Loading...
1/*
2 * Copyright 2016 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/kernel.h>
24#include <linux/firmware.h>
25#include <drm/drmP.h>
26#include "amdgpu.h"
27#include "amdgpu_gfx.h"
28#include "soc15.h"
29#include "soc15d.h"
30
31#include "gc/gc_9_0_offset.h"
32#include "gc/gc_9_0_sh_mask.h"
33#include "vega10_enum.h"
34#include "hdp/hdp_4_0_offset.h"
35
36#include "soc15_common.h"
37#include "clearstate_gfx9.h"
38#include "v9_structs.h"
39
40#define GFX9_NUM_GFX_RINGS 1
41#define GFX9_MEC_HPD_SIZE 2048
42#define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
43#define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
44#define GFX9_RLC_FORMAT_DIRECT_REG_LIST_LENGTH 34
45
46#define mmPWR_MISC_CNTL_STATUS 0x0183
47#define mmPWR_MISC_CNTL_STATUS_BASE_IDX 0
48#define PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN__SHIFT 0x0
49#define PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT 0x1
50#define PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK 0x00000001L
51#define PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK 0x00000006L
52
53MODULE_FIRMWARE("amdgpu/vega10_ce.bin");
54MODULE_FIRMWARE("amdgpu/vega10_pfp.bin");
55MODULE_FIRMWARE("amdgpu/vega10_me.bin");
56MODULE_FIRMWARE("amdgpu/vega10_mec.bin");
57MODULE_FIRMWARE("amdgpu/vega10_mec2.bin");
58MODULE_FIRMWARE("amdgpu/vega10_rlc.bin");
59
60MODULE_FIRMWARE("amdgpu/vega12_ce.bin");
61MODULE_FIRMWARE("amdgpu/vega12_pfp.bin");
62MODULE_FIRMWARE("amdgpu/vega12_me.bin");
63MODULE_FIRMWARE("amdgpu/vega12_mec.bin");
64MODULE_FIRMWARE("amdgpu/vega12_mec2.bin");
65MODULE_FIRMWARE("amdgpu/vega12_rlc.bin");
66
67MODULE_FIRMWARE("amdgpu/raven_ce.bin");
68MODULE_FIRMWARE("amdgpu/raven_pfp.bin");
69MODULE_FIRMWARE("amdgpu/raven_me.bin");
70MODULE_FIRMWARE("amdgpu/raven_mec.bin");
71MODULE_FIRMWARE("amdgpu/raven_mec2.bin");
72MODULE_FIRMWARE("amdgpu/raven_rlc.bin");
73
74static const struct soc15_reg_golden golden_settings_gc_9_0[] =
75{
76 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
77 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
78 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
79 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
80 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
81 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
82 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
83 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
84 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
85 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
86 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
87 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
88 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
89 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
90 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
91 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
92 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
93 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
94 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
95 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
96 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
97 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
98 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
99};
100
101static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
102{
103 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107),
104 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
105 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
106 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042),
107 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000),
108 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
109 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800)
110};
111
112static const struct soc15_reg_golden golden_settings_gc_9_1[] =
113{
114 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
115 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
116 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
117 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
118 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
119 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
120 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
121 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
122 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
123 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
124 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
125 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
126 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
127 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
128 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
129 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
130 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
131 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
132 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
133 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff),
134 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
135};
136
137static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
138{
139 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
140 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042),
141 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042),
142 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000),
143 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000),
144 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
145 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800)
146};
147
148static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
149{
150 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000),
151 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382)
152};
153
154static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
155{
156 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
157 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
158 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
159 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
160 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
161 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
162 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
163 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
164 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
165 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
166 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
167 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
168 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
169 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
170 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
171 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
172};
173
174static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
175{
176 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080),
177 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
178 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
179 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041),
180 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041),
181 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
182 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107),
183 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
184 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410),
185 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000)
186};
187
188#define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
189#define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
190#define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
191
192static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
193static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
194static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
195static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
196static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
197 struct amdgpu_cu_info *cu_info);
198static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
199static void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance);
200static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring);
201
202static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
203{
204 switch (adev->asic_type) {
205 case CHIP_VEGA10:
206 soc15_program_register_sequence(adev,
207 golden_settings_gc_9_0,
208 ARRAY_SIZE(golden_settings_gc_9_0));
209 soc15_program_register_sequence(adev,
210 golden_settings_gc_9_0_vg10,
211 ARRAY_SIZE(golden_settings_gc_9_0_vg10));
212 break;
213 case CHIP_VEGA12:
214 soc15_program_register_sequence(adev,
215 golden_settings_gc_9_2_1,
216 ARRAY_SIZE(golden_settings_gc_9_2_1));
217 soc15_program_register_sequence(adev,
218 golden_settings_gc_9_2_1_vg12,
219 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
220 break;
221 case CHIP_RAVEN:
222 soc15_program_register_sequence(adev,
223 golden_settings_gc_9_1,
224 ARRAY_SIZE(golden_settings_gc_9_1));
225 soc15_program_register_sequence(adev,
226 golden_settings_gc_9_1_rv1,
227 ARRAY_SIZE(golden_settings_gc_9_1_rv1));
228 break;
229 default:
230 break;
231 }
232
233 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
234 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
235}
236
237static void gfx_v9_0_scratch_init(struct amdgpu_device *adev)
238{
239 adev->gfx.scratch.num_reg = 8;
240 adev->gfx.scratch.reg_base = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
241 adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
242}
243
244static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
245 bool wc, uint32_t reg, uint32_t val)
246{
247 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
248 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
249 WRITE_DATA_DST_SEL(0) |
250 (wc ? WR_CONFIRM : 0));
251 amdgpu_ring_write(ring, reg);
252 amdgpu_ring_write(ring, 0);
253 amdgpu_ring_write(ring, val);
254}
255
256static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
257 int mem_space, int opt, uint32_t addr0,
258 uint32_t addr1, uint32_t ref, uint32_t mask,
259 uint32_t inv)
260{
261 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
262 amdgpu_ring_write(ring,
263 /* memory (1) or register (0) */
264 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
265 WAIT_REG_MEM_OPERATION(opt) | /* wait */
266 WAIT_REG_MEM_FUNCTION(3) | /* equal */
267 WAIT_REG_MEM_ENGINE(eng_sel)));
268
269 if (mem_space)
270 BUG_ON(addr0 & 0x3); /* Dword align */
271 amdgpu_ring_write(ring, addr0);
272 amdgpu_ring_write(ring, addr1);
273 amdgpu_ring_write(ring, ref);
274 amdgpu_ring_write(ring, mask);
275 amdgpu_ring_write(ring, inv); /* poll interval */
276}
277
278static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
279{
280 struct amdgpu_device *adev = ring->adev;
281 uint32_t scratch;
282 uint32_t tmp = 0;
283 unsigned i;
284 int r;
285
286 r = amdgpu_gfx_scratch_get(adev, &scratch);
287 if (r) {
288 DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r);
289 return r;
290 }
291 WREG32(scratch, 0xCAFEDEAD);
292 r = amdgpu_ring_alloc(ring, 3);
293 if (r) {
294 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
295 ring->idx, r);
296 amdgpu_gfx_scratch_free(adev, scratch);
297 return r;
298 }
299 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
300 amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
301 amdgpu_ring_write(ring, 0xDEADBEEF);
302 amdgpu_ring_commit(ring);
303
304 for (i = 0; i < adev->usec_timeout; i++) {
305 tmp = RREG32(scratch);
306 if (tmp == 0xDEADBEEF)
307 break;
308 DRM_UDELAY(1);
309 }
310 if (i < adev->usec_timeout) {
311 DRM_DEBUG("ring test on %d succeeded in %d usecs\n",
312 ring->idx, i);
313 } else {
314 DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
315 ring->idx, scratch, tmp);
316 r = -EINVAL;
317 }
318 amdgpu_gfx_scratch_free(adev, scratch);
319 return r;
320}
321
322static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
323{
324 struct amdgpu_device *adev = ring->adev;
325 struct amdgpu_ib ib;
326 struct dma_fence *f = NULL;
327
328 unsigned index;
329 uint64_t gpu_addr;
330 uint32_t tmp;
331 long r;
332
333 r = amdgpu_device_wb_get(adev, &index);
334 if (r) {
335 dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r);
336 return r;
337 }
338
339 gpu_addr = adev->wb.gpu_addr + (index * 4);
340 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
341 memset(&ib, 0, sizeof(ib));
342 r = amdgpu_ib_get(adev, NULL, 16, &ib);
343 if (r) {
344 DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
345 goto err1;
346 }
347 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
348 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
349 ib.ptr[2] = lower_32_bits(gpu_addr);
350 ib.ptr[3] = upper_32_bits(gpu_addr);
351 ib.ptr[4] = 0xDEADBEEF;
352 ib.length_dw = 5;
353
354 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
355 if (r)
356 goto err2;
357
358 r = dma_fence_wait_timeout(f, false, timeout);
359 if (r == 0) {
360 DRM_ERROR("amdgpu: IB test timed out.\n");
361 r = -ETIMEDOUT;
362 goto err2;
363 } else if (r < 0) {
364 DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
365 goto err2;
366 }
367
368 tmp = adev->wb.wb[index];
369 if (tmp == 0xDEADBEEF) {
370 DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx);
371 r = 0;
372 } else {
373 DRM_ERROR("ib test on ring %d failed\n", ring->idx);
374 r = -EINVAL;
375 }
376
377err2:
378 amdgpu_ib_free(adev, &ib, NULL);
379 dma_fence_put(f);
380err1:
381 amdgpu_device_wb_free(adev, index);
382 return r;
383}
384
385
386static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
387{
388 release_firmware(adev->gfx.pfp_fw);
389 adev->gfx.pfp_fw = NULL;
390 release_firmware(adev->gfx.me_fw);
391 adev->gfx.me_fw = NULL;
392 release_firmware(adev->gfx.ce_fw);
393 adev->gfx.ce_fw = NULL;
394 release_firmware(adev->gfx.rlc_fw);
395 adev->gfx.rlc_fw = NULL;
396 release_firmware(adev->gfx.mec_fw);
397 adev->gfx.mec_fw = NULL;
398 release_firmware(adev->gfx.mec2_fw);
399 adev->gfx.mec2_fw = NULL;
400
401 kfree(adev->gfx.rlc.register_list_format);
402}
403
404static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
405{
406 const char *chip_name;
407 char fw_name[30];
408 int err;
409 struct amdgpu_firmware_info *info = NULL;
410 const struct common_firmware_header *header = NULL;
411 const struct gfx_firmware_header_v1_0 *cp_hdr;
412 const struct rlc_firmware_header_v2_0 *rlc_hdr;
413 unsigned int *tmp = NULL;
414 unsigned int i = 0;
415
416 DRM_DEBUG("\n");
417
418 switch (adev->asic_type) {
419 case CHIP_VEGA10:
420 chip_name = "vega10";
421 break;
422 case CHIP_VEGA12:
423 chip_name = "vega12";
424 break;
425 case CHIP_RAVEN:
426 chip_name = "raven";
427 break;
428 default:
429 BUG();
430 }
431
432 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
433 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
434 if (err)
435 goto out;
436 err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
437 if (err)
438 goto out;
439 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
440 adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
441 adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
442
443 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
444 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
445 if (err)
446 goto out;
447 err = amdgpu_ucode_validate(adev->gfx.me_fw);
448 if (err)
449 goto out;
450 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
451 adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
452 adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
453
454 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
455 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
456 if (err)
457 goto out;
458 err = amdgpu_ucode_validate(adev->gfx.ce_fw);
459 if (err)
460 goto out;
461 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
462 adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
463 adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
464
465 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
466 err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
467 if (err)
468 goto out;
469 err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
470 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
471 adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
472 adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
473 adev->gfx.rlc.save_and_restore_offset =
474 le32_to_cpu(rlc_hdr->save_and_restore_offset);
475 adev->gfx.rlc.clear_state_descriptor_offset =
476 le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
477 adev->gfx.rlc.avail_scratch_ram_locations =
478 le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
479 adev->gfx.rlc.reg_restore_list_size =
480 le32_to_cpu(rlc_hdr->reg_restore_list_size);
481 adev->gfx.rlc.reg_list_format_start =
482 le32_to_cpu(rlc_hdr->reg_list_format_start);
483 adev->gfx.rlc.reg_list_format_separate_start =
484 le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
485 adev->gfx.rlc.starting_offsets_start =
486 le32_to_cpu(rlc_hdr->starting_offsets_start);
487 adev->gfx.rlc.reg_list_format_size_bytes =
488 le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
489 adev->gfx.rlc.reg_list_size_bytes =
490 le32_to_cpu(rlc_hdr->reg_list_size_bytes);
491 adev->gfx.rlc.register_list_format =
492 kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
493 adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
494 if (!adev->gfx.rlc.register_list_format) {
495 err = -ENOMEM;
496 goto out;
497 }
498
499 tmp = (unsigned int *)((uintptr_t)rlc_hdr +
500 le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
501 for (i = 0 ; i < (rlc_hdr->reg_list_format_size_bytes >> 2); i++)
502 adev->gfx.rlc.register_list_format[i] = le32_to_cpu(tmp[i]);
503
504 adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
505
506 tmp = (unsigned int *)((uintptr_t)rlc_hdr +
507 le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
508 for (i = 0 ; i < (rlc_hdr->reg_list_size_bytes >> 2); i++)
509 adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
510
511 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
512 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
513 if (err)
514 goto out;
515 err = amdgpu_ucode_validate(adev->gfx.mec_fw);
516 if (err)
517 goto out;
518 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
519 adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
520 adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
521
522
523 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
524 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
525 if (!err) {
526 err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
527 if (err)
528 goto out;
529 cp_hdr = (const struct gfx_firmware_header_v1_0 *)
530 adev->gfx.mec2_fw->data;
531 adev->gfx.mec2_fw_version =
532 le32_to_cpu(cp_hdr->header.ucode_version);
533 adev->gfx.mec2_feature_version =
534 le32_to_cpu(cp_hdr->ucode_feature_version);
535 } else {
536 err = 0;
537 adev->gfx.mec2_fw = NULL;
538 }
539
540 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
541 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
542 info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
543 info->fw = adev->gfx.pfp_fw;
544 header = (const struct common_firmware_header *)info->fw->data;
545 adev->firmware.fw_size +=
546 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
547
548 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
549 info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
550 info->fw = adev->gfx.me_fw;
551 header = (const struct common_firmware_header *)info->fw->data;
552 adev->firmware.fw_size +=
553 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
554
555 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
556 info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
557 info->fw = adev->gfx.ce_fw;
558 header = (const struct common_firmware_header *)info->fw->data;
559 adev->firmware.fw_size +=
560 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
561
562 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
563 info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
564 info->fw = adev->gfx.rlc_fw;
565 header = (const struct common_firmware_header *)info->fw->data;
566 adev->firmware.fw_size +=
567 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
568
569 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
570 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
571 info->fw = adev->gfx.mec_fw;
572 header = (const struct common_firmware_header *)info->fw->data;
573 cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
574 adev->firmware.fw_size +=
575 ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
576
577 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1_JT];
578 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1_JT;
579 info->fw = adev->gfx.mec_fw;
580 adev->firmware.fw_size +=
581 ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
582
583 if (adev->gfx.mec2_fw) {
584 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
585 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
586 info->fw = adev->gfx.mec2_fw;
587 header = (const struct common_firmware_header *)info->fw->data;
588 cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
589 adev->firmware.fw_size +=
590 ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
591 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2_JT];
592 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2_JT;
593 info->fw = adev->gfx.mec2_fw;
594 adev->firmware.fw_size +=
595 ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
596 }
597
598 }
599
600out:
601 if (err) {
602 dev_err(adev->dev,
603 "gfx9: Failed to load firmware \"%s\"\n",
604 fw_name);
605 release_firmware(adev->gfx.pfp_fw);
606 adev->gfx.pfp_fw = NULL;
607 release_firmware(adev->gfx.me_fw);
608 adev->gfx.me_fw = NULL;
609 release_firmware(adev->gfx.ce_fw);
610 adev->gfx.ce_fw = NULL;
611 release_firmware(adev->gfx.rlc_fw);
612 adev->gfx.rlc_fw = NULL;
613 release_firmware(adev->gfx.mec_fw);
614 adev->gfx.mec_fw = NULL;
615 release_firmware(adev->gfx.mec2_fw);
616 adev->gfx.mec2_fw = NULL;
617 }
618 return err;
619}
620
621static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
622{
623 u32 count = 0;
624 const struct cs_section_def *sect = NULL;
625 const struct cs_extent_def *ext = NULL;
626
627 /* begin clear state */
628 count += 2;
629 /* context control state */
630 count += 3;
631
632 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
633 for (ext = sect->section; ext->extent != NULL; ++ext) {
634 if (sect->id == SECT_CONTEXT)
635 count += 2 + ext->reg_count;
636 else
637 return 0;
638 }
639 }
640
641 /* end clear state */
642 count += 2;
643 /* clear state */
644 count += 2;
645
646 return count;
647}
648
649static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
650 volatile u32 *buffer)
651{
652 u32 count = 0, i;
653 const struct cs_section_def *sect = NULL;
654 const struct cs_extent_def *ext = NULL;
655
656 if (adev->gfx.rlc.cs_data == NULL)
657 return;
658 if (buffer == NULL)
659 return;
660
661 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
662 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
663
664 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
665 buffer[count++] = cpu_to_le32(0x80000000);
666 buffer[count++] = cpu_to_le32(0x80000000);
667
668 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
669 for (ext = sect->section; ext->extent != NULL; ++ext) {
670 if (sect->id == SECT_CONTEXT) {
671 buffer[count++] =
672 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
673 buffer[count++] = cpu_to_le32(ext->reg_index -
674 PACKET3_SET_CONTEXT_REG_START);
675 for (i = 0; i < ext->reg_count; i++)
676 buffer[count++] = cpu_to_le32(ext->extent[i]);
677 } else {
678 return;
679 }
680 }
681 }
682
683 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
684 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
685
686 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
687 buffer[count++] = cpu_to_le32(0);
688}
689
690static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
691{
692 uint32_t data;
693
694 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
695 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
696 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7);
697 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
698 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16));
699
700 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
701 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
702
703 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
704 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500);
705
706 mutex_lock(&adev->grbm_idx_mutex);
707 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
708 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
709 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
710
711 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
712 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
713 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
714 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
715 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
716
717 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
718 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
719 data &= 0x0000FFFF;
720 data |= 0x00C00000;
721 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
722
723 /* set RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF */
724 WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, 0xFFF);
725
726 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
727 * but used for RLC_LB_CNTL configuration */
728 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
729 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
730 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
731 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
732 mutex_unlock(&adev->grbm_idx_mutex);
733}
734
735static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
736{
737 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0);
738}
739
740static void rv_init_cp_jump_table(struct amdgpu_device *adev)
741{
742 const __le32 *fw_data;
743 volatile u32 *dst_ptr;
744 int me, i, max_me = 5;
745 u32 bo_offset = 0;
746 u32 table_offset, table_size;
747
748 /* write the cp table buffer */
749 dst_ptr = adev->gfx.rlc.cp_table_ptr;
750 for (me = 0; me < max_me; me++) {
751 if (me == 0) {
752 const struct gfx_firmware_header_v1_0 *hdr =
753 (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
754 fw_data = (const __le32 *)
755 (adev->gfx.ce_fw->data +
756 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
757 table_offset = le32_to_cpu(hdr->jt_offset);
758 table_size = le32_to_cpu(hdr->jt_size);
759 } else if (me == 1) {
760 const struct gfx_firmware_header_v1_0 *hdr =
761 (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
762 fw_data = (const __le32 *)
763 (adev->gfx.pfp_fw->data +
764 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
765 table_offset = le32_to_cpu(hdr->jt_offset);
766 table_size = le32_to_cpu(hdr->jt_size);
767 } else if (me == 2) {
768 const struct gfx_firmware_header_v1_0 *hdr =
769 (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
770 fw_data = (const __le32 *)
771 (adev->gfx.me_fw->data +
772 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
773 table_offset = le32_to_cpu(hdr->jt_offset);
774 table_size = le32_to_cpu(hdr->jt_size);
775 } else if (me == 3) {
776 const struct gfx_firmware_header_v1_0 *hdr =
777 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
778 fw_data = (const __le32 *)
779 (adev->gfx.mec_fw->data +
780 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
781 table_offset = le32_to_cpu(hdr->jt_offset);
782 table_size = le32_to_cpu(hdr->jt_size);
783 } else if (me == 4) {
784 const struct gfx_firmware_header_v1_0 *hdr =
785 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
786 fw_data = (const __le32 *)
787 (adev->gfx.mec2_fw->data +
788 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
789 table_offset = le32_to_cpu(hdr->jt_offset);
790 table_size = le32_to_cpu(hdr->jt_size);
791 }
792
793 for (i = 0; i < table_size; i ++) {
794 dst_ptr[bo_offset + i] =
795 cpu_to_le32(le32_to_cpu(fw_data[table_offset + i]));
796 }
797
798 bo_offset += table_size;
799 }
800}
801
802static void gfx_v9_0_rlc_fini(struct amdgpu_device *adev)
803{
804 /* clear state block */
805 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
806 &adev->gfx.rlc.clear_state_gpu_addr,
807 (void **)&adev->gfx.rlc.cs_ptr);
808
809 /* jump table block */
810 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
811 &adev->gfx.rlc.cp_table_gpu_addr,
812 (void **)&adev->gfx.rlc.cp_table_ptr);
813}
814
815static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
816{
817 volatile u32 *dst_ptr;
818 u32 dws;
819 const struct cs_section_def *cs_data;
820 int r;
821
822 adev->gfx.rlc.cs_data = gfx9_cs_data;
823
824 cs_data = adev->gfx.rlc.cs_data;
825
826 if (cs_data) {
827 /* clear state block */
828 adev->gfx.rlc.clear_state_size = dws = gfx_v9_0_get_csb_size(adev);
829 r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE,
830 AMDGPU_GEM_DOMAIN_VRAM,
831 &adev->gfx.rlc.clear_state_obj,
832 &adev->gfx.rlc.clear_state_gpu_addr,
833 (void **)&adev->gfx.rlc.cs_ptr);
834 if (r) {
835 dev_err(adev->dev, "(%d) failed to create rlc csb bo\n",
836 r);
837 gfx_v9_0_rlc_fini(adev);
838 return r;
839 }
840 /* set up the cs buffer */
841 dst_ptr = adev->gfx.rlc.cs_ptr;
842 gfx_v9_0_get_csb_buffer(adev, dst_ptr);
843 amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj);
844 amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
845 }
846
847 if (adev->asic_type == CHIP_RAVEN) {
848 /* TODO: double check the cp_table_size for RV */
849 adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
850 r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size,
851 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
852 &adev->gfx.rlc.cp_table_obj,
853 &adev->gfx.rlc.cp_table_gpu_addr,
854 (void **)&adev->gfx.rlc.cp_table_ptr);
855 if (r) {
856 dev_err(adev->dev,
857 "(%d) failed to create cp table bo\n", r);
858 gfx_v9_0_rlc_fini(adev);
859 return r;
860 }
861
862 rv_init_cp_jump_table(adev);
863 amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj);
864 amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
865
866 gfx_v9_0_init_lbpw(adev);
867 }
868
869 return 0;
870}
871
872static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
873{
874 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
875 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
876}
877
878static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
879{
880 int r;
881 u32 *hpd;
882 const __le32 *fw_data;
883 unsigned fw_size;
884 u32 *fw;
885 size_t mec_hpd_size;
886
887 const struct gfx_firmware_header_v1_0 *mec_hdr;
888
889 bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
890
891 /* take ownership of the relevant compute queues */
892 amdgpu_gfx_compute_queue_acquire(adev);
893 mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE;
894
895 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
896 AMDGPU_GEM_DOMAIN_GTT,
897 &adev->gfx.mec.hpd_eop_obj,
898 &adev->gfx.mec.hpd_eop_gpu_addr,
899 (void **)&hpd);
900 if (r) {
901 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
902 gfx_v9_0_mec_fini(adev);
903 return r;
904 }
905
906 memset(hpd, 0, adev->gfx.mec.hpd_eop_obj->tbo.mem.size);
907
908 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
909 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
910
911 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
912
913 fw_data = (const __le32 *)
914 (adev->gfx.mec_fw->data +
915 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
916 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
917
918 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
919 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
920 &adev->gfx.mec.mec_fw_obj,
921 &adev->gfx.mec.mec_fw_gpu_addr,
922 (void **)&fw);
923 if (r) {
924 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
925 gfx_v9_0_mec_fini(adev);
926 return r;
927 }
928
929 memcpy(fw, fw_data, fw_size);
930
931 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
932 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
933
934 return 0;
935}
936
937static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
938{
939 WREG32_SOC15(GC, 0, mmSQ_IND_INDEX,
940 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
941 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
942 (address << SQ_IND_INDEX__INDEX__SHIFT) |
943 (SQ_IND_INDEX__FORCE_READ_MASK));
944 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
945}
946
947static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
948 uint32_t wave, uint32_t thread,
949 uint32_t regno, uint32_t num, uint32_t *out)
950{
951 WREG32_SOC15(GC, 0, mmSQ_IND_INDEX,
952 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
953 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
954 (regno << SQ_IND_INDEX__INDEX__SHIFT) |
955 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
956 (SQ_IND_INDEX__FORCE_READ_MASK) |
957 (SQ_IND_INDEX__AUTO_INCR_MASK));
958 while (num--)
959 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
960}
961
962static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
963{
964 /* type 1 wave data */
965 dst[(*no_fields)++] = 1;
966 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
967 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
968 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
969 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
970 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
971 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
972 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
973 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
974 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
975 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
976 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
977 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
978 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
979 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
980}
981
982static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
983 uint32_t wave, uint32_t start,
984 uint32_t size, uint32_t *dst)
985{
986 wave_read_regs(
987 adev, simd, wave, 0,
988 start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
989}
990
991static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
992 uint32_t wave, uint32_t thread,
993 uint32_t start, uint32_t size,
994 uint32_t *dst)
995{
996 wave_read_regs(
997 adev, simd, wave, thread,
998 start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
999}
1000
1001static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
1002 u32 me, u32 pipe, u32 q)
1003{
1004 soc15_grbm_select(adev, me, pipe, q, 0);
1005}
1006
1007static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
1008 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
1009 .select_se_sh = &gfx_v9_0_select_se_sh,
1010 .read_wave_data = &gfx_v9_0_read_wave_data,
1011 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
1012 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
1013 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q
1014};
1015
1016static void gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
1017{
1018 u32 gb_addr_config;
1019
1020 adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
1021
1022 switch (adev->asic_type) {
1023 case CHIP_VEGA10:
1024 adev->gfx.config.max_hw_contexts = 8;
1025 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1026 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1027 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1028 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1029 gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
1030 break;
1031 case CHIP_VEGA12:
1032 adev->gfx.config.max_hw_contexts = 8;
1033 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1034 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1035 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1036 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1037 gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
1038 DRM_INFO("fix gfx.config for vega12\n");
1039 break;
1040 case CHIP_RAVEN:
1041 adev->gfx.config.max_hw_contexts = 8;
1042 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1043 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1044 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1045 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1046 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN;
1047 break;
1048 default:
1049 BUG();
1050 break;
1051 }
1052
1053 adev->gfx.config.gb_addr_config = gb_addr_config;
1054
1055 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
1056 REG_GET_FIELD(
1057 adev->gfx.config.gb_addr_config,
1058 GB_ADDR_CONFIG,
1059 NUM_PIPES);
1060
1061 adev->gfx.config.max_tile_pipes =
1062 adev->gfx.config.gb_addr_config_fields.num_pipes;
1063
1064 adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
1065 REG_GET_FIELD(
1066 adev->gfx.config.gb_addr_config,
1067 GB_ADDR_CONFIG,
1068 NUM_BANKS);
1069 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
1070 REG_GET_FIELD(
1071 adev->gfx.config.gb_addr_config,
1072 GB_ADDR_CONFIG,
1073 MAX_COMPRESSED_FRAGS);
1074 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
1075 REG_GET_FIELD(
1076 adev->gfx.config.gb_addr_config,
1077 GB_ADDR_CONFIG,
1078 NUM_RB_PER_SE);
1079 adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
1080 REG_GET_FIELD(
1081 adev->gfx.config.gb_addr_config,
1082 GB_ADDR_CONFIG,
1083 NUM_SHADER_ENGINES);
1084 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
1085 REG_GET_FIELD(
1086 adev->gfx.config.gb_addr_config,
1087 GB_ADDR_CONFIG,
1088 PIPE_INTERLEAVE_SIZE));
1089}
1090
1091static int gfx_v9_0_ngg_create_buf(struct amdgpu_device *adev,
1092 struct amdgpu_ngg_buf *ngg_buf,
1093 int size_se,
1094 int default_size_se)
1095{
1096 int r;
1097
1098 if (size_se < 0) {
1099 dev_err(adev->dev, "Buffer size is invalid: %d\n", size_se);
1100 return -EINVAL;
1101 }
1102 size_se = size_se ? size_se : default_size_se;
1103
1104 ngg_buf->size = size_se * adev->gfx.config.max_shader_engines;
1105 r = amdgpu_bo_create_kernel(adev, ngg_buf->size,
1106 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
1107 &ngg_buf->bo,
1108 &ngg_buf->gpu_addr,
1109 NULL);
1110 if (r) {
1111 dev_err(adev->dev, "(%d) failed to create NGG buffer\n", r);
1112 return r;
1113 }
1114 ngg_buf->bo_size = amdgpu_bo_size(ngg_buf->bo);
1115
1116 return r;
1117}
1118
1119static int gfx_v9_0_ngg_fini(struct amdgpu_device *adev)
1120{
1121 int i;
1122
1123 for (i = 0; i < NGG_BUF_MAX; i++)
1124 amdgpu_bo_free_kernel(&adev->gfx.ngg.buf[i].bo,
1125 &adev->gfx.ngg.buf[i].gpu_addr,
1126 NULL);
1127
1128 memset(&adev->gfx.ngg.buf[0], 0,
1129 sizeof(struct amdgpu_ngg_buf) * NGG_BUF_MAX);
1130
1131 adev->gfx.ngg.init = false;
1132
1133 return 0;
1134}
1135
1136static int gfx_v9_0_ngg_init(struct amdgpu_device *adev)
1137{
1138 int r;
1139
1140 if (!amdgpu_ngg || adev->gfx.ngg.init == true)
1141 return 0;
1142
1143 /* GDS reserve memory: 64 bytes alignment */
1144 adev->gfx.ngg.gds_reserve_size = ALIGN(5 * 4, 0x40);
1145 adev->gds.mem.total_size -= adev->gfx.ngg.gds_reserve_size;
1146 adev->gds.mem.gfx_partition_size -= adev->gfx.ngg.gds_reserve_size;
1147 adev->gfx.ngg.gds_reserve_addr = RREG32_SOC15(GC, 0, mmGDS_VMID0_BASE);
1148 adev->gfx.ngg.gds_reserve_addr += RREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE);
1149
1150 /* Primitive Buffer */
1151 r = gfx_v9_0_ngg_create_buf(adev, &adev->gfx.ngg.buf[NGG_PRIM],
1152 amdgpu_prim_buf_per_se,
1153 64 * 1024);
1154 if (r) {
1155 dev_err(adev->dev, "Failed to create Primitive Buffer\n");
1156 goto err;
1157 }
1158
1159 /* Position Buffer */
1160 r = gfx_v9_0_ngg_create_buf(adev, &adev->gfx.ngg.buf[NGG_POS],
1161 amdgpu_pos_buf_per_se,
1162 256 * 1024);
1163 if (r) {
1164 dev_err(adev->dev, "Failed to create Position Buffer\n");
1165 goto err;
1166 }
1167
1168 /* Control Sideband */
1169 r = gfx_v9_0_ngg_create_buf(adev, &adev->gfx.ngg.buf[NGG_CNTL],
1170 amdgpu_cntl_sb_buf_per_se,
1171 256);
1172 if (r) {
1173 dev_err(adev->dev, "Failed to create Control Sideband Buffer\n");
1174 goto err;
1175 }
1176
1177 /* Parameter Cache, not created by default */
1178 if (amdgpu_param_buf_per_se <= 0)
1179 goto out;
1180
1181 r = gfx_v9_0_ngg_create_buf(adev, &adev->gfx.ngg.buf[NGG_PARAM],
1182 amdgpu_param_buf_per_se,
1183 512 * 1024);
1184 if (r) {
1185 dev_err(adev->dev, "Failed to create Parameter Cache\n");
1186 goto err;
1187 }
1188
1189out:
1190 adev->gfx.ngg.init = true;
1191 return 0;
1192err:
1193 gfx_v9_0_ngg_fini(adev);
1194 return r;
1195}
1196
1197static int gfx_v9_0_ngg_en(struct amdgpu_device *adev)
1198{
1199 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
1200 int r;
1201 u32 data, base;
1202
1203 if (!amdgpu_ngg)
1204 return 0;
1205
1206 /* Program buffer size */
1207 data = REG_SET_FIELD(0, WD_BUF_RESOURCE_1, INDEX_BUF_SIZE,
1208 adev->gfx.ngg.buf[NGG_PRIM].size >> 8);
1209 data = REG_SET_FIELD(data, WD_BUF_RESOURCE_1, POS_BUF_SIZE,
1210 adev->gfx.ngg.buf[NGG_POS].size >> 8);
1211 WREG32_SOC15(GC, 0, mmWD_BUF_RESOURCE_1, data);
1212
1213 data = REG_SET_FIELD(0, WD_BUF_RESOURCE_2, CNTL_SB_BUF_SIZE,
1214 adev->gfx.ngg.buf[NGG_CNTL].size >> 8);
1215 data = REG_SET_FIELD(data, WD_BUF_RESOURCE_2, PARAM_BUF_SIZE,
1216 adev->gfx.ngg.buf[NGG_PARAM].size >> 10);
1217 WREG32_SOC15(GC, 0, mmWD_BUF_RESOURCE_2, data);
1218
1219 /* Program buffer base address */
1220 base = lower_32_bits(adev->gfx.ngg.buf[NGG_PRIM].gpu_addr);
1221 data = REG_SET_FIELD(0, WD_INDEX_BUF_BASE, BASE, base);
1222 WREG32_SOC15(GC, 0, mmWD_INDEX_BUF_BASE, data);
1223
1224 base = upper_32_bits(adev->gfx.ngg.buf[NGG_PRIM].gpu_addr);
1225 data = REG_SET_FIELD(0, WD_INDEX_BUF_BASE_HI, BASE_HI, base);
1226 WREG32_SOC15(GC, 0, mmWD_INDEX_BUF_BASE_HI, data);
1227
1228 base = lower_32_bits(adev->gfx.ngg.buf[NGG_POS].gpu_addr);
1229 data = REG_SET_FIELD(0, WD_POS_BUF_BASE, BASE, base);
1230 WREG32_SOC15(GC, 0, mmWD_POS_BUF_BASE, data);
1231
1232 base = upper_32_bits(adev->gfx.ngg.buf[NGG_POS].gpu_addr);
1233 data = REG_SET_FIELD(0, WD_POS_BUF_BASE_HI, BASE_HI, base);
1234 WREG32_SOC15(GC, 0, mmWD_POS_BUF_BASE_HI, data);
1235
1236 base = lower_32_bits(adev->gfx.ngg.buf[NGG_CNTL].gpu_addr);
1237 data = REG_SET_FIELD(0, WD_CNTL_SB_BUF_BASE, BASE, base);
1238 WREG32_SOC15(GC, 0, mmWD_CNTL_SB_BUF_BASE, data);
1239
1240 base = upper_32_bits(adev->gfx.ngg.buf[NGG_CNTL].gpu_addr);
1241 data = REG_SET_FIELD(0, WD_CNTL_SB_BUF_BASE_HI, BASE_HI, base);
1242 WREG32_SOC15(GC, 0, mmWD_CNTL_SB_BUF_BASE_HI, data);
1243
1244 /* Clear GDS reserved memory */
1245 r = amdgpu_ring_alloc(ring, 17);
1246 if (r) {
1247 DRM_ERROR("amdgpu: NGG failed to lock ring %d (%d).\n",
1248 ring->idx, r);
1249 return r;
1250 }
1251
1252 gfx_v9_0_write_data_to_reg(ring, 0, false,
1253 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE),
1254 (adev->gds.mem.total_size +
1255 adev->gfx.ngg.gds_reserve_size) >>
1256 AMDGPU_GDS_SHIFT);
1257
1258 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
1259 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
1260 PACKET3_DMA_DATA_DST_SEL(1) |
1261 PACKET3_DMA_DATA_SRC_SEL(2)));
1262 amdgpu_ring_write(ring, 0);
1263 amdgpu_ring_write(ring, 0);
1264 amdgpu_ring_write(ring, adev->gfx.ngg.gds_reserve_addr);
1265 amdgpu_ring_write(ring, 0);
1266 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
1267 adev->gfx.ngg.gds_reserve_size);
1268
1269 gfx_v9_0_write_data_to_reg(ring, 0, false,
1270 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE), 0);
1271
1272 amdgpu_ring_commit(ring);
1273
1274 return 0;
1275}
1276
1277static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1278 int mec, int pipe, int queue)
1279{
1280 int r;
1281 unsigned irq_type;
1282 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1283
1284 ring = &adev->gfx.compute_ring[ring_id];
1285
1286 /* mec0 is me1 */
1287 ring->me = mec + 1;
1288 ring->pipe = pipe;
1289 ring->queue = queue;
1290
1291 ring->ring_obj = NULL;
1292 ring->use_doorbell = true;
1293 ring->doorbell_index = (AMDGPU_DOORBELL_MEC_RING0 + ring_id) << 1;
1294 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1295 + (ring_id * GFX9_MEC_HPD_SIZE);
1296 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1297
1298 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1299 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1300 + ring->pipe;
1301
1302 /* type-2 packets are deprecated on MEC, use type-3 instead */
1303 r = amdgpu_ring_init(adev, ring, 1024,
1304 &adev->gfx.eop_irq, irq_type);
1305 if (r)
1306 return r;
1307
1308
1309 return 0;
1310}
1311
1312static int gfx_v9_0_sw_init(void *handle)
1313{
1314 int i, j, k, r, ring_id;
1315 struct amdgpu_ring *ring;
1316 struct amdgpu_kiq *kiq;
1317 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1318
1319 switch (adev->asic_type) {
1320 case CHIP_VEGA10:
1321 case CHIP_VEGA12:
1322 case CHIP_RAVEN:
1323 adev->gfx.mec.num_mec = 2;
1324 break;
1325 default:
1326 adev->gfx.mec.num_mec = 1;
1327 break;
1328 }
1329
1330 adev->gfx.mec.num_pipe_per_mec = 4;
1331 adev->gfx.mec.num_queue_per_pipe = 8;
1332
1333 /* KIQ event */
1334 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, 178, &adev->gfx.kiq.irq);
1335 if (r)
1336 return r;
1337
1338 /* EOP Event */
1339 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, 181, &adev->gfx.eop_irq);
1340 if (r)
1341 return r;
1342
1343 /* Privileged reg */
1344 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, 184,
1345 &adev->gfx.priv_reg_irq);
1346 if (r)
1347 return r;
1348
1349 /* Privileged inst */
1350 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, 185,
1351 &adev->gfx.priv_inst_irq);
1352 if (r)
1353 return r;
1354
1355 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1356
1357 gfx_v9_0_scratch_init(adev);
1358
1359 r = gfx_v9_0_init_microcode(adev);
1360 if (r) {
1361 DRM_ERROR("Failed to load gfx firmware!\n");
1362 return r;
1363 }
1364
1365 r = gfx_v9_0_rlc_init(adev);
1366 if (r) {
1367 DRM_ERROR("Failed to init rlc BOs!\n");
1368 return r;
1369 }
1370
1371 r = gfx_v9_0_mec_init(adev);
1372 if (r) {
1373 DRM_ERROR("Failed to init MEC BOs!\n");
1374 return r;
1375 }
1376
1377 /* set up the gfx ring */
1378 for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
1379 ring = &adev->gfx.gfx_ring[i];
1380 ring->ring_obj = NULL;
1381 if (!i)
1382 sprintf(ring->name, "gfx");
1383 else
1384 sprintf(ring->name, "gfx_%d", i);
1385 ring->use_doorbell = true;
1386 ring->doorbell_index = AMDGPU_DOORBELL64_GFX_RING0 << 1;
1387 r = amdgpu_ring_init(adev, ring, 1024,
1388 &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_EOP);
1389 if (r)
1390 return r;
1391 }
1392
1393 /* set up the compute queues - allocate horizontally across pipes */
1394 ring_id = 0;
1395 for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1396 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
1397 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
1398 if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
1399 continue;
1400
1401 r = gfx_v9_0_compute_ring_init(adev,
1402 ring_id,
1403 i, k, j);
1404 if (r)
1405 return r;
1406
1407 ring_id++;
1408 }
1409 }
1410 }
1411
1412 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE);
1413 if (r) {
1414 DRM_ERROR("Failed to init KIQ BOs!\n");
1415 return r;
1416 }
1417
1418 kiq = &adev->gfx.kiq;
1419 r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
1420 if (r)
1421 return r;
1422
1423 /* create MQD for all compute queues as wel as KIQ for SRIOV case */
1424 r = amdgpu_gfx_compute_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation));
1425 if (r)
1426 return r;
1427
1428 /* reserve GDS, GWS and OA resource for gfx */
1429 r = amdgpu_bo_create_kernel(adev, adev->gds.mem.gfx_partition_size,
1430 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GDS,
1431 &adev->gds.gds_gfx_bo, NULL, NULL);
1432 if (r)
1433 return r;
1434
1435 r = amdgpu_bo_create_kernel(adev, adev->gds.gws.gfx_partition_size,
1436 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GWS,
1437 &adev->gds.gws_gfx_bo, NULL, NULL);
1438 if (r)
1439 return r;
1440
1441 r = amdgpu_bo_create_kernel(adev, adev->gds.oa.gfx_partition_size,
1442 PAGE_SIZE, AMDGPU_GEM_DOMAIN_OA,
1443 &adev->gds.oa_gfx_bo, NULL, NULL);
1444 if (r)
1445 return r;
1446
1447 adev->gfx.ce_ram_size = 0x8000;
1448
1449 gfx_v9_0_gpu_early_init(adev);
1450
1451 r = gfx_v9_0_ngg_init(adev);
1452 if (r)
1453 return r;
1454
1455 return 0;
1456}
1457
1458
1459static int gfx_v9_0_sw_fini(void *handle)
1460{
1461 int i;
1462 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1463
1464 amdgpu_bo_free_kernel(&adev->gds.oa_gfx_bo, NULL, NULL);
1465 amdgpu_bo_free_kernel(&adev->gds.gws_gfx_bo, NULL, NULL);
1466 amdgpu_bo_free_kernel(&adev->gds.gds_gfx_bo, NULL, NULL);
1467
1468 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
1469 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
1470 for (i = 0; i < adev->gfx.num_compute_rings; i++)
1471 amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
1472
1473 amdgpu_gfx_compute_mqd_sw_fini(adev);
1474 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring, &adev->gfx.kiq.irq);
1475 amdgpu_gfx_kiq_fini(adev);
1476
1477 gfx_v9_0_mec_fini(adev);
1478 gfx_v9_0_ngg_fini(adev);
1479 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
1480 &adev->gfx.rlc.clear_state_gpu_addr,
1481 (void **)&adev->gfx.rlc.cs_ptr);
1482 if (adev->asic_type == CHIP_RAVEN) {
1483 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
1484 &adev->gfx.rlc.cp_table_gpu_addr,
1485 (void **)&adev->gfx.rlc.cp_table_ptr);
1486 }
1487 gfx_v9_0_free_microcode(adev);
1488
1489 return 0;
1490}
1491
1492
1493static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
1494{
1495 /* TODO */
1496}
1497
1498static void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance)
1499{
1500 u32 data;
1501
1502 if (instance == 0xffffffff)
1503 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1504 else
1505 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1506
1507 if (se_num == 0xffffffff)
1508 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
1509 else
1510 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
1511
1512 if (sh_num == 0xffffffff)
1513 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
1514 else
1515 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
1516
1517 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, data);
1518}
1519
1520static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1521{
1522 u32 data, mask;
1523
1524 data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
1525 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
1526
1527 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1528 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1529
1530 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1531 adev->gfx.config.max_sh_per_se);
1532
1533 return (~data) & mask;
1534}
1535
1536static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
1537{
1538 int i, j;
1539 u32 data;
1540 u32 active_rbs = 0;
1541 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1542 adev->gfx.config.max_sh_per_se;
1543
1544 mutex_lock(&adev->grbm_idx_mutex);
1545 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1546 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1547 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
1548 data = gfx_v9_0_get_rb_active_bitmap(adev);
