_set_prt - set PRT VM fault
 433   + *
 434   + * @adev: amdgpu_device pointer
 435   + * @enable: enable/disable VM fault handling for PRT
 436   +*/
 437static void gmc_v6_0_set_prt(struct amdgpu_device *adev, bool enable)
 438{
 439	u32 tmp;
 440
 441	if (enable && !adev->gmc.prt_warning) {
 442		dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
 443		adev->gmc.prt_warning = true;
 444	}
 445
 446	tmp = RREG32(mmVM_PRT_CNTL);
 447	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 448			    CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
 449			    enable);
 450	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 451			    TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS,
 452			    enable);
 453	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 454			    L2_CACHE_STORE_INVALID_ENTRIES,
 455			    enable);
 456	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 457			    L1_TLB_STORE_INVALID_ENTRIES,
 458			    enable);
 459	WREG32(mmVM_PRT_CNTL, tmp);
 460
 461	if (enable) {
 462		uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
 463		uint32_t high = adev->vm_manager.max_pfn -
 464			(AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
 465
 466		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
 467		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
 468		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
 469		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
 470		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
 471		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
 472		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
 473		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
 474	} else {
 475		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
 476		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
 477		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
 478		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
 479		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
 480		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
 481		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
 482		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
 483	}
 484}
 485
 486static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
 487{
 488	uint64_t table_addr;
 489	int r, i;
 490	u32 field;
 491
 492	if (adev->gart.bo == NULL) {
 493		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
 494		return -EINVAL;
 495	}
 496	r = amdgpu_gart_table_vram_pin(adev);
 497	if (r)
 498		return r;
 499
 500	table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
 501
 502	/* Setup TLB control */
 503	WREG32(mmMC_VM_MX_L1_TLB_CNTL,
 504	       (0xA << 7) |
 505	       MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
 506	       MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
 507	       MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
 508	       MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
 509	       (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
 510	/* Setup L2 cache */
 511	WREG32(mmVM_L2_CNTL,
 512	       VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
 513	       VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
 514	       VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 515	       VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 516	       (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
 517	       (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
 518	WREG32(mmVM_L2_CNTL2,
 519	       VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
 520	       VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
 521
 522	field = adev->vm_manager.fragment_size;
 523	WREG32(mmVM_L2_CNTL3,
 524	       VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
 525	       (field << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
 526	       (field << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
 527	/* setup context0 */
 528	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
 529	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
 530	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
 531	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 532			(u32)(adev->dummy_page_addr >> 12));
 533	WREG32(mmVM_CONTEXT0_CNTL2, 0);
 534	WREG32(mmVM_CONTEXT0_CNTL,
 535	       VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
 536	       (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
 537	       VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
 538
 539	WREG32(0x575, 0);
 540	WREG32(0x576, 0);
 541	WREG32(0x577, 0);
 542
 543	/* empty context1-15 */
 544	/* set vm size, must be a multiple of 4 */
 545	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
 546	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
 547	/* Assign the pt base to something valid for now; the pts used for
 548	 * the VMs are determined by the application and setup and assigned
 549	 * on the fly in the vm part of radeon_gart.c
 550	 */
 551	for (i = 1; i < 16; i++) {
 552		if (i < 8)
 553			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
 554			       table_addr >> 12);
 555		else
 556			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
 557			       table_addr >> 12);
 558	}
 559
 560	/* enable context1-15 */
 561	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
 562	       (u32)(adev->dummy_page_addr >> 12));
 563	WREG32(mmVM_CONTEXT1_CNTL2, 4);
 564	WREG32(mmVM_CONTEXT1_CNTL,
 565	       VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
 566	       (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
 567	       ((adev->vm_manager.block_size - 9)
 568	       << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
 569	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 570		gmc_v6_0_set_fault_enable_default(adev, false);
 571	else
 572		gmc_v6_0_set_fault_enable_default(adev, true);
 573
 574	gmc_v6_0_flush_gpu_tlb(adev, 0, 0, 0);
 575	dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
 576		 (unsigned)(adev->gmc.gart_size >> 20),
 577		 (unsigned long long)table_addr);
 578	adev->gart.ready = true;
 579	return 0;
 580}
 581
 582static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
 583{
 584	int r;
 585
 586	if (adev->gart.bo) {
 587		dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
 588		return 0;
 589	}
 590	r = amdgpu_gart_init(adev);
 591	if (r)
 592		return r;
 593	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
 594	adev->gart.gart_pte_flags = 0;
 595	return amdgpu_gart_table_vram_alloc(adev);
 596}
 597
 598static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
 599{
 600	/*unsigned i;
 601
 602	for (i = 1; i < 16; ++i) {
 603		uint32_t reg;
 604		if (i < 8)
 605			reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
 606		else
 607			reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
 608		adev->vm_manager.saved_table_addr[i] = RREG32(reg);
 609	}*/
 610
 611	/* Disable all tables */
 612	WREG32(mmVM_CONTEXT0_CNTL, 0);
 613	WREG32(mmVM_CONTEXT1_CNTL, 0);
 614	/* Setup TLB control */
 615	WREG32(mmMC_VM_MX_L1_TLB_CNTL,
 616	       MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
 617	       (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
 618	/* Setup L2 cache */
 619	WREG32(mmVM_L2_CNTL,
 620	       VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 621	       VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 622	       (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
 623	       (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
 624	WREG32(mmVM_L2_CNTL2, 0);
 625	WREG32(mmVM_L2_CNTL3,
 626	       VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
 627	       (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
 628	amdgpu_gart_table_vram_unpin(adev);
 629}
 630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 631static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
 632				     u32 status, u32 addr, u32 mc_client)
 633{
 634	u32 mc_id;
 635	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
 636	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 637					PROTECTIONS);
 638	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
 639		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
 640
 641	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 642			      MEMORY_CLIENT_ID);
 643
 644	dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
 645	       protections, vmid, addr,
 646	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 647			     MEMORY_CLIENT_RW) ?
 648	       "write" : "read", block, mc_client, mc_id);
 649}
 650
 651/*
 652static const u32 mc_cg_registers[] = {
 653	MC_HUB_MISC_HUB_CG,
 654	MC_HUB_MISC_SIP_CG,
 655	MC_HUB_MISC_VM_CG,
 656	MC_XPB_CLK_GAT,
 657	ATC_MISC_CG,
 658	MC_CITF_MISC_WR_CG,
 659	MC_CITF_MISC_RD_CG,
 660	MC_CITF_MISC_VM_CG,
 661	VM_L2_CG,
 662};
 663
 664static const u32 mc_cg_ls_en[] = {
 665	MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
 666	MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
 667	MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
 668	MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
 669	ATC_MISC_CG__MEM_LS_ENABLE_MASK,
 670	MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
 671	MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
 672	MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
 673	VM_L2_CG__MEM_LS_ENABLE_MASK,
 674};
 675
 676static const u32 mc_cg_en[] = {
 677	MC_HUB_MISC_HUB_CG__ENABLE_MASK,
 678	MC_HUB_MISC_SIP_CG__ENABLE_MASK,
 679	MC_HUB_MISC_VM_CG__ENABLE_MASK,
 680	MC_XPB_CLK_GAT__ENABLE_MASK,
 681	ATC_MISC_CG__ENABLE_MASK,
 682	MC_CITF_MISC_WR_CG__ENABLE_MASK,
 683	MC_CITF_MISC_RD_CG__ENABLE_MASK,
 684	MC_CITF_MISC_VM_CG__ENABLE_MASK,
 685	VM_L2_CG__ENABLE_MASK,
 686};
 687
 688static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
 689				  bool enable)
 690{
 691	int i;
 692	u32 orig, data;
 693
 694	for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
 695		orig = data = RREG32(mc_cg_registers[i]);
 696		if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
 697			data |= mc_cg_ls_en[i];
 698		else
 699			data &= ~mc_cg_ls_en[i];
 700		if (data != orig)
 701			WREG32(mc_cg_registers[i], data);
 702	}
 703}
 704
 705static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
 706				    bool enable)
 707{
 708	int i;
 709	u32 orig, data;
 710
 711	for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
 712		orig = data = RREG32(mc_cg_registers[i]);
 713		if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
 714			data |= mc_cg_en[i];
 715		else
 716			data &= ~mc_cg_en[i];
 717		if (data != orig)
 718			WREG32(mc_cg_registers[i], data);
 719	}
 720}
 721
 722static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
 723				     bool enable)
 724{
 725	u32 orig, data;
 726
 727	orig = data = RREG32_PCIE(ixPCIE_CNTL2);
 728
 729	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
 730		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
 731		data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
 732		data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
 733		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
 734	} else {
 735		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
 736		data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
 737		data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
 738		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
 739	}
 740
 741	if (orig != data)
 742		WREG32_PCIE(ixPCIE_CNTL2, data);
 743}
 744
 745static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
 746				     bool enable)
 747{
 748	u32 orig, data;
 749
 750	orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
 751
 752	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
 753		data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
 754	else
 755		data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
 756
 757	if (orig != data)
 758		WREG32(mmHDP_HOST_PATH_CNTL, data);
 759}
 760
 761static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
 762				   bool enable)
 763{
 764	u32 orig, data;
 765
 766	orig = data = RREG32(mmHDP_MEM_POWER_LS);
 767
 768	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
 769		data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
 770	else
 771		data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
 772
 773	if (orig != data)
 774		WREG32(mmHDP_MEM_POWER_LS, data);
 775}
 776*/
 777
 778static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
 779{
 780	switch (mc_seq_vram_type) {
 781	case MC_SEQ_MISC0__MT__GDDR1:
 782		return AMDGPU_VRAM_TYPE_GDDR1;
 783	case MC_SEQ_MISC0__MT__DDR2:
 784		return AMDGPU_VRAM_TYPE_DDR2;
 785	case MC_SEQ_MISC0__MT__GDDR3:
 786		return AMDGPU_VRAM_TYPE_GDDR3;
 787	case MC_SEQ_MISC0__MT__GDDR4:
 788		return AMDGPU_VRAM_TYPE_GDDR4;
 789	case MC_SEQ_MISC0__MT__GDDR5:
 790		return AMDGPU_VRAM_TYPE_GDDR5;
 791	case MC_SEQ_MISC0__MT__DDR3:
 792		return AMDGPU_VRAM_TYPE_DDR3;
 793	default:
 794		return AMDGPU_VRAM_TYPE_UNKNOWN;
 795	}
 796}
 797
 798static int gmc_v6_0_early_init(void *handle)
 799{
 800	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 801
 802	gmc_v6_0_set_gmc_funcs(adev);
 803	gmc_v6_0_set_irq_funcs(adev);
 804
 