1549 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1550 rb_bitmap_width_per_sh);
1551 }
1552 }
1553 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1554 mutex_unlock(&adev->grbm_idx_mutex);
1555
1556 adev->gfx.config.backend_enable_mask = active_rbs;
1557 adev->gfx.config.num_rbs = hweight32(active_rbs);
1558}
1559
1560#define DEFAULT_SH_MEM_BASES (0x6000)
1561#define FIRST_COMPUTE_VMID (8)
1562#define LAST_COMPUTE_VMID (16)
1563static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
1564{
1565 int i;
1566 uint32_t sh_mem_config;
1567 uint32_t sh_mem_bases;
1568
1569 /*
1570 * Configure apertures:
1571 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB)
1572 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB)
1573 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB)
1574 */
1575 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1576
1577 sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
1578 SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1579 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1580
1581 mutex_lock(&adev->srbm_mutex);
1582 for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1583 soc15_grbm_select(adev, 0, 0, 0, i);
1584 /* CP and shaders */
1585 WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
1586 WREG32_SOC15(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
1587 }
1588 soc15_grbm_select(adev, 0, 0, 0, 0);
1589 mutex_unlock(&adev->srbm_mutex);
1590}
1591
1592static void gfx_v9_0_gpu_init(struct amdgpu_device *adev)
1593{
1594 u32 tmp;
1595 int i;
1596
1597 WREG32_FIELD15(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
1598
1599 gfx_v9_0_tiling_mode_table_init(adev);
1600
1601 gfx_v9_0_setup_rb(adev);
1602 gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
1603
1604 /* XXX SH_MEM regs */
1605 /* where to put LDS, scratch, GPUVM in FSA64 space */
1606 mutex_lock(&adev->srbm_mutex);
1607 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB].num_ids; i++) {
1608 soc15_grbm_select(adev, 0, 0, 0, i);
1609 /* CP and shaders */
1610 if (i == 0) {
1611 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1612 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1613 WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, tmp);
1614 WREG32_SOC15(GC, 0, mmSH_MEM_BASES, 0);
1615 } else {
1616 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1617 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1618 WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, tmp);
1619 tmp = adev->gmc.shared_aperture_start >> 48;
1620 WREG32_SOC15(GC, 0, mmSH_MEM_BASES, tmp);
1621 }
1622 }
1623 soc15_grbm_select(adev, 0, 0, 0, 0);
1624
1625 mutex_unlock(&adev->srbm_mutex);
1626
1627 gfx_v9_0_init_compute_vmid(adev);
1628
1629 mutex_lock(&adev->grbm_idx_mutex);
1630 /*
1631 * making sure that the following register writes will be broadcasted
1632 * to all the shaders
1633 */
1634 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1635
1636 WREG32_SOC15(GC, 0, mmPA_SC_FIFO_SIZE,
1637 (adev->gfx.config.sc_prim_fifo_size_frontend <<
1638 PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1639 (adev->gfx.config.sc_prim_fifo_size_backend <<
1640 PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1641 (adev->gfx.config.sc_hiz_tile_fifo_size <<
1642 PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1643 (adev->gfx.config.sc_earlyz_tile_fifo_size <<
1644 PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
1645 mutex_unlock(&adev->grbm_idx_mutex);
1646
1647}
1648
1649static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
1650{
1651 u32 i, j, k;
1652 u32 mask;
1653
1654 mutex_lock(&adev->grbm_idx_mutex);
1655 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1656 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1657 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
1658 for (k = 0; k < adev->usec_timeout; k++) {
1659 if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0)
1660 break;
1661 udelay(1);
1662 }
1663 if (k == adev->usec_timeout) {
1664 gfx_v9_0_select_se_sh(adev, 0xffffffff,
1665 0xffffffff, 0xffffffff);
1666 mutex_unlock(&adev->grbm_idx_mutex);
1667 DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
1668 i, j);
1669 return;
1670 }
1671 }
1672 }
1673 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1674 mutex_unlock(&adev->grbm_idx_mutex);
1675
1676 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
1677 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
1678 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
1679 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
1680 for (k = 0; k < adev->usec_timeout; k++) {
1681 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
1682 break;
1683 udelay(1);
1684 }
1685}
1686
1687static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
1688 bool enable)
1689{
1690 u32 tmp = RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
1691
1692 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
1693 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
1694 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
1695 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
1696
1697 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
1698}
1699
1700static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
1701{
1702 /* csib */
1703 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),
1704 adev->gfx.rlc.clear_state_gpu_addr >> 32);
1705 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),
1706 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
1707 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),
1708 adev->gfx.rlc.clear_state_size);
1709}
1710
1711static void gfx_v9_0_parse_ind_reg_list(int *register_list_format,
1712 int indirect_offset,
1713 int list_size,
1714 int *unique_indirect_regs,
1715 int *unique_indirect_reg_count,
1716 int max_indirect_reg_count,
1717 int *indirect_start_offsets,
1718 int *indirect_start_offsets_count,
1719 int max_indirect_start_offsets_count)
1720{
1721 int idx;
1722 bool new_entry = true;
1723
1724 for (; indirect_offset < list_size; indirect_offset++) {
1725
1726 if (new_entry) {
1727 new_entry = false;
1728 indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
1729 *indirect_start_offsets_count = *indirect_start_offsets_count + 1;
1730 BUG_ON(*indirect_start_offsets_count >= max_indirect_start_offsets_count);
1731 }
1732
1733 if (register_list_format[indirect_offset] == 0xFFFFFFFF) {
1734 new_entry = true;
1735 continue;
1736 }
1737
1738 indirect_offset += 2;
1739
1740 /* look for the matching indice */
1741 for (idx = 0; idx < *unique_indirect_reg_count; idx++) {
1742 if (unique_indirect_regs[idx] ==
1743 register_list_format[indirect_offset])
1744 break;
1745 }
1746
1747 if (idx >= *unique_indirect_reg_count) {
1748 unique_indirect_regs[*unique_indirect_reg_count] =
1749 register_list_format[indirect_offset];
1750 idx = *unique_indirect_reg_count;
1751 *unique_indirect_reg_count = *unique_indirect_reg_count + 1;
1752 BUG_ON(*unique_indirect_reg_count >= max_indirect_reg_count);
1753 }
1754
1755 register_list_format[indirect_offset] = idx;
1756 }
1757}
1758
1759static int gfx_v9_0_init_rlc_save_restore_list(struct amdgpu_device *adev)
1760{
1761 int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
1762 int unique_indirect_reg_count = 0;
1763
1764 int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
1765 int indirect_start_offsets_count = 0;
1766
1767 int list_size = 0;
1768 int i = 0;
1769 u32 tmp = 0;
1770
1771 u32 *register_list_format =
1772 kmalloc(adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
1773 if (!register_list_format)
1774 return -ENOMEM;
1775 memcpy(register_list_format, adev->gfx.rlc.register_list_format,
1776 adev->gfx.rlc.reg_list_format_size_bytes);
1777
1778 /* setup unique_indirect_regs array and indirect_start_offsets array */
1779 gfx_v9_0_parse_ind_reg_list(register_list_format,
1780 GFX9_RLC_FORMAT_DIRECT_REG_LIST_LENGTH,
1781 adev->gfx.rlc.reg_list_format_size_bytes >> 2,
1782 unique_indirect_regs,
1783 &unique_indirect_reg_count,
1784 ARRAY_SIZE(unique_indirect_regs),
1785 indirect_start_offsets,
1786 &indirect_start_offsets_count,
1787 ARRAY_SIZE(indirect_start_offsets));
1788
1789 /* enable auto inc in case it is disabled */
1790 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
1791 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
1792 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
1793
1794 /* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
1795 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),
1796 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET);
1797 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
1798 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
1799 adev->gfx.rlc.register_restore[i]);
1800
1801 /* load direct register */
1802 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR), 0);
1803 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
1804 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
1805 adev->gfx.rlc.register_restore[i]);
1806
1807 /* load indirect register */
1808 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
1809 adev->gfx.rlc.reg_list_format_start);
1810 for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
1811 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
1812 register_list_format[i]);
1813
1814 /* set save/restore list size */
1815 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
1816 list_size = list_size >> 1;
1817 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
1818 adev->gfx.rlc.reg_restore_list_size);
1819 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size);
1820
1821 /* write the starting offsets to RLC scratch ram */
1822 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
1823 adev->gfx.rlc.starting_offsets_start);
1824 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
1825 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
1826 indirect_start_offsets[i]);
1827
1828 /* load unique indirect regs*/
1829 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
1830 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0) + i,
1831 unique_indirect_regs[i] & 0x3FFFF);
1832 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0) + i,
1833 unique_indirect_regs[i] >> 20);
1834 }
1835
1836 kfree(register_list_format);
1837 return 0;
1838}
1839
1840static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
1841{
1842 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
1843}
1844
1845static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
1846 bool enable)
1847{
1848 uint32_t data = 0;
1849 uint32_t default_data = 0;
1850
1851 default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS));
1852 if (enable == true) {
1853 /* enable GFXIP control over CGPG */
1854 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
1855 if(default_data != data)
1856 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
1857
1858 /* update status */
1859 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK;
1860 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT);
1861 if(default_data != data)
1862 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
1863 } else {
1864 /* restore GFXIP control over GCPG */
1865 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
1866 if(default_data != data)
1867 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
1868 }
1869}
1870
1871static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
1872{
1873 uint32_t data = 0;
1874
1875 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
1876 AMD_PG_SUPPORT_GFX_SMG |
1877 AMD_PG_SUPPORT_GFX_DMG)) {
1878 /* init IDLE_POLL_COUNT = 60 */
1879 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL));
1880 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
1881 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
1882 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data);
1883
1884 /* init RLC PG Delay */
1885 data = 0;
1886 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT);
1887 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT);
1888 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT);
1889 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT);
1890 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data);
1891
1892 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2));
1893 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK;
1894 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT);
1895 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data);
1896
1897 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3));
1898 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK;
1899 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT);
1900 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data);
1901
1902 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL));
1903 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
1904
1905 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
1906 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
1907 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data);
1908
1909 pwr_10_0_gfxip_control_over_cgpg(adev, true);
1910 }
1911}
1912
1913static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
1914 bool enable)
1915{
1916 uint32_t data = 0;
1917 uint32_t default_data = 0;
1918
1919 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
1920 data = REG_SET_FIELD(data, RLC_PG_CNTL,
1921 SMU_CLK_SLOWDOWN_ON_PU_ENABLE,
1922 enable ? 1 : 0);
1923 if (default_data != data)
1924 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
1925}
1926
1927static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
1928 bool enable)
1929{
1930 uint32_t data = 0;
1931 uint32_t default_data = 0;
1932
1933 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
1934 data = REG_SET_FIELD(data, RLC_PG_CNTL,
1935 SMU_CLK_SLOWDOWN_ON_PD_ENABLE,
1936 enable ? 1 : 0);
1937 if(default_data != data)
1938 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
1939}
1940
1941static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
1942 bool enable)
1943{
1944 uint32_t data = 0;
1945 uint32_t default_data = 0;
1946
1947 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
1948 data = REG_SET_FIELD(data, RLC_PG_CNTL,
1949 CP_PG_DISABLE,
1950 enable ? 0 : 1);
1951 if(default_data != data)
1952 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
1953}
1954
1955static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
1956 bool enable)
1957{
1958 uint32_t data, default_data;
1959
1960 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
1961 data = REG_SET_FIELD(data, RLC_PG_CNTL,
1962 GFX_POWER_GATING_ENABLE,
1963 enable ? 1 : 0);
1964 if(default_data != data)
1965 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
1966}
1967
1968static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
1969 bool enable)
1970{
1971 uint32_t data, default_data;
1972
1973 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
1974 data = REG_SET_FIELD(data, RLC_PG_CNTL,
1975 GFX_PIPELINE_PG_ENABLE,
1976 enable ? 1 : 0);
1977 if(default_data != data)
1978 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
1979
1980 if (!enable)
1981 /* read any GFX register to wake up GFX */
1982 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL));
1983}
1984
1985static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
1986 bool enable)
1987{
1988 uint32_t data, default_data;
1989
1990 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
1991 data = REG_SET_FIELD(data, RLC_PG_CNTL,
1992 STATIC_PER_CU_PG_ENABLE,
1993 enable ? 1 : 0);
1994 if(default_data != data)
1995 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
1996}
1997
1998static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
1999 bool enable)
2000{
2001 uint32_t data, default_data;
2002
2003 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2004 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2005 DYN_PER_CU_PG_ENABLE,
2006 enable ? 1 : 0);
2007 if(default_data != data)
2008 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2009}
2010
2011static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
2012{
2013 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2014 AMD_PG_SUPPORT_GFX_SMG |
2015 AMD_PG_SUPPORT_GFX_DMG |
2016 AMD_PG_SUPPORT_CP |
2017 AMD_PG_SUPPORT_GDS |
2018 AMD_PG_SUPPORT_RLC_SMU_HS)) {
2019 gfx_v9_0_init_csb(adev);
2020 gfx_v9_0_init_rlc_save_restore_list(adev);
2021 gfx_v9_0_enable_save_restore_machine(adev);
2022
2023 if (adev->asic_type == CHIP_RAVEN) {
2024 WREG32(mmRLC_JUMP_TABLE_RESTORE,
2025 adev->gfx.rlc.cp_table_gpu_addr >> 8);
2026 gfx_v9_0_init_gfx_power_gating(adev);
2027
2028 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
2029 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
2030 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
2031 } else {
2032 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
2033 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
2034 }
2035
2036 if (adev->pg_flags & AMD_PG_SUPPORT_CP)
2037 gfx_v9_0_enable_cp_power_gating(adev, true);
2038 else
2039 gfx_v9_0_enable_cp_power_gating(adev, false);
2040 }
2041 }
2042}
2043
2044void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
2045{
2046 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
2047 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
2048 gfx_v9_0_wait_for_rlc_serdes(adev);
2049}
2050
2051static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
2052{
2053 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
2054 udelay(50);
2055 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
2056 udelay(50);
2057}
2058
2059static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
2060{
2061#ifdef AMDGPU_RLC_DEBUG_RETRY
2062 u32 rlc_ucode_ver;
2063#endif
2064
2065 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
2066
2067 /* carrizo do enable cp interrupt after cp inited */
2068 if (!(adev->flags & AMD_IS_APU))
2069 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
2070
2071 udelay(50);
2072
2073#ifdef AMDGPU_RLC_DEBUG_RETRY
2074 /* RLC_GPM_GENERAL_6 : RLC Ucode version */
2075 rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6);
2076 if(rlc_ucode_ver == 0x108) {
2077 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
2078 rlc_ucode_ver, adev->gfx.rlc_fw_version);
2079 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
2080 * default is 0x9C4 to create a 100us interval */
2081 WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4);
2082 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
2083 * to disable the page fault retry interrupts, default is
2084 * 0x100 (256) */
2085 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100);
2086 }
2087#endif
2088}
2089
2090static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
2091{
2092 const struct rlc_firmware_header_v2_0 *hdr;
2093 const __le32 *fw_data;
2094 unsigned i, fw_size;
2095
2096 if (!adev->gfx.rlc_fw)
2097 return -EINVAL;
2098
2099 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
2100 amdgpu_ucode_print_rlc_hdr(&hdr->header);
2101
2102 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2103 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
2104 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
2105
2106 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
2107 RLCG_UCODE_LOADING_START_ADDRESS);
2108 for (i = 0; i < fw_size; i++)
2109 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
2110 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
2111
2112 return 0;
2113}
2114
2115static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
2116{
2117 int r;
2118
2119 if (amdgpu_sriov_vf(adev)) {
2120 gfx_v9_0_init_csb(adev);
2121 return 0;
2122 }
2123
2124 gfx_v9_0_rlc_stop(adev);
2125
2126 /* disable CG */
2127 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
2128
2129 /* disable PG */
2130 WREG32_SOC15(GC, 0, mmRLC_PG_CNTL, 0);
2131
2132 gfx_v9_0_rlc_reset(adev);
2133
2134 gfx_v9_0_init_pg(adev);
2135
2136 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
2137 /* legacy rlc firmware loading */
2138 r = gfx_v9_0_rlc_load_microcode(adev);
2139 if (r)
2140 return r;
2141 }
2142
2143 if (adev->asic_type == CHIP_RAVEN) {
2144 if (amdgpu_lbpw != 0)
2145 gfx_v9_0_enable_lbpw(adev, true);
2146 else
2147 gfx_v9_0_enable_lbpw(adev, false);
2148 }
2149
2150 gfx_v9_0_rlc_start(adev);
2151
2152 return 0;
2153}
2154
2155static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2156{
2157 int i;
2158 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
2159
2160 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2161 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2162 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
2163 if (!enable) {
2164 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2165 adev->gfx.gfx_ring[i].ready = false;
2166 }
2167 WREG32_SOC15(GC, 0, mmCP_ME_CNTL, tmp);
2168 udelay(50);
2169}
2170
2171static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2172{
2173 const struct gfx_firmware_header_v1_0 *pfp_hdr;
2174 const struct gfx_firmware_header_v1_0 *ce_hdr;
2175 const struct gfx_firmware_header_v1_0 *me_hdr;
2176 const __le32 *fw_data;
2177 unsigned i, fw_size;
2178
2179 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2180 return -EINVAL;
2181
2182 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2183 adev->gfx.pfp_fw->data;
2184 ce_hdr = (const struct gfx_firmware_header_v1_0 *)
2185 adev->gfx.ce_fw->data;
2186 me_hdr = (const struct gfx_firmware_header_v1_0 *)
2187 adev->gfx.me_fw->data;
2188
2189 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2190 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2191 amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2192
2193 gfx_v9_0_cp_gfx_enable(adev, false);
2194
2195 /* PFP */
2196 fw_data = (const __le32 *)
2197 (adev->gfx.pfp_fw->data +
2198 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2199 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2200 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0);
2201 for (i = 0; i < fw_size; i++)
2202 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2203 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2204
2205 /* CE */
2206 fw_data = (const __le32 *)
2207 (adev->gfx.ce_fw->data +
2208 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2209 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2210 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0);
2211 for (i = 0; i < fw_size; i++)
2212 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2213 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2214
2215 /* ME */
2216 fw_data = (const __le32 *)
2217 (adev->gfx.me_fw->data +
2218 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2219 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2220 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0);
2221 for (i = 0; i < fw_size; i++)
2222 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2223 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2224
2225 return 0;
2226}
2227
2228static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
2229{
2230 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2231 const struct cs_section_def *sect = NULL;
2232 const struct cs_extent_def *ext = NULL;
2233 int r, i, tmp;
2234
2235 /* init the CP */
2236 WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2237 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
2238
2239 gfx_v9_0_cp_gfx_enable(adev, true);
2240
2241 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
2242 if (r) {
2243 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2244 return r;
2245 }
2246
2247 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2248 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2249
2250 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2251 amdgpu_ring_write(ring, 0x80000000);
2252 amdgpu_ring_write(ring, 0x80000000);
2253
2254 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
2255 for (ext = sect->section; ext->extent != NULL; ++ext) {
2256 if (sect->id == SECT_CONTEXT) {
2257 amdgpu_ring_write(ring,
2258 PACKET3(PACKET3_SET_CONTEXT_REG,
2259 ext->reg_count));
2260 amdgpu_ring_write(ring,
2261 ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2262 for (i = 0; i < ext->reg_count; i++)
2263 amdgpu_ring_write(ring, ext->extent[i]);
2264 }
2265 }
2266 }
2267
2268 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2269 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2270
2271 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2272 amdgpu_ring_write(ring, 0);
2273
2274 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2275 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2276 amdgpu_ring_write(ring, 0x8000);
2277 amdgpu_ring_write(ring, 0x8000);
2278
2279 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
2280 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
2281 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
2282 amdgpu_ring_write(ring, tmp);
2283 amdgpu_ring_write(ring, 0);
2284
2285 amdgpu_ring_commit(ring);
2286
2287 return 0;
2288}
2289
2290static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
2291{
2292 struct amdgpu_ring *ring;
2293 u32 tmp;
2294 u32 rb_bufsz;
2295 u64 rb_addr, rptr_addr, wptr_gpu_addr;
2296
2297 /* Set the write pointer delay */
2298 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
2299
2300 /* set the RB to use vmid 0 */
2301 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
2302
2303 /* Set ring buffer size */
2304 ring = &adev->gfx.gfx_ring[0];
2305 rb_bufsz = order_base_2(ring->ring_size / 8);
2306 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
2307 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
2308#ifdef __BIG_ENDIAN
2309 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
2310#endif
2311 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
2312
2313 /* Initialize the ring buffer's write pointers */
2314 ring->wptr = 0;
2315 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2316 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
2317
2318 /* set the wb address wether it's enabled or not */
2319 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2320 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2321 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
2322
2323 wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2324 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
2325 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
2326
2327 mdelay(1);
2328 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
2329
2330 rb_addr = ring->gpu_addr >> 8;
2331 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
2332 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2333
2334 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
2335 if (ring->use_doorbell) {
2336 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
2337 DOORBELL_OFFSET, ring->doorbell_index);
2338 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
2339 DOORBELL_EN, 1);
2340 } else {
2341 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
2342 }
2343 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
2344
2345 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
2346 DOORBELL_RANGE_LOWER, ring->doorbell_index);
2347 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
2348
2349 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
2350 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
2351
2352
2353 /* start the ring */
2354 gfx_v9_0_cp_gfx_start(adev);
2355 ring->ready = true;
2356
2357 return 0;
2358}
2359
2360static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2361{
2362 int i;
2363
2364 if (enable) {
2365 WREG32_SOC15(GC, 0, mmCP_MEC_CNTL, 0);
2366 } else {
2367 WREG32_SOC15(GC, 0, mmCP_MEC_CNTL,
2368 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2369 for (i = 0; i < adev->gfx.num_compute_rings; i++)
2370 adev->gfx.compute_ring[i].ready = false;
2371 adev->gfx.kiq.ring.ready = false;
2372 }
2373 udelay(50);
2374}
2375
2376static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2377{
2378 const struct gfx_firmware_header_v1_0 *mec_hdr;
2379 const __le32 *fw_data;
2380 unsigned i;
2381 u32 tmp;
2382
2383 if (!adev->gfx.mec_fw)
2384 return -EINVAL;
2385
2386 gfx_v9_0_cp_compute_enable(adev, false);
2387
2388 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2389 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2390
2391 fw_data = (const __le32 *)
2392 (adev->gfx.mec_fw->data +
2393 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2394 tmp = 0;
2395 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
2396 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2397 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
2398
2399 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
2400 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
2401 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
2402 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
2403
2404 /* MEC1 */
2405 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
2406 mec_hdr->jt_offset);
2407 for (i = 0; i < mec_hdr->jt_size; i++)
2408 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
2409 le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
2410
2411 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
2412 adev->gfx.mec_fw_version);
2413 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
2414
2415 return 0;
2416}
2417
2418/* KIQ functions */
2419static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
2420{
2421 uint32_t tmp;
2422 struct amdgpu_device *adev = ring->adev;
2423
2424 /* tell RLC which is KIQ queue */
2425 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
2426 tmp &= 0xffffff00;
2427 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
2428 WREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
2429 tmp |= 0x80;
2430 WREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
2431}
2432
2433static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev)
2434{
2435 struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
2436 uint32_t scratch, tmp = 0;
2437 uint64_t queue_mask = 0;
2438 int r, i;
2439
2440 for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
2441 if (!test_bit(i, adev->gfx.mec.queue_bitmap))
2442 continue;
2443
2444 /* This situation may be hit in the future if a new HW
2445 * generation exposes more than 64 queues. If so, the
2446 * definition of queue_mask needs updating */
2447 if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
2448 DRM_ERROR("Invalid KCQ enabled: %d\n", i);
2449 break;
2450 }
2451
2452 queue_mask |= (1ull << i);
2453 }
2454
2455 r = amdgpu_gfx_scratch_get(adev, &scratch);
2456 if (r) {
2457 DRM_ERROR("Failed to get scratch reg (%d).\n", r);
2458 return r;
2459 }
2460 WREG32(scratch, 0xCAFEDEAD);
2461
2462 r = amdgpu_ring_alloc(kiq_ring, (7 * adev->gfx.num_compute_rings) + 11);
2463 if (r) {
2464 DRM_ERROR("Failed to lock KIQ (%d).\n", r);
2465 amdgpu_gfx_scratch_free(adev, scratch);
2466 return r;
2467 }
2468
2469 /* set resources */
2470 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
2471 amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
2472 PACKET3_SET_RESOURCES_QUEUE_TYPE(0)); /* vmid_mask:0 queue_type:0 (KIQ) */
2473 amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask)); /* queue mask lo */
2474 amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask)); /* queue mask hi */
2475 amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */
2476 amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */
2477 amdgpu_ring_write(kiq_ring, 0); /* oac mask */
2478 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */
2479 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2480 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2481 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
2482 uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2483
2484 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
2485 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
2486 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
2487 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
2488 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
2489 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
2490 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
2491 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
2492 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
2493 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
2494 PACKET3_MAP_QUEUES_ENGINE_SEL(0) | /* engine_sel: compute */
2495 PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
2496 amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
2497 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
2498 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
2499 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
2500 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
2501 }
2502 /* write to scratch for completion */
2503 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2504 amdgpu_ring_write(kiq_ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
2505 amdgpu_ring_write(kiq_ring, 0xDEADBEEF);
2506 amdgpu_ring_commit(kiq_ring);
2507
2508 for (i = 0; i < adev->usec_timeout; i++) {
2509 tmp = RREG32(scratch);
2510 if (tmp == 0xDEADBEEF)
2511 break;
2512 DRM_UDELAY(1);
2513 }
2514 if (i >= adev->usec_timeout) {
2515 DRM_ERROR("KCQ enable failed (scratch(0x%04X)=0x%08X)\n",
2516 scratch, tmp);
2517 r = -EINVAL;
2518 }
2519 amdgpu_gfx_scratch_free(adev, scratch);
2520
2521 return r;
2522}
2523
2524static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
2525{
2526 struct amdgpu_device *adev = ring->adev;
2527 struct v9_mqd *mqd = ring->mqd_ptr;
2528 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
2529 uint32_t tmp;
2530
2531 mqd->header = 0xC0310800;
2532 mqd->compute_pipelinestat_enable = 0x00000001;
2533 mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2534 mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2535 mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2536 mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2537 mqd->compute_misc_reserved = 0x00000003;
2538
2539 mqd->dynamic_cu_mask_addr_lo =
2540 lower_32_bits(ring->mqd_gpu_addr
2541 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
2542 mqd->dynamic_cu_mask_addr_hi =
2543 upper_32_bits(ring->mqd_gpu_addr
2544 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
2545
2546 eop_base_addr = ring->eop_gpu_addr >> 8;
2547 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
2548 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
2549
2550 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2551 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
2552 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
2553 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
2554
2555 mqd->cp_hqd_eop_control = tmp;
2556
2557 /* enable doorbell? */
2558 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
2559
2560 if (ring->use_doorbell) {
2561 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2562 DOORBELL_OFFSET, ring->doorbell_index);
2563 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2564 DOORBELL_EN, 1);
2565 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2566 DOORBELL_SOURCE, 0);
2567 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2568 DOORBELL_HIT, 0);
2569 } else {
2570 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2571 DOORBELL_EN, 0);
2572 }
2573
2574 mqd->cp_hqd_pq_doorbell_control = tmp;
2575
2576 /* disable the queue if it's active */
2577 ring->wptr = 0;
2578 mqd->cp_hqd_dequeue_request = 0;
2579 mqd->cp_hqd_pq_rptr = 0;
2580 mqd->cp_hqd_pq_wptr_lo = 0;
2581 mqd->cp_hqd_pq_wptr_hi = 0;
2582
2583 /* set the pointer to the MQD */
2584 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
2585 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
2586
2587 /* set MQD vmid to 0 */
2588 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
2589 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
2590 mqd->cp_mqd_control = tmp;
2591
2592 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2593 hqd_gpu_addr = ring->gpu_addr >> 8;
2594 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2595 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2596
2597 /* set up the HQD, this is similar to CP_RB0_CNTL */
2598 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
2599 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
2600 (order_base_2(ring->ring_size / 4) - 1));
2601 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
2602 ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
2603#ifdef __BIG_ENDIAN
2604 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
2605#endif
2606 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
2607 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
2608 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
2609 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
2610 mqd->cp_hqd_pq_control = tmp;
2611
2612 /* set the wb address whether it's enabled or not */
2613 wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2614 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2615 mqd->cp_hqd_pq_rptr_report_addr_hi =
2616 upper_32_bits(wb_gpu_addr) & 0xffff;
2617
2618 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2619 wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2620 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2621 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2622
2623 tmp = 0;
2624 /* enable the doorbell if requested */
2625 if (ring->use_doorbell) {
2626 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
2627 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2628 DOORBELL_OFFSET, ring->doorbell_index);
2629
2630 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2631 DOORBELL_EN, 1);
2632 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2633 DOORBELL_SOURCE, 0);
2634 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
2635 DOORBELL_HIT, 0);
2636 }
2637
2638 mqd->cp_hqd_pq_doorbell_control = tmp;
2639
2640 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2641 ring->wptr = 0;
2642 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
2643
2644 /* set the vmid for the queue */
2645 mqd->cp_hqd_vmid = 0;
2646
2647 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
2648 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
2649 mqd->cp_hqd_persistent_state = tmp;
2650
2651 /* set MIN_IB_AVAIL_SIZE */
2652 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
2653 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
2654 mqd->cp_hqd_ib_control = tmp;
2655
2656 /* activate the queue */
2657 mqd->cp_hqd_active = 1;
2658
2659 return 0;
2660}
2661
2662static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
2663{
2664 struct amdgpu_device *adev = ring->adev;
2665 struct v9_mqd *mqd = ring->mqd_ptr;
2666 int j;
2667
2668 /* disable wptr polling */
2669 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
2670
2671 WREG32_SOC15(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
2672 mqd->cp_hqd_eop_base_addr_lo);
2673 WREG32_SOC15(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
2674 mqd->cp_hqd_eop_base_addr_hi);
2675
2676 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2677 WREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL,
2678 mqd->cp_hqd_eop_control);
2679
2680 /* enable doorbell? */
2681 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
2682 mqd->cp_hqd_pq_doorbell_control);
2683
2684 /* disable the queue if it's active */
2685 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
2686 WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
2687 for (j = 0; j < adev->usec_timeout; j++) {
2688 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
2689 break;
2690 udelay(1);
2691 }
2692 WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
2693 mqd->cp_hqd_dequeue_request);
2694 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR,
2695 mqd->cp_hqd_pq_rptr);
2696 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO,
2697 mqd->cp_hqd_pq_wptr_lo);
2698 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI,
2699 mqd->cp_hqd_pq_wptr_hi);
2700 }
2701
2702 /* set the pointer to the MQD */
2703 WREG32_SOC15(GC, 0, mmCP_MQD_BASE_ADDR,
2704 mqd->cp_mqd_base_addr_lo);
2705 WREG32_SOC15(GC, 0, mmCP_MQD_BASE_ADDR_HI,
2706 mqd->cp_mqd_base_addr_hi);
2707
2708 /* set MQD vmid to 0 */
2709 WREG32_SOC15(GC, 0, mmCP_MQD_CONTROL,
2710 mqd->cp_mqd_control);
2711
2712 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2713 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE,
2714 mqd->cp_hqd_pq_base_lo);
2715 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE_HI,
2716 mqd->cp_hqd_pq_base_hi);
2717
2718 /* set up the HQD, this is similar to CP_RB0_CNTL */
2719 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL,
2720 mqd->cp_hqd_pq_control);
2721
2722 /* set the wb address whether it's enabled or not */
2723 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
2724 mqd->cp_hqd_pq_rptr_report_addr_lo);
2725 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
2726 mqd->cp_hqd_pq_rptr_report_addr_hi);
2727
2728 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2729 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
2730 mqd->cp_hqd_pq_wptr_poll_addr_lo);
2731 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
2732 mqd->cp_hqd_pq_wptr_poll_addr_hi);
2733
2734 /* enable the doorbell if requested */
2735 if (ring->use_doorbell) {
2736 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
2737 (AMDGPU_DOORBELL64_KIQ *2) << 2);
2738 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
2739 (AMDGPU_DOORBELL64_USERQUEUE_END * 2) << 2);
2740 }
2741
2742 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
2743 mqd->cp_hqd_pq_doorbell_control);
2744
2745 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2746 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO,
2747 mqd->cp_hqd_pq_wptr_lo);
2748 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI,
2749 mqd->cp_hqd_pq_wptr_hi);
2750
2751 /* set the vmid for the queue */
2752 WREG32_SOC15(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
2753
2754 WREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE,
2755 mqd->cp_hqd_persistent_state);
2756
2757 /* activate the queue */
2758 WREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE,
2759 mqd->cp_hqd_active);
2760
2761 if (ring->use_doorbell)
2762 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
2763
2764 return 0;
2765}
2766
2767static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
2768{
2769 struct amdgpu_device *adev = ring->adev;
2770 int j;
2771
2772 /* disable the queue if it's active */
2773 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
2774
2775 WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
2776
2777 for (j = 0; j < adev->usec_timeout; j++) {
2778 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
2779 break;
2780 udelay(1);
2781 }
2782
2783 if (j == AMDGPU_MAX_USEC_TIMEOUT) {
2784 DRM_DEBUG("KIQ dequeue request failed.\n");
2785
2786 /* Manual disable if dequeue request times out */
2787 WREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE, 0);
2788 }
2789
2790 WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
2791 0);
2792 }
2793
2794 WREG32_SOC15(GC, 0, mmCP_HQD_IQ_TIMER, 0);
2795 WREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL, 0);
2796 WREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
2797 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
2798 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
2799 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR, 0);
2800 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
2801 WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
2802
2803 return 0;
2804}
2805
2806static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
2807{
2808 struct amdgpu_device *adev = ring->adev;
2809 struct v9_mqd *mqd = ring->mqd_ptr;
2810 int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
2811
2812 gfx_v9_0_kiq_setting(ring);
2813
2814 if (adev->in_gpu_reset) { /* for GPU_RESET case */
2815 /* reset MQD to a clean status */
2816 if (adev->gfx.mec.mqd_backup[mqd_idx])
2817 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
2818
2819 /* reset ring buffer */
2820 ring->wptr = 0;
2821 amdgpu_ring_clear_ring(ring);
2822
2823 mutex_lock(&adev->srbm_mutex);
2824 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
2825 gfx_v9_0_kiq_init_register(ring);
2826 soc15_grbm_select(adev, 0, 0, 0, 0);
2827 mutex_unlock(&adev->srbm_mutex);
2828 } else {
2829 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
2830 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
2831 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
2832 mutex_lock(&adev->srbm_mutex);
2833 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
2834 gfx_v9_0_mqd_init(ring);
2835 gfx_v9_0_kiq_init_register(ring);
2836 soc15_grbm_select(adev, 0, 0, 0, 0);
2837 mutex_unlock(&adev->srbm_mutex);
2838
2839 if (adev->gfx.mec.mqd_backup[mqd_idx])
2840 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
2841 }
2842
2843 return 0;
2844}
2845
2846static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
2847{
2848 struct amdgpu_device *adev = ring->adev;
2849 struct v9_mqd *mqd = ring->mqd_ptr;
2850 int mqd_idx = ring - &adev->gfx.compute_ring[0];
2851
2852 if (!adev->in_gpu_reset && !adev->gfx.in_suspend) {
2853 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
2854 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
2855 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
2856 mutex_lock(&adev->srbm_mutex);
2857 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
2858 gfx_v9_0_mqd_init(ring);
2859 soc15_grbm_select(adev, 0, 0, 0, 0);
2860 mutex_unlock(&adev->srbm_mutex);
2861
2862 if (adev->gfx.mec.mqd_backup[mqd_idx])
2863 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
2864 } else if (adev->in_gpu_reset) { /* for GPU_RESET case */
2865 /* reset MQD to a clean status */
2866 if (adev->gfx.mec.mqd_backup[mqd_idx])
2867 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
2868
2869 /* reset ring buffer */
2870 ring->wptr = 0;
2871 amdgpu_ring_clear_ring(ring);
2872 } else {
2873 amdgpu_ring_clear_ring(ring);
2874 }
2875
2876 return 0;
2877}
2878
2879static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
2880{
2881 struct amdgpu_ring *ring = NULL;
2882 int r = 0, i;
2883
2884 gfx_v9_0_cp_compute_enable(adev, true);
2885
2886 ring = &adev->gfx.kiq.ring;
2887
2888 r = amdgpu_bo_reserve(ring->mqd_obj, false);
2889 if (unlikely(r != 0))
2890 goto done;
2891
2892 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
2893 if (!r) {
2894 r = gfx_v9_0_kiq_init_queue(ring);
2895 amdgpu_bo_kunmap(ring->mqd_obj);
2896 ring->mqd_ptr = NULL;
2897 }
2898 amdgpu_bo_unreserve(ring->mqd_obj);
2899 if (r)
2900 goto done;
2901
2902 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2903 ring = &adev->gfx.compute_ring[i];
2904
2905 r = amdgpu_bo_reserve(ring->mqd_obj, false);
2906 if (unlikely(r != 0))
2907 goto done;
2908 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
2909 if (!r) {
2910 r = gfx_v9_0_kcq_init_queue(ring);
2911 amdgpu_bo_kunmap(ring->mqd_obj);
2912 ring->mqd_ptr = NULL;
2913 }
2914 amdgpu_bo_unreserve(ring->mqd_obj);
2915 if (r)
2916 goto done;
2917 }
2918
2919 r = gfx_v9_0_kiq_kcq_enable(adev);
2920done:
2921 return r;
2922}
2923
2924static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
2925{
2926 int r, i;
2927 struct amdgpu_ring *ring;
2928
2929 if (!(adev->flags & AMD_IS_APU))
2930 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
2931
2932 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
2933 /* legacy firmware loading */
2934 r = gfx_v9_0_cp_gfx_load_microcode(adev);
2935 if (r)
2936 return r;
2937
2938 r = gfx_v9_0_cp_compute_load_microcode(adev);
2939 if (r)
2940 return r;
2941 }
2942
2943 r = gfx_v9_0_cp_gfx_resume(adev);
2944 if (r)
2945 return r;
2946
2947 r = gfx_v9_0_kiq_resume(adev);
2948 if (r)
2949 return r;
2950
2951 ring = &adev->gfx.gfx_ring[0];
2952 r = amdgpu_ring_test_ring(ring);
2953 if (r) {
2954 ring->ready = false;
2955 return r;
2956 }
2957
2958 ring = &adev->gfx.kiq.ring;
2959 ring->ready = true;
2960 r = amdgpu_ring_test_ring(ring);
2961 if (r)
2962 ring->ready = false;
2963
2964 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2965 ring = &adev->gfx.compute_ring[i];
2966
2967 ring->ready = true;
2968 r = amdgpu_ring_test_ring(ring);
2969 if (r)
2970 ring->ready = false;
2971 }
2972
2973 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
2974
2975 return 0;
2976}
2977
2978static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
2979{
2980 gfx_v9_0_cp_gfx_enable(adev, enable);
2981 gfx_v9_0_cp_compute_enable(adev, enable);
2982}
2983
2984static int gfx_v9_0_hw_init(void *handle)
2985{
2986 int r;
2987 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2988
2989 gfx_v9_0_init_golden_registers(adev);
2990
2991 gfx_v9_0_gpu_init(adev);
2992
2993 r = gfx_v9_0_rlc_resume(adev);
2994 if (r)
2995 return r;
2996
2997 r = gfx_v9_0_cp_resume(adev);
2998 if (r)
2999 return r;
3000
3001 r = gfx_v9_0_ngg_en(adev);
3002 if (r)
3003 return r;
3004
3005 return r;
3006}
3007
3008static int gfx_v9_0_kcq_disable(struct amdgpu_ring *kiq_ring,struct amdgpu_ring *ring)
3009{
3010 struct amdgpu_device *adev = kiq_ring->adev;
3011 uint32_t scratch, tmp = 0;
3012 int r, i;
3013
3014 r = amdgpu_gfx_scratch_get(adev, &scratch);
3015 if (r) {
3016 DRM_ERROR("Failed to get scratch reg (%d).\n", r);
3017 return r;
3018 }
3019 WREG32(scratch, 0xCAFEDEAD);
3020
3021 r = amdgpu_ring_alloc(kiq_ring, 10);
3022 if (r) {
3023 DRM_ERROR("Failed to lock KIQ (%d).\n", r);
3024 amdgpu_gfx_scratch_free(adev, scratch);
3025 return r;
3026 }
3027
3028 /* unmap queues */
3029 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
3030 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
3031 PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
3032 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
3033 PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
3034 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
3035 amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
3036 amdgpu_ring_write(kiq_ring, 0);
3037 amdgpu_ring_write(kiq_ring, 0);
3038 amdgpu_ring_write(kiq_ring, 0);
3039 /* write to scratch for completion */
3040 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
3041 amdgpu_ring_write(kiq_ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
3042 amdgpu_ring_write(kiq_ring, 0xDEADBEEF);
3043 amdgpu_ring_commit(kiq_ring);
3044
3045 for (i = 0; i < adev->usec_timeout; i++) {
3046 tmp = RREG32(scratch);
3047 if (tmp == 0xDEADBEEF)
3048 break;
3049 DRM_UDELAY(1);
3050 }
3051 if (i >= adev->usec_timeout) {
3052 DRM_ERROR("KCQ disabled failed (scratch(0x%04X)=0x%08X)\n", scratch, tmp);
3053 r = -EINVAL;
3054 }
3055 amdgpu_gfx_scratch_free(adev, scratch);
3056 return r;
3057}
3058
3059static int gfx_v9_0_hw_fini(void *handle)
3060{
3061 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3062 int i;
3063
3064 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3065 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3066
3067 /* disable KCQ to avoid CPC touch memory not valid anymore */
3068 for (i = 0; i < adev->gfx.num_compute_rings; i++)
3069 gfx_v9_0_kcq_disable(&adev->gfx.kiq.ring, &adev->gfx.compute_ring[i]);
3070
3071 if (amdgpu_sriov_vf(adev)) {
3072 gfx_v9_0_cp_gfx_enable(adev, false);
3073 /* must disable polling for SRIOV when hw finished, otherwise
3074 * CPC engine may still keep fetching WB address which is already
3075 * invalid after sw finished and trigger DMAR reading error in
3076 * hypervisor side.