 
 
 
 
 
 
 
 805	return 0;
 806}
 807
 808static int gmc_v6_0_late_init(void *handle)
 809{
 810	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 811
 812	amdgpu_bo_late_init(adev);
 813
 814	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
 815		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
 816	else
 817		return 0;
 818}
 819
 820static unsigned gmc_v6_0_get_vbios_fb_size(struct amdgpu_device *adev)
 821{
 822	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
 823	unsigned size;
 824
 825	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
 826		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
 827	} else {
 828		u32 viewport = RREG32(mmVIEWPORT_SIZE);
 829		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
 830			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
 831			4);
 832	}
 833	/* return 0 if the pre-OS buffer uses up most of vram */
 834	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
 835		return 0;
 836	return size;
 837}
 838
 839static int gmc_v6_0_sw_init(void *handle)
 840{
 841	int r;
 
 842	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 843
 844	adev->num_vmhubs = 1;
 845
 846	if (adev->flags & AMD_IS_APU) {
 847		adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
 848	} else {
 849		u32 tmp = RREG32(mmMC_SEQ_MISC0);
 850		tmp &= MC_SEQ_MISC0__MT__MASK;
 851		adev->gmc.vram_type = gmc_v6_0_convert_vram_type(tmp);
 852	}
 853
 854	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 146, &adev->gmc.vm_fault);
 855	if (r)
 856		return r;
 857
 858	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 147, &adev->gmc.vm_fault);
 859	if (r)
 860		return r;
 861
 862	amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
 863
 864	adev->gmc.mc_mask = 0xffffffffffULL;
 865
 866	r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(44));
 
 
 867	if (r) {
 
 
 868		dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
 869		return r;
 870	}
 871	adev->need_swiotlb = drm_need_swiotlb(44);
 
 
 
 
 872
 873	r = gmc_v6_0_init_microcode(adev);
 874	if (r) {
 875		dev_err(adev->dev, "Failed to load mc firmware!\n");
 876		return r;
 877	}
 878
 879	r = gmc_v6_0_mc_init(adev);
 880	if (r)
 881		return r;
 882
 883	adev->gmc.stolen_size = gmc_v6_0_get_vbios_fb_size(adev);
 884
 885	r = amdgpu_bo_init(adev);
 886	if (r)
 887		return r;
 888
 889	r = gmc_v6_0_gart_init(adev);
 890	if (r)
 891		return r;
 892
 893	/*
 894	 * number of VMs
 895	 * VMID 0 is reserved for System
 896	 * amdgpu graphics/compute will use VMIDs 1-7
 897	 * amdkfd will use VMIDs 8-15
 898	 */
 899	adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
 900	amdgpu_vm_manager_init(adev);
 901
 902	/* base offset of vram pages */
 903	if (adev->flags & AMD_IS_APU) {
 904		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
 905
 906		tmp <<= 22;
 907		adev->vm_manager.vram_base_offset = tmp;
 908	} else {
 909		adev->vm_manager.vram_base_offset = 0;
 910	}
 911
 912	return 0;
 913}
 914
 915static int gmc_v6_0_sw_fini(void *handle)
 916{
 917	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 918
 