3077 */
3078 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3079 return 0;
3080 }
3081
3082 /* Use deinitialize sequence from CAIL when unbinding device from driver,
3083 * otherwise KIQ is hanging when binding back
3084 */
3085 if (!adev->in_gpu_reset && !adev->gfx.in_suspend) {
3086 mutex_lock(&adev->srbm_mutex);
3087 soc15_grbm_select(adev, adev->gfx.kiq.ring.me,
3088 adev->gfx.kiq.ring.pipe,
3089 adev->gfx.kiq.ring.queue, 0);
3090 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring);
3091 soc15_grbm_select(adev, 0, 0, 0, 0);
3092 mutex_unlock(&adev->srbm_mutex);
3093 }
3094
3095 gfx_v9_0_cp_enable(adev, false);
3096 gfx_v9_0_rlc_stop(adev);
3097
3098 return 0;
3099}
3100
3101static int gfx_v9_0_suspend(void *handle)
3102{
3103 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3104
3105 adev->gfx.in_suspend = true;
3106 return gfx_v9_0_hw_fini(adev);
3107}
3108
3109static int gfx_v9_0_resume(void *handle)
3110{
3111 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3112 int r;
3113
3114 r = gfx_v9_0_hw_init(adev);
3115 adev->gfx.in_suspend = false;
3116 return r;
3117}
3118
3119static bool gfx_v9_0_is_idle(void *handle)
3120{
3121 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3122
3123 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
3124 GRBM_STATUS, GUI_ACTIVE))
3125 return false;
3126 else
3127 return true;
3128}
3129
3130static int gfx_v9_0_wait_for_idle(void *handle)
3131{
3132 unsigned i;
3133 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3134
3135 for (i = 0; i < adev->usec_timeout; i++) {
3136 if (gfx_v9_0_is_idle(handle))
3137 return 0;
3138 udelay(1);
3139 }
3140 return -ETIMEDOUT;
3141}
3142
3143static int gfx_v9_0_soft_reset(void *handle)
3144{
3145 u32 grbm_soft_reset = 0;
3146 u32 tmp;
3147 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3148
3149 /* GRBM_STATUS */
3150 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
3151 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
3152 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
3153 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
3154 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
3155 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
3156 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
3157 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3158 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3159 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3160 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
3161 }
3162
3163 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
3164 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3165 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3166 }
3167
3168 /* GRBM_STATUS2 */
3169 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
3170 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
3171 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3172 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
3173
3174
3175 if (grbm_soft_reset) {
3176 /* stop the rlc */
3177 gfx_v9_0_rlc_stop(adev);
3178
3179 /* Disable GFX parsing/prefetching */
3180 gfx_v9_0_cp_gfx_enable(adev, false);
3181
3182 /* Disable MEC parsing/prefetching */
3183 gfx_v9_0_cp_compute_enable(adev, false);
3184
3185 if (grbm_soft_reset) {
3186 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3187 tmp |= grbm_soft_reset;
3188 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
3189 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3190 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3191
3192 udelay(50);
3193
3194 tmp &= ~grbm_soft_reset;
3195 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3196 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3197 }
3198
3199 /* Wait a little for things to settle down */
3200 udelay(50);
3201 }
3202 return 0;
3203}
3204
3205static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
3206{
3207 uint64_t clock;
3208
3209 mutex_lock(&adev->gfx.gpu_clock_mutex);
3210 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
3211 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
3212 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
3213 mutex_unlock(&adev->gfx.gpu_clock_mutex);
3214 return clock;
3215}
3216
3217static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
3218 uint32_t vmid,
3219 uint32_t gds_base, uint32_t gds_size,
3220 uint32_t gws_base, uint32_t gws_size,
3221 uint32_t oa_base, uint32_t oa_size)
3222{
3223 struct amdgpu_device *adev = ring->adev;
3224
3225 gds_base = gds_base >> AMDGPU_GDS_SHIFT;
3226 gds_size = gds_size >> AMDGPU_GDS_SHIFT;
3227
3228 gws_base = gws_base >> AMDGPU_GWS_SHIFT;
3229 gws_size = gws_size >> AMDGPU_GWS_SHIFT;
3230
3231 oa_base = oa_base >> AMDGPU_OA_SHIFT;
3232 oa_size = oa_size >> AMDGPU_OA_SHIFT;
3233
3234 /* GDS Base */
3235 gfx_v9_0_write_data_to_reg(ring, 0, false,
3236 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
3237 gds_base);
3238
3239 /* GDS Size */
3240 gfx_v9_0_write_data_to_reg(ring, 0, false,
3241 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
3242 gds_size);
3243
3244 /* GWS */
3245 gfx_v9_0_write_data_to_reg(ring, 0, false,
3246 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
3247 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
3248
3249 /* OA */
3250 gfx_v9_0_write_data_to_reg(ring, 0, false,
3251 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
3252 (1 << (oa_size + oa_base)) - (1 << oa_base));
3253}
3254
3255static int gfx_v9_0_early_init(void *handle)
3256{
3257 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3258
3259 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
3260 adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
3261 gfx_v9_0_set_ring_funcs(adev);
3262 gfx_v9_0_set_irq_funcs(adev);
3263 gfx_v9_0_set_gds_init(adev);
3264 gfx_v9_0_set_rlc_funcs(adev);
3265
3266 return 0;
3267}
3268
3269static int gfx_v9_0_late_init(void *handle)
3270{
3271 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3272 int r;
3273
3274 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
3275 if (r)
3276 return r;
3277
3278 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
3279 if (r)
3280 return r;
3281
3282 return 0;
3283}
3284
3285static void gfx_v9_0_enter_rlc_safe_mode(struct amdgpu_device *adev)
3286{
3287 uint32_t rlc_setting, data;
3288 unsigned i;
3289
3290 if (adev->gfx.rlc.in_safe_mode)
3291 return;
3292
3293 /* if RLC is not enabled, do nothing */
3294 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
3295 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
3296 return;
3297
3298 if (adev->cg_flags &
3299 (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG |
3300 AMD_CG_SUPPORT_GFX_3D_CGCG)) {
3301 data = RLC_SAFE_MODE__CMD_MASK;
3302 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
3303 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
3304
3305 /* wait for RLC_SAFE_MODE */
3306 for (i = 0; i < adev->usec_timeout; i++) {
3307 if (!REG_GET_FIELD(SOC15_REG_OFFSET(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
3308 break;
3309 udelay(1);
3310 }
3311 adev->gfx.rlc.in_safe_mode = true;
3312 }
3313}
3314
3315static void gfx_v9_0_exit_rlc_safe_mode(struct amdgpu_device *adev)
3316{
3317 uint32_t rlc_setting, data;
3318
3319 if (!adev->gfx.rlc.in_safe_mode)
3320 return;
3321
3322 /* if RLC is not enabled, do nothing */
3323 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
3324 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
3325 return;
3326
3327 if (adev->cg_flags &
3328 (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) {
3329 /*
3330 * Try to exit safe mode only if it is already in safe
3331 * mode.
3332 */
3333 data = RLC_SAFE_MODE__CMD_MASK;
3334 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
3335 adev->gfx.rlc.in_safe_mode = false;
3336 }
3337}
3338
3339static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
3340 bool enable)
3341{
3342 /* TODO: double check if we need to perform under safe mdoe */
3343 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
3344
3345 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
3346 gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
3347 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
3348 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
3349 } else {
3350 gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
3351 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
3352 }
3353
3354 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
3355}
3356
3357static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
3358 bool enable)
3359{
3360 /* TODO: double check if we need to perform under safe mode */
3361 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
3362
3363 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
3364 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
3365 else
3366 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
3367
3368 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
3369 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
3370 else
3371 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
3372
3373 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
3374}
3375
3376static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
3377 bool enable)
3378{
3379 uint32_t data, def;
3380
3381 /* It is disabled by HW by default */
3382 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3383 /* 1 - RLC_CGTT_MGCG_OVERRIDE */
3384 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
3385 data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK |
3386 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
3387 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
3388 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
3389
3390 /* only for Vega10 & Raven1 */
3391 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
3392
3393 if (def != data)
3394 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
3395
3396 /* MGLS is a global flag to control all MGLS in GFX */
3397 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3398 /* 2 - RLC memory Light sleep */
3399 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
3400 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
3401 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3402 if (def != data)
3403 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
3404 }
3405 /* 3 - CP memory Light sleep */
3406 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3407 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
3408 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3409 if (def != data)
3410 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
3411 }
3412 }
3413 } else {
3414 /* 1 - MGCG_OVERRIDE */
3415 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
3416 data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK |
3417 RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
3418 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
3419 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
3420 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
3421 if (def != data)
3422 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
3423
3424 /* 2 - disable MGLS in RLC */
3425 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
3426 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3427 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3428 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
3429 }
3430
3431 /* 3 - disable MGLS in CP */
3432 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
3433 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3434 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3435 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
3436 }
3437 }
3438}
3439
3440static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
3441 bool enable)
3442{
3443 uint32_t data, def;
3444
3445 adev->gfx.rlc.funcs->enter_safe_mode(adev);
3446
3447 /* Enable 3D CGCG/CGLS */
3448 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)) {
3449 /* write cmd to clear cgcg/cgls ov */
3450 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
3451 /* unset CGCG override */
3452 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
3453 /* update CGCG and CGLS override bits */
3454 if (def != data)
3455 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
3456 /* enable 3Dcgcg FSM(0x0020003f) */
3457 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
3458 data = (0x2000 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
3459 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
3460 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
3461 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
3462 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
3463 if (def != data)
3464 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
3465
3466 /* set IDLE_POLL_COUNT(0x00900100) */
3467 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
3468 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
3469 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3470 if (def != data)
3471 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
3472 } else {
3473 /* Disable CGCG/CGLS */
3474 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
3475 /* disable cgcg, cgls should be disabled */
3476 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
3477 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
3478 /* disable cgcg and cgls in FSM */
3479 if (def != data)
3480 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
3481 }
3482
3483 adev->gfx.rlc.funcs->exit_safe_mode(adev);
3484}
3485
3486static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
3487 bool enable)
3488{
3489 uint32_t def, data;
3490
3491 adev->gfx.rlc.funcs->enter_safe_mode(adev);
3492
3493 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3494 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
3495 /* unset CGCG override */
3496 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
3497 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
3498 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
3499 else
3500 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
3501 /* update CGCG and CGLS override bits */
3502 if (def != data)
3503 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
3504
3505 /* enable cgcg FSM(0x0020003F) */
3506 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
3507 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
3508 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
3509 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
3510 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
3511 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3512 if (def != data)
3513 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
3514
3515 /* set IDLE_POLL_COUNT(0x00900100) */
3516 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
3517 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
3518 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3519 if (def != data)
3520 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
3521 } else {
3522 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
3523 /* reset CGCG/CGLS bits */
3524 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3525 /* disable cgcg and cgls in FSM */
3526 if (def != data)
3527 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
3528 }
3529
3530 adev->gfx.rlc.funcs->exit_safe_mode(adev);
3531}
3532
3533static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
3534 bool enable)
3535{
3536 if (enable) {
3537 /* CGCG/CGLS should be enabled after MGCG/MGLS
3538 * === MGCG + MGLS ===
3539 */
3540 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
3541 /* === CGCG /CGLS for GFX 3D Only === */
3542 gfx_v9_0_update_3d_clock_gating(adev, enable);
3543 /* === CGCG + CGLS === */
3544 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
3545 } else {
3546 /* CGCG/CGLS should be disabled before MGCG/MGLS
3547 * === CGCG + CGLS ===
3548 */
3549 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
3550 /* === CGCG /CGLS for GFX 3D Only === */
3551 gfx_v9_0_update_3d_clock_gating(adev, enable);
3552 /* === MGCG + MGLS === */
3553 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
3554 }
3555 return 0;
3556}
3557
3558static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
3559 .enter_safe_mode = gfx_v9_0_enter_rlc_safe_mode,
3560 .exit_safe_mode = gfx_v9_0_exit_rlc_safe_mode
3561};
3562
3563static int gfx_v9_0_set_powergating_state(void *handle,
3564 enum amd_powergating_state state)
3565{
3566 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3567 bool enable = (state == AMD_PG_STATE_GATE) ? true : false;
3568
3569 switch (adev->asic_type) {
3570 case CHIP_RAVEN:
3571 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
3572 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
3573 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
3574 } else {
3575 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
3576 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
3577 }
3578
3579 if (adev->pg_flags & AMD_PG_SUPPORT_CP)
3580 gfx_v9_0_enable_cp_power_gating(adev, true);
3581 else
3582 gfx_v9_0_enable_cp_power_gating(adev, false);
3583
3584 /* update gfx cgpg state */
3585 gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
3586
3587 /* update mgcg state */
3588 gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
3589 break;
3590 default:
3591 break;
3592 }
3593
3594 return 0;
3595}
3596
3597static int gfx_v9_0_set_clockgating_state(void *handle,
3598 enum amd_clockgating_state state)
3599{
3600 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3601
3602 if (amdgpu_sriov_vf(adev))
3603 return 0;
3604
3605 switch (adev->asic_type) {
3606 case CHIP_VEGA10:
3607 case CHIP_VEGA12:
3608 case CHIP_RAVEN:
3609 gfx_v9_0_update_gfx_clock_gating(adev,
3610 state == AMD_CG_STATE_GATE ? true : false);
3611 break;
3612 default:
3613 break;
3614 }
3615 return 0;
3616}
3617
3618static void gfx_v9_0_get_clockgating_state(void *handle, u32 *flags)
3619{
3620 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3621 int data;
3622
3623 if (amdgpu_sriov_vf(adev))
3624 *flags = 0;
3625
3626 /* AMD_CG_SUPPORT_GFX_MGCG */
3627 data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
3628 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
3629 *flags |= AMD_CG_SUPPORT_GFX_MGCG;
3630
3631 /* AMD_CG_SUPPORT_GFX_CGCG */
3632 data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
3633 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
3634 *flags |= AMD_CG_SUPPORT_GFX_CGCG;
3635
3636 /* AMD_CG_SUPPORT_GFX_CGLS */
3637 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
3638 *flags |= AMD_CG_SUPPORT_GFX_CGLS;
3639
3640 /* AMD_CG_SUPPORT_GFX_RLC_LS */
3641 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
3642 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
3643 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
3644
3645 /* AMD_CG_SUPPORT_GFX_CP_LS */
3646 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
3647 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
3648 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
3649
3650 /* AMD_CG_SUPPORT_GFX_3D_CGCG */
3651 data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
3652 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
3653 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
3654
3655 /* AMD_CG_SUPPORT_GFX_3D_CGLS */
3656 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
3657 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
3658}
3659
3660static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
3661{
3662 return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 is 32bit rptr*/
3663}
3664
3665static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
3666{
3667 struct amdgpu_device *adev = ring->adev;
3668 u64 wptr;
3669
3670 /* XXX check if swapping is necessary on BE */
3671 if (ring->use_doorbell) {
3672 wptr = atomic64_read((atomic64_t *)&adev->wb.wb[ring->wptr_offs]);
3673 } else {
3674 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
3675 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
3676 }
3677
3678 return wptr;
3679}
3680
3681static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
3682{
3683 struct amdgpu_device *adev = ring->adev;
3684
3685 if (ring->use_doorbell) {
3686 /* XXX check if swapping is necessary on BE */
3687 atomic64_set((atomic64_t*)&adev->wb.wb[ring->wptr_offs], ring->wptr);
3688 WDOORBELL64(ring->doorbell_index, ring->wptr);
3689 } else {
3690 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3691 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3692 }
3693}
3694
3695static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
3696{
3697 struct amdgpu_device *adev = ring->adev;
3698 u32 ref_and_mask, reg_mem_engine;
3699 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio_funcs->hdp_flush_reg;
3700
3701 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3702 switch (ring->me) {
3703 case 1:
3704 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
3705 break;
3706 case 2:
3707 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
3708 break;
3709 default:
3710 return;
3711 }
3712 reg_mem_engine = 0;
3713 } else {
3714 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
3715 reg_mem_engine = 1; /* pfp */
3716 }
3717
3718 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
3719 adev->nbio_funcs->get_hdp_flush_req_offset(adev),
3720 adev->nbio_funcs->get_hdp_flush_done_offset(adev),
3721 ref_and_mask, ref_and_mask, 0x20);
3722}
3723
3724static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
3725 struct amdgpu_ib *ib,
3726 unsigned vmid, bool ctx_switch)
3727{
3728 u32 header, control = 0;
3729
3730 if (ib->flags & AMDGPU_IB_FLAG_CE)
3731 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
3732 else
3733 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
3734
3735 control |= ib->length_dw | (vmid << 24);
3736
3737 if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
3738 control |= INDIRECT_BUFFER_PRE_ENB(1);
3739
3740 if (!(ib->flags & AMDGPU_IB_FLAG_CE))
3741 gfx_v9_0_ring_emit_de_meta(ring);
3742 }
3743
3744 amdgpu_ring_write(ring, header);
3745BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
3746 amdgpu_ring_write(ring,
3747#ifdef __BIG_ENDIAN
3748 (2 << 0) |
3749#endif
3750 lower_32_bits(ib->gpu_addr));
3751 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
3752 amdgpu_ring_write(ring, control);
3753}
3754
3755static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
3756 struct amdgpu_ib *ib,
3757 unsigned vmid, bool ctx_switch)
3758{
3759 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
3760
3761 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
3762 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
3763 amdgpu_ring_write(ring,
3764#ifdef __BIG_ENDIAN
3765 (2 << 0) |
3766#endif
3767 lower_32_bits(ib->gpu_addr));
3768 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
3769 amdgpu_ring_write(ring, control);
3770}
3771
3772static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
3773 u64 seq, unsigned flags)
3774{
3775 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
3776 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
3777
3778 /* RELEASE_MEM - flush caches, send int */
3779 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
3780 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
3781 EOP_TC_ACTION_EN |
3782 EOP_TC_WB_ACTION_EN |
3783 EOP_TC_MD_ACTION_EN |
3784 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
3785 EVENT_INDEX(5)));
3786 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
3787
3788 /*
3789 * the address should be Qword aligned if 64bit write, Dword
3790 * aligned if only send 32bit data low (discard data high)
3791 */
3792 if (write64bit)
3793 BUG_ON(addr & 0x7);
3794 else
3795 BUG_ON(addr & 0x3);
3796 amdgpu_ring_write(ring, lower_32_bits(addr));
3797 amdgpu_ring_write(ring, upper_32_bits(addr));
3798 amdgpu_ring_write(ring, lower_32_bits(seq));
3799 amdgpu_ring_write(ring, upper_32_bits(seq));
3800 amdgpu_ring_write(ring, 0);
3801}
3802
3803static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3804{
3805 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3806 uint32_t seq = ring->fence_drv.sync_seq;
3807 uint64_t addr = ring->fence_drv.gpu_addr;
3808
3809 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
3810 lower_32_bits(addr), upper_32_bits(addr),
3811 seq, 0xffffffff, 4);
3812}
3813
3814static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3815 unsigned vmid, uint64_t pd_addr)
3816{
3817 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3818
3819 /* compute doesn't have PFP */
3820 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
3821 /* sync PFP to ME, otherwise we might get invalid PFP reads */
3822 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3823 amdgpu_ring_write(ring, 0x0);
3824 }
3825}
3826
3827static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
3828{
3829 return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 hardware is 32bit rptr */
3830}
3831
3832static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
3833{
3834 u64 wptr;
3835
3836 /* XXX check if swapping is necessary on BE */
3837 if (ring->use_doorbell)
3838 wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]);
3839 else
3840 BUG();
3841 return wptr;
3842}
3843
3844static void gfx_v9_0_ring_set_pipe_percent(struct amdgpu_ring *ring,
3845 bool acquire)
3846{
3847 struct amdgpu_device *adev = ring->adev;
3848 int pipe_num, tmp, reg;
3849 int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1;
3850
3851 pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe;
3852
3853 /* first me only has 2 entries, GFX and HP3D */
3854 if (ring->me > 0)
3855 pipe_num -= 2;
3856
3857 reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX) + pipe_num;
3858 tmp = RREG32(reg);
3859 tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent);
3860 WREG32(reg, tmp);
3861}
3862
3863static void gfx_v9_0_pipe_reserve_resources(struct amdgpu_device *adev,
3864 struct amdgpu_ring *ring,
3865 bool acquire)
3866{
3867 int i, pipe;
3868 bool reserve;
3869 struct amdgpu_ring *iring;
3870
3871 mutex_lock(&adev->gfx.pipe_reserve_mutex);
3872 pipe = amdgpu_gfx_queue_to_bit(adev, ring->me, ring->pipe, 0);
3873 if (acquire)
3874 set_bit(pipe, adev->gfx.pipe_reserve_bitmap);
3875 else
3876 clear_bit(pipe, adev->gfx.pipe_reserve_bitmap);
3877
3878 if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) {
3879 /* Clear all reservations - everyone reacquires all resources */
3880 for (i = 0; i < adev->gfx.num_gfx_rings; ++i)
3881 gfx_v9_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i],
3882 true);
3883
3884 for (i = 0; i < adev->gfx.num_compute_rings; ++i)
3885 gfx_v9_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i],
3886 true);
3887 } else {
3888 /* Lower all pipes without a current reservation */
3889 for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
3890 iring = &adev->gfx.gfx_ring[i];
3891 pipe = amdgpu_gfx_queue_to_bit(adev,
3892 iring->me,
3893 iring->pipe,
3894 0);
3895 reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
3896 gfx_v9_0_ring_set_pipe_percent(iring, reserve);
3897 }
3898
3899 for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
3900 iring = &adev->gfx.compute_ring[i];
3901 pipe = amdgpu_gfx_queue_to_bit(adev,
3902 iring->me,
3903 iring->pipe,
3904 0);
3905 reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
3906 gfx_v9_0_ring_set_pipe_percent(iring, reserve);
3907 }
3908 }
3909
3910 mutex_unlock(&adev->gfx.pipe_reserve_mutex);
3911}
3912
3913static void gfx_v9_0_hqd_set_priority(struct amdgpu_device *adev,
3914 struct amdgpu_ring *ring,
3915 bool acquire)
3916{
3917 uint32_t pipe_priority = acquire ? 0x2 : 0x0;
3918 uint32_t queue_priority = acquire ? 0xf : 0x0;
3919
3920 mutex_lock(&adev->srbm_mutex);
3921 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3922
3923 WREG32_SOC15(GC, 0, mmCP_HQD_PIPE_PRIORITY, pipe_priority);
3924 WREG32_SOC15(GC, 0, mmCP_HQD_QUEUE_PRIORITY, queue_priority);
3925
3926 soc15_grbm_select(adev, 0, 0, 0, 0);
3927 mutex_unlock(&adev->srbm_mutex);
3928}
3929
3930static void gfx_v9_0_ring_set_priority_compute(struct amdgpu_ring *ring,
3931 enum drm_sched_priority priority)
3932{
3933 struct amdgpu_device *adev = ring->adev;
3934 bool acquire = priority == DRM_SCHED_PRIORITY_HIGH_HW;
3935
3936 if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
3937 return;
3938
3939 gfx_v9_0_hqd_set_priority(adev, ring, acquire);
3940 gfx_v9_0_pipe_reserve_resources(adev, ring, acquire);
3941}
3942
3943static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
3944{
3945 struct amdgpu_device *adev = ring->adev;
3946
3947 /* XXX check if swapping is necessary on BE */
3948 if (ring->use_doorbell) {
3949 atomic64_set((atomic64_t*)&adev->wb.wb[ring->wptr_offs], ring->wptr);
3950 WDOORBELL64(ring->doorbell_index, ring->wptr);
3951 } else{
3952 BUG(); /* only DOORBELL method supported on gfx9 now */
3953 }
3954}
3955
3956static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
3957 u64 seq, unsigned int flags)
3958{
3959 struct amdgpu_device *adev = ring->adev;
3960
3961 /* we only allocate 32bit for each seq wb address */
3962 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
3963
3964 /* write fence seq to the "addr" */
3965 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3966 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3967 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
3968 amdgpu_ring_write(ring, lower_32_bits(addr));
3969 amdgpu_ring_write(ring, upper_32_bits(addr));
3970 amdgpu_ring_write(ring, lower_32_bits(seq));
3971
3972 if (flags & AMDGPU_FENCE_FLAG_INT) {
3973 /* set register to trigger INT */
3974 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3975 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3976 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
3977 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
3978 amdgpu_ring_write(ring, 0);
3979 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
3980 }
3981}
3982
3983static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
3984{
3985 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3986 amdgpu_ring_write(ring, 0);
3987}
3988
3989static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
3990{
3991 struct v9_ce_ib_state ce_payload = {0};
3992 uint64_t csa_addr;
3993 int cnt;
3994
3995 cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
3996 csa_addr = amdgpu_csa_vaddr(ring->adev);
3997
3998 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
3999 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
4000 WRITE_DATA_DST_SEL(8) |
4001 WR_CONFIRM) |
4002 WRITE_DATA_CACHE_POLICY(0));
4003 amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
4004 amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
4005 amdgpu_ring_write_multiple(ring, (void *)&ce_payload, sizeof(ce_payload) >> 2);
4006}
4007
4008static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring)
4009{
4010 struct v9_de_ib_state de_payload = {0};
4011 uint64_t csa_addr, gds_addr;
4012 int cnt;
4013
4014 csa_addr = amdgpu_csa_vaddr(ring->adev);
4015 gds_addr = csa_addr + 4096;
4016 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
4017 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
4018
4019 cnt = (sizeof(de_payload) >> 2) + 4 - 2;
4020 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
4021 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
4022 WRITE_DATA_DST_SEL(8) |
4023 WR_CONFIRM) |
4024 WRITE_DATA_CACHE_POLICY(0));
4025 amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
4026 amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
4027 amdgpu_ring_write_multiple(ring, (void *)&de_payload, sizeof(de_payload) >> 2);
4028}
4029
4030static void gfx_v9_0_ring_emit_tmz(struct amdgpu_ring *ring, bool start)
4031{
4032 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
4033 amdgpu_ring_write(ring, FRAME_CMD(start ? 0 : 1)); /* frame_end */
4034}
4035
4036static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
4037{
4038 uint32_t dw2 = 0;
4039
4040 if (amdgpu_sriov_vf(ring->adev))
4041 gfx_v9_0_ring_emit_ce_meta(ring);
4042
4043 gfx_v9_0_ring_emit_tmz(ring, true);
4044
4045 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
4046 if (flags & AMDGPU_HAVE_CTX_SWITCH) {
4047 /* set load_global_config & load_global_uconfig */
4048 dw2 |= 0x8001;
4049 /* set load_cs_sh_regs */
4050 dw2 |= 0x01000000;
4051 /* set load_per_context_state & load_gfx_sh_regs for GFX */
4052 dw2 |= 0x10002;
4053
4054 /* set load_ce_ram if preamble presented */
4055 if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
4056 dw2 |= 0x10000000;
4057 } else {
4058 /* still load_ce_ram if this is the first time preamble presented
4059 * although there is no context switch happens.
4060 */
4061 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
4062 dw2 |= 0x10000000;
4063 }
4064
4065 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4066 amdgpu_ring_write(ring, dw2);
4067 amdgpu_ring_write(ring, 0);
4068}
4069
4070static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
4071{
4072 unsigned ret;
4073 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
4074 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
4075 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
4076 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
4077 ret = ring->wptr & ring->buf_mask;
4078 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
4079 return ret;
4080}
4081
4082static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
4083{
4084 unsigned cur;
4085 BUG_ON(offset > ring->buf_mask);
4086 BUG_ON(ring->ring[offset] != 0x55aa55aa);
4087
4088 cur = (ring->wptr & ring->buf_mask) - 1;
4089 if (likely(cur > offset))
4090 ring->ring[offset] = cur - offset;
4091 else
4092 ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
4093}
4094
4095static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg)
4096{
4097 struct amdgpu_device *adev = ring->adev;
4098
4099 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
4100 amdgpu_ring_write(ring, 0 | /* src: register*/
4101 (5 << 8) | /* dst: memory */
4102 (1 << 20)); /* write confirm */
4103 amdgpu_ring_write(ring, reg);
4104 amdgpu_ring_write(ring, 0);
4105 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
4106 adev->virt.reg_val_offs * 4));
4107 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
4108 adev->virt.reg_val_offs * 4));
4109}
4110
4111static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
4112 uint32_t val)
4113{
4114 uint32_t cmd = 0;
4115
4116 switch (ring->funcs->type) {
4117 case AMDGPU_RING_TYPE_GFX:
4118 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
4119 break;
4120 case AMDGPU_RING_TYPE_KIQ:
4121 cmd = (1 << 16); /* no inc addr */
4122 break;
4123 default:
4124 cmd = WR_CONFIRM;
4125 break;
4126 }
4127 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4128 amdgpu_ring_write(ring, cmd);
4129 amdgpu_ring_write(ring, reg);
4130 amdgpu_ring_write(ring, 0);
4131 amdgpu_ring_write(ring, val);
4132}
4133
4134static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
4135 uint32_t val, uint32_t mask)
4136{
4137 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
4138}
4139
4140static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4141 enum amdgpu_interrupt_state state)
4142{
4143 switch (state) {
4144 case AMDGPU_IRQ_STATE_DISABLE:
4145 case AMDGPU_IRQ_STATE_ENABLE:
4146 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
4147 TIME_STAMP_INT_ENABLE,
4148 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
4149 break;
4150 default:
4151 break;
4152 }
4153}
4154
4155static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4156 int me, int pipe,
4157 enum amdgpu_interrupt_state state)
4158{
4159 u32 mec_int_cntl, mec_int_cntl_reg;
4160
4161 /*
4162 * amdgpu controls only the first MEC. That's why this function only
4163 * handles the setting of interrupts for this specific MEC. All other
4164 * pipes' interrupts are set by amdkfd.
4165 */
4166
4167 if (me == 1) {
4168 switch (pipe) {
4169 case 0:
4170 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
4171 break;
4172 case 1:
4173 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
4174 break;
4175 case 2:
4176 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
4177 break;
4178 case 3:
4179 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
4180 break;
4181 default:
4182 DRM_DEBUG("invalid pipe %d\n", pipe);
4183 return;
4184 }
4185 } else {
4186 DRM_DEBUG("invalid me %d\n", me);
4187 return;
4188 }
4189
4190 switch (state) {
4191 case AMDGPU_IRQ_STATE_DISABLE:
4192 mec_int_cntl = RREG32(mec_int_cntl_reg);
4193 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
4194 TIME_STAMP_INT_ENABLE, 0);
4195 WREG32(mec_int_cntl_reg, mec_int_cntl);
4196 break;
4197 case AMDGPU_IRQ_STATE_ENABLE:
4198 mec_int_cntl = RREG32(mec_int_cntl_reg);
4199 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
4200 TIME_STAMP_INT_ENABLE, 1);
4201 WREG32(mec_int_cntl_reg, mec_int_cntl);
4202 break;
4203 default:
4204 break;
4205 }
4206}
4207
4208static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4209 struct amdgpu_irq_src *source,
4210 unsigned type,
4211 enum amdgpu_interrupt_state state)
4212{
4213 switch (state) {
4214 case AMDGPU_IRQ_STATE_DISABLE:
4215 case AMDGPU_IRQ_STATE_ENABLE:
4216 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
4217 PRIV_REG_INT_ENABLE,
4218 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
4219 break;
4220 default:
4221 break;
4222 }
4223
4224 return 0;
4225}
4226
4227static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4228 struct amdgpu_irq_src *source,
4229 unsigned type,
4230 enum amdgpu_interrupt_state state)
4231{
4232 switch (state) {
4233 case AMDGPU_IRQ_STATE_DISABLE:
4234 case AMDGPU_IRQ_STATE_ENABLE:
4235 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
4236 PRIV_INSTR_INT_ENABLE,
4237 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
4238 default:
4239 break;
4240 }
4241
4242 return 0;
4243}
4244
4245static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4246 struct amdgpu_irq_src *src,
4247 unsigned type,
4248 enum amdgpu_interrupt_state state)
4249{
4250 switch (type) {
4251 case AMDGPU_CP_IRQ_GFX_EOP:
4252 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
4253 break;
4254 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4255 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4256 break;
4257 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4258 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4259 break;
4260 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4261 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4262 break;
4263 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4264 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4265 break;
4266 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4267 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4268 break;
4269 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4270 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4271 break;
4272 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4273 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4274 break;
4275 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4276 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4277 break;
4278 default:
4279 break;
4280 }
4281 return 0;
4282}
4283
4284static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
4285 struct amdgpu_irq_src *source,
4286 struct amdgpu_iv_entry *entry)
4287{
4288 int i;
4289 u8 me_id, pipe_id, queue_id;
4290 struct amdgpu_ring *ring;
4291
4292 DRM_DEBUG("IH: CP EOP\n");
4293 me_id = (entry->ring_id & 0x0c) >> 2;
4294 pipe_id = (entry->ring_id & 0x03) >> 0;
4295 queue_id = (entry->ring_id & 0x70) >> 4;
4296
4297 switch (me_id) {
4298 case 0:
4299 amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4300 break;
4301 case 1:
4302 case 2:
4303 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4304 ring = &adev->gfx.compute_ring[i];
4305 /* Per-queue interrupt is supported for MEC starting from VI.
4306 * The interrupt can only be enabled/disabled per pipe instead of per queue.
4307 */
4308 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
4309 amdgpu_fence_process(ring);
4310 }
4311 break;
4312 }
4313 return 0;
4314}
4315
4316static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
4317 struct amdgpu_irq_src *source,
4318 struct amdgpu_iv_entry *entry)
4319{
4320 DRM_ERROR("Illegal register access in command stream\n");
4321 schedule_work(&adev->reset_work);
4322 return 0;
4323}
4324
4325static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
4326 struct amdgpu_irq_src *source,
4327 struct amdgpu_iv_entry *entry)
4328{
4329 DRM_ERROR("Illegal instruction in command stream\n");
4330 schedule_work(&adev->reset_work);
4331 return 0;
4332}
4333
4334static int gfx_v9_0_kiq_set_interrupt_state(struct amdgpu_device *adev,
4335 struct amdgpu_irq_src *src,
4336 unsigned int type,
4337 enum amdgpu_interrupt_state state)
4338{
4339 uint32_t tmp, target;
4340 struct amdgpu_ring *ring = &(adev->gfx.kiq.ring);
4341
4342 if (ring->me == 1)
4343 target = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
4344 else
4345 target = SOC15_REG_OFFSET(GC, 0, mmCP_ME2_PIPE0_INT_CNTL);
4346 target += ring->pipe;
4347
4348 switch (type) {
4349 case AMDGPU_CP_KIQ_IRQ_DRIVER0:
4350 if (state == AMDGPU_IRQ_STATE_DISABLE) {
4351 tmp = RREG32_SOC15(GC, 0, mmCPC_INT_CNTL);
4352 tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
4353 GENERIC2_INT_ENABLE, 0);
4354 WREG32_SOC15(GC, 0, mmCPC_INT_CNTL, tmp);
4355
4356 tmp = RREG32(target);
4357 tmp = REG_SET_FIELD(tmp, CP_ME2_PIPE0_INT_CNTL,
4358 GENERIC2_INT_ENABLE, 0);
4359 WREG32(target, tmp);
4360 } else {
4361 tmp = RREG32_SOC15(GC, 0, mmCPC_INT_CNTL);
4362 tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
4363 GENERIC2_INT_ENABLE, 1);
4364 WREG32_SOC15(GC, 0, mmCPC_INT_CNTL, tmp);
4365
4366 tmp = RREG32(target);
4367 tmp = REG_SET_FIELD(tmp, CP_ME2_PIPE0_INT_CNTL,
4368 GENERIC2_INT_ENABLE, 1);
4369 WREG32(target, tmp);
4370 }
4371 break;
4372 default:
4373 BUG(); /* kiq only support GENERIC2_INT now */
4374 break;
4375 }
4376 return 0;
4377}
4378
4379static int gfx_v9_0_kiq_irq(struct amdgpu_device *adev,
4380 struct amdgpu_irq_src *source,
4381 struct amdgpu_iv_entry *entry)
4382{
4383 u8 me_id, pipe_id, queue_id;
4384 struct amdgpu_ring *ring = &(adev->gfx.kiq.ring);
4385
4386 me_id = (entry->ring_id & 0x0c) >> 2;
4387 pipe_id = (entry->ring_id & 0x03) >> 0;
4388 queue_id = (entry->ring_id & 0x70) >> 4;
4389 DRM_DEBUG("IH: CPC GENERIC2_INT, me:%d, pipe:%d, queue:%d\n",
4390 me_id, pipe_id, queue_id);
4391
4392 amdgpu_fence_process(ring);
4393 return 0;
4394}
4395
4396static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
4397 .name = "gfx_v9_0",
4398 .early_init = gfx_v9_0_early_init,
4399 .late_init = gfx_v9_0_late_init,
4400 .sw_init = gfx_v9_0_sw_init,
4401 .sw_fini = gfx_v9_0_sw_fini,
4402 .hw_init = gfx_v9_0_hw_init,
4403 .hw_fini = gfx_v9_0_hw_fini,
4404 .suspend = gfx_v9_0_suspend,
4405 .resume = gfx_v9_0_resume,
4406 .is_idle = gfx_v9_0_is_idle,
4407 .wait_for_idle = gfx_v9_0_wait_for_idle,
4408 .soft_reset = gfx_v9_0_soft_reset,
4409 .set_clockgating_state = gfx_v9_0_set_clockgating_state,
4410 .set_powergating_state = gfx_v9_0_set_powergating_state,
4411 .get_clockgating_state = gfx_v9_0_get_clockgating_state,
4412};
4413
4414static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
4415 .type = AMDGPU_RING_TYPE_GFX,
4416 .align_mask = 0xff,
4417 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
4418 .support_64bit_ptrs = true,
4419 .vmhub = AMDGPU_GFXHUB,
4420 .get_rptr = gfx_v9_0_ring_get_rptr_gfx,
4421 .get_wptr = gfx_v9_0_ring_get_wptr_gfx,
4422 .set_wptr = gfx_v9_0_ring_set_wptr_gfx,
4423 .emit_frame_size = /* totally 242 maximum if 16 IBs */
4424 5 + /* COND_EXEC */
4425 7 + /* PIPELINE_SYNC */
4426 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
4427 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
4428 2 + /* VM_FLUSH */
4429 8 + /* FENCE for VM_FLUSH */
4430 20 + /* GDS switch */
4431 4 + /* double SWITCH_BUFFER,
4432 the first COND_EXEC jump to the place just
4433 prior to this double SWITCH_BUFFER */
4434 5 + /* COND_EXEC */
4435 7 + /* HDP_flush */
4436 4 + /* VGT_flush */
4437 14 + /* CE_META */
4438 31 + /* DE_META */
4439 3 + /* CNTX_CTRL */
4440 5 + /* HDP_INVL */
4441 8 + 8 + /* FENCE x2 */
4442 2, /* SWITCH_BUFFER */
4443 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
4444 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
4445 .emit_fence = gfx_v9_0_ring_emit_fence,
4446 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
4447 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
4448 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
4449 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
4450 .test_ring = gfx_v9_0_ring_test_ring,
4451 .test_ib = gfx_v9_0_ring_test_ib,
4452 .insert_nop = amdgpu_ring_insert_nop,
4453 .pad_ib = amdgpu_ring_generic_pad_ib,
4454 .emit_switch_buffer = gfx_v9_ring_emit_sb,
4455 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
4456 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
4457 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
4458 .emit_tmz = gfx_v9_0_ring_emit_tmz,
4459 .emit_wreg = gfx_v9_0_ring_emit_wreg,
4460 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
4461};
4462
4463static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
4464 .type = AMDGPU_RING_TYPE_COMPUTE,
4465 .align_mask = 0xff,
4466 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
4467 .support_64bit_ptrs = true,
4468 .vmhub = AMDGPU_GFXHUB,
4469 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
4470 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
4471 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
4472 .emit_frame_size =
4473 20 + /* gfx_v9_0_ring_emit_gds_switch */
4474 7 + /* gfx_v9_0_ring_emit_hdp_flush */
4475 5 + /* hdp invalidate */
4476 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
4477 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
4478 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
4479 2 + /* gfx_v9_0_ring_emit_vm_flush */
4480 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
4481 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_compute */
4482 .emit_ib = gfx_v9_0_ring_emit_ib_compute,
4483 .emit_fence = gfx_v9_0_ring_emit_fence,
4484 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
4485 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
4486 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
4487 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
4488 .test_ring = gfx_v9_0_ring_test_ring,
4489 .test_ib = gfx_v9_0_ring_test_ib,
4490 .insert_nop = amdgpu_ring_insert_nop,
4491 .pad_ib = amdgpu_ring_generic_pad_ib,
4492 .set_priority = gfx_v9_0_ring_set_priority_compute,
4493 .emit_wreg = gfx_v9_0_ring_emit_wreg,
4494 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
4495};
4496
4497static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
4498 .type = AMDGPU_RING_TYPE_KIQ,
4499 .align_mask = 0xff,
4500 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
4501 .support_64bit_ptrs = true,
4502 .vmhub = AMDGPU_GFXHUB,
4503 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
4504 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
4505 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
4506 .emit_frame_size =
4507 20 + /* gfx_v9_0_ring_emit_gds_switch */
4508 7 + /* gfx_v9_0_ring_emit_hdp_flush */
4509 5 + /* hdp invalidate */
4510 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
4511 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
4512 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
4513 2 + /* gfx_v9_0_ring_emit_vm_flush */
4514 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
4515 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_compute */
4516 .emit_ib = gfx_v9_0_ring_emit_ib_compute,
4517 .emit_fence = gfx_v9_0_ring_emit_fence_kiq,
4518 .test_ring = gfx_v9_0_ring_test_ring,
4519 .test_ib = gfx_v9_0_ring_test_ib,
4520 .insert_nop = amdgpu_ring_insert_nop,
4521 .pad_ib = amdgpu_ring_generic_pad_ib,
4522 .emit_rreg = gfx_v9_0_ring_emit_rreg,
4523 .emit_wreg = gfx_v9_0_ring_emit_wreg,
4524 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
4525};
4526
4527static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
4528{
4529 int i;
4530
4531 adev->gfx.kiq.ring.funcs = &gfx_v9_0_ring_funcs_kiq;
4532
4533 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4534 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
4535
4536 for (i = 0; i < adev->gfx.num_compute_rings; i++)
4537 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
4538}
4539
4540static const struct amdgpu_irq_src_funcs gfx_v9_0_kiq_irq_funcs = {
4541 .set = gfx_v9_0_kiq_set_interrupt_state,
4542 .process = gfx_v9_0_kiq_irq,
4543};
4544
4545static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
4546 .set = gfx_v9_0_set_eop_interrupt_state,
4547 .process = gfx_v9_0_eop_irq,
4548};
4549
4550static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
4551 .set = gfx_v9_0_set_priv_reg_fault_state,
4552 .process = gfx_v9_0_priv_reg_irq,
4553};
4554
4555static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
4556 .set = gfx_v9_0_set_priv_inst_fault_state,
4557 .process = gfx_v9_0_priv_inst_irq,
4558};
4559
4560static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
4561{
4562 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
4563 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
4564
4565 adev->gfx.priv_reg_irq.num_types = 1;
4566 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
4567
4568 adev->gfx.priv_inst_irq.num_types = 1;
4569 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
4570
4571 adev->gfx.kiq.irq.num_types = AMDGPU_CP_KIQ_IRQ_LAST;
4572 adev->gfx.kiq.irq.funcs = &gfx_v9_0_kiq_irq_funcs;
4573}
4574
4575static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
4576{
4577 switch (adev->asic_type) {
4578 case CHIP_VEGA10:
4579 case CHIP_VEGA12:
4580 case CHIP_RAVEN:
4581 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
4582 break;
4583 default:
4584 break;
4585 }
4586}
4587
4588static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
4589{
4590 /* init asci gds info */
4591 adev->gds.mem.total_size = RREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE);
4592 adev->gds.gws.total_size = 64;
4593 adev->gds.oa.total_size = 16;
4594
4595 if (adev->gds.mem.total_size == 64 * 1024) {
4596 adev->gds.mem.gfx_partition_size = 4096;
4597 adev->gds.mem.cs_partition_size = 4096;
4598
4599 adev->gds.gws.gfx_partition_size = 4;
4600 adev->gds.gws.cs_partition_size = 4;
4601
4602 adev->gds.oa.gfx_partition_size = 4;
4603 adev->gds.oa.cs_partition_size = 1;
4604 } else {
4605 adev->gds.mem.gfx_partition_size = 1024;
4606 adev->gds.mem.cs_partition_size = 1024;
4607
4608 adev->gds.gws.gfx_partition_size = 16;
4609 adev->gds.gws.cs_partition_size = 16;
4610
4611 adev->gds.oa.gfx_partition_size = 4;
4612 adev->gds.oa.cs_partition_size = 4;
4613 }
4614}
4615
4616static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
4617 u32 bitmap)
4618{
4619 u32 data;
4620
4621 if (!bitmap)
4622 return;
4623
4624 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4625 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4626
4627 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
4628}
4629
4630static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
4631{
4632 u32 data, mask;
4633
4634 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
4635 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
4636
4637 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4638 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4639
4640 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
4641
4642 return (~data) & mask;
4643}
4644
4645static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
4646 struct amdgpu_cu_info *cu_info)
4647{
4648 int i, j, k, counter, active_cu_number = 0;
4649 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
4650 unsigned disable_masks[4 * 2];
4651
4652 if (!adev || !cu_info)
4653 return -EINVAL;
4654
4655 amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
4656
4657 mutex_lock(&adev->grbm_idx_mutex);
4658 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
4659 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
4660 mask = 1;
4661 ao_bitmap = 0;
4662 counter = 0;
4663 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
4664 if (i < 4 && j < 2)
4665 gfx_v9_0_set_user_cu_inactive_bitmap(
4666 adev, disable_masks[i * 2 + j]);
4667 bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
4668 cu_info->bitmap[i][j] = bitmap;
4669
4670 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
4671 if (bitmap & mask) {
4672 if (counter < adev->gfx.config.max_cu_per_sh)
4673 ao_bitmap |= mask;
4674 counter ++;
4675 }
4676 mask <<= 1;
4677 }
4678 active_cu_number += counter;
4679 if (i < 2 && j < 2)
4680 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
4681 cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
4682 }
4683 }
4684 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
4685 mutex_unlock(&adev->grbm_idx_mutex);
4686
4687 cu_info->number = active_cu_number;
4688 cu_info->ao_cu_mask = ao_cu_mask;
4689
4690 return 0;
4691}
4692
4693const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
4694{
4695 .type = AMD_IP_BLOCK_TYPE_GFX,
4696 .major = 9,
4697 .minor = 0,
4698 .rev = 0,
4699 .funcs = &gfx_v9_0_ip_funcs,
4700};
1/*
2 * Copyright 2016 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 "soc15.h"
33#include "soc15d.h"
34#include "amdgpu_atomfirmware.h"
35#include "amdgpu_pm.h"
36
37#include "gc/gc_9_0_offset.h"
38#include "gc/gc_9_0_sh_mask.h"
39
40#include "vega10_enum.h"
41
42#include "soc15_common.h"
43#include "clearstate_gfx9.h"
44#include "v9_structs.h"
45
46#include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
47
48#include "amdgpu_ras.h"
49
50#include "amdgpu_ring_mux.h"
51#include "gfx_v9_4.h"
52#include "gfx_v9_0.h"
53#include "gfx_v9_0_cleaner_shader.h"
54#include "gfx_v9_4_2.h"
55
56#include "asic_reg/pwr/pwr_10_0_offset.h"
57#include "asic_reg/pwr/pwr_10_0_sh_mask.h"
58#include "asic_reg/gc/gc_9_0_default.h"
59
60#define GFX9_NUM_GFX_RINGS 1
61#define GFX9_NUM_SW_GFX_RINGS 2
62#define GFX9_MEC_HPD_SIZE 4096
63#define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
64#define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
65
66#define mmGCEA_PROBE_MAP 0x070c
67#define mmGCEA_PROBE_MAP_BASE_IDX 0
68
69MODULE_FIRMWARE("amdgpu/vega10_ce.bin");
70MODULE_FIRMWARE("amdgpu/vega10_pfp.bin");
71MODULE_FIRMWARE("amdgpu/vega10_me.bin");
72MODULE_FIRMWARE("amdgpu/vega10_mec.bin");
73MODULE_FIRMWARE("amdgpu/vega10_mec2.bin");
74MODULE_FIRMWARE("amdgpu/vega10_rlc.bin");
75
76MODULE_FIRMWARE("amdgpu/vega12_ce.bin");
77MODULE_FIRMWARE("amdgpu/vega12_pfp.bin");
78MODULE_FIRMWARE("amdgpu/vega12_me.bin");
79MODULE_FIRMWARE("amdgpu/vega12_mec.bin");
80MODULE_FIRMWARE("amdgpu/vega12_mec2.bin");
81MODULE_FIRMWARE("amdgpu/vega12_rlc.bin");
82
83MODULE_FIRMWARE("amdgpu/vega20_ce.bin");
84MODULE_FIRMWARE("amdgpu/vega20_pfp.bin");
85MODULE_FIRMWARE("amdgpu/vega20_me.bin");
86MODULE_FIRMWARE("amdgpu/vega20_mec.bin");
87MODULE_FIRMWARE("amdgpu/vega20_mec2.bin");
88MODULE_FIRMWARE("amdgpu/vega20_rlc.bin");
89
90MODULE_FIRMWARE("amdgpu/raven_ce.bin");
91MODULE_FIRMWARE("amdgpu/raven_pfp.bin");
92MODULE_FIRMWARE("amdgpu/raven_me.bin");
93MODULE_FIRMWARE("amdgpu/raven_mec.bin");
94MODULE_FIRMWARE("amdgpu/raven_mec2.bin");
95MODULE_FIRMWARE("amdgpu/raven_rlc.bin");
96
97MODULE_FIRMWARE("amdgpu/picasso_ce.bin");
98MODULE_FIRMWARE("amdgpu/picasso_pfp.bin");
99MODULE_FIRMWARE("amdgpu/picasso_me.bin");
100MODULE_FIRMWARE("amdgpu/picasso_mec.bin");
101MODULE_FIRMWARE("amdgpu/picasso_mec2.bin");
102MODULE_FIRMWARE("amdgpu/picasso_rlc.bin");
103MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin");
104
105MODULE_FIRMWARE("amdgpu/raven2_ce.bin");
106MODULE_FIRMWARE("amdgpu/raven2_pfp.bin");
107MODULE_FIRMWARE("amdgpu/raven2_me.bin");
108MODULE_FIRMWARE("amdgpu/raven2_mec.bin");
109MODULE_FIRMWARE("amdgpu/raven2_mec2.bin");
110MODULE_FIRMWARE("amdgpu/raven2_rlc.bin");
111MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin");
112
113MODULE_FIRMWARE("amdgpu/arcturus_mec.bin");
114MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin");
115
116MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
117MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
118MODULE_FIRMWARE("amdgpu/renoir_me.bin");
119MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
120MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
121
122MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin");
123MODULE_FIRMWARE("amdgpu/green_sardine_pfp.bin");
124MODULE_FIRMWARE("amdgpu/green_sardine_me.bin");
125MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin");
126MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin");
127MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin");
128
129MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin");
130MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin");
131MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin");
132MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec.bin");
133MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec2.bin");
134
135#define mmTCP_CHAN_STEER_0_ARCT 0x0b03
136#define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX 0
137#define mmTCP_CHAN_STEER_1_ARCT 0x0b04
138#define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX 0
139#define mmTCP_CHAN_STEER_2_ARCT 0x0b09
140#define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX 0
141#define mmTCP_CHAN_STEER_3_ARCT 0x0b0a
142#define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX 0
143#define mmTCP_CHAN_STEER_4_ARCT 0x0b0b
144#define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX 0
145#define mmTCP_CHAN_STEER_5_ARCT 0x0b0c
146#define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX 0
147
148#define mmGOLDEN_TSC_COUNT_UPPER_Renoir 0x0025
149#define mmGOLDEN_TSC_COUNT_UPPER_Renoir_BASE_IDX 1
150#define mmGOLDEN_TSC_COUNT_LOWER_Renoir 0x0026
151#define mmGOLDEN_TSC_COUNT_LOWER_Renoir_BASE_IDX 1
152
153static const struct amdgpu_hwip_reg_entry gc_reg_list_9[] = {
154 SOC15_REG_ENTRY_STR(GC, 0, mmGRBM_STATUS),
155 SOC15_REG_ENTRY_STR(GC, 0, mmGRBM_STATUS2),
156 SOC15_REG_ENTRY_STR(GC, 0, mmCP_STALLED_STAT1),
157 SOC15_REG_ENTRY_STR(GC, 0, mmCP_STALLED_STAT2),
158 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CPC_STALLED_STAT1),
159 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CPF_STALLED_STAT1),
160 SOC15_REG_ENTRY_STR(GC, 0, mmCP_BUSY_STAT),
161 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CPC_BUSY_STAT),
162 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CPF_BUSY_STAT),
163 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CPF_STATUS),
164 SOC15_REG_ENTRY_STR(GC, 0, mmCP_GFX_ERROR),
165 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB_BASE),
166 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB_RPTR),
167 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB_WPTR),
168 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB0_BASE),
169 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB0_RPTR),
170 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB0_WPTR),
171 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB1_BASE),
172 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB1_RPTR),
173 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB1_WPTR),
174 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB2_BASE),
175 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB2_WPTR),
176 SOC15_REG_ENTRY_STR(GC, 0, mmCP_RB2_WPTR),
177 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB1_CMD_BUFSZ),
178 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB2_CMD_BUFSZ),
179 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB1_CMD_BUFSZ),
180 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB2_CMD_BUFSZ),
181 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB1_BASE_LO),
182 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB1_BASE_HI),
183 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB1_BUFSZ),
184 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB2_BASE_LO),
185 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB2_BASE_HI),
186 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_IB2_BUFSZ),
187 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB1_BASE_LO),
188 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB1_BASE_HI),
189 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB1_BUFSZ),
190 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB2_BASE_LO),
191 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB2_BASE_HI),
192 SOC15_REG_ENTRY_STR(GC, 0, mmCP_IB2_BUFSZ),
193 SOC15_REG_ENTRY_STR(GC, 0, mmCPF_UTCL1_STATUS),
194 SOC15_REG_ENTRY_STR(GC, 0, mmCPC_UTCL1_STATUS),
195 SOC15_REG_ENTRY_STR(GC, 0, mmCPG_UTCL1_STATUS),
196 SOC15_REG_ENTRY_STR(GC, 0, mmGDS_PROTECTION_FAULT),
197 SOC15_REG_ENTRY_STR(GC, 0, mmGDS_VM_PROTECTION_FAULT),
198 SOC15_REG_ENTRY_STR(GC, 0, mmIA_UTCL1_STATUS),
199 SOC15_REG_ENTRY_STR(GC, 0, mmIA_UTCL1_CNTL),
200 SOC15_REG_ENTRY_STR(GC, 0, mmPA_CL_CNTL_STATUS),
201 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_UTCL1_STATUS),
202 SOC15_REG_ENTRY_STR(GC, 0, mmRMI_UTCL1_STATUS),
203 SOC15_REG_ENTRY_STR(GC, 0, mmSQC_DCACHE_UTCL1_STATUS),
204 SOC15_REG_ENTRY_STR(GC, 0, mmSQC_ICACHE_UTCL1_STATUS),
205 SOC15_REG_ENTRY_STR(GC, 0, mmSQ_UTCL1_STATUS),
206 SOC15_REG_ENTRY_STR(GC, 0, mmTCP_UTCL1_STATUS),
207 SOC15_REG_ENTRY_STR(GC, 0, mmWD_UTCL1_STATUS),
208 SOC15_REG_ENTRY_STR(GC, 0, mmVM_L2_PROTECTION_FAULT_CNTL),
209 SOC15_REG_ENTRY_STR(GC, 0, mmVM_L2_PROTECTION_FAULT_STATUS),
210 SOC15_REG_ENTRY_STR(GC, 0, mmCP_DEBUG),
211 SOC15_REG_ENTRY_STR(GC, 0, mmCP_MEC_CNTL),
212 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_INSTR_PNTR),
213 SOC15_REG_ENTRY_STR(GC, 0, mmCP_MEC1_INSTR_PNTR),
214 SOC15_REG_ENTRY_STR(GC, 0, mmCP_MEC2_INSTR_PNTR),
215 SOC15_REG_ENTRY_STR(GC, 0, mmCP_ME_INSTR_PNTR),
216 SOC15_REG_ENTRY_STR(GC, 0, mmCP_PFP_INSTR_PNTR),
217 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CPC_STATUS),
218 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_STAT),
219 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_SMU_COMMAND),
220 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_SMU_MESSAGE),
221 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_SMU_ARGUMENT_1),
222 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_SMU_ARGUMENT_2),
223 SOC15_REG_ENTRY_STR(GC, 0, mmSMU_RLC_RESPONSE),
224 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_SAFE_MODE),
225 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_SMU_SAFE_MODE),
226 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_INT_STAT),
227 SOC15_REG_ENTRY_STR(GC, 0, mmRLC_GPM_GENERAL_6),
228 /* cp header registers */
229 SOC15_REG_ENTRY_STR(GC, 0, mmCP_CE_HEADER_DUMP),
230 SOC15_REG_ENTRY_STR(GC, 0, mmCP_MEC_ME1_HEADER_DUMP),
231 SOC15_REG_ENTRY_STR(GC, 0, mmCP_MEC_ME2_HEADER_DUMP),
232 SOC15_REG_ENTRY_STR(GC, 0, mmCP_PFP_HEADER_DUMP),
233 SOC15_REG_ENTRY_STR(GC, 0, mmCP_ME_HEADER_DUMP),
234 /* SE status registers */
235 SOC15_REG_ENTRY_STR(GC, 0, mmGRBM_STATUS_SE0),
236 SOC15_REG_ENTRY_STR(GC, 0, mmGRBM_STATUS_SE1),
237 SOC15_REG_ENTRY_STR(GC, 0, mmGRBM_STATUS_SE2),
238 SOC15_REG_ENTRY_STR(GC, 0, mmGRBM_STATUS_SE3)
239};
240
241static const struct amdgpu_hwip_reg_entry gc_cp_reg_list_9[] = {
242 /* compute queue registers */
243 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_VMID),
244 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_ACTIVE),
245 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PERSISTENT_STATE),
246 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PIPE_PRIORITY),
247 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_QUEUE_PRIORITY),
248 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_QUANTUM),
249 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_BASE),
250 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_BASE_HI),
251 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_RPTR),
252 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR),
253 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI),
254 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL),
255 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_CONTROL),
256 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_IB_BASE_ADDR),
257 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_IB_BASE_ADDR_HI),
258 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_IB_RPTR),
259 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_IB_CONTROL),
260 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_DEQUEUE_REQUEST),
261 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_BASE_ADDR),
262 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI),
263 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_CONTROL),
264 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_RPTR),
265 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_WPTR),
266 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_EVENTS),
267 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_CTX_SAVE_BASE_ADDR_LO),
268 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_CTX_SAVE_BASE_ADDR_HI),
269 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_CTX_SAVE_CONTROL),
270 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_CNTL_STACK_OFFSET),
271 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_CNTL_STACK_SIZE),
272 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_WG_STATE_OFFSET),
273 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_CTX_SAVE_SIZE),
274 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_GDS_RESOURCE_STATE),
275 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_ERROR),
276 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_EOP_WPTR_MEM),
277 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_WPTR_LO),
278 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_PQ_WPTR_HI),
279 SOC15_REG_ENTRY_STR(GC, 0, mmCP_HQD_GFX_STATUS),
280};
281
282enum ta_ras_gfx_subblock {
283 /*CPC*/
284 TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
285 TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START,
286 TA_RAS_BLOCK__GFX_CPC_UCODE,
287 TA_RAS_BLOCK__GFX_DC_STATE_ME1,
288 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
289 TA_RAS_BLOCK__GFX_DC_RESTORE_ME1,
290 TA_RAS_BLOCK__GFX_DC_STATE_ME2,
291 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
292 TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
293 TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
294 /* CPF*/
295 TA_RAS_BLOCK__GFX_CPF_INDEX_START,
296 TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START,
297 TA_RAS_BLOCK__GFX_CPF_ROQ_ME1,
298 TA_RAS_BLOCK__GFX_CPF_TAG,
299 TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG,
300 /* CPG*/
301 TA_RAS_BLOCK__GFX_CPG_INDEX_START,
302 TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START,
303 TA_RAS_BLOCK__GFX_CPG_DMA_TAG,
304 TA_RAS_BLOCK__GFX_CPG_TAG,
305 TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG,
306 /* GDS*/
307 TA_RAS_BLOCK__GFX_GDS_INDEX_START,
308 TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START,
309 TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
310 TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
311 TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
312 TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
313 TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
314 /* SPI*/
315 TA_RAS_BLOCK__GFX_SPI_SR_MEM,
316 /* SQ*/
317 TA_RAS_BLOCK__GFX_SQ_INDEX_START,
318 TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START,
319 TA_RAS_BLOCK__GFX_SQ_LDS_D,
320 TA_RAS_BLOCK__GFX_SQ_LDS_I,
321 TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/
322 TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR,
323 /* SQC (3 ranges)*/
324 TA_RAS_BLOCK__GFX_SQC_INDEX_START,
325 /* SQC range 0*/
326 TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START,
327 TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
328 TA_RAS_BLOCK__GFX_SQC_INDEX0_START,
329 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
330 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
331 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
332 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
333 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
334 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
335 TA_RAS_BLOCK__GFX_SQC_INDEX0_END =
336 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
337 /* SQC range 1*/
338 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
339 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
340 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
341 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
342 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
343 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
344 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
345 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
346 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
347 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
348 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
349 TA_RAS_BLOCK__GFX_SQC_INDEX1_END =
350 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
351 /* SQC range 2*/
352 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
353 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
354 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
355 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
356 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
357 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
358 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
359 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
360 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
361 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
362 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
363 TA_RAS_BLOCK__GFX_SQC_INDEX2_END =
364 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
365 TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END,
366 /* TA*/
367 TA_RAS_BLOCK__GFX_TA_INDEX_START,
368 TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START,
369 TA_RAS_BLOCK__GFX_TA_FS_AFIFO,
370 TA_RAS_BLOCK__GFX_TA_FL_LFIFO,
371 TA_RAS_BLOCK__GFX_TA_FX_LFIFO,
372 TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
373 TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
374 /* TCA*/
375 TA_RAS_BLOCK__GFX_TCA_INDEX_START,
376 TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START,
377 TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
378 TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
379 /* TCC (5 sub-ranges)*/
380 TA_RAS_BLOCK__GFX_TCC_INDEX_START,
381 /* TCC range 0*/
382 TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START,
383 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START,
384 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
385 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
386 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
387 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
388 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
389 TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
390 TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
391 TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
392 /* TCC range 1*/
393 TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
394 TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
395 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
396 TA_RAS_BLOCK__GFX_TCC_INDEX1_END =
397 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
398 /* TCC range 2*/
399 TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
400 TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
401 TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
402 TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
403 TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
404 TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
405 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO,
406 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
407 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
408 TA_RAS_BLOCK__GFX_TCC_INDEX2_END =
409 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
410 /* TCC range 3*/
411 TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
412 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
413 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
414 TA_RAS_BLOCK__GFX_TCC_INDEX3_END =
415 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
416 /* TCC range 4*/
417 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
418 TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
419 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
420 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
421 TA_RAS_BLOCK__GFX_TCC_INDEX4_END =
422 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
423 TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END,
424 /* TCI*/
425 TA_RAS_BLOCK__GFX_TCI_WRITE_RAM,
426 /* TCP*/
427 TA_RAS_BLOCK__GFX_TCP_INDEX_START,
428 TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START,
429 TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
430 TA_RAS_BLOCK__GFX_TCP_CMD_FIFO,
431 TA_RAS_BLOCK__GFX_TCP_VM_FIFO,
432 TA_RAS_BLOCK__GFX_TCP_DB_RAM,
433 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
434 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
435 TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
436 /* TD*/
437 TA_RAS_BLOCK__GFX_TD_INDEX_START,
438 TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START,
439 TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
440 TA_RAS_BLOCK__GFX_TD_CS_FIFO,
441 TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO,
442 /* EA (3 sub-ranges)*/
443 TA_RAS_BLOCK__GFX_EA_INDEX_START,
444 /* EA range 0*/
445 TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START,
446 TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START,
447 TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
448 TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
449 TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
450 TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
451 TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
452 TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
453 TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
454 TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
455 /* EA range 1*/
456 TA_RAS_BLOCK__GFX_EA_INDEX1_START,
457 TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START,
458 TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
459 TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
460 TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
461 TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
462 TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
463 TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
464 TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
465 /* EA range 2*/
466 TA_RAS_BLOCK__GFX_EA_INDEX2_START,
467 TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START,
468 TA_RAS_BLOCK__GFX_EA_MAM_D1MEM,
469 TA_RAS_BLOCK__GFX_EA_MAM_D2MEM,
470 TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
471 TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
472 TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END,
473 /* UTC VM L2 bank*/
474 TA_RAS_BLOCK__UTC_VML2_BANK_CACHE,
475 /* UTC VM walker*/
476 TA_RAS_BLOCK__UTC_VML2_WALKER,
477 /* UTC ATC L2 2MB cache*/
478 TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
479 /* UTC ATC L2 4KB cache*/
480 TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
481 TA_RAS_BLOCK__GFX_MAX
482};
483
484struct ras_gfx_subblock {
485 unsigned char *name;
486 int ta_subblock;
487 int hw_supported_error_type;
488 int sw_supported_error_type;
489};
490
491#define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h) \
492 [AMDGPU_RAS_BLOCK__##subblock] = { \
493 #subblock, \
494 TA_RAS_BLOCK__##subblock, \
495 ((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)), \
496 (((e) << 1) | ((f) << 3) | (g) | ((h) << 2)), \
497 }
498
499static const struct ras_gfx_subblock ras_gfx_subblocks[] = {
500 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1),
501 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1),
502 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
503 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
504 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
505 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
506 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
507 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
508 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
509 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
510 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1),
511 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0),
512 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1),
513 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1),
514 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
515 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0),
516 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0,
517 0),
518 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0,
519 0),
520 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
521 AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0),
522 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0),
523 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1),
524 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0),
525 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0),
526 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1),
527 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
528 0, 0),
529 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
530 0),
531 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
532 0, 0),
533 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0,
534 0),
535 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
536 0, 0),
537 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
538 0),
539 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
540 1),
541 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
542 0, 0, 0),
543 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
544 0),
545 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
546 0),
547 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
548 0),
549 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
550 0),
551 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
552 0),
553 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
554 0, 0),
555 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
556 0),
557 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
558 0),
559 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
560 0, 0, 0),
561 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
562 0),
563 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
564 0),
565 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
566 0),
567 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
568 0),
569 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
570 0),
571 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
572 0, 0),
573 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
574 0),
575 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1),
576 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
577 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
578 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
579 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
580 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0),
581 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
582 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1),
583 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0,
584 1),
585 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0,
586 1),
587 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0,
588 1),
589 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0,
590 0),
591 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0,
592 0),
593 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
594 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
595 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0),
596 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0),
597 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0),
598 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0),
599 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
600 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0),
601 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0),
602 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
603 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0),
604 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0,
605 0),
606 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
607 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0,
608 0),
609 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0,
610 0, 0),
611 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0,
612 0),
613 AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
614 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1),
615 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0),
616 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
617 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
618 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
619 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0),
620 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0),
621 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1),
622 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0),
623 AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
624 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1),
625 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
626 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
627 AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
628 AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
629 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
630 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
631 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
632 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
633 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
634 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
635 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
636 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0),
637 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
638 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