 
 
 
 
 919	amdgpu_gem_force_release(adev);
 920	amdgpu_vm_manager_fini(adev);
 921	amdgpu_gart_table_vram_free(adev);
 922	amdgpu_bo_fini(adev);
 923	amdgpu_gart_fini(adev);
 924	release_firmware(adev->gmc.fw);
 925	adev->gmc.fw = NULL;
 926
 927	return 0;
 928}
 929
 930static int gmc_v6_0_hw_init(void *handle)
 931{
 932	int r;
 933	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 934
 935	gmc_v6_0_mc_program(adev);
 936
 937	if (!(adev->flags & AMD_IS_APU)) {
 938		r = gmc_v6_0_mc_load_microcode(adev);
 939		if (r) {
 940			dev_err(adev->dev, "Failed to load MC firmware!\n");
 941			return r;
 942		}
 943	}
 944
 945	r = gmc_v6_0_gart_enable(adev);
 946	if (r)
 947		return r;
 948
 949	return r;
 950}
 951
 952static int gmc_v6_0_hw_fini(void *handle)
 953{
 954	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 955
 956	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
 957	gmc_v6_0_gart_disable(adev);
 958
 959	return 0;
 960}
 961
 962static int gmc_v6_0_suspend(void *handle)
 963{
 964	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 965
 
 
 
 
 966	gmc_v6_0_hw_fini(adev);
 967
 968	return 0;
 969}
 970
 971static int gmc_v6_0_resume(void *handle)
 972{
 973	int r;
 974	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 975
 976	r = gmc_v6_0_hw_init(adev);
 977	if (r)
 978		return r;
 979
 980	amdgpu_vmid_reset_all(adev);
 
 
 
 
 
 
 
 981
 982	return 0;
 983}
 984
 985static bool gmc_v6_0_is_idle(void *handle)
 986{
 987	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 988	u32 tmp = RREG32(mmSRBM_STATUS);
 989
 990	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 991		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
 992		return false;
 993
 994	return true;
 995}
 996
 997static int gmc_v6_0_wait_for_idle(void *handle)
 998{
 999	unsigned i;
 
1000	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1001
1002	for (i = 0; i < adev->usec_timeout; i++) {
1003		if (gmc_v6_0_is_idle(handle))
 
 
 
 
 
1004			return 0;
1005		udelay(1);
1006	}
1007	return -ETIMEDOUT;
1008
1009}
1010
1011static int gmc_v6_0_soft_reset(void *handle)
1012{
1013	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
1014	u32 srbm_soft_reset = 0;
1015	u32 tmp = RREG32(mmSRBM_STATUS);
1016
1017	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1018		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1019						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1020
1021	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1022		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1023		if (!(adev->flags & AMD_IS_APU))
1024			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1025							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1026	}
1027
1028	if (srbm_soft_reset) {
1029		gmc_v6_0_mc_stop(adev);
1030		if (gmc_v6_0_wait_for_idle(adev)) {
1031			dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1032		}
1033
1034
1035		tmp = RREG32(mmSRBM_SOFT_RESET);
1036		tmp |= srbm_soft_reset;
1037		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1038		WREG32(mmSRBM_SOFT_RESET, tmp);
1039		tmp = RREG32(mmSRBM_SOFT_RESET);
1040
1041		udelay(50);
1042
1043		tmp &= ~srbm_soft_reset;
1044		WREG32(mmSRBM_SOFT_RESET, tmp);
1045		tmp = RREG32(mmSRBM_SOFT_RESET);
1046
1047		udelay(50);
1048
1049		gmc_v6_0_mc_resume(adev);
1050		udelay(50);
1051	}
1052
1053	return 0;
1054}
1055
1056static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1057					     struct amdgpu_irq_src *src,
1058					     unsigned type,
1059					     enum amdgpu_interrupt_state state)
1060{
1061	u32 tmp;
1062	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1063		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1064		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1065		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1066		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1067		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1068
1069	switch (state) {
1070	case AMDGPU_IRQ_STATE_DISABLE:
1071		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1072		tmp &= ~bits;
1073		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1074		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1075		tmp &= ~bits;
1076		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1077		break;
1078	case AMDGPU_IRQ_STATE_ENABLE:
1079		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1080		tmp |= bits;
1081		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1082		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1083		tmp |= bits;
1084		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1085		break;
1086	default:
1087		break;
1088	}
1089
1090	return 0;
1091}
1092
1093static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1094				      struct amdgpu_irq_src *source,
1095				      struct amdgpu_iv_entry *entry)
1096{
1097	u32 addr, status;
1098
1099	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1100	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1101	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1102
1103	if (!addr && !status)
1104		return 0;
1105
1106	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1107		gmc_v6_0_set_fault_enable_default(adev, false);
1108
1109	if (printk_ratelimit()) {
1110		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1111			entry->src_id, entry->src_data[0]);
1112		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1113			addr);
1114		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1115			status);
1116		gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1117	}
1118
1119	return 0;
1120}
1121
1122static int gmc_v6_0_set_clockgating_state(void *handle,
1123					  enum amd_clockgating_state state)
1124{
1125	return 0;
1126}
1127
1128static int gmc_v6_0_set_powergating_state(void *handle,
1129					  enum amd_powergating_state state)
1130{
1131	return 0;
1132}
1133
1134static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1135	.name = "gmc_v6_0",
1136	.early_init = gmc_v6_0_early_init,
1137	.late_init = gmc_v6_0_late_init,
1138	.sw_init = gmc_v6_0_sw_init,
1139	.sw_fini = gmc_v6_0_sw_fini,
1140	.hw_init = gmc_v6_0_hw_init,
1141	.hw_fini = gmc_v6_0_hw_fini,
1142	.suspend = gmc_v6_0_suspend,
1143	.resume = gmc_v6_0_resume,
1144	.is_idle = gmc_v6_0_is_idle,
1145	.wait_for_idle = gmc_v6_0_wait_for_idle,
1146	.soft_reset = gmc_v6_0_soft_reset,
1147	.set_clockgating_state = gmc_v6_0_set_clockgating_state,
1148	.set_powergating_state = gmc_v6_0_set_powergating_state,
1149};
1150
1151static const struct amdgpu_gmc_funcs gmc_v6_0_gmc_funcs = {
1152	.flush_gpu_tlb = gmc_v6_0_flush_gpu_tlb,
1153	.emit_flush_gpu_tlb = gmc_v6_0_emit_flush_gpu_tlb,
1154	.set_prt = gmc_v6_0_set_prt,
1155	.get_vm_pde = gmc_v6_0_get_vm_pde,
1156	.get_vm_pte_flags = gmc_v6_0_get_vm_pte_flags
1157};
1158
1159static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1160	.set = gmc_v6_0_vm_fault_interrupt_state,
1161	.process = gmc_v6_0_process_interrupt,
1162};
1163
1164static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device *adev)
1165{
1166	adev->gmc.gmc_funcs = &gmc_v6_0_gmc_funcs;
 
1167}
1168
1169static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1170{
1171	adev->gmc.vm_fault.num_types = 1;
1172	adev->gmc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1173}
1174
1175const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1176{
1177	.type = AMD_IP_BLOCK_TYPE_GMC,
1178	.major = 6,
1179	.minor = 0,
1180	.rev = 0,
1181	.funcs = &gmc_v6_0_ip_funcs,
1182};
v4.10.11
   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#include <linux/firmware.h>
  24#include "drmP.h"
 
 
 
  25#include "amdgpu.h"
  26#include "gmc_v6_0.h"
  27#include "amdgpu_ucode.h"
 
  28
  29#include "bif/bif_3_0_d.h"
  30#include "bif/bif_3_0_sh_mask.h"
  31#include "oss/oss_1_0_d.h"
  32#include "oss/oss_1_0_sh_mask.h"
  33#include "gmc/gmc_6_0_d.h"
  34#include "gmc/gmc_6_0_sh_mask.h"
  35#include "dce/dce_6_0_d.h"
  36#include "dce/dce_6_0_sh_mask.h"
  37#include "si_enums.h"
  38
  39static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev);
  40static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev);
  41static int gmc_v6_0_wait_for_idle(void *handle);
  42
  43MODULE_FIRMWARE("radeon/tahiti_mc.bin");
  44MODULE_FIRMWARE("radeon/pitcairn_mc.bin");
  45MODULE_FIRMWARE("radeon/verde_mc.bin");
  46MODULE_FIRMWARE("radeon/oland_mc.bin");
  47MODULE_FIRMWARE("radeon/si58_mc.bin");
 
  48
  49#define MC_SEQ_MISC0__MT__MASK   0xf0000000
  50#define MC_SEQ_MISC0__MT__GDDR1  0x10000000
  51#define MC_SEQ_MISC0__MT__DDR2   0x20000000
  52#define MC_SEQ_MISC0__MT__GDDR3  0x30000000
  53#define MC_SEQ_MISC0__MT__GDDR4  0x40000000
  54#define MC_SEQ_MISC0__MT__GDDR5  0x50000000
  55#define MC_SEQ_MISC0__MT__HBM    0x60000000
  56#define MC_SEQ_MISC0__MT__DDR3   0xB0000000
  57
  58
  59static const u32 crtc_offsets[6] =
  60{
  61	SI_CRTC0_REGISTER_OFFSET,
  62	SI_CRTC1_REGISTER_OFFSET,
  63	SI_CRTC2_REGISTER_OFFSET,
  64	SI_CRTC3_REGISTER_OFFSET,
  65	SI_CRTC4_REGISTER_OFFSET,
  66	SI_CRTC5_REGISTER_OFFSET
  67};
  68
  69static void gmc_v6_0_mc_stop(struct amdgpu_device *adev,
  70			     struct amdgpu_mode_mc_save *save)
  71{
  72	u32 blackout;
  73
  74	if (adev->mode_info.num_crtc)
  75		amdgpu_display_stop_mc_access(adev, save);
  76
  77	gmc_v6_0_wait_for_idle((void *)adev);
  78
  79	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
  80	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
  81		/* Block CPU access */
  82		WREG32(mmBIF_FB_EN, 0);
  83		/* blackout the MC */
  84		blackout = REG_SET_FIELD(blackout,
  85					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
  86		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
  87	}
  88	/* wait for the MC to settle */
  89	udelay(100);
  90
  91}
  92
  93static void gmc_v6_0_mc_resume(struct amdgpu_device *adev,
  94			       struct amdgpu_mode_mc_save *save)
  95{
  96	u32 tmp;
  97
  98	/* unblackout the MC */
  99	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 100	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
 101	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
 102	/* allow CPU access */
 103	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
 104	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
 105	WREG32(mmBIF_FB_EN, tmp);
 106
 107	if (adev->mode_info.num_crtc)
 108		amdgpu_display_resume_mc_access(adev, save);
 109
 110}
 111
 112static int gmc_v6_0_init_microcode(struct amdgpu_device *adev)
 113{
 114	const char *chip_name;
 115	char fw_name[30];
 116	int err;
 117	bool is_58_fw = false;
 118
 119	DRM_DEBUG("\n");
 120
 121	switch (adev->asic_type) {
 122	case CHIP_TAHITI:
 123		chip_name = "tahiti";
 124		break;
 125	case CHIP_PITCAIRN:
 126		chip_name = "pitcairn";
 127		break;
 128	case CHIP_VERDE:
 129		chip_name = "verde";
 130		break;
 131	case CHIP_OLAND:
 132		chip_name = "oland";
 133		break;
 134	case CHIP_HAINAN:
 135		chip_name = "hainan";
 136		break;
 137	default: BUG();
 138	}
 139
 140	/* this memory configuration requires special firmware */
 141	if (((RREG32(mmMC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58)
 142		is_58_fw = true;
 143
 144	if (is_58_fw)
 145		snprintf(fw_name, sizeof(fw_name), "radeon/si58_mc.bin");
 146	else
 147		snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
 148	err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
 149	if (err)
 150		goto out;
 151
 152	err = amdgpu_ucode_validate(adev->mc.fw);
 153
 154out:
 155	if (err) {
 156		dev_err(adev->dev,
 157		       "si_mc: Failed to load firmware \"%s\"\n",
 158		       fw_name);
 159		release_firmware(adev->mc.fw);
 160		adev->mc.fw = NULL;
 161	}
 162	return err;
 163}
 164
 165static int gmc_v6_0_mc_load_microcode(struct amdgpu_device *adev)
 166{
 167	const __le32 *new_fw_data = NULL;
 168	u32 running;
 169	const __le32 *new_io_mc_regs = NULL;
 170	int i, regs_size, ucode_size;
 171	const struct mc_firmware_header_v1_0 *hdr;
 172
 173	if (!adev->mc.fw)
 174		return -EINVAL;
 175
 176	hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
 177
 178	amdgpu_ucode_print_mc_hdr(&hdr->header);
 179
 180	adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 181	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 182	new_io_mc_regs = (const __le32 *)
 183		(adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 184	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 185	new_fw_data = (const __le32 *)
 186		(adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 187
 188	running = RREG32(mmMC_SEQ_SUP_CNTL) & MC_SEQ_SUP_CNTL__RUN_MASK;
 189
 190	if (running == 0) {
 191
 192		/* reset the engine and set to writable */
 193		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 194		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 195
 196		/* load mc io regs */
 197		for (i = 0; i < regs_size; i++) {
 198			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++));
 199			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++));
 200		}
 201		/* load the MC ucode */
 202		for (i = 0; i < ucode_size; i++) {
 203			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++));
 204		}
 205
 206		/* put the engine back into the active state */
 207		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 208		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 209		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 210
 211		/* wait for training to complete */
 212		for (i = 0; i < adev->usec_timeout; i++) {
 213			if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK)
 214				break;
 215			udelay(1);
 216		}
 217		for (i = 0; i < adev->usec_timeout; i++) {
 218			if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK)
 219				break;
 220			udelay(1);
 221		}
 222
 223	}
 224
 225	return 0;
 226}
 227
 228static void gmc_v6_0_vram_gtt_location(struct amdgpu_device *adev,
 229				       struct amdgpu_mc *mc)
 230{
 231	if (mc->mc_vram_size > 0xFFC0000000ULL) {
 232		dev_warn(adev->dev, "limiting VRAM\n");
 233		mc->real_vram_size = 0xFFC0000000ULL;
 234		mc->mc_vram_size = 0xFFC0000000ULL;
 235	}
 236	amdgpu_vram_location(adev, &adev->mc, 0);
 237	adev->mc.gtt_base_align = 0;
 238	amdgpu_gtt_location(adev, mc);
 239}
 240
 241static void gmc_v6_0_mc_program(struct amdgpu_device *adev)
 242{
 243	struct amdgpu_mode_mc_save save;
 244	u32 tmp;
 245	int i, j;
 246
 247	/* Initialize HDP */
 248	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
 249		WREG32((0xb05 + j), 0x00000000);
 250		WREG32((0xb06 + j), 0x00000000);
 251		WREG32((0xb07 + j), 0x00000000);
 252		WREG32((0xb08 + j), 0x00000000);
 253		WREG32((0xb09 + j), 0x00000000);
 254	}
 255	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
 256
 257	if (adev->mode_info.num_crtc)
 258		amdgpu_display_set_vga_render_state(adev, false);
 259
 260	gmc_v6_0_mc_stop(adev, &save);
 261
 262	if (gmc_v6_0_wait_for_idle((void *)adev)) {
 263		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 264	}
 265
 266	WREG32(mmVGA_HDP_CONTROL, VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK);
 