639 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0),
640 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0),
641 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0),
642 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0),
643 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0),
644 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0),
645 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0),
646 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0),
647};
648
649static const struct soc15_reg_golden golden_settings_gc_9_0[] =
650{
651 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
652 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000),
653 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
654 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
655 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
656 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
657 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
658 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
659 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
660 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x00ffff87),
661 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x00ffff8f),
662 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
663 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
664 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
665 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
666 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
667 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
668 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
669 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
670 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
671};
672
673static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
674{
675 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107),
676 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
677 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
678 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
679 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
680 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
681 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042),
682 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
683 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000),
684 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
685 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
686 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
687 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
688 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
689 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
690 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
691 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800),
692 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
693};
694
695static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
696{
697 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080),
698 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
699 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
700 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042),
701 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042),
702 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400),
703 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000),
704 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000),
705 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107),
706 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000),
707 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000)
708};
709
710static const struct soc15_reg_golden golden_settings_gc_9_1[] =
711{
712 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
713 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
714 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
715 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
716 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
717 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
718 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
719 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
720 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
721 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
722 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
723 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
724 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
725 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
726 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
727 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
728 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
729 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
730 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
731 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff),
732 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080),
733 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
734 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
735 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
736};
737
738static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
739{
740 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
741 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042),
742 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042),
743 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000),
744 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000),
745 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
746 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800)
747};
748
749static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] =
750{
751 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000),
752 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
753 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
754 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080),
755 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080),
756 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080),
757 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041),
758 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041),
759 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
760 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
761 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080),
762 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080),
763 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080),
764 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080),
765 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080),
766 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
767 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010),
768 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
769 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
770};
771
772static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] =
773{
774 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
775 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
776 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
777 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042),
778 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042),
779 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
780 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
781 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
782 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
783 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
784 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
785 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc),
786};
787
788static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
789{
790 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff),
791 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000),
792 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382)
793};
794
795static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
796{
797 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
798 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
799 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
800 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
801 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
802 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
803 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
804 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
805 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
806 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
807 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
808 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
809 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
810 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
811 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
812 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
813};
814
815static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
816{
817 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080),
818 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
819 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
820 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041),
821 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041),
822 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
823 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107),
824 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
825 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410),
826 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
827 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
828 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
829 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
830};
831
832static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] =
833{
834 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
835 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000),
836 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e),
837 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca),
838 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098),
839 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3),
840 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1),
841 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135),
842 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000),
843 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00),
844 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_UTCL1_CNTL1, 0x30000000, 0x30000000)
845};
846
847static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = {
848 {SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)},
849 {SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)},
850};
851
852static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
853{
854 mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
855 mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
856 mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
857 mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
858 mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
859 mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
860 mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
861 mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
862};
863
864static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
865{
866 mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0,
867 mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0,
868 mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0,
869 mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0,
870 mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0,
871 mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0,
872 mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0,
873 mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0,
874};
875
876#define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
877#define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
878#define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
879#define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041
880
881static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
882static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
883static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
884static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
885static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
886 struct amdgpu_cu_info *cu_info);
887static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
888static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume, bool usegds);
889static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring);
890static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
891 void *ras_error_status);
892static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
893 void *inject_if, uint32_t instance_mask);
894static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev);
895static void gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device *adev,
896 unsigned int vmid);
897static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id);
898static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id);
899
900static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
901 uint64_t queue_mask)
902{
903 struct amdgpu_device *adev = kiq_ring->adev;
904 u64 shader_mc_addr;
905
906 /* Cleaner shader MC address */
907 shader_mc_addr = adev->gfx.cleaner_shader_gpu_addr >> 8;
908
909 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
910 amdgpu_ring_write(kiq_ring,
911 PACKET3_SET_RESOURCES_VMID_MASK(0) |
912 /* vmid_mask:0* queue_type:0 (KIQ) */
913 PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
914 amdgpu_ring_write(kiq_ring,
915 lower_32_bits(queue_mask)); /* queue mask lo */
916 amdgpu_ring_write(kiq_ring,
917 upper_32_bits(queue_mask)); /* queue mask hi */
918 amdgpu_ring_write(kiq_ring, lower_32_bits(shader_mc_addr)); /* cleaner shader addr lo */
919 amdgpu_ring_write(kiq_ring, upper_32_bits(shader_mc_addr)); /* cleaner shader addr hi */
920 amdgpu_ring_write(kiq_ring, 0); /* oac mask */
921 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */
922}
923
924static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
925 struct amdgpu_ring *ring)
926{
927 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
928 uint64_t wptr_addr = ring->wptr_gpu_addr;
929 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
930
931 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
932 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
933 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
934 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
935 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
936 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
937 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
938 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
939 /*queue_type: normal compute queue */
940 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
941 /* alloc format: all_on_one_pipe */
942 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
943 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
944 /* num_queues: must be 1 */
945 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
946 amdgpu_ring_write(kiq_ring,
947 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
948 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
949 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
950 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
951 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
952}
953
954static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
955 struct amdgpu_ring *ring,
956 enum amdgpu_unmap_queues_action action,
957 u64 gpu_addr, u64 seq)
958{
959 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
960
961 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
962 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
963 PACKET3_UNMAP_QUEUES_ACTION(action) |
964 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
965 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
966 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
967 amdgpu_ring_write(kiq_ring,
968 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
969
970 if (action == PREEMPT_QUEUES_NO_UNMAP) {
971 amdgpu_ring_write(kiq_ring, lower_32_bits(ring->wptr & ring->buf_mask));
972 amdgpu_ring_write(kiq_ring, 0);
973 amdgpu_ring_write(kiq_ring, 0);
974
975 } else {
976 amdgpu_ring_write(kiq_ring, 0);
977 amdgpu_ring_write(kiq_ring, 0);
978 amdgpu_ring_write(kiq_ring, 0);
979 }
980}
981
982static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
983 struct amdgpu_ring *ring,
984 u64 addr,
985 u64 seq)
986{
987 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
988
989 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
990 amdgpu_ring_write(kiq_ring,
991 PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
992 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
993 PACKET3_QUERY_STATUS_COMMAND(2));
994 /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
995 amdgpu_ring_write(kiq_ring,
996 PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
997 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
998 amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
999 amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
1000 amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
1001 amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
1002}
1003
1004static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
1005 uint16_t pasid, uint32_t flush_type,
1006 bool all_hub)
1007{
1008 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
1009 amdgpu_ring_write(kiq_ring,
1010 PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
1011 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
1012 PACKET3_INVALIDATE_TLBS_PASID(pasid) |
1013 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
1014}
1015
1016
1017static void gfx_v9_0_kiq_reset_hw_queue(struct amdgpu_ring *kiq_ring, uint32_t queue_type,
1018 uint32_t me_id, uint32_t pipe_id, uint32_t queue_id,
1019 uint32_t xcc_id, uint32_t vmid)
1020{
1021 struct amdgpu_device *adev = kiq_ring->adev;
1022 unsigned i;
1023
1024 /* enter save mode */
1025 amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id);
1026 mutex_lock(&adev->srbm_mutex);
1027 soc15_grbm_select(adev, me_id, pipe_id, queue_id, 0, 0);
1028
1029 if (queue_type == AMDGPU_RING_TYPE_COMPUTE) {
1030 WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 0x2);
1031 WREG32_SOC15(GC, 0, mmSPI_COMPUTE_QUEUE_RESET, 0x1);
1032 /* wait till dequeue take effects */
1033 for (i = 0; i < adev->usec_timeout; i++) {
1034 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
1035 break;
1036 udelay(1);
1037 }
1038 if (i >= adev->usec_timeout)
1039 dev_err(adev->dev, "fail to wait on hqd deactive\n");
1040 } else {
1041 dev_err(adev->dev, "reset queue_type(%d) not supported\n", queue_type);
1042 }
1043
1044 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
1045 mutex_unlock(&adev->srbm_mutex);
1046 /* exit safe mode */
1047 amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id);
1048}
1049
1050static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
1051 .kiq_set_resources = gfx_v9_0_kiq_set_resources,
1052 .kiq_map_queues = gfx_v9_0_kiq_map_queues,
1053 .kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
1054 .kiq_query_status = gfx_v9_0_kiq_query_status,
1055 .kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
1056 .kiq_reset_hw_queue = gfx_v9_0_kiq_reset_hw_queue,
1057 .set_resources_size = 8,
1058 .map_queues_size = 7,
1059 .unmap_queues_size = 6,
1060 .query_status_size = 7,
1061 .invalidate_tlbs_size = 2,
1062};
1063
1064static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
1065{
1066 adev->gfx.kiq[0].pmf = &gfx_v9_0_kiq_pm4_funcs;
1067}
1068
1069static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
1070{
1071 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
1072 case IP_VERSION(9, 0, 1):
1073 soc15_program_register_sequence(adev,
1074 golden_settings_gc_9_0,
1075 ARRAY_SIZE(golden_settings_gc_9_0));
1076 soc15_program_register_sequence(adev,
1077 golden_settings_gc_9_0_vg10,
1078 ARRAY_SIZE(golden_settings_gc_9_0_vg10));
1079 break;
1080 case IP_VERSION(9, 2, 1):
1081 soc15_program_register_sequence(adev,
1082 golden_settings_gc_9_2_1,
1083 ARRAY_SIZE(golden_settings_gc_9_2_1));
1084 soc15_program_register_sequence(adev,
1085 golden_settings_gc_9_2_1_vg12,
1086 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
1087 break;
1088 case IP_VERSION(9, 4, 0):
1089 soc15_program_register_sequence(adev,
1090 golden_settings_gc_9_0,
1091 ARRAY_SIZE(golden_settings_gc_9_0));
1092 soc15_program_register_sequence(adev,
1093 golden_settings_gc_9_0_vg20,
1094 ARRAY_SIZE(golden_settings_gc_9_0_vg20));
1095 break;
1096 case IP_VERSION(9, 4, 1):
1097 soc15_program_register_sequence(adev,
1098 golden_settings_gc_9_4_1_arct,
1099 ARRAY_SIZE(golden_settings_gc_9_4_1_arct));
1100 break;
1101 case IP_VERSION(9, 2, 2):
1102 case IP_VERSION(9, 1, 0):
1103 soc15_program_register_sequence(adev, golden_settings_gc_9_1,
1104 ARRAY_SIZE(golden_settings_gc_9_1));
1105 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
1106 soc15_program_register_sequence(adev,
1107 golden_settings_gc_9_1_rv2,
1108 ARRAY_SIZE(golden_settings_gc_9_1_rv2));
1109 else
1110 soc15_program_register_sequence(adev,
1111 golden_settings_gc_9_1_rv1,
1112 ARRAY_SIZE(golden_settings_gc_9_1_rv1));
1113 break;
1114 case IP_VERSION(9, 3, 0):
1115 soc15_program_register_sequence(adev,
1116 golden_settings_gc_9_1_rn,
1117 ARRAY_SIZE(golden_settings_gc_9_1_rn));
1118 return; /* for renoir, don't need common goldensetting */
1119 case IP_VERSION(9, 4, 2):
1120 gfx_v9_4_2_init_golden_registers(adev,
1121 adev->smuio.funcs->get_die_id(adev));
1122 break;
1123 default:
1124 break;
1125 }
1126
1127 if ((amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 1)) &&
1128 (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 2)))
1129 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
1130 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
1131}
1132
1133static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
1134 bool wc, uint32_t reg, uint32_t val)
1135{
1136 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
1137 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
1138 WRITE_DATA_DST_SEL(0) |
1139 (wc ? WR_CONFIRM : 0));
1140 amdgpu_ring_write(ring, reg);
1141 amdgpu_ring_write(ring, 0);
1142 amdgpu_ring_write(ring, val);
1143}
1144
1145static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
1146 int mem_space, int opt, uint32_t addr0,
1147 uint32_t addr1, uint32_t ref, uint32_t mask,
1148 uint32_t inv)
1149{
1150 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
1151 amdgpu_ring_write(ring,
1152 /* memory (1) or register (0) */
1153 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
1154 WAIT_REG_MEM_OPERATION(opt) | /* wait */
1155 WAIT_REG_MEM_FUNCTION(3) | /* equal */
1156 WAIT_REG_MEM_ENGINE(eng_sel)));
1157
1158 if (mem_space)
1159 BUG_ON(addr0 & 0x3); /* Dword align */
1160 amdgpu_ring_write(ring, addr0);
1161 amdgpu_ring_write(ring, addr1);
1162 amdgpu_ring_write(ring, ref);
1163 amdgpu_ring_write(ring, mask);
1164 amdgpu_ring_write(ring, inv); /* poll interval */
1165}
1166
1167static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
1168{
1169 struct amdgpu_device *adev = ring->adev;
1170 uint32_t scratch = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
1171 uint32_t tmp = 0;
1172 unsigned i;
1173 int r;
1174
1175 WREG32(scratch, 0xCAFEDEAD);
1176 r = amdgpu_ring_alloc(ring, 3);
1177 if (r)
1178 return r;
1179
1180 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
1181 amdgpu_ring_write(ring, scratch - PACKET3_SET_UCONFIG_REG_START);
1182 amdgpu_ring_write(ring, 0xDEADBEEF);
1183 amdgpu_ring_commit(ring);
1184
1185 for (i = 0; i < adev->usec_timeout; i++) {
1186 tmp = RREG32(scratch);
1187 if (tmp == 0xDEADBEEF)
1188 break;
1189 udelay(1);
1190 }
1191
1192 if (i >= adev->usec_timeout)
1193 r = -ETIMEDOUT;
1194 return r;
1195}
1196
1197static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1198{
1199 struct amdgpu_device *adev = ring->adev;
1200 struct amdgpu_ib ib;
1201 struct dma_fence *f = NULL;
1202
1203 unsigned index;
1204 uint64_t gpu_addr;
1205 uint32_t tmp;
1206 long r;
1207
1208 r = amdgpu_device_wb_get(adev, &index);
1209 if (r)
1210 return r;
1211
1212 gpu_addr = adev->wb.gpu_addr + (index * 4);
1213 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
1214 memset(&ib, 0, sizeof(ib));
1215
1216 r = amdgpu_ib_get(adev, NULL, 20, AMDGPU_IB_POOL_DIRECT, &ib);
1217 if (r)
1218 goto err1;
1219
1220 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
1221 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
1222 ib.ptr[2] = lower_32_bits(gpu_addr);
1223 ib.ptr[3] = upper_32_bits(gpu_addr);
1224 ib.ptr[4] = 0xDEADBEEF;
1225 ib.length_dw = 5;
1226
1227 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1228 if (r)
1229 goto err2;
1230
1231 r = dma_fence_wait_timeout(f, false, timeout);
1232 if (r == 0) {
1233 r = -ETIMEDOUT;
1234 goto err2;
1235 } else if (r < 0) {
1236 goto err2;
1237 }
1238
1239 tmp = adev->wb.wb[index];
1240 if (tmp == 0xDEADBEEF)
1241 r = 0;
1242 else
1243 r = -EINVAL;
1244
1245err2:
1246 amdgpu_ib_free(adev, &ib, NULL);
1247 dma_fence_put(f);
1248err1:
1249 amdgpu_device_wb_free(adev, index);
1250 return r;
1251}
1252
1253
1254static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
1255{
1256 amdgpu_ucode_release(&adev->gfx.pfp_fw);
1257 amdgpu_ucode_release(&adev->gfx.me_fw);
1258 amdgpu_ucode_release(&adev->gfx.ce_fw);
1259 amdgpu_ucode_release(&adev->gfx.rlc_fw);
1260 amdgpu_ucode_release(&adev->gfx.mec_fw);
1261 amdgpu_ucode_release(&adev->gfx.mec2_fw);
1262
1263 kfree(adev->gfx.rlc.register_list_format);
1264}
1265
1266static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev)
1267{
1268 adev->gfx.me_fw_write_wait = false;
1269 adev->gfx.mec_fw_write_wait = false;
1270
1271 if ((amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 1)) &&
1272 ((adev->gfx.mec_fw_version < 0x000001a5) ||
1273 (adev->gfx.mec_feature_version < 46) ||
1274 (adev->gfx.pfp_fw_version < 0x000000b7) ||
1275 (adev->gfx.pfp_feature_version < 46)))
1276 DRM_WARN_ONCE("CP firmware version too old, please update!");
1277
1278 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
1279 case IP_VERSION(9, 0, 1):
1280 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1281 (adev->gfx.me_feature_version >= 42) &&
1282 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1283 (adev->gfx.pfp_feature_version >= 42))
1284 adev->gfx.me_fw_write_wait = true;
1285
1286 if ((adev->gfx.mec_fw_version >= 0x00000193) &&
1287 (adev->gfx.mec_feature_version >= 42))
1288 adev->gfx.mec_fw_write_wait = true;
1289 break;
1290 case IP_VERSION(9, 2, 1):
1291 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1292 (adev->gfx.me_feature_version >= 44) &&
1293 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1294 (adev->gfx.pfp_feature_version >= 44))
1295 adev->gfx.me_fw_write_wait = true;
1296
1297 if ((adev->gfx.mec_fw_version >= 0x00000196) &&
1298 (adev->gfx.mec_feature_version >= 44))
1299 adev->gfx.mec_fw_write_wait = true;
1300 break;
1301 case IP_VERSION(9, 4, 0):
1302 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1303 (adev->gfx.me_feature_version >= 44) &&
1304 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1305 (adev->gfx.pfp_feature_version >= 44))
1306 adev->gfx.me_fw_write_wait = true;
1307
1308 if ((adev->gfx.mec_fw_version >= 0x00000197) &&
1309 (adev->gfx.mec_feature_version >= 44))
1310 adev->gfx.mec_fw_write_wait = true;
1311 break;
1312 case IP_VERSION(9, 1, 0):
1313 case IP_VERSION(9, 2, 2):
1314 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1315 (adev->gfx.me_feature_version >= 42) &&
1316 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1317 (adev->gfx.pfp_feature_version >= 42))
1318 adev->gfx.me_fw_write_wait = true;
1319
1320 if ((adev->gfx.mec_fw_version >= 0x00000192) &&
1321 (adev->gfx.mec_feature_version >= 42))
1322 adev->gfx.mec_fw_write_wait = true;
1323 break;
1324 default:
1325 adev->gfx.me_fw_write_wait = true;
1326 adev->gfx.mec_fw_write_wait = true;
1327 break;
1328 }
1329}
1330
1331struct amdgpu_gfxoff_quirk {
1332 u16 chip_vendor;
1333 u16 chip_device;
1334 u16 subsys_vendor;
1335 u16 subsys_device;
1336 u8 revision;
1337};
1338
1339static const struct amdgpu_gfxoff_quirk amdgpu_gfxoff_quirk_list[] = {
1340 /* https://bugzilla.kernel.org/show_bug.cgi?id=204689 */
1341 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc8 },
1342 /* https://bugzilla.kernel.org/show_bug.cgi?id=207171 */
1343 { 0x1002, 0x15dd, 0x103c, 0x83e7, 0xd3 },
1344 /* GFXOFF is unstable on C6 parts with a VBIOS 113-RAVEN-114 */
1345 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc6 },
1346 /* Apple MacBook Pro (15-inch, 2019) Radeon Pro Vega 20 4 GB */
1347 { 0x1002, 0x69af, 0x106b, 0x019a, 0xc0 },
1348 /* https://bbs.openkylin.top/t/topic/171497 */
1349 { 0x1002, 0x15d8, 0x19e5, 0x3e14, 0xc2 },
1350 /* HP 705G4 DM with R5 2400G */
1351 { 0x1002, 0x15dd, 0x103c, 0x8464, 0xd6 },
1352 { 0, 0, 0, 0, 0 },
1353};
1354
1355static bool gfx_v9_0_should_disable_gfxoff(struct pci_dev *pdev)
1356{
1357 const struct amdgpu_gfxoff_quirk *p = amdgpu_gfxoff_quirk_list;
1358
1359 while (p && p->chip_device != 0) {
1360 if (pdev->vendor == p->chip_vendor &&
1361 pdev->device == p->chip_device &&
1362 pdev->subsystem_vendor == p->subsys_vendor &&
1363 pdev->subsystem_device == p->subsys_device &&
1364 pdev->revision == p->revision) {
1365 return true;
1366 }
1367 ++p;
1368 }
1369 return false;
1370}
1371
1372static bool is_raven_kicker(struct amdgpu_device *adev)
1373{
1374 if (adev->pm.fw_version >= 0x41e2b)
1375 return true;
1376 else
1377 return false;
1378}
1379
1380static bool check_if_enlarge_doorbell_range(struct amdgpu_device *adev)
1381{
1382 if ((amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 3, 0)) &&
1383 (adev->gfx.me_fw_version >= 0x000000a5) &&
1384 (adev->gfx.me_feature_version >= 52))
1385 return true;
1386 else
1387 return false;
1388}
1389
1390static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
1391{
1392 if (gfx_v9_0_should_disable_gfxoff(adev->pdev))
1393 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1394
1395 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
1396 case IP_VERSION(9, 0, 1):
1397 case IP_VERSION(9, 2, 1):
1398 case IP_VERSION(9, 4, 0):
1399 break;
1400 case IP_VERSION(9, 2, 2):
1401 case IP_VERSION(9, 1, 0):
1402 if (!((adev->apu_flags & AMD_APU_IS_RAVEN2) ||
1403 (adev->apu_flags & AMD_APU_IS_PICASSO)) &&
1404 ((!is_raven_kicker(adev) &&
1405 adev->gfx.rlc_fw_version < 531) ||
1406 (adev->gfx.rlc_feature_version < 1) ||
1407 !adev->gfx.rlc.is_rlc_v2_1))
1408 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1409
1410 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1411 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1412 AMD_PG_SUPPORT_CP |
1413 AMD_PG_SUPPORT_RLC_SMU_HS;
1414 break;
1415 case IP_VERSION(9, 3, 0):
1416 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1417 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1418 AMD_PG_SUPPORT_CP |
1419 AMD_PG_SUPPORT_RLC_SMU_HS;
1420 break;
1421 default:
1422 break;
1423 }
1424}
1425
1426static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev,
1427 char *chip_name)
1428{
1429 int err;
1430
1431 err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
1432 "amdgpu/%s_pfp.bin", chip_name);
1433 if (err)
1434 goto out;
1435 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_PFP);
1436
1437 err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1438 "amdgpu/%s_me.bin", chip_name);
1439 if (err)
1440 goto out;
1441 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_ME);
1442
1443 err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1444 "amdgpu/%s_ce.bin", chip_name);
1445 if (err)
1446 goto out;
1447 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_CE);
1448
1449out:
1450 if (err) {
1451 amdgpu_ucode_release(&adev->gfx.pfp_fw);
1452 amdgpu_ucode_release(&adev->gfx.me_fw);
1453 amdgpu_ucode_release(&adev->gfx.ce_fw);
1454 }
1455 return err;
1456}
1457
1458static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev,
1459 char *chip_name)
1460{
1461 int err;
1462 const struct rlc_firmware_header_v2_0 *rlc_hdr;
1463 uint16_t version_major;
1464 uint16_t version_minor;
1465 uint32_t smu_version;
1466
1467 /*
1468 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin
1469 * instead of picasso_rlc.bin.
1470 * Judgment method:
1471 * PCO AM4: revision >= 0xC8 && revision <= 0xCF
1472 * or revision >= 0xD8 && revision <= 0xDF
1473 * otherwise is PCO FP5
1474 */
1475 if (!strcmp(chip_name, "picasso") &&
1476 (((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) ||
1477 ((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF))))
1478 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw,
1479 "amdgpu/%s_rlc_am4.bin", chip_name);
1480 else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) &&
1481 (smu_version >= 0x41e2b))
1482 /**
1483 *SMC is loaded by SBIOS on APU and it's able to get the SMU version directly.
1484 */
1485 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw,
1486 "amdgpu/%s_kicker_rlc.bin", chip_name);
1487 else
1488 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw,
1489 "amdgpu/%s_rlc.bin", chip_name);
1490 if (err)
1491 goto out;
1492
1493 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1494 version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1495 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1496 err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor);
1497out:
1498 if (err)
1499 amdgpu_ucode_release(&adev->gfx.rlc_fw);
1500
1501 return err;
1502}
1503
1504static bool gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device *adev)
1505{
1506 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2) ||
1507 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1) ||
1508 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 3, 0))
1509 return false;
1510
1511 return true;
1512}
1513
1514static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
1515 char *chip_name)
1516{
1517 int err;
1518
1519 if (amdgpu_sriov_vf(adev) && (adev->asic_type == CHIP_ALDEBARAN))
1520 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1521 "amdgpu/%s_sjt_mec.bin", chip_name);
1522 else
1523 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1524 "amdgpu/%s_mec.bin", chip_name);
1525 if (err)
1526 goto out;
1527
1528 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);
1529 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT);
1530
1531 if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
1532 if (amdgpu_sriov_vf(adev) && (adev->asic_type == CHIP_ALDEBARAN))
1533 err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1534 "amdgpu/%s_sjt_mec2.bin", chip_name);
1535 else
1536 err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1537 "amdgpu/%s_mec2.bin", chip_name);
1538 if (!err) {
1539 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2);
1540 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2_JT);
1541 } else {
1542 err = 0;
1543 amdgpu_ucode_release(&adev->gfx.mec2_fw);
1544 }
1545 } else {
1546 adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version;
1547 adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version;
1548 }
1549
1550 gfx_v9_0_check_if_need_gfxoff(adev);
1551 gfx_v9_0_check_fw_write_wait(adev);
1552
1553out:
1554 if (err)
1555 amdgpu_ucode_release(&adev->gfx.mec_fw);
1556 return err;
1557}
1558
1559static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
1560{
1561 char ucode_prefix[30];
1562 int r;
1563
1564 DRM_DEBUG("\n");
1565 amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
1566
1567 /* No CPG in Arcturus */
1568 if (adev->gfx.num_gfx_rings) {
1569 r = gfx_v9_0_init_cp_gfx_microcode(adev, ucode_prefix);
1570 if (r)
1571 return r;
1572 }
1573
1574 r = gfx_v9_0_init_rlc_microcode(adev, ucode_prefix);
1575 if (r)
1576 return r;
1577
1578 r = gfx_v9_0_init_cp_compute_microcode(adev, ucode_prefix);
1579 if (r)
1580 return r;
1581
1582 return r;
1583}
1584
1585static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
1586{
1587 u32 count = 0;
1588 const struct cs_section_def *sect = NULL;
1589 const struct cs_extent_def *ext = NULL;
1590
1591 /* begin clear state */
1592 count += 2;
1593 /* context control state */
1594 count += 3;
1595
1596 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
1597 for (ext = sect->section; ext->extent != NULL; ++ext) {
1598 if (sect->id == SECT_CONTEXT)
1599 count += 2 + ext->reg_count;
1600 else
1601 return 0;
1602 }
1603 }
1604
1605 /* end clear state */
1606 count += 2;
1607 /* clear state */
1608 count += 2;
1609
1610 return count;
1611}
1612
1613static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
1614 volatile u32 *buffer)
1615{
1616 u32 count = 0, i;
1617 const struct cs_section_def *sect = NULL;
1618 const struct cs_extent_def *ext = NULL;
1619
1620 if (adev->gfx.rlc.cs_data == NULL)
1621 return;
1622 if (buffer == NULL)
1623 return;
1624
1625 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1626 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1627
1628 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1629 buffer[count++] = cpu_to_le32(0x80000000);
1630 buffer[count++] = cpu_to_le32(0x80000000);
1631
1632 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1633 for (ext = sect->section; ext->extent != NULL; ++ext) {
1634 if (sect->id == SECT_CONTEXT) {
1635 buffer[count++] =
1636 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1637 buffer[count++] = cpu_to_le32(ext->reg_index -
1638 PACKET3_SET_CONTEXT_REG_START);
1639 for (i = 0; i < ext->reg_count; i++)
1640 buffer[count++] = cpu_to_le32(ext->extent[i]);
1641 } else {
1642 return;
1643 }
1644 }
1645 }
1646
1647 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1648 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1649
1650 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1651 buffer[count++] = cpu_to_le32(0);
1652}
1653
1654static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev)
1655{
1656 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1657 uint32_t pg_always_on_cu_num = 2;
1658 uint32_t always_on_cu_num;
1659 uint32_t i, j, k;
1660 uint32_t mask, cu_bitmap, counter;
1661
1662 if (adev->flags & AMD_IS_APU)
1663 always_on_cu_num = 4;
1664 else if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 2, 1))
1665 always_on_cu_num = 8;
1666 else
1667 always_on_cu_num = 12;
1668
1669 mutex_lock(&adev->grbm_idx_mutex);
1670 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1671 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1672 mask = 1;
1673 cu_bitmap = 0;
1674 counter = 0;
1675 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
1676
1677 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
1678 if (cu_info->bitmap[0][i][j] & mask) {
1679 if (counter == pg_always_on_cu_num)
1680 WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap);
1681 if (counter < always_on_cu_num)
1682 cu_bitmap |= mask;
1683 else
1684 break;
1685 counter++;
1686 }
1687 mask <<= 1;
1688 }
1689
1690 WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap);
1691 cu_info->ao_cu_bitmap[i][j] = cu_bitmap;
1692 }
1693 }
1694 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1695 mutex_unlock(&adev->grbm_idx_mutex);
1696}
1697
1698static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
1699{
1700 uint32_t data;
1701
1702 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1703 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1704 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7);
1705 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1706 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16));
1707
1708 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1709 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1710
1711 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1712 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500);
1713
1714 mutex_lock(&adev->grbm_idx_mutex);
1715 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1716 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1717 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1718
1719 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1720 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1721 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1722 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1723 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1724
1725 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1726 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1727 data &= 0x0000FFFF;
1728 data |= 0x00C00000;
1729 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1730
1731 /*
1732 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven),
1733 * programmed in gfx_v9_0_init_always_on_cu_mask()
1734 */
1735
1736 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1737 * but used for RLC_LB_CNTL configuration */
1738 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1739 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1740 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1741 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1742 mutex_unlock(&adev->grbm_idx_mutex);
1743
1744 gfx_v9_0_init_always_on_cu_mask(adev);
1745}
1746
1747static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev)
1748{
1749 uint32_t data;
1750
1751 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1752 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1753 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8);
1754 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1755 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16));
1756
1757 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1758 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1759
1760 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1761 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800);
1762
1763 mutex_lock(&adev->grbm_idx_mutex);
1764 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1765 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1766 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1767
1768 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1769 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1770 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1771 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1772 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1773
1774 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1775 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1776 data &= 0x0000FFFF;
1777 data |= 0x00C00000;
1778 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1779
1780 /*
1781 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON),
1782 * programmed in gfx_v9_0_init_always_on_cu_mask()
1783 */
1784
1785 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1786 * but used for RLC_LB_CNTL configuration */
1787 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1788 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1789 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1790 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1791 mutex_unlock(&adev->grbm_idx_mutex);
1792
1793 gfx_v9_0_init_always_on_cu_mask(adev);
1794}
1795
1796static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
1797{
1798 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0);
1799}
1800
1801static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
1802{
1803 if (gfx_v9_0_load_mec2_fw_bin_support(adev))
1804 return 5;
1805 else
1806 return 4;
1807}
1808
1809static void gfx_v9_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
1810{
1811 struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
1812
1813 reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl[0];
1814 reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
1815 reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG1);
1816 reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG2);
1817 reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG3);
1818 reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_CNTL);
1819 reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX);
1820 reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, mmRLC_SPARE_INT);
1821 adev->gfx.rlc.rlcg_reg_access_supported = true;
1822}
1823
1824static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
1825{
1826 const struct cs_section_def *cs_data;
1827 int r;
1828
1829 adev->gfx.rlc.cs_data = gfx9_cs_data;
1830
1831 cs_data = adev->gfx.rlc.cs_data;
1832
1833 if (cs_data) {
1834 /* init clear state block */
1835 r = amdgpu_gfx_rlc_init_csb(adev);
1836 if (r)
1837 return r;
1838 }
1839
1840 if (adev->flags & AMD_IS_APU) {
1841 /* TODO: double check the cp_table_size for RV */
1842 adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1843 r = amdgpu_gfx_rlc_init_cpt(adev);
1844 if (r)
1845 return r;
1846 }
1847
1848 return 0;
1849}
1850
1851static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
1852{
1853 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1854 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1855}
1856
1857static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
1858{
1859 int r;
1860 u32 *hpd;
1861 const __le32 *fw_data;
1862 unsigned fw_size;
1863 u32 *fw;
1864 size_t mec_hpd_size;
1865
1866 const struct gfx_firmware_header_v1_0 *mec_hdr;
1867
1868 bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1869
1870 /* take ownership of the relevant compute queues */
1871 amdgpu_gfx_compute_queue_acquire(adev);
1872 mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE;
1873 if (mec_hpd_size) {
1874 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1875 AMDGPU_GEM_DOMAIN_VRAM |
1876 AMDGPU_GEM_DOMAIN_GTT,
1877 &adev->gfx.mec.hpd_eop_obj,
1878 &adev->gfx.mec.hpd_eop_gpu_addr,
1879 (void **)&hpd);
1880 if (r) {
1881 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1882 gfx_v9_0_mec_fini(adev);
1883 return r;
1884 }
1885
1886 memset(hpd, 0, mec_hpd_size);
1887
1888 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1889 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1890 }
1891
1892 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1893
1894 fw_data = (const __le32 *)
1895 (adev->gfx.mec_fw->data +
1896 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1897 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
1898
1899 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1900 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
1901 &adev->gfx.mec.mec_fw_obj,
1902 &adev->gfx.mec.mec_fw_gpu_addr,
1903 (void **)&fw);
1904 if (r) {