 
 
 
 
 
 
 
 
 
 
 
 267	/* Update configuration */
 268	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
 269	       adev->mc.vram_start >> 12);
 270	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 271	       adev->mc.vram_end >> 12);
 272	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
 273	       adev->vram_scratch.gpu_addr >> 12);
 274	tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
 275	tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
 276	WREG32(mmMC_VM_FB_LOCATION, tmp);
 277	/* XXX double check these! */
 278	WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
 279	WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
 280	WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
 281	WREG32(mmMC_VM_AGP_BASE, 0);
 282	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
 283	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
 284
 285	if (gmc_v6_0_wait_for_idle((void *)adev)) {
 286		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 287	}
 288	gmc_v6_0_mc_resume(adev, &save);
 289}
 290
 291static int gmc_v6_0_mc_init(struct amdgpu_device *adev)
 292{
 293
 294	u32 tmp;
 295	int chansize, numchan;
 
 296
 297	tmp = RREG32(mmMC_ARB_RAMCFG);
 298	if (tmp & (1 << 11)) {
 299		chansize = 16;
 300	} else if (tmp & MC_ARB_RAMCFG__CHANSIZE_MASK) {
 301		chansize = 64;
 302	} else {
 303		chansize = 32;
 304	}
 305	tmp = RREG32(mmMC_SHARED_CHMAP);
 306	switch ((tmp & MC_SHARED_CHMAP__NOOFCHAN_MASK) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT) {
 307	case 0:
 308	default:
 309		numchan = 1;
 310		break;
 311	case 1:
 312		numchan = 2;
 313		break;
 314	case 2:
 315		numchan = 4;
 316		break;
 317	case 3:
 318		numchan = 8;
 319		break;
 320	case 4:
 321		numchan = 3;
 322		break;
 323	case 5:
 324		numchan = 6;
 325		break;
 326	case 6:
 327		numchan = 10;
 328		break;
 329	case 7:
 330		numchan = 12;
 331		break;
 332	case 8:
 333		numchan = 16;
 334		break;
 335	}
 336	adev->mc.vram_width = numchan * chansize;
 337	/* Could aper size report 0 ? */
 338	adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
 339	adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
 340	/* size in MB on si */
 341	adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 342	adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 343	adev->mc.visible_vram_size = adev->mc.aper_size;
 344
 345	/* unless the user had overridden it, set the gart
 346	 * size equal to the 1024 or vram, whichever is larger.
 347	 */
 348	if (amdgpu_gart_size == -1)
 349		adev->mc.gtt_size = max((1024ULL << 20), adev->mc.mc_vram_size);
 350	else
 351		adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 352
 353	gmc_v6_0_vram_gtt_location(adev, &adev->mc);
 354
 355	return 0;
 356}
 357
 358static void gmc_v6_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
 359					uint32_t vmid)
 
 
 
 
 
 
 360{
 361	WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
 362
 363	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 
 
 
 
 
 
 
 
 
 
 364}
 365
 366static int gmc_v6_0_gart_set_pte_pde(struct amdgpu_device *adev,
 367				     void *cpu_pt_addr,
 368				     uint32_t gpu_page_idx,
 369				     uint64_t addr,
 370				     uint32_t flags)
 371{
 372	void __iomem *ptr = (void *)cpu_pt_addr;
 373	uint64_t value;
 374
 375	value = addr & 0xFFFFFFFFFFFFF000ULL;
 376	value |= flags;
 377	writeq(value, ptr + (gpu_page_idx * 8));
 
 
 378
 379	return 0;
 
 
 
 380}
 381
 382static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device *adev,
 383					      bool value)
 384{
 385	u32 tmp;
 386
 387	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 388	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 389			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 390	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 391			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 392	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 393			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 394	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 395			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 396	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 397			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 398	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 399			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 400	WREG32(mmVM_CONTEXT1_CNTL, tmp);
 401}
 402
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 403static int gmc_v6_0_gart_enable(struct amdgpu_device *adev)
 404{
 
 405	int r, i;
 
 406
 407	if (adev->gart.robj == NULL) {
 408		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
 409		return -EINVAL;
 410	}
 411	r = amdgpu_gart_table_vram_pin(adev);
 412	if (r)
 413		return r;
 
 
 
 414	/* Setup TLB control */
 415	WREG32(mmMC_VM_MX_L1_TLB_CNTL,
 416	       (0xA << 7) |
 417	       MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK |
 418	       MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK |
 419	       MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
 420	       MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK |
 421	       (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
 422	/* Setup L2 cache */
 423	WREG32(mmVM_L2_CNTL,
 424	       VM_L2_CNTL__ENABLE_L2_CACHE_MASK |
 425	       VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK |
 426	       VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 427	       VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 428	       (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
 429	       (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
 430	WREG32(mmVM_L2_CNTL2,
 431	       VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK |
 432	       VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK);
 
 
 433	WREG32(mmVM_L2_CNTL3,
 434	       VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
 435	       (4UL << VM_L2_CNTL3__BANK_SELECT__SHIFT) |
 436	       (4UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
 437	/* setup context0 */
 438	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
 439	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
 440	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
 441	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 442			(u32)(adev->dummy_page.addr >> 12));
 443	WREG32(mmVM_CONTEXT0_CNTL2, 0);
 444	WREG32(mmVM_CONTEXT0_CNTL,
 445	       VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK |
 446	       (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
 447	       VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK);
 448
 449	WREG32(0x575, 0);
 450	WREG32(0x576, 0);
 451	WREG32(0x577, 0);
 452
 453	/* empty context1-15 */
 454	/* set vm size, must be a multiple of 4 */
 455	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
 456	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
 457	/* Assign the pt base to something valid for now; the pts used for
 458	 * the VMs are determined by the application and setup and assigned
 459	 * on the fly in the vm part of radeon_gart.c
 460	 */
 461	for (i = 1; i < 16; i++) {
 462		if (i < 8)
 463			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
 464			       adev->gart.table_addr >> 12);
 465		else
 466			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
 467			       adev->gart.table_addr >> 12);
 468	}
 469
 470	/* enable context1-15 */
 471	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
 472	       (u32)(adev->dummy_page.addr >> 12));
 473	WREG32(mmVM_CONTEXT1_CNTL2, 4);
 474	WREG32(mmVM_CONTEXT1_CNTL,
 475	       VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
 476	       (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
 477	       ((amdgpu_vm_block_size - 9) << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
 
 478	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 479		gmc_v6_0_set_fault_enable_default(adev, false);
 480	else
 481		gmc_v6_0_set_fault_enable_default(adev, true);
 482
 483	gmc_v6_0_gart_flush_gpu_tlb(adev, 0);
 484	dev_info(adev->dev, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
 485		 (unsigned)(adev->mc.gtt_size >> 20),
 486		 (unsigned long long)adev->gart.table_addr);
 487	adev->gart.ready = true;
 488	return 0;
 489}
 490
 491static int gmc_v6_0_gart_init(struct amdgpu_device *adev)
 492{
 493	int r;
 494
 495	if (adev->gart.robj) {
 496		dev_warn(adev->dev, "gmc_v6_0 PCIE GART already initialized\n");
 497		return 0;
 498	}
 499	r = amdgpu_gart_init(adev);
 500	if (r)
 501		return r;
 502	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
 
 503	return amdgpu_gart_table_vram_alloc(adev);
 504}
 505
 506static void gmc_v6_0_gart_disable(struct amdgpu_device *adev)
 507{
 508	/*unsigned i;
 509
 510	for (i = 1; i < 16; ++i) {
 511		uint32_t reg;
 512		if (i < 8)
 513			reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
 514		else
 515			reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
 516		adev->vm_manager.saved_table_addr[i] = RREG32(reg);
 517	}*/
 518
 519	/* Disable all tables */
 520	WREG32(mmVM_CONTEXT0_CNTL, 0);
 521	WREG32(mmVM_CONTEXT1_CNTL, 0);
 522	/* Setup TLB control */
 523	WREG32(mmMC_VM_MX_L1_TLB_CNTL,
 524	       MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK |
 525	       (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT));
 526	/* Setup L2 cache */
 527	WREG32(mmVM_L2_CNTL,
 528	       VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 529	       VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK |
 530	       (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT) |
 531	       (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT));
 532	WREG32(mmVM_L2_CNTL2, 0);
 533	WREG32(mmVM_L2_CNTL3,
 534	       VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK |
 535	       (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT));
 536	amdgpu_gart_table_vram_unpin(adev);
 537}
 538
 539static void gmc_v6_0_gart_fini(struct amdgpu_device *adev)
 540{
 541	amdgpu_gart_table_vram_free(adev);
 542	amdgpu_gart_fini(adev);
 543}
 544
 545static int gmc_v6_0_vm_init(struct amdgpu_device *adev)
 546{
 547	/*
 548	 * number of VMs
 549	 * VMID 0 is reserved for System
 550	 * amdgpu graphics/compute will use VMIDs 1-7
 551	 * amdkfd will use VMIDs 8-15
 552	 */
 553	adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS;
 554	amdgpu_vm_manager_init(adev);
 555
 556	/* base offset of vram pages */
 557	if (adev->flags & AMD_IS_APU) {
 558		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
 559		tmp <<= 22;
 560		adev->vm_manager.vram_base_offset = tmp;
 561	} else
 562		adev->vm_manager.vram_base_offset = 0;
 563
 564	return 0;
 565}
 566
 567static void gmc_v6_0_vm_fini(struct amdgpu_device *adev)
 568{
 569}
 570
 571static void gmc_v6_0_vm_decode_fault(struct amdgpu_device *adev,
 572				     u32 status, u32 addr, u32 mc_client)
 573{
 574	u32 mc_id;
 575	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
 576	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 577					PROTECTIONS);
 578	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
 579		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
 580
 581	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 582			      MEMORY_CLIENT_ID);
 583
 584	dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
 585	       protections, vmid, addr,
 586	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 587			     MEMORY_CLIENT_RW) ?
 588	       "write" : "read", block, mc_client, mc_id);
 589}
 590
 591/*
 592static const u32 mc_cg_registers[] = {
 593	MC_HUB_MISC_HUB_CG,
 594	MC_HUB_MISC_SIP_CG,
 595	MC_HUB_MISC_VM_CG,
 596	MC_XPB_CLK_GAT,
 597	ATC_MISC_CG,
 598	MC_CITF_MISC_WR_CG,
 599	MC_CITF_MISC_RD_CG,
 600	MC_CITF_MISC_VM_CG,
 601	VM_L2_CG,
 602};
 603
 604static const u32 mc_cg_ls_en[] = {
 605	MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
 606	MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
 607	MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
 608	MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
 609	ATC_MISC_CG__MEM_LS_ENABLE_MASK,
 610	MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
 611	MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
 612	MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
 613	VM_L2_CG__MEM_LS_ENABLE_MASK,
 614};
 615
 616static const u32 mc_cg_en[] = {
 617	MC_HUB_MISC_HUB_CG__ENABLE_MASK,
 618	MC_HUB_MISC_SIP_CG__ENABLE_MASK,
 619	MC_HUB_MISC_VM_CG__ENABLE_MASK,
 620	MC_XPB_CLK_GAT__ENABLE_MASK,
 621	ATC_MISC_CG__ENABLE_MASK,
 622	MC_CITF_MISC_WR_CG__ENABLE_MASK,
 623	MC_CITF_MISC_RD_CG__ENABLE_MASK,
 624	MC_CITF_MISC_VM_CG__ENABLE_MASK,
 625	VM_L2_CG__ENABLE_MASK,
 626};
 627
 628static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
 629				  bool enable)
 630{
 631	int i;
 632	u32 orig, data;
 633
 634	for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
 635		orig = data = RREG32(mc_cg_registers[i]);
 636		if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
 637			data |= mc_cg_ls_en[i];
 638		else
 639			data &= ~mc_cg_ls_en[i];
 640		if (data != orig)
 641			WREG32(mc_cg_registers[i], data);
 642	}
 643}
 644
 645static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
 646				    bool enable)
 647{
 648	int i;
 649	u32 orig, data;
 650
 651	for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
 652		orig = data = RREG32(mc_cg_registers[i]);
 653		if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
 654			data |= mc_cg_en[i];
 655		else
 656			data &= ~mc_cg_en[i];
 657		if (data != orig)
 658			WREG32(mc_cg_registers[i], data);
 659	}
 660}
 661
 662static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
 663				     bool enable)
 664{
 665	u32 orig, data;
 666
 667	orig = data = RREG32_PCIE(ixPCIE_CNTL2);
 668
 669	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
 670		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
 671		data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
 672		data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
 673		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
 674	} else {
 675		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
 676		data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
 677		data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
 678		data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
 679	}
 680
 681	if (orig != data)
 682		WREG32_PCIE(ixPCIE_CNTL2, data);
 683}
 684
 685static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
 686				     bool enable)
 687{
 688	u32 orig, data;
 689
 690	orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
 691
 692	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
 693		data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
 694	else
 695		data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
 696
 697	if (orig != data)
 698		WREG32(mmHDP_HOST_PATH_CNTL, data);
 699}
 700
 701static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
 702				   bool enable)
 703{
 704	u32 orig, data;
 705
 706	orig = data = RREG32(mmHDP_MEM_POWER_LS);
 707
 708	if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
 709		data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
 710	else
 711		data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
 712
 713	if (orig != data)
 714		WREG32(mmHDP_MEM_POWER_LS, data);
 715}
 716*/
 717
 718static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type)
 719{
 720	switch (mc_seq_vram_type) {
 721	case MC_SEQ_MISC0__MT__GDDR1:
 722		return AMDGPU_VRAM_TYPE_GDDR1;
 723	case MC_SEQ_MISC0__MT__DDR2:
 724		return AMDGPU_VRAM_TYPE_DDR2;
 725	case MC_SEQ_MISC0__MT__GDDR3:
 726		return AMDGPU_VRAM_TYPE_GDDR3;
 727	case MC_SEQ_MISC0__MT__GDDR4:
 728		return AMDGPU_VRAM_TYPE_GDDR4;
 729	case MC_SEQ_MISC0__MT__GDDR5:
 730		return AMDGPU_VRAM_TYPE_GDDR5;
 731	case MC_SEQ_MISC0__MT__DDR3:
 732		return AMDGPU_VRAM_TYPE_DDR3;
 733	default:
 734		return AMDGPU_VRAM_TYPE_UNKNOWN;
 735	}
 736}
 737
 738static int gmc_v6_0_early_init(void *handle)
 739{
 740	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 741
 742	gmc_v6_0_set_gart_funcs(adev);
 743	gmc_v6_0_set_irq_funcs(adev);
 744
 745	if (adev->flags & AMD_IS_APU) {
 746		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
 747	} else {
 748		u32 tmp = RREG32(mmMC_SEQ_MISC0);
 749		tmp &= MC_SEQ_MISC0__MT__MASK;
 750		adev->mc.vram_type = gmc_v6_0_convert_vram_type(tmp);
 751	}
 752
 753	return 0;
 754}
 755
 756static int gmc_v6_0_late_init(void *handle)
 757{
 758	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 759
 