1905 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
1906 gfx_v9_0_mec_fini(adev);
1907 return r;
1908 }
1909
1910 memcpy(fw, fw_data, fw_size);
1911
1912 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
1913 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
1914
1915 return 0;
1916}
1917
1918static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
1919{
1920 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
1921 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1922 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1923 (address << SQ_IND_INDEX__INDEX__SHIFT) |
1924 (SQ_IND_INDEX__FORCE_READ_MASK));
1925 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1926}
1927
1928static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
1929 uint32_t wave, uint32_t thread,
1930 uint32_t regno, uint32_t num, uint32_t *out)
1931{
1932 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
1933 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1934 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1935 (regno << SQ_IND_INDEX__INDEX__SHIFT) |
1936 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
1937 (SQ_IND_INDEX__FORCE_READ_MASK) |
1938 (SQ_IND_INDEX__AUTO_INCR_MASK));
1939 while (num--)
1940 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1941}
1942
1943static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
1944{
1945 /* type 1 wave data */
1946 dst[(*no_fields)++] = 1;
1947 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
1948 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
1949 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
1950 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
1951 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
1952 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
1953 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
1954 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
1955 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
1956 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
1957 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
1958 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
1959 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
1960 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
1961 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
1962}
1963
1964static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
1965 uint32_t wave, uint32_t start,
1966 uint32_t size, uint32_t *dst)
1967{
1968 wave_read_regs(
1969 adev, simd, wave, 0,
1970 start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
1971}
1972
1973static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
1974 uint32_t wave, uint32_t thread,
1975 uint32_t start, uint32_t size,
1976 uint32_t *dst)
1977{
1978 wave_read_regs(
1979 adev, simd, wave, thread,
1980 start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
1981}
1982
1983static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
1984 u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
1985{
1986 soc15_grbm_select(adev, me, pipe, q, vm, 0);
1987}
1988
1989static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
1990 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
1991 .select_se_sh = &gfx_v9_0_select_se_sh,
1992 .read_wave_data = &gfx_v9_0_read_wave_data,
1993 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
1994 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
1995 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
1996};
1997
1998const struct amdgpu_ras_block_hw_ops gfx_v9_0_ras_ops = {
1999 .ras_error_inject = &gfx_v9_0_ras_error_inject,
2000 .query_ras_error_count = &gfx_v9_0_query_ras_error_count,
2001 .reset_ras_error_count = &gfx_v9_0_reset_ras_error_count,
2002};
2003
2004static struct amdgpu_gfx_ras gfx_v9_0_ras = {
2005 .ras_block = {
2006 .hw_ops = &gfx_v9_0_ras_ops,
2007 },
2008};
2009
2010static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
2011{
2012 u32 gb_addr_config;
2013 int err;
2014
2015 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
2016 case IP_VERSION(9, 0, 1):
2017 adev->gfx.config.max_hw_contexts = 8;
2018 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2019 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2020 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2021 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2022 gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
2023 break;
2024 case IP_VERSION(9, 2, 1):
2025 adev->gfx.config.max_hw_contexts = 8;
2026 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2027 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2028 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2029 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2030 gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
2031 DRM_INFO("fix gfx.config for vega12\n");
2032 break;
2033 case IP_VERSION(9, 4, 0):
2034 adev->gfx.ras = &gfx_v9_0_ras;
2035 adev->gfx.config.max_hw_contexts = 8;
2036 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2037 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2038 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2039 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2040 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2041 gb_addr_config &= ~0xf3e777ff;
2042 gb_addr_config |= 0x22014042;
2043 /* check vbios table if gpu info is not available */
2044 err = amdgpu_atomfirmware_get_gfx_info(adev);
2045 if (err)
2046 return err;
2047 break;
2048 case IP_VERSION(9, 2, 2):
2049 case IP_VERSION(9, 1, 0):
2050 adev->gfx.config.max_hw_contexts = 8;
2051 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2052 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2053 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2054 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2055 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
2056 gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN;
2057 else
2058 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN;
2059 break;
2060 case IP_VERSION(9, 4, 1):
2061 adev->gfx.ras = &gfx_v9_4_ras;
2062 adev->gfx.config.max_hw_contexts = 8;
2063 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2064 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2065 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2066 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2067 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2068 gb_addr_config &= ~0xf3e777ff;
2069 gb_addr_config |= 0x22014042;
2070 break;
2071 case IP_VERSION(9, 3, 0):
2072 adev->gfx.config.max_hw_contexts = 8;
2073 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2074 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2075 adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
2076 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2077 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2078 gb_addr_config &= ~0xf3e777ff;
2079 gb_addr_config |= 0x22010042;
2080 break;
2081 case IP_VERSION(9, 4, 2):
2082 adev->gfx.ras = &gfx_v9_4_2_ras;
2083 adev->gfx.config.max_hw_contexts = 8;
2084 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2085 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2086 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2087 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2088 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2089 gb_addr_config &= ~0xf3e777ff;
2090 gb_addr_config |= 0x22014042;
2091 /* check vbios table if gpu info is not available */
2092 err = amdgpu_atomfirmware_get_gfx_info(adev);
2093 if (err)
2094 return err;
2095 break;
2096 default:
2097 BUG();
2098 break;
2099 }
2100
2101 adev->gfx.config.gb_addr_config = gb_addr_config;
2102
2103 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
2104 REG_GET_FIELD(
2105 adev->gfx.config.gb_addr_config,
2106 GB_ADDR_CONFIG,
2107 NUM_PIPES);
2108
2109 adev->gfx.config.max_tile_pipes =
2110 adev->gfx.config.gb_addr_config_fields.num_pipes;
2111
2112 adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
2113 REG_GET_FIELD(
2114 adev->gfx.config.gb_addr_config,
2115 GB_ADDR_CONFIG,
2116 NUM_BANKS);
2117 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
2118 REG_GET_FIELD(
2119 adev->gfx.config.gb_addr_config,
2120 GB_ADDR_CONFIG,
2121 MAX_COMPRESSED_FRAGS);
2122 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
2123 REG_GET_FIELD(
2124 adev->gfx.config.gb_addr_config,
2125 GB_ADDR_CONFIG,
2126 NUM_RB_PER_SE);
2127 adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
2128 REG_GET_FIELD(
2129 adev->gfx.config.gb_addr_config,
2130 GB_ADDR_CONFIG,
2131 NUM_SHADER_ENGINES);
2132 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
2133 REG_GET_FIELD(
2134 adev->gfx.config.gb_addr_config,
2135 GB_ADDR_CONFIG,
2136 PIPE_INTERLEAVE_SIZE));
2137
2138 return 0;
2139}
2140
2141static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
2142 int mec, int pipe, int queue)
2143{
2144 unsigned irq_type;
2145 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
2146 unsigned int hw_prio;
2147
2148 ring = &adev->gfx.compute_ring[ring_id];
2149
2150 /* mec0 is me1 */
2151 ring->me = mec + 1;
2152 ring->pipe = pipe;
2153 ring->queue = queue;
2154
2155 ring->ring_obj = NULL;
2156 ring->use_doorbell = true;
2157 ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
2158 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
2159 + (ring_id * GFX9_MEC_HPD_SIZE);
2160 ring->vm_hub = AMDGPU_GFXHUB(0);
2161 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
2162
2163 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
2164 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
2165 + ring->pipe;
2166 hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
2167 AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
2168 /* type-2 packets are deprecated on MEC, use type-3 instead */
2169 return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
2170 hw_prio, NULL);
2171}
2172
2173static void gfx_v9_0_alloc_ip_dump(struct amdgpu_device *adev)
2174{
2175 uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9);
2176 uint32_t *ptr;
2177 uint32_t inst;
2178
2179 ptr = kcalloc(reg_count, sizeof(uint32_t), GFP_KERNEL);
2180 if (!ptr) {
2181 DRM_ERROR("Failed to allocate memory for GFX IP Dump\n");
2182 adev->gfx.ip_dump_core = NULL;
2183 } else {
2184 adev->gfx.ip_dump_core = ptr;
2185 }
2186
2187 /* Allocate memory for compute queue registers for all the instances */
2188 reg_count = ARRAY_SIZE(gc_cp_reg_list_9);
2189 inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec *
2190 adev->gfx.mec.num_queue_per_pipe;
2191
2192 ptr = kcalloc(reg_count * inst, sizeof(uint32_t), GFP_KERNEL);
2193 if (!ptr) {
2194 DRM_ERROR("Failed to allocate memory for Compute Queues IP Dump\n");
2195 adev->gfx.ip_dump_compute_queues = NULL;
2196 } else {
2197 adev->gfx.ip_dump_compute_queues = ptr;
2198 }
2199}
2200
2201static int gfx_v9_0_sw_init(struct amdgpu_ip_block *ip_block)
2202{
2203 int i, j, k, r, ring_id;
2204 int xcc_id = 0;
2205 struct amdgpu_ring *ring;
2206 struct amdgpu_device *adev = ip_block->adev;
2207 unsigned int hw_prio;
2208
2209 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
2210 case IP_VERSION(9, 0, 1):
2211 case IP_VERSION(9, 2, 1):
2212 case IP_VERSION(9, 4, 0):
2213 case IP_VERSION(9, 2, 2):
2214 case IP_VERSION(9, 1, 0):
2215 case IP_VERSION(9, 4, 1):
2216 case IP_VERSION(9, 3, 0):
2217 case IP_VERSION(9, 4, 2):
2218 adev->gfx.mec.num_mec = 2;
2219 break;
2220 default:
2221 adev->gfx.mec.num_mec = 1;
2222 break;
2223 }
2224
2225 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
2226 case IP_VERSION(9, 4, 2):
2227 adev->gfx.cleaner_shader_ptr = gfx_9_4_2_cleaner_shader_hex;
2228 adev->gfx.cleaner_shader_size = sizeof(gfx_9_4_2_cleaner_shader_hex);
2229 if (adev->gfx.mec_fw_version >= 88) {
2230 adev->gfx.enable_cleaner_shader = true;
2231 r = amdgpu_gfx_cleaner_shader_sw_init(adev, adev->gfx.cleaner_shader_size);
2232 if (r) {
2233 adev->gfx.enable_cleaner_shader = false;
2234 dev_err(adev->dev, "Failed to initialize cleaner shader\n");
2235 }
2236 }
2237 break;
2238 default:
2239 adev->gfx.enable_cleaner_shader = false;
2240 break;
2241 }
2242
2243 adev->gfx.mec.num_pipe_per_mec = 4;
2244 adev->gfx.mec.num_queue_per_pipe = 8;
2245
2246 /* EOP Event */
2247 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq);
2248 if (r)
2249 return r;
2250
2251 /* Bad opcode Event */
2252 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP,
2253 GFX_9_0__SRCID__CP_BAD_OPCODE_ERROR,
2254 &adev->gfx.bad_op_irq);
2255 if (r)
2256 return r;
2257
2258 /* Privileged reg */
2259 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT,
2260 &adev->gfx.priv_reg_irq);
2261 if (r)
2262 return r;
2263
2264 /* Privileged inst */
2265 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT,
2266 &adev->gfx.priv_inst_irq);
2267 if (r)
2268 return r;
2269
2270 /* ECC error */
2271 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR,
2272 &adev->gfx.cp_ecc_error_irq);
2273 if (r)
2274 return r;
2275
2276 /* FUE error */
2277 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR,
2278 &adev->gfx.cp_ecc_error_irq);
2279 if (r)
2280 return r;
2281
2282 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2283
2284 if (adev->gfx.rlc.funcs) {
2285 if (adev->gfx.rlc.funcs->init) {
2286 r = adev->gfx.rlc.funcs->init(adev);
2287 if (r) {
2288 dev_err(adev->dev, "Failed to init rlc BOs!\n");
2289 return r;
2290 }
2291 }
2292 }
2293
2294 r = gfx_v9_0_mec_init(adev);
2295 if (r) {
2296 DRM_ERROR("Failed to init MEC BOs!\n");
2297 return r;
2298 }
2299
2300 /* set up the gfx ring */
2301 for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2302 ring = &adev->gfx.gfx_ring[i];
2303 ring->ring_obj = NULL;
2304 if (!i)
2305 sprintf(ring->name, "gfx");
2306 else
2307 sprintf(ring->name, "gfx_%d", i);
2308 ring->use_doorbell = true;
2309 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2310
2311 /* disable scheduler on the real ring */
2312 ring->no_scheduler = adev->gfx.mcbp;
2313 ring->vm_hub = AMDGPU_GFXHUB(0);
2314 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2315 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
2316 AMDGPU_RING_PRIO_DEFAULT, NULL);
2317 if (r)
2318 return r;
2319 }
2320
2321 /* set up the software rings */
2322 if (adev->gfx.mcbp && adev->gfx.num_gfx_rings) {
2323 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) {
2324 ring = &adev->gfx.sw_gfx_ring[i];
2325 ring->ring_obj = NULL;
2326 sprintf(ring->name, amdgpu_sw_ring_name(i));
2327 ring->use_doorbell = true;
2328 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2329 ring->is_sw_ring = true;
2330 hw_prio = amdgpu_sw_ring_priority(i);
2331 ring->vm_hub = AMDGPU_GFXHUB(0);
2332 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2333 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, hw_prio,
2334 NULL);
2335 if (r)
2336 return r;
2337 ring->wptr = 0;
2338 }
2339
2340 /* init the muxer and add software rings */
2341 r = amdgpu_ring_mux_init(&adev->gfx.muxer, &adev->gfx.gfx_ring[0],
2342 GFX9_NUM_SW_GFX_RINGS);
2343 if (r) {
2344 DRM_ERROR("amdgpu_ring_mux_init failed(%d)\n", r);
2345 return r;
2346 }
2347 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) {
2348 r = amdgpu_ring_mux_add_sw_ring(&adev->gfx.muxer,
2349 &adev->gfx.sw_gfx_ring[i]);
2350 if (r) {
2351 DRM_ERROR("amdgpu_ring_mux_add_sw_ring failed(%d)\n", r);
2352 return r;
2353 }
2354 }
2355 }
2356
2357 /* set up the compute queues - allocate horizontally across pipes */
2358 ring_id = 0;
2359 for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2360 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2361 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2362 if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
2363 k, j))
2364 continue;
2365
2366 r = gfx_v9_0_compute_ring_init(adev,
2367 ring_id,
2368 i, k, j);
2369 if (r)
2370 return r;
2371
2372 ring_id++;
2373 }
2374 }
2375 }
2376
2377 /* TODO: Add queue reset mask when FW fully supports it */
2378 adev->gfx.gfx_supported_reset =
2379 amdgpu_get_soft_full_reset_mask(&adev->gfx.gfx_ring[0]);
2380 adev->gfx.compute_supported_reset =
2381 amdgpu_get_soft_full_reset_mask(&adev->gfx.compute_ring[0]);
2382
2383 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE, 0);
2384 if (r) {
2385 DRM_ERROR("Failed to init KIQ BOs!\n");
2386 return r;
2387 }
2388
2389 r = amdgpu_gfx_kiq_init_ring(adev, xcc_id);
2390 if (r)
2391 return r;
2392
2393 /* create MQD for all compute queues as wel as KIQ for SRIOV case */
2394 r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation), 0);
2395 if (r)
2396 return r;
2397
2398 adev->gfx.ce_ram_size = 0x8000;
2399
2400 r = gfx_v9_0_gpu_early_init(adev);
2401 if (r)
2402 return r;
2403
2404 if (amdgpu_gfx_ras_sw_init(adev)) {
2405 dev_err(adev->dev, "Failed to initialize gfx ras block!\n");
2406 return -EINVAL;
2407 }
2408
2409 gfx_v9_0_alloc_ip_dump(adev);
2410
2411 r = amdgpu_gfx_sysfs_init(adev);
2412 if (r)
2413 return r;
2414
2415 return 0;
2416}
2417
2418
2419static int gfx_v9_0_sw_fini(struct amdgpu_ip_block *ip_block)
2420{
2421 int i;
2422 struct amdgpu_device *adev = ip_block->adev;
2423
2424 if (adev->gfx.mcbp && adev->gfx.num_gfx_rings) {
2425 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
2426 amdgpu_ring_fini(&adev->gfx.sw_gfx_ring[i]);
2427 amdgpu_ring_mux_fini(&adev->gfx.muxer);
2428 }
2429
2430 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2431 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2432 for (i = 0; i < adev->gfx.num_compute_rings; i++)
2433 amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2434
2435 amdgpu_gfx_mqd_sw_fini(adev, 0);
2436 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
2437 amdgpu_gfx_kiq_fini(adev, 0);
2438
2439 amdgpu_gfx_cleaner_shader_sw_fini(adev);
2440
2441 gfx_v9_0_mec_fini(adev);
2442 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2443 &adev->gfx.rlc.clear_state_gpu_addr,
2444 (void **)&adev->gfx.rlc.cs_ptr);
2445 if (adev->flags & AMD_IS_APU) {
2446 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2447 &adev->gfx.rlc.cp_table_gpu_addr,
2448 (void **)&adev->gfx.rlc.cp_table_ptr);
2449 }
2450 gfx_v9_0_free_microcode(adev);
2451
2452 amdgpu_gfx_sysfs_fini(adev);
2453
2454 kfree(adev->gfx.ip_dump_core);
2455 kfree(adev->gfx.ip_dump_compute_queues);
2456
2457 return 0;
2458}
2459
2460
2461static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
2462{
2463 /* TODO */
2464}
2465
2466void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num,
2467 u32 instance, int xcc_id)
2468{
2469 u32 data;
2470
2471 if (instance == 0xffffffff)
2472 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
2473 else
2474 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
2475
2476 if (se_num == 0xffffffff)
2477 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
2478 else
2479 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
2480
2481 if (sh_num == 0xffffffff)
2482 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
2483 else
2484 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
2485
2486 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data);
2487}
2488
2489static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
2490{
2491 u32 data, mask;
2492
2493 data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
2494 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
2495
2496 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
2497 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
2498
2499 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
2500 adev->gfx.config.max_sh_per_se);
2501
2502 return (~data) & mask;
2503}
2504
2505static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
2506{
2507 int i, j;
2508 u32 data;
2509 u32 active_rbs = 0;
2510 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
2511 adev->gfx.config.max_sh_per_se;
2512
2513 mutex_lock(&adev->grbm_idx_mutex);
2514 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2515 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2516 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
2517 data = gfx_v9_0_get_rb_active_bitmap(adev);
2518 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
2519 rb_bitmap_width_per_sh);
2520 }
2521 }
2522 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
2523 mutex_unlock(&adev->grbm_idx_mutex);
2524
2525 adev->gfx.config.backend_enable_mask = active_rbs;
2526 adev->gfx.config.num_rbs = hweight32(active_rbs);
2527}
2528
2529static void gfx_v9_0_debug_trap_config_init(struct amdgpu_device *adev,
2530 uint32_t first_vmid,
2531 uint32_t last_vmid)
2532{
2533 uint32_t data;
2534 uint32_t trap_config_vmid_mask = 0;
2535 int i;
2536
2537 /* Calculate trap config vmid mask */
2538 for (i = first_vmid; i < last_vmid; i++)
2539 trap_config_vmid_mask |= (1 << i);
2540
2541 data = REG_SET_FIELD(0, SPI_GDBG_TRAP_CONFIG,
2542 VMID_SEL, trap_config_vmid_mask);
2543 data = REG_SET_FIELD(data, SPI_GDBG_TRAP_CONFIG,
2544 TRAP_EN, 1);
2545 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_CONFIG), data);
2546 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
2547
2548 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA0), 0);
2549 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA1), 0);
2550}
2551
2552#define DEFAULT_SH_MEM_BASES (0x6000)
2553static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
2554{
2555 int i;
2556 uint32_t sh_mem_config;
2557 uint32_t sh_mem_bases;
2558
2559 /*
2560 * Configure apertures:
2561 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB)
2562 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB)
2563 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB)
2564 */
2565 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
2566
2567 sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
2568 SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
2569 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
2570
2571 mutex_lock(&adev->srbm_mutex);
2572 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
2573 soc15_grbm_select(adev, 0, 0, 0, i, 0);
2574 /* CP and shaders */
2575 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
2576 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
2577 }
2578 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
2579 mutex_unlock(&adev->srbm_mutex);
2580
2581 /* Initialize all compute VMIDs to have no GDS, GWS, or OA
2582 access. These should be enabled by FW for target VMIDs. */
2583 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
2584 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0);
2585 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0);
2586 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0);
2587 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0);
2588 }
2589}
2590
2591static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev)
2592{
2593 int vmid;
2594
2595 /*
2596 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
2597 * access. Compute VMIDs should be enabled by FW for target VMIDs,
2598 * the driver can enable them for graphics. VMID0 should maintain
2599 * access so that HWS firmware can save/restore entries.
2600 */
2601 for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
2602 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0);
2603 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0);
2604 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0);
2605 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0);
2606 }
2607}
2608
2609static void gfx_v9_0_init_sq_config(struct amdgpu_device *adev)
2610{
2611 uint32_t tmp;
2612
2613 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
2614 case IP_VERSION(9, 4, 1):
2615 tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG);
2616 tmp = REG_SET_FIELD(tmp, SQ_CONFIG, DISABLE_BARRIER_WAITCNT,
2617 !READ_ONCE(adev->barrier_has_auto_waitcnt));
2618 WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp);
2619 break;
2620 default:
2621 break;
2622 }
2623}
2624
2625static void gfx_v9_0_constants_init(struct amdgpu_device *adev)
2626{
2627 u32 tmp;
2628 int i;
2629
2630 WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
2631
2632 gfx_v9_0_tiling_mode_table_init(adev);
2633
2634 if (adev->gfx.num_gfx_rings)
2635 gfx_v9_0_setup_rb(adev);
2636 gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
2637 adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2);
2638
2639 /* XXX SH_MEM regs */
2640 /* where to put LDS, scratch, GPUVM in FSA64 space */
2641 mutex_lock(&adev->srbm_mutex);
2642 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB(0)].num_ids; i++) {
2643 soc15_grbm_select(adev, 0, 0, 0, i, 0);
2644 /* CP and shaders */
2645 if (i == 0) {
2646 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2647 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2648 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2649 !!adev->gmc.noretry);
2650 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2651 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0);
2652 } else {
2653 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2654 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2655 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2656 !!adev->gmc.noretry);
2657 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2658 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
2659 (adev->gmc.private_aperture_start >> 48));
2660 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
2661 (adev->gmc.shared_aperture_start >> 48));
2662 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp);
2663 }
2664 }
2665 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
2666
2667 mutex_unlock(&adev->srbm_mutex);
2668
2669 gfx_v9_0_init_compute_vmid(adev);
2670 gfx_v9_0_init_gds_vmid(adev);
2671 gfx_v9_0_init_sq_config(adev);
2672}
2673
2674static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
2675{
2676 u32 i, j, k;
2677 u32 mask;
2678
2679 mutex_lock(&adev->grbm_idx_mutex);
2680 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2681 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2682 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
2683 for (k = 0; k < adev->usec_timeout; k++) {
2684 if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0)
2685 break;
2686 udelay(1);
2687 }
2688 if (k == adev->usec_timeout) {
2689 amdgpu_gfx_select_se_sh(adev, 0xffffffff,
2690 0xffffffff, 0xffffffff, 0);
2691 mutex_unlock(&adev->grbm_idx_mutex);
2692 DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
2693 i, j);
2694 return;
2695 }
2696 }
2697 }
2698 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
2699 mutex_unlock(&adev->grbm_idx_mutex);
2700
2701 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
2702 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
2703 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
2704 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
2705 for (k = 0; k < adev->usec_timeout; k++) {
2706 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
2707 break;
2708 udelay(1);
2709 }
2710}
2711
2712static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
2713 bool enable)
2714{
2715 u32 tmp;
2716
2717 /* These interrupts should be enabled to drive DS clock */
2718
2719 tmp= RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
2720
2721 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
2722 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
2723 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
2724 if (adev->gfx.num_gfx_rings)
2725 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
2726
2727 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
2728}
2729
2730static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
2731{
2732 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
2733 /* csib */
2734 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),
2735 adev->gfx.rlc.clear_state_gpu_addr >> 32);
2736 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),
2737 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
2738 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),
2739 adev->gfx.rlc.clear_state_size);
2740}
2741
2742static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
2743 int indirect_offset,
2744 int list_size,
2745 int *unique_indirect_regs,
2746 int unique_indirect_reg_count,
2747 int *indirect_start_offsets,
2748 int *indirect_start_offsets_count,
2749 int max_start_offsets_count)
2750{
2751 int idx;
2752
2753 for (; indirect_offset < list_size; indirect_offset++) {
2754 WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count);
2755 indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
2756 *indirect_start_offsets_count = *indirect_start_offsets_count + 1;
2757
2758 while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
2759 indirect_offset += 2;
2760
2761 /* look for the matching indice */
2762 for (idx = 0; idx < unique_indirect_reg_count; idx++) {
2763 if (unique_indirect_regs[idx] ==
2764 register_list_format[indirect_offset] ||
2765 !unique_indirect_regs[idx])
2766 break;
2767 }
2768
2769 BUG_ON(idx >= unique_indirect_reg_count);
2770
2771 if (!unique_indirect_regs[idx])
2772 unique_indirect_regs[idx] = register_list_format[indirect_offset];
2773
2774 indirect_offset++;
2775 }
2776 }
2777}
2778
2779static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
2780{
2781 int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2782 int unique_indirect_reg_count = 0;
2783
2784 int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2785 int indirect_start_offsets_count = 0;
2786
2787 int list_size = 0;
2788 int i = 0, j = 0;
2789 u32 tmp = 0;
2790
2791 u32 *register_list_format =
2792 kmemdup(adev->gfx.rlc.register_list_format,
2793 adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
2794 if (!register_list_format)
2795 return -ENOMEM;
2796
2797 /* setup unique_indirect_regs array and indirect_start_offsets array */
2798 unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs);
2799 gfx_v9_1_parse_ind_reg_list(register_list_format,
2800 adev->gfx.rlc.reg_list_format_direct_reg_list_length,
2801 adev->gfx.rlc.reg_list_format_size_bytes >> 2,
2802 unique_indirect_regs,
2803 unique_indirect_reg_count,
2804 indirect_start_offsets,
2805 &indirect_start_offsets_count,
2806 ARRAY_SIZE(indirect_start_offsets));
2807
2808 /* enable auto inc in case it is disabled */
2809 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
2810 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
2811 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
2812
2813 /* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
2814 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),
2815 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET);
2816 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
2817 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
2818 adev->gfx.rlc.register_restore[i]);
2819
2820 /* load indirect register */
2821 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2822 adev->gfx.rlc.reg_list_format_start);
2823
2824 /* direct register portion */
2825 for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
2826 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2827 register_list_format[i]);
2828
2829 /* indirect register portion */
2830 while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
2831 if (register_list_format[i] == 0xFFFFFFFF) {
2832 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2833 continue;
2834 }
2835
2836 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2837 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2838
2839 for (j = 0; j < unique_indirect_reg_count; j++) {
2840 if (register_list_format[i] == unique_indirect_regs[j]) {
2841 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j);
2842 break;
2843 }
2844 }
2845
2846 BUG_ON(j >= unique_indirect_reg_count);
2847
2848 i++;
2849 }
2850
2851 /* set save/restore list size */
2852 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
2853 list_size = list_size >> 1;
2854 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2855 adev->gfx.rlc.reg_restore_list_size);
2856 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size);
2857
2858 /* write the starting offsets to RLC scratch ram */
2859 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2860 adev->gfx.rlc.starting_offsets_start);
2861 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
2862 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2863 indirect_start_offsets[i]);
2864
2865 /* load unique indirect regs*/
2866 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
2867 if (unique_indirect_regs[i] != 0) {
2868 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)
2869 + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],
2870 unique_indirect_regs[i] & 0x3FFFF);
2871
2872 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)
2873 + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],
2874 unique_indirect_regs[i] >> 20);
2875 }
2876 }
2877
2878 kfree(register_list_format);
2879 return 0;
2880}
2881
2882static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
2883{
2884 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
2885}
2886
2887static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
2888 bool enable)
2889{
2890 uint32_t data = 0;
2891 uint32_t default_data = 0;
2892
2893 default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS));
2894 if (enable) {
2895 /* enable GFXIP control over CGPG */
2896 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2897 if(default_data != data)
2898 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2899
2900 /* update status */
2901 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK;
2902 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT);
2903 if(default_data != data)
2904 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2905 } else {
2906 /* restore GFXIP control over GCPG */
2907 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2908 if(default_data != data)
2909 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2910 }
2911}
2912
2913static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
2914{
2915 uint32_t data = 0;
2916
2917 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2918 AMD_PG_SUPPORT_GFX_SMG |
2919 AMD_PG_SUPPORT_GFX_DMG)) {
2920 /* init IDLE_POLL_COUNT = 60 */
2921 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL));
2922 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
2923 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
2924 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data);
2925
2926 /* init RLC PG Delay */
2927 data = 0;
2928 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT);
2929 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT);
2930 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT);
2931 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT);
2932 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data);
2933
2934 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2));
2935 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK;
2936 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT);
2937 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data);
2938
2939 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3));
2940 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK;
2941 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT);
2942 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data);
2943
2944 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL));
2945 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
2946
2947 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
2948 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
2949 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data);
2950 if (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 3, 0))
2951 pwr_10_0_gfxip_control_over_cgpg(adev, true);
2952 }
2953}
2954
2955static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
2956 bool enable)
2957{
2958 uint32_t data = 0;
2959 uint32_t default_data = 0;
2960
2961 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2962 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2963 SMU_CLK_SLOWDOWN_ON_PU_ENABLE,
2964 enable ? 1 : 0);
2965 if (default_data != data)
2966 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2967}
2968
2969static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
2970 bool enable)
2971{
2972 uint32_t data = 0;
2973 uint32_t default_data = 0;
2974
2975 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2976 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2977 SMU_CLK_SLOWDOWN_ON_PD_ENABLE,
2978 enable ? 1 : 0);
2979 if(default_data != data)
2980 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2981}
2982
2983static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
2984 bool enable)
2985{
2986 uint32_t data = 0;
2987 uint32_t default_data = 0;
2988
2989 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2990 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2991 CP_PG_DISABLE,
2992 enable ? 0 : 1);
2993 if(default_data != data)
2994 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2995}
2996
2997static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
2998 bool enable)
2999{
3000 uint32_t data, default_data;
3001
3002 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
3003 data = REG_SET_FIELD(data, RLC_PG_CNTL,
3004 GFX_POWER_GATING_ENABLE,
3005 enable ? 1 : 0);
3006 if(default_data != data)
3007 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
3008}
3009
3010static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
3011 bool enable)
3012{
3013 uint32_t data, default_data;
3014
3015 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
3016 data = REG_SET_FIELD(data, RLC_PG_CNTL,
3017 GFX_PIPELINE_PG_ENABLE,
3018 enable ? 1 : 0);
3019 if(default_data != data)
3020 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
3021
3022 if (!enable)
3023 /* read any GFX register to wake up GFX */
3024 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL));
3025}
3026
3027static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
3028 bool enable)
3029{
3030 uint32_t data, default_data;
3031
3032 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
3033 data = REG_SET_FIELD(data, RLC_PG_CNTL,
3034 STATIC_PER_CU_PG_ENABLE,
3035 enable ? 1 : 0);
3036 if(default_data != data)
3037 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
3038}
3039
3040static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
3041 bool enable)
3042{
3043 uint32_t data, default_data;
3044
3045 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
3046 data = REG_SET_FIELD(data, RLC_PG_CNTL,
3047 DYN_PER_CU_PG_ENABLE,
3048 enable ? 1 : 0);
3049 if(default_data != data)
3050 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
3051}
3052
3053static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
3054{
3055 gfx_v9_0_init_csb(adev);
3056
3057 /*
3058 * Rlc save restore list is workable since v2_1.
3059 * And it's needed by gfxoff feature.
3060 */
3061 if (adev->gfx.rlc.is_rlc_v2_1) {
3062 if (amdgpu_ip_version(adev, GC_HWIP, 0) ==
3063 IP_VERSION(9, 2, 1) ||
3064 (adev->apu_flags & AMD_APU_IS_RAVEN2))
3065 gfx_v9_1_init_rlc_save_restore_list(adev);
3066 gfx_v9_0_enable_save_restore_machine(adev);
3067 }
3068
3069 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
3070 AMD_PG_SUPPORT_GFX_SMG |
3071 AMD_PG_SUPPORT_GFX_DMG |
3072 AMD_PG_SUPPORT_CP |
3073 AMD_PG_SUPPORT_GDS |
3074 AMD_PG_SUPPORT_RLC_SMU_HS)) {
3075 WREG32_SOC15(GC, 0, mmRLC_JUMP_TABLE_RESTORE,
3076 adev->gfx.rlc.cp_table_gpu_addr >> 8);
3077 gfx_v9_0_init_gfx_power_gating(adev);
3078 }
3079}
3080
3081static void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
3082{
3083 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
3084 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3085 gfx_v9_0_wait_for_rlc_serdes(adev);
3086}
3087
3088static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
3089{
3090 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
3091 udelay(50);
3092 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
3093 udelay(50);
3094}
3095
3096static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
3097{
3098#ifdef AMDGPU_RLC_DEBUG_RETRY
3099 u32 rlc_ucode_ver;
3100#endif
3101
3102 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
3103 udelay(50);
3104
3105 /* carrizo do enable cp interrupt after cp inited */
3106 if (!(adev->flags & AMD_IS_APU)) {
3107 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3108 udelay(50);
3109 }
3110
3111#ifdef AMDGPU_RLC_DEBUG_RETRY
3112 /* RLC_GPM_GENERAL_6 : RLC Ucode version */
3113 rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6);
3114 if(rlc_ucode_ver == 0x108) {
3115 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
3116 rlc_ucode_ver, adev->gfx.rlc_fw_version);
3117 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
3118 * default is 0x9C4 to create a 100us interval */
3119 WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4);
3120 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
3121 * to disable the page fault retry interrupts, default is
3122 * 0x100 (256) */
3123 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100);
3124 }
3125#endif
3126}
3127
3128static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
3129{
3130 const struct rlc_firmware_header_v2_0 *hdr;
3131 const __le32 *fw_data;
3132 unsigned i, fw_size;
3133
3134 if (!adev->gfx.rlc_fw)
3135 return -EINVAL;
3136
3137 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
3138 amdgpu_ucode_print_rlc_hdr(&hdr->header);
3139
3140 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
3141 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3142 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3143
3144 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
3145 RLCG_UCODE_LOADING_START_ADDRESS);
3146 for (i = 0; i < fw_size; i++)
3147 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3148 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3149
3150 return 0;
3151}
3152
3153static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
3154{
3155 int r;
3156
3157 if (amdgpu_sriov_vf(adev)) {
3158 gfx_v9_0_init_csb(adev);
3159 return 0;
3160 }
3161
3162 adev->gfx.rlc.funcs->stop(adev);
3163
3164 /* disable CG */
3165 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
3166
3167 gfx_v9_0_init_pg(adev);
3168
3169 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3170 /* legacy rlc firmware loading */
3171 r = gfx_v9_0_rlc_load_microcode(adev);
3172 if (r)
3173 return r;
3174 }
3175
3176 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
3177 case IP_VERSION(9, 2, 2):
3178 case IP_VERSION(9, 1, 0):
3179 gfx_v9_0_init_lbpw(adev);
3180 if (amdgpu_lbpw == 0)
3181 gfx_v9_0_enable_lbpw(adev, false);
3182 else
3183 gfx_v9_0_enable_lbpw(adev, true);
3184 break;
3185 case IP_VERSION(9, 4, 0):
3186 gfx_v9_4_init_lbpw(adev);
3187 if (amdgpu_lbpw > 0)
3188 gfx_v9_0_enable_lbpw(adev, true);
3189 else
3190 gfx_v9_0_enable_lbpw(adev, false);
3191 break;
3192 default:
3193 break;
3194 }
3195
3196 gfx_v9_0_update_spm_vmid_internal(adev, 0xf);
3197
3198 adev->gfx.rlc.funcs->start(adev);
3199
3200 return 0;
3201}
3202
3203static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
3204{
3205 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
3206
3207 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_INVALIDATE_ICACHE, enable ? 0 : 1);
3208 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_INVALIDATE_ICACHE, enable ? 0 : 1);
3209 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_INVALIDATE_ICACHE, enable ? 0 : 1);
3210 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_PIPE0_RESET, enable ? 0 : 1);
3211 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_PIPE1_RESET, enable ? 0 : 1);
3212 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, enable ? 0 : 1);
3213 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, enable ? 0 : 1);
3214 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, enable ? 0 : 1);
3215 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, enable ? 0 : 1);
3216 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
3217 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
3218 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
3219 WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp);
3220 udelay(50);
3221}
3222
3223static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
3224{
3225 const struct gfx_firmware_header_v1_0 *pfp_hdr;
3226 const struct gfx_firmware_header_v1_0 *ce_hdr;
3227 const struct gfx_firmware_header_v1_0 *me_hdr;
3228 const __le32 *fw_data;
3229 unsigned i, fw_size;
3230
3231 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
3232 return -EINVAL;
3233
3234 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
3235 adev->gfx.pfp_fw->data;
3236 ce_hdr = (const struct gfx_firmware_header_v1_0 *)
3237 adev->gfx.ce_fw->data;
3238 me_hdr = (const struct gfx_firmware_header_v1_0 *)
3239 adev->gfx.me_fw->data;
3240
3241 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
3242 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
3243 amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
3244
3245 gfx_v9_0_cp_gfx_enable(adev, false);
3246
3247 /* PFP */
3248 fw_data = (const __le32 *)
3249 (adev->gfx.pfp_fw->data +
3250 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
3251 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
3252 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0);
3253 for (i = 0; i < fw_size; i++)
3254 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
3255 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
3256
3257 /* CE */
3258 fw_data = (const __le32 *)
3259 (adev->gfx.ce_fw->data +
3260 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
3261 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
3262 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0);
3263 for (i = 0; i < fw_size; i++)
3264 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
3265 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
3266
3267 /* ME */
3268 fw_data = (const __le32 *)
3269 (adev->gfx.me_fw->data +
3270 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
3271 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
3272 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0);
3273 for (i = 0; i < fw_size; i++)
3274 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
3275 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
3276
3277 return 0;
3278}
3279
3280static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
3281{
3282 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
3283 const struct cs_section_def *sect = NULL;
3284 const struct cs_extent_def *ext = NULL;
3285 int r, i, tmp;
3286
3287 /* init the CP */
3288 WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
3289 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
3290
3291 gfx_v9_0_cp_gfx_enable(adev, true);
3292
3293 /* Now only limit the quirk on the APU gfx9 series and already
3294 * confirmed that the APU gfx10/gfx11 needn't such update.