 
 760	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
 761		return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
 762	else
 763		return 0;
 764}
 765
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 766static int gmc_v6_0_sw_init(void *handle)
 767{
 768	int r;
 769	int dma_bits;
 770	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 771
 772	r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault);
 
 
 
 
 
 
 
 
 
 
 773	if (r)
 774		return r;
 775
 776	r = amdgpu_irq_add_id(adev, 147, &adev->mc.vm_fault);
 777	if (r)
 778		return r;
 779
 780	adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
 781
 782	adev->mc.mc_mask = 0xffffffffffULL;
 783
 784	adev->need_dma32 = false;
 785	dma_bits = adev->need_dma32 ? 32 : 40;
 786	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
 787	if (r) {
 788		adev->need_dma32 = true;
 789		dma_bits = 32;
 790		dev_warn(adev->dev, "amdgpu: No suitable DMA available.\n");
 
 791	}
 792	r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
 793	if (r) {
 794		pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
 795		dev_warn(adev->dev, "amdgpu: No coherent DMA available.\n");
 796	}
 797
 798	r = gmc_v6_0_init_microcode(adev);
 799	if (r) {
 800		dev_err(adev->dev, "Failed to load mc firmware!\n");
 801		return r;
 802	}
 803
 804	r = gmc_v6_0_mc_init(adev);
 805	if (r)
 806		return r;
 807
 
 
 808	r = amdgpu_bo_init(adev);
 809	if (r)
 810		return r;
 811
 812	r = gmc_v6_0_gart_init(adev);
 813	if (r)
 814		return r;
 815
 816	if (!adev->vm_manager.enabled) {
 817		r = gmc_v6_0_vm_init(adev);
 818		if (r) {
 819			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
 820			return r;
 821		}
 822		adev->vm_manager.enabled = true;
 
 
 
 
 
 
 
 
 
 
 823	}
 824
 825	return r;
 826}
 827
 828static int gmc_v6_0_sw_fini(void *handle)
 829{
 830	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 831
 832	if (adev->vm_manager.enabled) {
 833		gmc_v6_0_vm_fini(adev);
 834		adev->vm_manager.enabled = false;
 835	}
 836	gmc_v6_0_gart_fini(adev);
 837	amdgpu_gem_force_release(adev);
 
 
 838	amdgpu_bo_fini(adev);
 
 
 
 839
 840	return 0;
 841}
 842
 843static int gmc_v6_0_hw_init(void *handle)
 844{
 845	int r;
 846	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 847
 848	gmc_v6_0_mc_program(adev);
 849
 850	if (!(adev->flags & AMD_IS_APU)) {
 851		r = gmc_v6_0_mc_load_microcode(adev);
 852		if (r) {
 853			dev_err(adev->dev, "Failed to load MC firmware!\n");
 854			return r;
 855		}
 856	}
 857
 858	r = gmc_v6_0_gart_enable(adev);
 859	if (r)
 860		return r;
 861
 862	return r;
 863}
 864
 865static int gmc_v6_0_hw_fini(void *handle)
 866{
 867	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 868
 869	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
 870	gmc_v6_0_gart_disable(adev);
 871
 872	return 0;
 873}
 874
 875static int gmc_v6_0_suspend(void *handle)
 876{
 877	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 878
 879	if (adev->vm_manager.enabled) {
 880		gmc_v6_0_vm_fini(adev);
 881		adev->vm_manager.enabled = false;
 882	}
 883	gmc_v6_0_hw_fini(adev);
 884
 885	return 0;
 886}
 887
 888static int gmc_v6_0_resume(void *handle)
 889{
 890	int r;
 891	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 892
 893	r = gmc_v6_0_hw_init(adev);
 894	if (r)
 895		return r;
 896
 897	if (!adev->vm_manager.enabled) {
 898		r = gmc_v6_0_vm_init(adev);
 899		if (r) {
 900			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
 901			return r;
 902		}
 903		adev->vm_manager.enabled = true;
 904	}
 905
 906	return r;
 907}
 908
 909static bool gmc_v6_0_is_idle(void *handle)
 910{
 911	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 912	u32 tmp = RREG32(mmSRBM_STATUS);
 913
 914	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 915		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
 916		return false;
 917
 918	return true;
 919}
 920
 921static int gmc_v6_0_wait_for_idle(void *handle)
 922{
 923	unsigned i;
 924	u32 tmp;
 925	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 926
 927	for (i = 0; i < adev->usec_timeout; i++) {
 928		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
 929					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 930					       SRBM_STATUS__MCC_BUSY_MASK |
 931					       SRBM_STATUS__MCD_BUSY_MASK |
 932					       SRBM_STATUS__VMC_BUSY_MASK);
 933		if (!tmp)
 934			return 0;
 935		udelay(1);
 936	}
 937	return -ETIMEDOUT;
 938
 939}
 940
 941static int gmc_v6_0_soft_reset(void *handle)
 942{
 943	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 944	struct amdgpu_mode_mc_save save;
 945	u32 srbm_soft_reset = 0;
 946	u32 tmp = RREG32(mmSRBM_STATUS);
 947
 948	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
 949		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
 950						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
 951
 952	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 953		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
 954		if (!(adev->flags & AMD_IS_APU))
 955			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
 956							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
 957	}
 958
 959	if (srbm_soft_reset) {
 960		gmc_v6_0_mc_stop(adev, &save);
 961		if (gmc_v6_0_wait_for_idle(adev)) {
 962			dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
 963		}
 964
 965
 966		tmp = RREG32(mmSRBM_SOFT_RESET);
 967		tmp |= srbm_soft_reset;
 968		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
 969		WREG32(mmSRBM_SOFT_RESET, tmp);
 970		tmp = RREG32(mmSRBM_SOFT_RESET);
 971
 972		udelay(50);
 973
 974		tmp &= ~srbm_soft_reset;
 975		WREG32(mmSRBM_SOFT_RESET, tmp);
 976		tmp = RREG32(mmSRBM_SOFT_RESET);
 977
 978		udelay(50);
 979
 980		gmc_v6_0_mc_resume(adev, &save);
 981		udelay(50);
 982	}
 983
 984	return 0;
 985}
 986
 987static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
 988					     struct amdgpu_irq_src *src,
 989					     unsigned type,
 990					     enum amdgpu_interrupt_state state)
 991{
 992	u32 tmp;
 993	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
 994		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
 995		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
 996		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
 997		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
 998		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
 999
1000	switch (state) {
1001	case AMDGPU_IRQ_STATE_DISABLE:
1002		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1003		tmp &= ~bits;
1004		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1005		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1006		tmp &= ~bits;
1007		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1008		break;
1009	case AMDGPU_IRQ_STATE_ENABLE:
1010		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1011		tmp |= bits;
1012		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1013		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1014		tmp |= bits;
1015		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1016		break;
1017	default:
1018		break;
1019	}
1020
1021	return 0;
1022}
1023
1024static int gmc_v6_0_process_interrupt(struct amdgpu_device *adev,
1025				      struct amdgpu_irq_src *source,
1026				      struct amdgpu_iv_entry *entry)
1027{
1028	u32 addr, status;
1029
1030	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1031	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1032	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1033
1034	if (!addr && !status)
1035		return 0;
1036
1037	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1038		gmc_v6_0_set_fault_enable_default(adev, false);
1039
1040	if (printk_ratelimit()) {
1041		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1042			entry->src_id, entry->src_data);
1043		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1044			addr);
1045		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1046			status);
1047		gmc_v6_0_vm_decode_fault(adev, status, addr, 0);
1048	}
1049
1050	return 0;
1051}
1052
1053static int gmc_v6_0_set_clockgating_state(void *handle,
1054					  enum amd_clockgating_state state)
1055{
1056	return 0;
1057}
1058
1059static int gmc_v6_0_set_powergating_state(void *handle,
1060					  enum amd_powergating_state state)
1061{
1062	return 0;
1063}
1064
1065static const struct amd_ip_funcs gmc_v6_0_ip_funcs = {
1066	.name = "gmc_v6_0",
1067	.early_init = gmc_v6_0_early_init,
1068	.late_init = gmc_v6_0_late_init,
1069	.sw_init = gmc_v6_0_sw_init,
1070	.sw_fini = gmc_v6_0_sw_fini,
1071	.hw_init = gmc_v6_0_hw_init,
1072	.hw_fini = gmc_v6_0_hw_fini,
1073	.suspend = gmc_v6_0_suspend,
1074	.resume = gmc_v6_0_resume,
1075	.is_idle = gmc_v6_0_is_idle,
1076	.wait_for_idle = gmc_v6_0_wait_for_idle,
1077	.soft_reset = gmc_v6_0_soft_reset,
1078	.set_clockgating_state = gmc_v6_0_set_clockgating_state,
1079	.set_powergating_state = gmc_v6_0_set_powergating_state,
1080};
1081
1082static const struct amdgpu_gart_funcs gmc_v6_0_gart_funcs = {
1083	.flush_gpu_tlb = gmc_v6_0_gart_flush_gpu_tlb,
1084	.set_pte_pde = gmc_v6_0_gart_set_pte_pde,
 
 
 
1085};
1086
1087static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs = {
1088	.set = gmc_v6_0_vm_fault_interrupt_state,
1089	.process = gmc_v6_0_process_interrupt,
1090};
1091
1092static void gmc_v6_0_set_gart_funcs(struct amdgpu_device *adev)
1093{
1094	if (adev->gart.gart_funcs == NULL)
1095		adev->gart.gart_funcs = &gmc_v6_0_gart_funcs;
1096}
1097
1098static void gmc_v6_0_set_irq_funcs(struct amdgpu_device *adev)
1099{
1100	adev->mc.vm_fault.num_types = 1;
1101	adev->mc.vm_fault.funcs = &gmc_v6_0_irq_funcs;
1102}
1103
1104const struct amdgpu_ip_block_version gmc_v6_0_ip_block =
1105{
1106	.type = AMD_IP_BLOCK_TYPE_GMC,
1107	.major = 6,
1108	.minor = 0,
1109	.rev = 0,
1110	.funcs = &gmc_v6_0_ip_funcs,
1111};