3295 */
3296 if (adev->flags & AMD_IS_APU &&
3297 adev->in_s3 && !pm_resume_via_firmware()) {
3298 DRM_INFO("Will skip the CSB packet resubmit\n");
3299 return 0;
3300 }
3301 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
3302 if (r) {
3303 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3304 return r;
3305 }
3306
3307 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3308 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3309
3310 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3311 amdgpu_ring_write(ring, 0x80000000);
3312 amdgpu_ring_write(ring, 0x80000000);
3313
3314 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
3315 for (ext = sect->section; ext->extent != NULL; ++ext) {
3316 if (sect->id == SECT_CONTEXT) {
3317 amdgpu_ring_write(ring,
3318 PACKET3(PACKET3_SET_CONTEXT_REG,
3319 ext->reg_count));
3320 amdgpu_ring_write(ring,
3321 ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3322 for (i = 0; i < ext->reg_count; i++)
3323 amdgpu_ring_write(ring, ext->extent[i]);
3324 }
3325 }
3326 }
3327
3328 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3329 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3330
3331 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3332 amdgpu_ring_write(ring, 0);
3333
3334 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3335 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3336 amdgpu_ring_write(ring, 0x8000);
3337 amdgpu_ring_write(ring, 0x8000);
3338
3339 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
3340 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
3341 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
3342 amdgpu_ring_write(ring, tmp);
3343 amdgpu_ring_write(ring, 0);
3344
3345 amdgpu_ring_commit(ring);
3346
3347 return 0;
3348}
3349
3350static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
3351{
3352 struct amdgpu_ring *ring;
3353 u32 tmp;
3354 u32 rb_bufsz;
3355 u64 rb_addr, rptr_addr, wptr_gpu_addr;
3356
3357 /* Set the write pointer delay */
3358 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
3359
3360 /* set the RB to use vmid 0 */
3361 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
3362
3363 /* Set ring buffer size */
3364 ring = &adev->gfx.gfx_ring[0];
3365 rb_bufsz = order_base_2(ring->ring_size / 8);
3366 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3367 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3368#ifdef __BIG_ENDIAN
3369 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
3370#endif
3371 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3372
3373 /* Initialize the ring buffer's write pointers */
3374 ring->wptr = 0;
3375 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3376 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3377
3378 /* set the wb address whether it's enabled or not */
3379 rptr_addr = ring->rptr_gpu_addr;
3380 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3381 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3382
3383 wptr_gpu_addr = ring->wptr_gpu_addr;
3384 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
3385 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
3386
3387 mdelay(1);
3388 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3389
3390 rb_addr = ring->gpu_addr >> 8;
3391 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
3392 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3393
3394 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3395 if (ring->use_doorbell) {
3396 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3397 DOORBELL_OFFSET, ring->doorbell_index);
3398 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3399 DOORBELL_EN, 1);
3400 } else {
3401 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
3402 }
3403 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
3404
3405 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3406 DOORBELL_RANGE_LOWER, ring->doorbell_index);
3407 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
3408
3409 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
3410 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3411
3412
3413 /* start the ring */
3414 gfx_v9_0_cp_gfx_start(adev);
3415
3416 return 0;
3417}
3418
3419static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3420{
3421 if (enable) {
3422 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0);
3423 } else {
3424 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,
3425 (CP_MEC_CNTL__MEC_INVALIDATE_ICACHE_MASK |
3426 CP_MEC_CNTL__MEC_ME1_PIPE0_RESET_MASK |
3427 CP_MEC_CNTL__MEC_ME1_PIPE1_RESET_MASK |
3428 CP_MEC_CNTL__MEC_ME1_PIPE2_RESET_MASK |
3429 CP_MEC_CNTL__MEC_ME1_PIPE3_RESET_MASK |
3430 CP_MEC_CNTL__MEC_ME2_PIPE0_RESET_MASK |
3431 CP_MEC_CNTL__MEC_ME2_PIPE1_RESET_MASK |
3432 CP_MEC_CNTL__MEC_ME1_HALT_MASK |
3433 CP_MEC_CNTL__MEC_ME2_HALT_MASK));
3434 adev->gfx.kiq[0].ring.sched.ready = false;
3435 }
3436 udelay(50);
3437}
3438
3439static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3440{
3441 const struct gfx_firmware_header_v1_0 *mec_hdr;
3442 const __le32 *fw_data;
3443 unsigned i;
3444 u32 tmp;
3445
3446 if (!adev->gfx.mec_fw)
3447 return -EINVAL;
3448
3449 gfx_v9_0_cp_compute_enable(adev, false);
3450
3451 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3452 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3453
3454 fw_data = (const __le32 *)
3455 (adev->gfx.mec_fw->data +
3456 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3457 tmp = 0;
3458 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3459 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3460 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
3461
3462 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
3463 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
3464 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
3465 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3466
3467 /* MEC1 */
3468 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3469 mec_hdr->jt_offset);
3470 for (i = 0; i < mec_hdr->jt_size; i++)
3471 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
3472 le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3473
3474 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3475 adev->gfx.mec_fw_version);
3476 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3477
3478 return 0;
3479}
3480
3481/* KIQ functions */
3482static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3483{
3484 uint32_t tmp;
3485 struct amdgpu_device *adev = ring->adev;
3486
3487 /* tell RLC which is KIQ queue */
3488 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
3489 tmp &= 0xffffff00;
3490 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3491 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3492 tmp |= 0x80;
3493 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3494}
3495
3496static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
3497{
3498 struct amdgpu_device *adev = ring->adev;
3499
3500 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3501 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
3502 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
3503 mqd->cp_hqd_queue_priority =
3504 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
3505 }
3506 }
3507}
3508
3509static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3510{
3511 struct amdgpu_device *adev = ring->adev;
3512 struct v9_mqd *mqd = ring->mqd_ptr;
3513 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3514 uint32_t tmp;
3515
3516 mqd->header = 0xC0310800;
3517 mqd->compute_pipelinestat_enable = 0x00000001;
3518 mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3519 mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3520 mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3521 mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3522 mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3523 mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3524 mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3525 mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3526 mqd->compute_misc_reserved = 0x00000003;
3527
3528 mqd->dynamic_cu_mask_addr_lo =
3529 lower_32_bits(ring->mqd_gpu_addr
3530 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3531 mqd->dynamic_cu_mask_addr_hi =
3532 upper_32_bits(ring->mqd_gpu_addr
3533 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3534
3535 eop_base_addr = ring->eop_gpu_addr >> 8;
3536 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3537 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3538
3539 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3540 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3541 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3542 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
3543
3544 mqd->cp_hqd_eop_control = tmp;
3545
3546 /* enable doorbell? */
3547 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3548
3549 if (ring->use_doorbell) {
3550 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3551 DOORBELL_OFFSET, ring->doorbell_index);
3552 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3553 DOORBELL_EN, 1);
3554 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3555 DOORBELL_SOURCE, 0);
3556 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3557 DOORBELL_HIT, 0);
3558 } else {
3559 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3560 DOORBELL_EN, 0);
3561 }
3562
3563 mqd->cp_hqd_pq_doorbell_control = tmp;
3564
3565 /* disable the queue if it's active */
3566 ring->wptr = 0;
3567 mqd->cp_hqd_dequeue_request = 0;
3568 mqd->cp_hqd_pq_rptr = 0;
3569 mqd->cp_hqd_pq_wptr_lo = 0;
3570 mqd->cp_hqd_pq_wptr_hi = 0;
3571
3572 /* set the pointer to the MQD */
3573 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3574 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3575
3576 /* set MQD vmid to 0 */
3577 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3578 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3579 mqd->cp_mqd_control = tmp;
3580
3581 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3582 hqd_gpu_addr = ring->gpu_addr >> 8;
3583 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3584 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3585
3586 /* set up the HQD, this is similar to CP_RB0_CNTL */
3587 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3588 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3589 (order_base_2(ring->ring_size / 4) - 1));
3590 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3591 (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
3592#ifdef __BIG_ENDIAN
3593 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3594#endif
3595 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3596 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
3597 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3598 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3599 mqd->cp_hqd_pq_control = tmp;
3600
3601 /* set the wb address whether it's enabled or not */
3602 wb_gpu_addr = ring->rptr_gpu_addr;
3603 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3604 mqd->cp_hqd_pq_rptr_report_addr_hi =
3605 upper_32_bits(wb_gpu_addr) & 0xffff;
3606
3607 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3608 wb_gpu_addr = ring->wptr_gpu_addr;
3609 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3610 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3611
3612 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3613 ring->wptr = 0;
3614 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3615
3616 /* set the vmid for the queue */
3617 mqd->cp_hqd_vmid = 0;
3618
3619 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3620 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3621 mqd->cp_hqd_persistent_state = tmp;
3622
3623 /* set MIN_IB_AVAIL_SIZE */
3624 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3625 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3626 mqd->cp_hqd_ib_control = tmp;
3627
3628 /* set static priority for a queue/ring */
3629 gfx_v9_0_mqd_set_priority(ring, mqd);
3630 mqd->cp_hqd_quantum = RREG32_SOC15(GC, 0, mmCP_HQD_QUANTUM);
3631
3632 /* map_queues packet doesn't need activate the queue,
3633 * so only kiq need set this field.
3634 */
3635 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
3636 mqd->cp_hqd_active = 1;
3637
3638 return 0;
3639}
3640
3641static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3642{
3643 struct amdgpu_device *adev = ring->adev;
3644 struct v9_mqd *mqd = ring->mqd_ptr;
3645 int j;
3646
3647 /* disable wptr polling */
3648 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3649
3650 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3651 mqd->cp_hqd_eop_base_addr_lo);
3652 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3653 mqd->cp_hqd_eop_base_addr_hi);
3654
3655 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3656 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,
3657 mqd->cp_hqd_eop_control);
3658
3659 /* enable doorbell? */
3660 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3661 mqd->cp_hqd_pq_doorbell_control);
3662
3663 /* disable the queue if it's active */
3664 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3665 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3666 for (j = 0; j < adev->usec_timeout; j++) {
3667 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3668 break;
3669 udelay(1);
3670 }
3671 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3672 mqd->cp_hqd_dequeue_request);
3673 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,
3674 mqd->cp_hqd_pq_rptr);
3675 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3676 mqd->cp_hqd_pq_wptr_lo);
3677 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3678 mqd->cp_hqd_pq_wptr_hi);
3679 }
3680
3681 /* set the pointer to the MQD */
3682 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,
3683 mqd->cp_mqd_base_addr_lo);
3684 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3685 mqd->cp_mqd_base_addr_hi);
3686
3687 /* set MQD vmid to 0 */
3688 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,
3689 mqd->cp_mqd_control);
3690
3691 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3692 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,
3693 mqd->cp_hqd_pq_base_lo);
3694 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,
3695 mqd->cp_hqd_pq_base_hi);
3696
3697 /* set up the HQD, this is similar to CP_RB0_CNTL */
3698 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,
3699 mqd->cp_hqd_pq_control);
3700
3701 /* set the wb address whether it's enabled or not */
3702 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3703 mqd->cp_hqd_pq_rptr_report_addr_lo);
3704 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3705 mqd->cp_hqd_pq_rptr_report_addr_hi);
3706
3707 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3708 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3709 mqd->cp_hqd_pq_wptr_poll_addr_lo);
3710 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3711 mqd->cp_hqd_pq_wptr_poll_addr_hi);
3712
3713 /* enable the doorbell if requested */
3714 if (ring->use_doorbell) {
3715 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3716 (adev->doorbell_index.kiq * 2) << 2);
3717 /* If GC has entered CGPG, ringing doorbell > first page
3718 * doesn't wakeup GC. Enlarge CP_MEC_DOORBELL_RANGE_UPPER to
3719 * workaround this issue. And this change has to align with firmware
3720 * update.
3721 */
3722 if (check_if_enlarge_doorbell_range(adev))
3723 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3724 (adev->doorbell.size - 4));
3725 else
3726 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3727 (adev->doorbell_index.userqueue_end * 2) << 2);
3728 }
3729
3730 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3731 mqd->cp_hqd_pq_doorbell_control);
3732
3733 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3734 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3735 mqd->cp_hqd_pq_wptr_lo);
3736 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3737 mqd->cp_hqd_pq_wptr_hi);
3738
3739 /* set the vmid for the queue */
3740 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3741
3742 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3743 mqd->cp_hqd_persistent_state);
3744
3745 /* activate the queue */
3746 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,
3747 mqd->cp_hqd_active);
3748
3749 if (ring->use_doorbell)
3750 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3751
3752 return 0;
3753}
3754
3755static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3756{
3757 struct amdgpu_device *adev = ring->adev;
3758 int j;
3759
3760 /* disable the queue if it's active */
3761 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3762
3763 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3764
3765 for (j = 0; j < adev->usec_timeout; j++) {
3766 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3767 break;
3768 udelay(1);
3769 }
3770
3771 if (j == AMDGPU_MAX_USEC_TIMEOUT) {
3772 DRM_DEBUG("KIQ dequeue request failed.\n");
3773
3774 /* Manual disable if dequeue request times out */
3775 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0);
3776 }
3777
3778 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3779 0);
3780 }
3781
3782 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0);
3783 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0);
3784 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
3785 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
3786 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
3787 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0);
3788 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
3789 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
3790
3791 return 0;
3792}
3793
3794static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3795{
3796 struct amdgpu_device *adev = ring->adev;
3797 struct v9_mqd *mqd = ring->mqd_ptr;
3798 struct v9_mqd *tmp_mqd;
3799
3800 gfx_v9_0_kiq_setting(ring);
3801
3802 /* GPU could be in bad state during probe, driver trigger the reset
3803 * after load the SMU, in this case , the mqd is not be initialized.
3804 * driver need to re-init the mqd.
3805 * check mqd->cp_hqd_pq_control since this value should not be 0
3806 */
3807 tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[0].mqd_backup;
3808 if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){
3809 /* for GPU_RESET case , reset MQD to a clean status */
3810 if (adev->gfx.kiq[0].mqd_backup)
3811 memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(struct v9_mqd_allocation));
3812
3813 /* reset ring buffer */
3814 ring->wptr = 0;
3815 amdgpu_ring_clear_ring(ring);
3816
3817 mutex_lock(&adev->srbm_mutex);
3818 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3819 gfx_v9_0_kiq_init_register(ring);
3820 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3821 mutex_unlock(&adev->srbm_mutex);
3822 } else {
3823 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3824 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3825 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3826 if (amdgpu_sriov_vf(adev) && adev->in_suspend)
3827 amdgpu_ring_clear_ring(ring);
3828 mutex_lock(&adev->srbm_mutex);
3829 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3830 gfx_v9_0_mqd_init(ring);
3831 gfx_v9_0_kiq_init_register(ring);
3832 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3833 mutex_unlock(&adev->srbm_mutex);
3834
3835 if (adev->gfx.kiq[0].mqd_backup)
3836 memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(struct v9_mqd_allocation));
3837 }
3838
3839 return 0;
3840}
3841
3842static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring, bool restore)
3843{
3844 struct amdgpu_device *adev = ring->adev;
3845 struct v9_mqd *mqd = ring->mqd_ptr;
3846 int mqd_idx = ring - &adev->gfx.compute_ring[0];
3847 struct v9_mqd *tmp_mqd;
3848
3849 /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
3850 * is not be initialized before
3851 */
3852 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3853
3854 if (!restore && (!tmp_mqd->cp_hqd_pq_control ||
3855 (!amdgpu_in_reset(adev) && !adev->in_suspend))) {
3856 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3857 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3858 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3859 mutex_lock(&adev->srbm_mutex);
3860 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3861 gfx_v9_0_mqd_init(ring);
3862 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3863 mutex_unlock(&adev->srbm_mutex);
3864
3865 if (adev->gfx.mec.mqd_backup[mqd_idx])
3866 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3867 } else {
3868 /* restore MQD to a clean status */
3869 if (adev->gfx.mec.mqd_backup[mqd_idx])
3870 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3871 /* reset ring buffer */
3872 ring->wptr = 0;
3873 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
3874 amdgpu_ring_clear_ring(ring);
3875 }
3876
3877 return 0;
3878}
3879
3880static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3881{
3882 struct amdgpu_ring *ring;
3883 int r;
3884
3885 ring = &adev->gfx.kiq[0].ring;
3886
3887 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3888 if (unlikely(r != 0))
3889 return r;
3890
3891 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3892 if (unlikely(r != 0)) {
3893 amdgpu_bo_unreserve(ring->mqd_obj);
3894 return r;
3895 }
3896
3897 gfx_v9_0_kiq_init_queue(ring);
3898 amdgpu_bo_kunmap(ring->mqd_obj);
3899 ring->mqd_ptr = NULL;
3900 amdgpu_bo_unreserve(ring->mqd_obj);
3901 return 0;
3902}
3903
3904static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3905{
3906 struct amdgpu_ring *ring = NULL;
3907 int r = 0, i;
3908
3909 gfx_v9_0_cp_compute_enable(adev, true);
3910
3911 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3912 ring = &adev->gfx.compute_ring[i];
3913
3914 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3915 if (unlikely(r != 0))
3916 goto done;
3917 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3918 if (!r) {
3919 r = gfx_v9_0_kcq_init_queue(ring, false);
3920 amdgpu_bo_kunmap(ring->mqd_obj);
3921 ring->mqd_ptr = NULL;
3922 }
3923 amdgpu_bo_unreserve(ring->mqd_obj);
3924 if (r)
3925 goto done;
3926 }
3927
3928 r = amdgpu_gfx_enable_kcq(adev, 0);
3929done:
3930 return r;
3931}
3932
3933static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3934{
3935 int r, i;
3936 struct amdgpu_ring *ring;
3937
3938 if (!(adev->flags & AMD_IS_APU))
3939 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3940
3941 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3942 if (adev->gfx.num_gfx_rings) {
3943 /* legacy firmware loading */
3944 r = gfx_v9_0_cp_gfx_load_microcode(adev);
3945 if (r)
3946 return r;
3947 }
3948
3949 r = gfx_v9_0_cp_compute_load_microcode(adev);
3950 if (r)
3951 return r;
3952 }
3953
3954 if (adev->gfx.num_gfx_rings)
3955 gfx_v9_0_cp_gfx_enable(adev, false);
3956 gfx_v9_0_cp_compute_enable(adev, false);
3957
3958 r = gfx_v9_0_kiq_resume(adev);
3959 if (r)
3960 return r;
3961
3962 if (adev->gfx.num_gfx_rings) {
3963 r = gfx_v9_0_cp_gfx_resume(adev);
3964 if (r)
3965 return r;
3966 }
3967
3968 r = gfx_v9_0_kcq_resume(adev);
3969 if (r)
3970 return r;
3971
3972 if (adev->gfx.num_gfx_rings) {
3973 ring = &adev->gfx.gfx_ring[0];
3974 r = amdgpu_ring_test_helper(ring);
3975 if (r)
3976 return r;
3977 }
3978
3979 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3980 ring = &adev->gfx.compute_ring[i];
3981 amdgpu_ring_test_helper(ring);
3982 }
3983
3984 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3985
3986 return 0;
3987}
3988
3989static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev)
3990{
3991 u32 tmp;
3992
3993 if (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 1) &&
3994 amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 2))
3995 return;
3996
3997 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG);
3998 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH,
3999 adev->df.hash_status.hash_64k);
4000 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH,
4001 adev->df.hash_status.hash_2m);
4002 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH,
4003 adev->df.hash_status.hash_1g);
4004 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp);
4005}
4006
4007static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
4008{
4009 if (adev->gfx.num_gfx_rings)
4010 gfx_v9_0_cp_gfx_enable(adev, enable);
4011 gfx_v9_0_cp_compute_enable(adev, enable);
4012}
4013
4014static int gfx_v9_0_hw_init(struct amdgpu_ip_block *ip_block)
4015{
4016 int r;
4017 struct amdgpu_device *adev = ip_block->adev;
4018
4019 amdgpu_gfx_cleaner_shader_init(adev, adev->gfx.cleaner_shader_size,
4020 adev->gfx.cleaner_shader_ptr);
4021
4022 if (!amdgpu_sriov_vf(adev))
4023 gfx_v9_0_init_golden_registers(adev);
4024
4025 gfx_v9_0_constants_init(adev);
4026
4027 gfx_v9_0_init_tcp_config(adev);
4028
4029 r = adev->gfx.rlc.funcs->resume(adev);
4030 if (r)
4031 return r;
4032
4033 r = gfx_v9_0_cp_resume(adev);
4034 if (r)
4035 return r;
4036
4037 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2))
4038 gfx_v9_4_2_set_power_brake_sequence(adev);
4039
4040 return r;
4041}
4042
4043static int gfx_v9_0_hw_fini(struct amdgpu_ip_block *ip_block)
4044{
4045 struct amdgpu_device *adev = ip_block->adev;
4046
4047 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4048 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4049 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4050 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4051 amdgpu_irq_put(adev, &adev->gfx.bad_op_irq, 0);
4052
4053 /* DF freeze and kcq disable will fail */
4054 if (!amdgpu_ras_intr_triggered())
4055 /* disable KCQ to avoid CPC touch memory not valid anymore */
4056 amdgpu_gfx_disable_kcq(adev, 0);
4057
4058 if (amdgpu_sriov_vf(adev)) {
4059 gfx_v9_0_cp_gfx_enable(adev, false);
4060 /* must disable polling for SRIOV when hw finished, otherwise
4061 * CPC engine may still keep fetching WB address which is already
4062 * invalid after sw finished and trigger DMAR reading error in
4063 * hypervisor side.
4064 */
4065 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
4066 return 0;
4067 }
4068
4069 /* Use deinitialize sequence from CAIL when unbinding device from driver,
4070 * otherwise KIQ is hanging when binding back
4071 */
4072 if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4073 mutex_lock(&adev->srbm_mutex);
4074 soc15_grbm_select(adev, adev->gfx.kiq[0].ring.me,
4075 adev->gfx.kiq[0].ring.pipe,
4076 adev->gfx.kiq[0].ring.queue, 0, 0);
4077 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq[0].ring);
4078 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
4079 mutex_unlock(&adev->srbm_mutex);
4080 }
4081
4082 gfx_v9_0_cp_enable(adev, false);
4083
4084 /* Skip stopping RLC with A+A reset or when RLC controls GFX clock */
4085 if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) ||
4086 (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(9, 4, 2))) {
4087 dev_dbg(adev->dev, "Skipping RLC halt\n");
4088 return 0;
4089 }
4090
4091 adev->gfx.rlc.funcs->stop(adev);
4092 return 0;
4093}
4094
4095static int gfx_v9_0_suspend(struct amdgpu_ip_block *ip_block)
4096{
4097 return gfx_v9_0_hw_fini(ip_block);
4098}
4099
4100static int gfx_v9_0_resume(struct amdgpu_ip_block *ip_block)
4101{
4102 return gfx_v9_0_hw_init(ip_block);
4103}
4104
4105static bool gfx_v9_0_is_idle(void *handle)
4106{
4107 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4108
4109 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
4110 GRBM_STATUS, GUI_ACTIVE))
4111 return false;
4112 else
4113 return true;
4114}
4115
4116static int gfx_v9_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
4117{
4118 unsigned i;
4119 struct amdgpu_device *adev = ip_block->adev;
4120
4121 for (i = 0; i < adev->usec_timeout; i++) {
4122 if (gfx_v9_0_is_idle(adev))
4123 return 0;
4124 udelay(1);
4125 }
4126 return -ETIMEDOUT;
4127}
4128
4129static int gfx_v9_0_soft_reset(struct amdgpu_ip_block *ip_block)
4130{
4131 u32 grbm_soft_reset = 0;
4132 u32 tmp;
4133 struct amdgpu_device *adev = ip_block->adev;
4134
4135 /* GRBM_STATUS */
4136 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
4137 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4138 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4139 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4140 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4141 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4142 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
4143 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4144 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4145 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4146 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4147 }
4148
4149 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4150 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4151 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4152 }
4153
4154 /* GRBM_STATUS2 */
4155 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
4156 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4157 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4158 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4159
4160
4161 if (grbm_soft_reset) {
4162 /* stop the rlc */
4163 adev->gfx.rlc.funcs->stop(adev);
4164
4165 if (adev->gfx.num_gfx_rings)
4166 /* Disable GFX parsing/prefetching */
4167 gfx_v9_0_cp_gfx_enable(adev, false);
4168
4169 /* Disable MEC parsing/prefetching */
4170 gfx_v9_0_cp_compute_enable(adev, false);
4171
4172 if (grbm_soft_reset) {
4173 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4174 tmp |= grbm_soft_reset;
4175 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4176 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
4177 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4178
4179 udelay(50);
4180
4181 tmp &= ~grbm_soft_reset;
4182 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
4183 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4184 }
4185
4186 /* Wait a little for things to settle down */
4187 udelay(50);
4188 }
4189 return 0;
4190}
4191
4192static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev)
4193{
4194 signed long r, cnt = 0;
4195 unsigned long flags;
4196 uint32_t seq, reg_val_offs = 0;
4197 uint64_t value = 0;
4198 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
4199 struct amdgpu_ring *ring = &kiq->ring;
4200
4201 BUG_ON(!ring->funcs->emit_rreg);
4202
4203 spin_lock_irqsave(&kiq->ring_lock, flags);
4204 if (amdgpu_device_wb_get(adev, ®_val_offs)) {
4205 pr_err("critical bug! too many kiq readers\n");
4206 goto failed_unlock;
4207 }
4208 amdgpu_ring_alloc(ring, 32);
4209 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
4210 amdgpu_ring_write(ring, 9 | /* src: register*/
4211 (5 << 8) | /* dst: memory */
4212 (1 << 16) | /* count sel */
4213 (1 << 20)); /* write confirm */
4214 amdgpu_ring_write(ring, 0);
4215 amdgpu_ring_write(ring, 0);
4216 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
4217 reg_val_offs * 4));
4218 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
4219 reg_val_offs * 4));
4220 r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
4221 if (r)
4222 goto failed_undo;
4223
4224 amdgpu_ring_commit(ring);
4225 spin_unlock_irqrestore(&kiq->ring_lock, flags);
4226
4227 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
4228
4229 /* don't wait anymore for gpu reset case because this way may
4230 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
4231 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
4232 * never return if we keep waiting in virt_kiq_rreg, which cause
4233 * gpu_recover() hang there.
4234 *
4235 * also don't wait anymore for IRQ context
4236 * */
4237 if (r < 1 && (amdgpu_in_reset(adev)))
4238 goto failed_kiq_read;
4239
4240 might_sleep();
4241 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
4242 msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
4243 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
4244 }
4245
4246 if (cnt > MAX_KIQ_REG_TRY)
4247 goto failed_kiq_read;
4248
4249 mb();
4250 value = (uint64_t)adev->wb.wb[reg_val_offs] |
4251 (uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL;
4252 amdgpu_device_wb_free(adev, reg_val_offs);
4253 return value;
4254
4255failed_undo:
4256 amdgpu_ring_undo(ring);
4257failed_unlock:
4258 spin_unlock_irqrestore(&kiq->ring_lock, flags);
4259failed_kiq_read:
4260 if (reg_val_offs)
4261 amdgpu_device_wb_free(adev, reg_val_offs);
4262 pr_err("failed to read gpu clock\n");
4263 return ~0;
4264}
4265
4266static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4267{
4268 uint64_t clock, clock_lo, clock_hi, hi_check;
4269
4270 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
4271 case IP_VERSION(9, 3, 0):
4272 preempt_disable();
4273 clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
4274 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4275 hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
4276 /* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over
4277 * roughly every 42 seconds.
4278 */
4279 if (hi_check != clock_hi) {
4280 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4281 clock_hi = hi_check;
4282 }
4283 preempt_enable();
4284 clock = clock_lo | (clock_hi << 32ULL);
4285 break;
4286 default:
4287 amdgpu_gfx_off_ctrl(adev, false);
4288 mutex_lock(&adev->gfx.gpu_clock_mutex);
4289 if (amdgpu_ip_version(adev, GC_HWIP, 0) ==
4290 IP_VERSION(9, 0, 1) &&
4291 amdgpu_sriov_runtime(adev)) {
4292 clock = gfx_v9_0_kiq_read_clock(adev);
4293 } else {
4294 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4295 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
4296 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4297 }
4298 mutex_unlock(&adev->gfx.gpu_clock_mutex);
4299 amdgpu_gfx_off_ctrl(adev, true);
4300 break;
4301 }
4302 return clock;
4303}
4304
4305static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4306 uint32_t vmid,
4307 uint32_t gds_base, uint32_t gds_size,
4308 uint32_t gws_base, uint32_t gws_size,
4309 uint32_t oa_base, uint32_t oa_size)
4310{
4311 struct amdgpu_device *adev = ring->adev;
4312
4313 /* GDS Base */
4314 gfx_v9_0_write_data_to_reg(ring, 0, false,
4315 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
4316 gds_base);
4317
4318 /* GDS Size */
4319 gfx_v9_0_write_data_to_reg(ring, 0, false,
4320 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
4321 gds_size);
4322
4323 /* GWS */
4324 gfx_v9_0_write_data_to_reg(ring, 0, false,
4325 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
4326 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4327
4328 /* OA */
4329 gfx_v9_0_write_data_to_reg(ring, 0, false,
4330 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
4331 (1 << (oa_size + oa_base)) - (1 << oa_base));
4332}
4333
4334static const u32 vgpr_init_compute_shader[] =
4335{
4336 0xb07c0000, 0xbe8000ff,
4337 0x000000f8, 0xbf110800,
4338 0x7e000280, 0x7e020280,
4339 0x7e040280, 0x7e060280,
4340 0x7e080280, 0x7e0a0280,
4341 0x7e0c0280, 0x7e0e0280,
4342 0x80808800, 0xbe803200,
4343 0xbf84fff5, 0xbf9c0000,
4344 0xd28c0001, 0x0001007f,
4345 0xd28d0001, 0x0002027e,
4346 0x10020288, 0xb8810904,
4347 0xb7814000, 0xd1196a01,
4348 0x00000301, 0xbe800087,
4349 0xbefc00c1, 0xd89c4000,
4350 0x00020201, 0xd89cc080,
4351 0x00040401, 0x320202ff,
4352 0x00000800, 0x80808100,
4353 0xbf84fff8, 0x7e020280,
4354 0xbf810000, 0x00000000,
4355};
4356
4357static const u32 sgpr_init_compute_shader[] =
4358{
4359 0xb07c0000, 0xbe8000ff,
4360 0x0000005f, 0xbee50080,
4361 0xbe812c65, 0xbe822c65,
4362 0xbe832c65, 0xbe842c65,
4363 0xbe852c65, 0xb77c0005,
4364 0x80808500, 0xbf84fff8,
4365 0xbe800080, 0xbf810000,
4366};
4367
4368static const u32 vgpr_init_compute_shader_arcturus[] = {
4369 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080,
4370 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080,
4371 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080,
4372 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080,
4373 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080,
4374 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080,
4375 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080,
4376 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080,
4377 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080,
4378 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080,
4379 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080,
4380 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080,
4381 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080,
4382 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080,
4383 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080,
4384 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080,
4385 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080,
4386 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080,
4387 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080,
4388 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080,
4389 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080,
4390 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080,
4391 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080,
4392 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080,
4393 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080,
4394 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080,
4395 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080,
4396 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080,
4397 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080,
4398 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080,
4399 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080,
4400 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080,
4401 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080,
4402 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080,
4403 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080,
4404 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080,
4405 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080,
4406 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080,
4407 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080,
4408 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080,
4409 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080,
4410 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080,
4411 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080,
4412 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080,
4413 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080,
4414 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080,
4415 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080,
4416 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080,
4417 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080,
4418 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080,
4419 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080,
4420 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080,
4421 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080,
4422 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080,
4423 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080,
4424 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080,
4425 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080,
4426 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080,
4427 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080,
4428 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080,
4429 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080,
4430 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080,
4431 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080,
4432 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080,
4433 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080,
4434 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080,
4435 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080,
4436 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080,
4437 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080,
4438 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080,
4439 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080,
4440 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080,
4441 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080,
4442 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080,
4443 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080,
4444 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080,
4445 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080,
4446 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080,
4447 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080,
4448 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080,
4449 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080,
4450 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080,
4451 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080,
4452 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080,
4453 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080,
4454 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a,
4455 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280,
4456 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000,
4457 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904,
4458 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000,
4459 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a,
4460 0xbf84fff8, 0xbf810000,
4461};
4462
4463/* When below register arrays changed, please update gpr_reg_size,
4464 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds,
4465 to cover all gfx9 ASICs */
4466static const struct soc15_reg_entry vgpr_init_regs[] = {
4467 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4468 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4469 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4470 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4471 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f },
4472 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4473 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4474 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4475 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4476 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4477 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4478 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4479 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4480 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4481};
4482
4483static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = {
4484 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4485 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4486 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4487 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4488 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf },
4489 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4490 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4491 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4492 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4493 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4494 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4495 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4496 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4497 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4498};
4499
4500static const struct soc15_reg_entry sgpr1_init_regs[] = {
4501 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4502 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4503 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4504 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4505 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4506 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4507 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff },
4508 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff },
4509 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff },
4510 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff },
4511 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff },
4512 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff },
4513 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff },
4514 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff },
4515};
4516
4517static const struct soc15_reg_entry sgpr2_init_regs[] = {
4518 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4519 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4520 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4521 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4522 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4523 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4524 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 },
4525 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 },
4526 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 },
4527 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 },
4528 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 },
4529 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 },
4530 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 },
4531 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 },
4532};
4533
4534static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = {
4535 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1},
4536 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1},
4537 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1},
4538 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1},
4539 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1},
4540 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1},
4541 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1},
4542 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1},
4543 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1},
4544 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1},
4545 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1},
4546 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1},
4547 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1},
4548 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6},
4549 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16},
4550 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16},
4551 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16},
4552 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16},
4553 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16},
4554 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16},
4555 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16},
4556 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16},
4557 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6},
4558 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16},
4559 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16},
4560 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1},
4561 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1},
4562 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32},
4563 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32},
4564 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72},
4565 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16},
4566 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2},
4567 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6},
4568};
4569
4570static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4571{
4572 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4573 int i, r;
4574
4575 /* only support when RAS is enabled */
4576 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4577 return 0;
4578
4579 r = amdgpu_ring_alloc(ring, 7);
4580 if (r) {
4581 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",
4582 ring->name, r);
4583 return r;
4584 }
4585
4586 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000);
4587 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size);
4588
4589 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4590 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
4591 PACKET3_DMA_DATA_DST_SEL(1) |
4592 PACKET3_DMA_DATA_SRC_SEL(2) |
4593 PACKET3_DMA_DATA_ENGINE(0)));
4594 amdgpu_ring_write(ring, 0);
4595 amdgpu_ring_write(ring, 0);
4596 amdgpu_ring_write(ring, 0);
4597 amdgpu_ring_write(ring, 0);
4598 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
4599 adev->gds.gds_size);
4600
4601 amdgpu_ring_commit(ring);
4602
4603 for (i = 0; i < adev->usec_timeout; i++) {
4604 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4605 break;
4606 udelay(1);
4607 }
4608
4609 if (i >= adev->usec_timeout)
4610 r = -ETIMEDOUT;
4611
4612 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000);
4613
4614 return r;
4615}
4616
4617static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4618{
4619 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4620 struct amdgpu_ib ib;
4621 struct dma_fence *f = NULL;
4622 int r, i;
4623 unsigned total_size, vgpr_offset, sgpr_offset;
4624 u64 gpu_addr;
4625
4626 int compute_dim_x = adev->gfx.config.max_shader_engines *
4627 adev->gfx.config.max_cu_per_sh *
4628 adev->gfx.config.max_sh_per_se;
4629 int sgpr_work_group_size = 5;
4630 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6;
4631 int vgpr_init_shader_size;
4632 const u32 *vgpr_init_shader_ptr;
4633 const struct soc15_reg_entry *vgpr_init_regs_ptr;
4634
4635 /* only support when RAS is enabled */
4636 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4637 return 0;
4638
4639 /* bail if the compute ring is not ready */
4640 if (!ring->sched.ready)
4641 return 0;
4642
4643 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1)) {
4644 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
4645 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
4646 vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
4647 } else {
4648 vgpr_init_shader_ptr = vgpr_init_compute_shader;
4649 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader);
4650 vgpr_init_regs_ptr = vgpr_init_regs;
4651 }
4652
4653 total_size =
4654 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */
4655 total_size +=
4656 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */
4657 total_size +=
4658 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */
4659 total_size = ALIGN(total_size, 256);
4660 vgpr_offset = total_size;
4661 total_size += ALIGN(vgpr_init_shader_size, 256);
4662 sgpr_offset = total_size;
4663 total_size += sizeof(sgpr_init_compute_shader);
4664
4665 /* allocate an indirect buffer to put the commands in */
4666 memset(&ib, 0, sizeof(ib));
4667 r = amdgpu_ib_get(adev, NULL, total_size,
4668 AMDGPU_IB_POOL_DIRECT, &ib);
4669 if (r) {
4670 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
4671 return r;
4672 }
4673
4674 /* load the compute shaders */
4675 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++)
4676 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i];
4677
4678 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
4679 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4680
4681 /* init the ib length to 0 */
4682 ib.length_dw = 0;
4683
4684 /* VGPR */
4685 /* write the register state for the compute dispatch */
4686 for (i = 0; i < gpr_reg_size; i++) {
4687 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4688 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i])
4689 - PACKET3_SET_SH_REG_START;
4690 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value;
4691 }
4692 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4693 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4694 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4695 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4696 - PACKET3_SET_SH_REG_START;
4697 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4698 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4699
4700 /* write dispatch packet */
4701 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4702 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */
4703 ib.ptr[ib.length_dw++] = 1; /* y */
4704 ib.ptr[ib.length_dw++] = 1; /* z */
4705 ib.ptr[ib.length_dw++] =
4706 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4707
4708 /* write CS partial flush packet */
4709 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4710 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4711
4712 /* SGPR1 */
4713 /* write the register state for the compute dispatch */
4714 for (i = 0; i < gpr_reg_size; i++) {
4715 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4716 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i])
4717 - PACKET3_SET_SH_REG_START;
4718 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value;
4719 }
4720 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4721 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4722 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4723 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4724 - PACKET3_SET_SH_REG_START;
4725 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4726 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4727
4728 /* write dispatch packet */
4729 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4730 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4731 ib.ptr[ib.length_dw++] = 1; /* y */
4732 ib.ptr[ib.length_dw++] = 1; /* z */
4733 ib.ptr[ib.length_dw++] =
4734 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4735
4736 /* write CS partial flush packet */
4737 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4738 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4739
4740 /* SGPR2 */
4741 /* write the register state for the compute dispatch */
4742 for (i = 0; i < gpr_reg_size; i++) {
4743 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4744 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i])
4745 - PACKET3_SET_SH_REG_START;
4746 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value;
4747 }
4748 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4749 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4750 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4751 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4752 - PACKET3_SET_SH_REG_START;
4753 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4754 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4755
4756 /* write dispatch packet */
4757 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4758 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4759 ib.ptr[ib.length_dw++] = 1; /* y */
4760 ib.ptr[ib.length_dw++] = 1; /* z */
4761 ib.ptr[ib.length_dw++] =
4762 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4763
4764 /* write CS partial flush packet */
4765 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4766 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4767
4768 /* shedule the ib on the ring */
4769 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
4770 if (r) {
4771 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
4772 goto fail;
4773 }
4774
4775 /* wait for the GPU to finish processing the IB */
4776 r = dma_fence_wait(f, false);
4777 if (r) {
4778 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
4779 goto fail;
4780 }
4781
4782fail:
4783 amdgpu_ib_free(adev, &ib, NULL);
4784 dma_fence_put(f);
4785
4786 return r;
4787}
4788
4789static int gfx_v9_0_early_init(struct amdgpu_ip_block *ip_block)
4790{
4791 struct amdgpu_device *adev = ip_block->adev;
4792
4793 adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
4794
4795 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1) ||
4796 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2))
4797 adev->gfx.num_gfx_rings = 0;
4798 else
4799 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
4800 adev->gfx.xcc_mask = 1;
4801 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4802 AMDGPU_MAX_COMPUTE_RINGS);
4803 gfx_v9_0_set_kiq_pm4_funcs(adev);
4804 gfx_v9_0_set_ring_funcs(adev);
4805 gfx_v9_0_set_irq_funcs(adev);
4806 gfx_v9_0_set_gds_init(adev);
4807 gfx_v9_0_set_rlc_funcs(adev);
4808
4809 /* init rlcg reg access ctrl */
4810 gfx_v9_0_init_rlcg_reg_access_ctrl(adev);
4811
4812 return gfx_v9_0_init_microcode(adev);
4813}
4814
4815static int gfx_v9_0_ecc_late_init(struct amdgpu_ip_block *ip_block)
4816{
4817 struct amdgpu_device *adev = ip_block->adev;
4818 int r;
4819
4820 /*
4821 * Temp workaround to fix the issue that CP firmware fails to
4822 * update read pointer when CPDMA is writing clearing operation
4823 * to GDS in suspend/resume sequence on several cards. So just
4824 * limit this operation in cold boot sequence.
4825 */
4826 if ((!adev->in_suspend) &&
4827 (adev->gds.gds_size)) {
4828 r = gfx_v9_0_do_edc_gds_workarounds(adev);
4829 if (r)
4830 return r;
4831 }
4832
4833 /* requires IBs so do in late init after IB pool is initialized */
4834 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2))
4835 r = gfx_v9_4_2_do_edc_gpr_workarounds(adev);
4836 else
4837 r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4838
4839 if (r)
4840 return r;
4841
4842 if (adev->gfx.ras &&
4843 adev->gfx.ras->enable_watchdog_timer)
4844 adev->gfx.ras->enable_watchdog_timer(adev);
4845
4846 return 0;
4847}
4848
4849static int gfx_v9_0_late_init(struct amdgpu_ip_block *ip_block)
4850{
4851 struct amdgpu_device *adev = ip_block->adev;
4852 int r;
4853
4854 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4855 if (r)
4856 return r;
4857
4858 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4859 if (r)
4860 return r;
4861
4862 r = amdgpu_irq_get(adev, &adev->gfx.bad_op_irq, 0);
4863 if (r)
4864 return r;
4865
4866 r = gfx_v9_0_ecc_late_init(ip_block);
4867 if (r)
4868 return r;
4869
4870 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2))
4871 gfx_v9_4_2_debug_trap_config_init(adev,
4872 adev->vm_manager.first_kfd_vmid, AMDGPU_NUM_VMID);
4873 else
4874 gfx_v9_0_debug_trap_config_init(adev,
4875 adev->vm_manager.first_kfd_vmid, AMDGPU_NUM_VMID);
4876
4877 return 0;
4878}
4879
4880static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4881{
4882 uint32_t rlc_setting;
4883
4884 /* if RLC is not enabled, do nothing */
4885 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
4886 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
4887 return false;
4888
4889 return true;
4890}
4891
4892static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
4893{
4894 uint32_t data;
4895 unsigned i;
4896
4897 data = RLC_SAFE_MODE__CMD_MASK;
4898 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4899 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4900
4901 /* wait for RLC_SAFE_MODE */
4902 for (i = 0; i < adev->usec_timeout; i++) {
4903 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
4904 break;
4905 udelay(1);
4906 }
4907}
4908
4909static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
4910{
4911 uint32_t data;
4912
4913 data = RLC_SAFE_MODE__CMD_MASK;
4914 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4915}
4916
4917static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4918 bool enable)
4919{
4920 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4921
4922 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
4923 gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
4924 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4925 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
4926 } else {
4927 gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
4928 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4929 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
4930 }
4931
4932 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4933}
4934
4935static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4936 bool enable)
4937{
4938 /* TODO: double check if we need to perform under safe mode */
4939 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
4940
4941 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
4942 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
4943 else
4944 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
4945
4946 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
4947 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
4948 else
4949 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
4950
4951 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
4952}
4953
4954static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4955 bool enable)
4956{
4957 uint32_t data, def;
4958
4959 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4960
4961 /* It is disabled by HW by default */
4962 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4963 /* 1 - RLC_CGTT_MGCG_OVERRIDE */
4964 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4965
4966 if (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 2, 1))
4967 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4968
4969 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4970 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4971 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4972
4973 /* only for Vega10 & Raven1 */
4974 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4975
4976 if (def != data)
4977 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4978
4979 /* MGLS is a global flag to control all MGLS in GFX */
4980 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4981 /* 2 - RLC memory Light sleep */
4982 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4983 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4984 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4985 if (def != data)
4986 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4987 }
4988 /* 3 - CP memory Light sleep */
4989 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4990 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4991 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4992 if (def != data)
4993 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4994 }
4995 }
4996 } else {
4997 /* 1 - MGCG_OVERRIDE */
4998 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4999
5000 if (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 2, 1))
5001 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
5002
5003 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
5004 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
5005 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
5006 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
5007
5008 if (def != data)
5009 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
5010
5011 /* 2 - disable MGLS in RLC */
5012 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
5013 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5014 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5015 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
5016 }
5017
5018 /* 3 - disable MGLS in CP */
5019 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
5020 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5021 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5022 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
5023 }
5024 }
5025
5026 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5027}
5028
5029static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
5030 bool enable)
5031{
5032 uint32_t data, def;
5033
5034 if (!adev->gfx.num_gfx_rings)
5035 return;
5036
5037 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5038
5039 /* Enable 3D CGCG/CGLS */
5040 if (enable) {
5041 /* write cmd to clear cgcg/cgls ov */
5042 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
5043 /* unset CGCG override */
5044 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
5045 /* update CGCG and CGLS override bits */
5046 if (def != data)
5047 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
5048
5049 /* enable 3Dcgcg FSM(0x0000363f) */
5050 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
5051
5052 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
5053 data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5054 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
5055 else
5056 data = 0x0 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT;
5057
5058 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
5059 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
5060 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
5061 if (def != data)
5062 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
5063
5064 /* set IDLE_POLL_COUNT(0x00900100) */
5065 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
5066 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
5067 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
5068 if (def != data)
5069 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
5070 } else {
5071 /* Disable CGCG/CGLS */
5072 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
5073 /* disable cgcg, cgls should be disabled */
5074 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
5075 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
5076 /* disable cgcg and cgls in FSM */
5077 if (def != data)
5078 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
5079 }
5080
5081 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5082}
5083
5084static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5085 bool enable)
5086{
5087 uint32_t def, data;
5088
5089 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5090
5091 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5092 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
5093 /* unset CGCG override */
5094 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
5095 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
5096 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
5097 else
5098 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
5099 /* update CGCG and CGLS override bits */
5100 if (def != data)
5101 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
5102
5103 /* enable cgcg FSM(0x0000363F) */
5104 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
5105
5106 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1))
5107 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5108 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5109 else
5110 data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5111 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5112 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
5113 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
5114 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5115 if (def != data)
5116 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
5117
5118 /* set IDLE_POLL_COUNT(0x00900100) */
5119 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
5120 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
5121 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
5122 if (def != data)
5123 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
5124 } else {
5125 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
5126 /* reset CGCG/CGLS bits */
5127 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5128 /* disable cgcg and cgls in FSM */
5129 if (def != data)
5130 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
5131 }
5132
5133 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5134}
5135
5136static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5137 bool enable)
5138{
5139 if (enable) {
5140 /* CGCG/CGLS should be enabled after MGCG/MGLS
5141 * === MGCG + MGLS ===
5142 */
5143 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
5144 /* === CGCG /CGLS for GFX 3D Only === */
5145 gfx_v9_0_update_3d_clock_gating(adev, enable);
5146 /* === CGCG + CGLS === */
5147 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
5148 } else {
5149 /* CGCG/CGLS should be disabled before MGCG/MGLS
5150 * === CGCG + CGLS ===
5151 */
5152 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
5153 /* === CGCG /CGLS for GFX 3D Only === */
5154 gfx_v9_0_update_3d_clock_gating(adev, enable);
5155 /* === MGCG + MGLS === */
5156 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
5157 }
5158 return 0;
5159}
5160
5161static void gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device *adev,
5162 unsigned int vmid)
5163{
5164 u32 reg, data;
5165
5166 reg = SOC15_REG_OFFSET(GC, 0, mmRLC_SPM_MC_CNTL);
5167 if (amdgpu_sriov_is_pp_one_vf(adev))
5168 data = RREG32_NO_KIQ(reg);
5169 else
5170 data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL);
5171
5172 data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
5173 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
5174
5175 if (amdgpu_sriov_is_pp_one_vf(adev))
5176 WREG32_SOC15_NO_KIQ(GC, 0, mmRLC_SPM_MC_CNTL, data);
5177 else
5178 WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data);
5179}
5180
5181static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, unsigned int vmid)
5182{
5183 amdgpu_gfx_off_ctrl(adev, false);
5184
5185 gfx_v9_0_update_spm_vmid_internal(adev, vmid);
5186
5187 amdgpu_gfx_off_ctrl(adev, true);
5188}
5189
5190static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev,
5191 uint32_t offset,
5192 struct soc15_reg_rlcg *entries, int arr_size)
5193{
5194 int i;
5195 uint32_t reg;
5196
5197 if (!entries)
5198 return false;
5199
5200 for (i = 0; i < arr_size; i++) {
5201 const struct soc15_reg_rlcg *entry;
5202
5203 entry = &entries[i];
5204 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg;
5205 if (offset == reg)
5206 return true;
5207 }
5208
5209 return false;
5210}
5211
5212static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset)
5213{
5214 return gfx_v9_0_check_rlcg_range(adev, offset,
5215 (void *)rlcg_access_gc_9_0,
5216 ARRAY_SIZE(rlcg_access_gc_9_0));
5217}
5218
5219static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
5220 .is_rlc_enabled = gfx_v9_0_is_rlc_enabled,
5221 .set_safe_mode = gfx_v9_0_set_safe_mode,
5222 .unset_safe_mode = gfx_v9_0_unset_safe_mode,
5223 .init = gfx_v9_0_rlc_init,
5224 .get_csb_size = gfx_v9_0_get_csb_size,
5225 .get_csb_buffer = gfx_v9_0_get_csb_buffer,
5226 .get_cp_table_num = gfx_v9_0_cp_jump_table_num,
5227 .resume = gfx_v9_0_rlc_resume,
5228 .stop = gfx_v9_0_rlc_stop,
5229 .reset = gfx_v9_0_rlc_reset,
5230 .start = gfx_v9_0_rlc_start,
5231 .update_spm_vmid = gfx_v9_0_update_spm_vmid,
5232 .is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range,
5233};
5234
5235static int gfx_v9_0_set_powergating_state(void *handle,
5236 enum amd_powergating_state state)
5237{
5238 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5239 bool enable = (state == AMD_PG_STATE_GATE);
5240
5241 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
5242 case IP_VERSION(9, 2, 2):
5243 case IP_VERSION(9, 1, 0):
5244 case IP_VERSION(9, 3, 0):
5245 if (!enable)
5246 amdgpu_gfx_off_ctrl(adev, false);
5247
5248 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5249 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
5250 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
5251 } else {
5252 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
5253 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
5254 }
5255
5256 if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5257 gfx_v9_0_enable_cp_power_gating(adev, true);
5258 else
5259 gfx_v9_0_enable_cp_power_gating(adev, false);
5260
5261 /* update gfx cgpg state */
5262 gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
5263
5264 /* update mgcg state */
5265 gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
5266
5267 if (enable)
5268 amdgpu_gfx_off_ctrl(adev, true);
5269 break;
5270 case IP_VERSION(9, 2, 1):
5271 amdgpu_gfx_off_ctrl(adev, enable);
5272 break;
5273 default:
5274 break;
5275 }
5276
5277 return 0;
5278}
5279
5280static int gfx_v9_0_set_clockgating_state(void *handle,
5281 enum amd_clockgating_state state)
5282{
5283 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5284
5285 if (amdgpu_sriov_vf(adev))
5286 return 0;
5287
5288 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
5289 case IP_VERSION(9, 0, 1):
5290 case IP_VERSION(9, 2, 1):
5291 case IP_VERSION(9, 4, 0):
5292 case IP_VERSION(9, 2, 2):
5293 case IP_VERSION(9, 1, 0):
5294 case IP_VERSION(9, 4, 1):
5295 case IP_VERSION(9, 3, 0):
5296 case IP_VERSION(9, 4, 2):
5297 gfx_v9_0_update_gfx_clock_gating(adev,
5298 state == AMD_CG_STATE_GATE);
5299 break;
5300 default:
5301 break;
5302 }
5303 return 0;
5304}
5305
5306static void gfx_v9_0_get_clockgating_state(void *handle, u64 *flags)
5307{
5308 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5309 int data;
5310
5311 if (amdgpu_sriov_vf(adev))
5312 *flags = 0;
5313
5314 /* AMD_CG_SUPPORT_GFX_MGCG */
5315 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE));
5316 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5317 *flags |= AMD_CG_SUPPORT_GFX_MGCG;
5318
5319 /* AMD_CG_SUPPORT_GFX_CGCG */
5320 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL));
5321 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5322 *flags |= AMD_CG_SUPPORT_GFX_CGCG;
5323
5324 /* AMD_CG_SUPPORT_GFX_CGLS */
5325 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5326 *flags |= AMD_CG_SUPPORT_GFX_CGLS;
5327
5328 /* AMD_CG_SUPPORT_GFX_RLC_LS */
5329 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL));
5330 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5331 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5332
5333 /* AMD_CG_SUPPORT_GFX_CP_LS */
5334 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL));
5335 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5336 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5337
5338 if (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 1)) {
5339 /* AMD_CG_SUPPORT_GFX_3D_CGCG */
5340 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D));
5341 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5342 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5343
5344 /* AMD_CG_SUPPORT_GFX_3D_CGLS */
5345 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5346 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5347 }
5348}
5349
5350static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5351{
5352 return *ring->rptr_cpu_addr; /* gfx9 is 32bit rptr*/
5353}
5354
5355static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5356{
5357 struct amdgpu_device *adev = ring->adev;
5358 u64 wptr;
5359
5360 /* XXX check if swapping is necessary on BE */
5361 if (ring->use_doorbell) {
5362 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5363 } else {
5364 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
5365 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
5366 }
5367
5368 return wptr;
5369}
5370
5371static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5372{
5373 struct amdgpu_device *adev = ring->adev;
5374
5375 if (ring->use_doorbell) {
5376 /* XXX check if swapping is necessary on BE */
5377 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5378 WDOORBELL64(ring->doorbell_index, ring->wptr);
5379 } else {
5380 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
5381 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
5382 }
5383}
5384
5385static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5386{
5387 struct amdgpu_device *adev = ring->adev;
5388 u32 ref_and_mask, reg_mem_engine;
5389 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5390
5391 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5392 switch (ring->me) {
5393 case 1:
5394 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5395 break;
5396 case 2:
5397 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5398 break;
5399 default:
5400 return;
5401 }
5402 reg_mem_engine = 0;
5403 } else {
5404 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5405 reg_mem_engine = 1; /* pfp */
5406 }
5407
5408 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5409 adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5410 adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5411 ref_and_mask, ref_and_mask, 0x20);
5412}
5413
5414static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5415 struct amdgpu_job *job,
5416 struct amdgpu_ib *ib,
5417 uint32_t flags)
5418{
5419 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5420 u32 header, control = 0;
5421
5422 if (ib->flags & AMDGPU_IB_FLAG_CE)
5423 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
5424 else
5425 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5426
5427 control |= ib->length_dw | (vmid << 24);
5428
5429 if (ib->flags & AMDGPU_IB_FLAG_PREEMPT) {
5430 control |= INDIRECT_BUFFER_PRE_ENB(1);
5431
5432 if (flags & AMDGPU_IB_PREEMPTED)
5433 control |= INDIRECT_BUFFER_PRE_RESUME(1);
5434
5435 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
5436 gfx_v9_0_ring_emit_de_meta(ring,
5437 (!amdgpu_sriov_vf(ring->adev) &&
5438 flags & AMDGPU_IB_PREEMPTED) ?
5439 true : false,
5440 job->gds_size > 0 && job->gds_base != 0);
5441 }
5442
5443 amdgpu_ring_write(ring, header);
5444 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5445 amdgpu_ring_write(ring,
5446#ifdef __BIG_ENDIAN
5447 (2 << 0) |
5448#endif
5449 lower_32_bits(ib->gpu_addr));
5450 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5451 amdgpu_ring_ib_on_emit_cntl(ring);
5452 amdgpu_ring_write(ring, control);
5453}
5454
5455static void gfx_v9_0_ring_patch_cntl(struct amdgpu_ring *ring,
5456 unsigned offset)
5457{
5458 u32 control = ring->ring[offset];
5459
5460 control |= INDIRECT_BUFFER_PRE_RESUME(1);
5461 ring->ring[offset] = control;
5462}
5463
5464static void gfx_v9_0_ring_patch_ce_meta(struct amdgpu_ring *ring,
5465 unsigned offset)
5466{
5467 struct amdgpu_device *adev = ring->adev;
5468 void *ce_payload_cpu_addr;
5469 uint64_t payload_offset, payload_size;
5470
5471 payload_size = sizeof(struct v9_ce_ib_state);
5472
5473 if (ring->is_mes_queue) {
5474 payload_offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5475 gfx[0].gfx_meta_data) +
5476 offsetof(struct v9_gfx_meta_data, ce_payload);
5477 ce_payload_cpu_addr =
5478 amdgpu_mes_ctx_get_offs_cpu_addr(ring, payload_offset);
5479 } else {
5480 payload_offset = offsetof(struct v9_gfx_meta_data, ce_payload);
5481 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + payload_offset;
5482 }
5483
5484 if (offset + (payload_size >> 2) <= ring->buf_mask + 1) {
5485 memcpy((void *)&ring->ring[offset], ce_payload_cpu_addr, payload_size);
5486 } else {
5487 memcpy((void *)&ring->ring[offset], ce_payload_cpu_addr,
5488 (ring->buf_mask + 1 - offset) << 2);
5489 payload_size -= (ring->buf_mask + 1 - offset) << 2;
5490 memcpy((void *)&ring->ring[0],
5491 ce_payload_cpu_addr + ((ring->buf_mask + 1 - offset) << 2),
5492 payload_size);
5493 }
5494}
5495
5496static void gfx_v9_0_ring_patch_de_meta(struct amdgpu_ring *ring,
5497 unsigned offset)
5498{
5499 struct amdgpu_device *adev = ring->adev;
5500 void *de_payload_cpu_addr;
5501 uint64_t payload_offset, payload_size;
5502
5503 payload_size = sizeof(struct v9_de_ib_state);
5504
5505 if (ring->is_mes_queue) {
5506 payload_offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5507 gfx[0].gfx_meta_data) +
5508 offsetof(struct v9_gfx_meta_data, de_payload);
5509 de_payload_cpu_addr =
5510 amdgpu_mes_ctx_get_offs_cpu_addr(ring, payload_offset);
5511 } else {
5512 payload_offset = offsetof(struct v9_gfx_meta_data, de_payload);
5513 de_payload_cpu_addr = adev->virt.csa_cpu_addr + payload_offset;
5514 }
5515
5516 ((struct v9_de_ib_state *)de_payload_cpu_addr)->ib_completion_status =
5517 IB_COMPLETION_STATUS_PREEMPTED;
5518
5519 if (offset + (payload_size >> 2) <= ring->buf_mask + 1) {
5520 memcpy((void *)&ring->ring[offset], de_payload_cpu_addr, payload_size);
5521 } else {
5522 memcpy((void *)&ring->ring[offset], de_payload_cpu_addr,
5523 (ring->buf_mask + 1 - offset) << 2);
5524 payload_size -= (ring->buf_mask + 1 - offset) << 2;
5525 memcpy((void *)&ring->ring[0],
5526 de_payload_cpu_addr + ((ring->buf_mask + 1 - offset) << 2),
5527 payload_size);
5528 }
5529}
5530
5531static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5532 struct amdgpu_job *job,
5533 struct amdgpu_ib *ib,
5534 uint32_t flags)
5535{
5536 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5537 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5538
5539 /* Currently, there is a high possibility to get wave ID mismatch
5540 * between ME and GDS, leading to a hw deadlock, because ME generates
5541 * different wave IDs than the GDS expects. This situation happens
5542 * randomly when at least 5 compute pipes use GDS ordered append.
5543 * The wave IDs generated by ME are also wrong after suspend/resume.
5544 * Those are probably bugs somewhere else in the kernel driver.
5545 *
5546 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5547 * GDS to 0 for this ring (me/pipe).
5548 */
5549 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5550 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5551 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
5552 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5553 }
5554
5555 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5556 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5557 amdgpu_ring_write(ring,
5558#ifdef __BIG_ENDIAN
5559 (2 << 0) |
5560#endif
5561 lower_32_bits(ib->gpu_addr));
5562 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5563 amdgpu_ring_write(ring, control);
5564}
5565
5566static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5567 u64 seq, unsigned flags)
5568{
5569 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5570 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5571 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
5572 bool exec = flags & AMDGPU_FENCE_FLAG_EXEC;
5573 uint32_t dw2 = 0;
5574
5575 /* RELEASE_MEM - flush caches, send int */
5576 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5577
5578 if (writeback) {
5579 dw2 = EOP_TC_NC_ACTION_EN;
5580 } else {
5581 dw2 = EOP_TCL1_ACTION_EN | EOP_TC_ACTION_EN |
5582 EOP_TC_MD_ACTION_EN;
5583 }
5584 dw2 |= EOP_TC_WB_ACTION_EN | EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5585 EVENT_INDEX(5);
5586 if (exec)
5587 dw2 |= EOP_EXEC;
5588
5589 amdgpu_ring_write(ring, dw2);
5590 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
5591
5592 /*
5593 * the address should be Qword aligned if 64bit write, Dword
5594 * aligned if only send 32bit data low (discard data high)
5595 */
5596 if (write64bit)
5597 BUG_ON(addr & 0x7);
5598 else
5599 BUG_ON(addr & 0x3);
5600 amdgpu_ring_write(ring, lower_32_bits(addr));
5601 amdgpu_ring_write(ring, upper_32_bits(addr));
5602 amdgpu_ring_write(ring, lower_32_bits(seq));
5603 amdgpu_ring_write(ring, upper_32_bits(seq));
5604 amdgpu_ring_write(ring, 0);
5605}
5606
5607static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5608{
5609 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5610 uint32_t seq = ring->fence_drv.sync_seq;
5611 uint64_t addr = ring->fence_drv.gpu_addr;
5612
5613 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
5614 lower_32_bits(addr), upper_32_bits(addr),
5615 seq, 0xffffffff, 4);
5616}
5617
5618static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5619 unsigned vmid, uint64_t pd_addr)
5620{
5621 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5622
5623 /* compute doesn't have PFP */
5624 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5625 /* sync PFP to ME, otherwise we might get invalid PFP reads */
5626 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5627 amdgpu_ring_write(ring, 0x0);
5628 }
5629}
5630
5631static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5632{
5633 return *ring->rptr_cpu_addr; /* gfx9 hardware is 32bit rptr */
5634}
5635
5636static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5637{
5638 u64 wptr;
5639
5640 /* XXX check if swapping is necessary on BE */
5641 if (ring->use_doorbell)
5642 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5643 else
5644 BUG();
5645 return wptr;
5646}
5647
5648static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5649{
5650 struct amdgpu_device *adev = ring->adev;
5651
5652 /* XXX check if swapping is necessary on BE */
5653 if (ring->use_doorbell) {
5654 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5655 WDOORBELL64(ring->doorbell_index, ring->wptr);
5656 } else{
5657 BUG(); /* only DOORBELL method supported on gfx9 now */
5658 }
5659}
5660
5661static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5662 u64 seq, unsigned int flags)
5663{
5664 struct amdgpu_device *adev = ring->adev;
5665
5666 /* we only allocate 32bit for each seq wb address */
5667 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5668
5669 /* write fence seq to the "addr" */
5670 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5671 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5672 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5673 amdgpu_ring_write(ring, lower_32_bits(addr));
5674 amdgpu_ring_write(ring, upper_32_bits(addr));
5675 amdgpu_ring_write(ring, lower_32_bits(seq));
5676
5677 if (flags & AMDGPU_FENCE_FLAG_INT) {
5678 /* set register to trigger INT */
5679 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5680 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5681 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5682 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
5683 amdgpu_ring_write(ring, 0);
5684 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5685 }
5686}
5687
5688static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
5689{
5690 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
5691 amdgpu_ring_write(ring, 0);
5692}
5693
5694static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring, bool resume)
5695{
5696 struct amdgpu_device *adev = ring->adev;
5697 struct v9_ce_ib_state ce_payload = {0};
5698 uint64_t offset, ce_payload_gpu_addr;
5699 void *ce_payload_cpu_addr;
5700 int cnt;
5701
5702 cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
5703
5704 if (ring->is_mes_queue) {
5705 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5706 gfx[0].gfx_meta_data) +
5707 offsetof(struct v9_gfx_meta_data, ce_payload);
5708 ce_payload_gpu_addr =
5709 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5710 ce_payload_cpu_addr =
5711 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5712 } else {
5713 offset = offsetof(struct v9_gfx_meta_data, ce_payload);
5714 ce_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5715 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5716 }
5717
5718 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5719 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
5720 WRITE_DATA_DST_SEL(8) |
5721 WR_CONFIRM) |
5722 WRITE_DATA_CACHE_POLICY(0));
5723 amdgpu_ring_write(ring, lower_32_bits(ce_payload_gpu_addr));
5724 amdgpu_ring_write(ring, upper_32_bits(ce_payload_gpu_addr));
5725
5726 amdgpu_ring_ib_on_emit_ce(ring);
5727
5728 if (resume)
5729 amdgpu_ring_write_multiple(ring, ce_payload_cpu_addr,
5730 sizeof(ce_payload) >> 2);
5731 else
5732 amdgpu_ring_write_multiple(ring, (void *)&ce_payload,
5733 sizeof(ce_payload) >> 2);
5734}
5735
5736static int gfx_v9_0_ring_preempt_ib(struct amdgpu_ring *ring)
5737{
5738 int i, r = 0;
5739 struct amdgpu_device *adev = ring->adev;
5740 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
5741 struct amdgpu_ring *kiq_ring = &kiq->ring;
5742 unsigned long flags;
5743
5744 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
5745 return -EINVAL;
5746
5747 spin_lock_irqsave(&kiq->ring_lock, flags);
5748
5749 if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
5750 spin_unlock_irqrestore(&kiq->ring_lock, flags);
5751 return -ENOMEM;
5752 }
5753
5754 /* assert preemption condition */
5755 amdgpu_ring_set_preempt_cond_exec(ring, false);
5756
5757 ring->trail_seq += 1;
5758 amdgpu_ring_alloc(ring, 13);
5759 gfx_v9_0_ring_emit_fence(ring, ring->trail_fence_gpu_addr,
5760 ring->trail_seq, AMDGPU_FENCE_FLAG_EXEC | AMDGPU_FENCE_FLAG_INT);
5761
5762 /* assert IB preemption, emit the trailing fence */
5763 kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
5764 ring->trail_fence_gpu_addr,
5765 ring->trail_seq);
5766
5767 amdgpu_ring_commit(kiq_ring);
5768 spin_unlock_irqrestore(&kiq->ring_lock, flags);
5769
5770 /* poll the trailing fence */
5771 for (i = 0; i < adev->usec_timeout; i++) {
5772 if (ring->trail_seq ==
5773 le32_to_cpu(*ring->trail_fence_cpu_addr))
5774 break;
5775 udelay(1);
5776 }
5777
5778 if (i >= adev->usec_timeout) {
5779 r = -EINVAL;
5780 DRM_WARN("ring %d timeout to preempt ib\n", ring->idx);
5781 }
5782
5783 /*reset the CP_VMID_PREEMPT after trailing fence*/
5784 amdgpu_ring_emit_wreg(ring,
5785 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_PREEMPT),
5786 0x0);
5787 amdgpu_ring_commit(ring);
5788
5789 /* deassert preemption condition */
5790 amdgpu_ring_set_preempt_cond_exec(ring, true);
5791 return r;
5792}
5793
5794static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume, bool usegds)
5795{
5796 struct amdgpu_device *adev = ring->adev;
5797 struct v9_de_ib_state de_payload = {0};
5798 uint64_t offset, gds_addr, de_payload_gpu_addr;
5799 void *de_payload_cpu_addr;
5800 int cnt;
5801
5802 if (ring->is_mes_queue) {
5803 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5804 gfx[0].gfx_meta_data) +
5805 offsetof(struct v9_gfx_meta_data, de_payload);
5806 de_payload_gpu_addr =
5807 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5808 de_payload_cpu_addr =
5809 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5810
5811 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5812 gfx[0].gds_backup) +
5813 offsetof(struct v9_gfx_meta_data, de_payload);
5814 gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5815 } else {
5816 offset = offsetof(struct v9_gfx_meta_data, de_payload);
5817 de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5818 de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5819
5820 gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) +
5821 AMDGPU_CSA_SIZE - adev->gds.gds_size,
5822 PAGE_SIZE);
5823 }
5824
5825 if (usegds) {
5826 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5827 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5828 }
5829
5830 cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5831 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5832 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5833 WRITE_DATA_DST_SEL(8) |
5834 WR_CONFIRM) |
5835 WRITE_DATA_CACHE_POLICY(0));
5836 amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr));
5837 amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr));
5838
5839 amdgpu_ring_ib_on_emit_de(ring);
5840 if (resume)
5841 amdgpu_ring_write_multiple(ring, de_payload_cpu_addr,
5842 sizeof(de_payload) >> 2);
5843 else
5844 amdgpu_ring_write_multiple(ring, (void *)&de_payload,
5845 sizeof(de_payload) >> 2);
5846}
5847
5848static void gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5849 bool secure)
5850{
5851 uint32_t v = secure ? FRAME_TMZ : 0;
5852
5853 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5854 amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5855}
5856
5857static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
5858{
5859 uint32_t dw2 = 0;
5860
5861 gfx_v9_0_ring_emit_ce_meta(ring,
5862 (!amdgpu_sriov_vf(ring->adev) &&
5863 flags & AMDGPU_IB_PREEMPTED) ? true : false);
5864
5865 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5866 if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5867 /* set load_global_config & load_global_uconfig */
5868 dw2 |= 0x8001;
5869 /* set load_cs_sh_regs */
5870 dw2 |= 0x01000000;
5871 /* set load_per_context_state & load_gfx_sh_regs for GFX */
5872 dw2 |= 0x10002;
5873
5874 /* set load_ce_ram if preamble presented */
5875 if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
5876 dw2 |= 0x10000000;
5877 } else {
5878 /* still load_ce_ram if this is the first time preamble presented
5879 * although there is no context switch happens.
5880 */
5881 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
5882 dw2 |= 0x10000000;
5883 }
5884
5885 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5886 amdgpu_ring_write(ring, dw2);
5887 amdgpu_ring_write(ring, 0);
5888}
5889
5890static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
5891 uint64_t addr)
5892{
5893 unsigned ret;
5894 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5895 amdgpu_ring_write(ring, lower_32_bits(addr));
5896 amdgpu_ring_write(ring, upper_32_bits(addr));
5897 /* discard following DWs if *cond_exec_gpu_addr==0 */
5898 amdgpu_ring_write(ring, 0);
5899 ret = ring->wptr & ring->buf_mask;
5900 /* patch dummy value later */
5901 amdgpu_ring_write(ring, 0);
5902 return ret;
5903}
5904
5905static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5906 uint32_t reg_val_offs)
5907{
5908 struct amdgpu_device *adev = ring->adev;
5909
5910 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5911 amdgpu_ring_write(ring, 0 | /* src: register*/
5912 (5 << 8) | /* dst: memory */
5913 (1 << 20)); /* write confirm */
5914 amdgpu_ring_write(ring, reg);
5915 amdgpu_ring_write(ring, 0);
5916 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5917 reg_val_offs * 4));
5918 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5919 reg_val_offs * 4));
5920}
5921
5922static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5923 uint32_t val)
5924{
5925 uint32_t cmd = 0;
5926
5927 switch (ring->funcs->type) {
5928 case AMDGPU_RING_TYPE_GFX:
5929 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5930 break;
5931 case AMDGPU_RING_TYPE_KIQ:
5932 cmd = (1 << 16); /* no inc addr */
5933 break;
5934 default:
5935 cmd = WR_CONFIRM;
5936 break;
5937 }
5938 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5939 amdgpu_ring_write(ring, cmd);
5940 amdgpu_ring_write(ring, reg);
5941 amdgpu_ring_write(ring, 0);
5942 amdgpu_ring_write(ring, val);
5943}
5944
5945static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5946 uint32_t val, uint32_t mask)
5947{
5948 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5949}
5950
5951static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5952 uint32_t reg0, uint32_t reg1,
5953 uint32_t ref, uint32_t mask)
5954{
5955 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5956 struct amdgpu_device *adev = ring->adev;
5957 bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ?
5958 adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait;
5959
5960 if (fw_version_ok)
5961 gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5962 ref, mask, 0x20);
5963 else
5964 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
5965 ref, mask);
5966}
5967
5968static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
5969{
5970 struct amdgpu_device *adev = ring->adev;
5971 uint32_t value = 0;
5972
5973 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5974 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5975 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5976 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5977 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5978 WREG32_SOC15(GC, 0, mmSQ_CMD, value);
5979 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5980}
5981
5982static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5983 enum amdgpu_interrupt_state state)
5984{
5985 switch (state) {
5986 case AMDGPU_IRQ_STATE_DISABLE:
5987 case AMDGPU_IRQ_STATE_ENABLE:
5988 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5989 TIME_STAMP_INT_ENABLE,
5990 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5991 break;
5992 default:
5993 break;
5994 }
5995}
5996
5997static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5998 int me, int pipe,
5999 enum amdgpu_interrupt_state state)
6000{
6001 u32 mec_int_cntl, mec_int_cntl_reg;
6002
6003 /*
6004 * amdgpu controls only the first MEC. That's why this function only
6005 * handles the setting of interrupts for this specific MEC. All other
6006 * pipes' interrupts are set by amdkfd.
6007 */
6008
6009 if (me == 1) {
6010 switch (pipe) {
6011 case 0:
6012 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
6013 break;
6014 case 1:
6015 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
6016 break;
6017 case 2:
6018 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
6019 break;
6020 case 3:
6021 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
6022 break;
6023 default:
6024 DRM_DEBUG("invalid pipe %d\n", pipe);
6025 return;
6026 }
6027 } else {
6028 DRM_DEBUG("invalid me %d\n", me);
6029 return;
6030 }
6031
6032 switch (state) {
6033 case AMDGPU_IRQ_STATE_DISABLE:
6034 mec_int_cntl = RREG32_SOC15_IP(GC,mec_int_cntl_reg);
6035 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
6036 TIME_STAMP_INT_ENABLE, 0);
6037 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
6038 break;
6039 case AMDGPU_IRQ_STATE_ENABLE:
6040 mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
6041 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
6042 TIME_STAMP_INT_ENABLE, 1);
6043 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
6044 break;
6045 default:
6046 break;
6047 }
6048}
6049
6050static u32 gfx_v9_0_get_cpc_int_cntl(struct amdgpu_device *adev,
6051 int me, int pipe)
6052{
6053 /*
6054 * amdgpu controls only the first MEC. That's why this function only
6055 * handles the setting of interrupts for this specific MEC. All other
6056 * pipes' interrupts are set by amdkfd.
6057 */
6058 if (me != 1)
6059 return 0;
6060
6061 switch (pipe) {
6062 case 0:
6063 return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
6064 case 1:
6065 return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
6066 case 2:
6067 return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
6068 case 3:
6069 return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
6070 default:
6071 return 0;
6072 }
6073}
6074
6075static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6076 struct amdgpu_irq_src *source,
6077 unsigned type,
6078 enum amdgpu_interrupt_state state)
6079{
6080 u32 cp_int_cntl_reg, cp_int_cntl;
6081 int i, j;
6082
6083 switch (state) {
6084 case AMDGPU_IRQ_STATE_DISABLE:
6085 case AMDGPU_IRQ_STATE_ENABLE:
6086 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
6087 PRIV_REG_INT_ENABLE,
6088 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
6089 for (i = 0; i < adev->gfx.mec.num_mec; i++) {
6090 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
6091 /* MECs start at 1 */
6092 cp_int_cntl_reg = gfx_v9_0_get_cpc_int_cntl(adev, i + 1, j);
6093
6094 if (cp_int_cntl_reg) {
6095 cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
6096 cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_ME1_PIPE0_INT_CNTL,
6097 PRIV_REG_INT_ENABLE,
6098 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
6099 WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
6100 }
6101 }
6102 }
6103 break;
6104 default:
6105 break;
6106 }
6107
6108 return 0;
6109}
6110
6111static int gfx_v9_0_set_bad_op_fault_state(struct amdgpu_device *adev,
6112 struct amdgpu_irq_src *source,
6113 unsigned type,
6114 enum amdgpu_interrupt_state state)
6115{
6116 u32 cp_int_cntl_reg, cp_int_cntl;
6117 int i, j;
6118
6119 switch (state) {
6120 case AMDGPU_IRQ_STATE_DISABLE:
6121 case AMDGPU_IRQ_STATE_ENABLE:
6122 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
6123 OPCODE_ERROR_INT_ENABLE,
6124 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
6125 for (i = 0; i < adev->gfx.mec.num_mec; i++) {
6126 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
6127 /* MECs start at 1 */
6128 cp_int_cntl_reg = gfx_v9_0_get_cpc_int_cntl(adev, i + 1, j);
6129
6130 if (cp_int_cntl_reg) {
6131 cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
6132 cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_ME1_PIPE0_INT_CNTL,
6133 OPCODE_ERROR_INT_ENABLE,
6134 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
6135 WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
6136 }
6137 }
6138 }
6139 break;
6140 default:
6141 break;
6142 }
6143
6144 return 0;
6145}
6146
6147static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6148 struct amdgpu_irq_src *source,
6149 unsigned type,
6150 enum amdgpu_interrupt_state state)
6151{
6152 switch (state) {
6153 case AMDGPU_IRQ_STATE_DISABLE:
6154 case AMDGPU_IRQ_STATE_ENABLE:
6155 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
6156 PRIV_INSTR_INT_ENABLE,
6157 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
6158 break;
6159 default:
6160 break;
6161 }
6162
6163 return 0;
6164}
6165
6166#define ENABLE_ECC_ON_ME_PIPE(me, pipe) \
6167 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
6168 CP_ECC_ERROR_INT_ENABLE, 1)
6169
6170#define DISABLE_ECC_ON_ME_PIPE(me, pipe) \
6171 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
6172 CP_ECC_ERROR_INT_ENABLE, 0)
6173
6174static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
6175 struct amdgpu_irq_src *source,
6176 unsigned type,
6177 enum amdgpu_interrupt_state state)
6178{
6179 switch (state) {
6180 case AMDGPU_IRQ_STATE_DISABLE:
6181 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
6182 CP_ECC_ERROR_INT_ENABLE, 0);
6183 DISABLE_ECC_ON_ME_PIPE(1, 0);
6184 DISABLE_ECC_ON_ME_PIPE(1, 1);
6185 DISABLE_ECC_ON_ME_PIPE(1, 2);
6186 DISABLE_ECC_ON_ME_PIPE(1, 3);
6187 break;
6188
6189 case AMDGPU_IRQ_STATE_ENABLE:
6190 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
6191 CP_ECC_ERROR_INT_ENABLE, 1);
6192 ENABLE_ECC_ON_ME_PIPE(1, 0);
6193 ENABLE_ECC_ON_ME_PIPE(1, 1);
6194 ENABLE_ECC_ON_ME_PIPE(1, 2);
6195 ENABLE_ECC_ON_ME_PIPE(1, 3);
6196 break;
6197 default:
6198 break;
6199 }
6200
6201 return 0;
6202}
6203
6204
6205static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6206 struct amdgpu_irq_src *src,
6207 unsigned type,
6208 enum amdgpu_interrupt_state state)
6209{
6210 switch (type) {
6211 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6212 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
6213 break;
6214 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6215 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6216 break;
6217 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6218 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6219 break;
6220 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6221 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6222 break;
6223 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6224 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6225 break;
6226 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6227 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6228 break;
6229 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6230 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6231 break;
6232 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6233 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6234 break;
6235 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6236 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6237 break;
6238 default:
6239 break;
6240 }
6241 return 0;
6242}
6243
6244static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
6245 struct amdgpu_irq_src *source,
6246 struct amdgpu_iv_entry *entry)
6247{
6248 int i;
6249 u8 me_id, pipe_id, queue_id;
6250 struct amdgpu_ring *ring;
6251
6252 DRM_DEBUG("IH: CP EOP\n");
6253 me_id = (entry->ring_id & 0x0c) >> 2;
6254 pipe_id = (entry->ring_id & 0x03) >> 0;
6255 queue_id = (entry->ring_id & 0x70) >> 4;
6256
6257 switch (me_id) {
6258 case 0:
6259 if (adev->gfx.num_gfx_rings) {
6260 if (!adev->gfx.mcbp) {
6261 amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6262 } else if (!amdgpu_mcbp_handle_trailing_fence_irq(&adev->gfx.muxer)) {
6263 /* Fence signals are handled on the software rings*/
6264 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
6265 amdgpu_fence_process(&adev->gfx.sw_gfx_ring[i]);
6266 }
6267 }
6268 break;
6269 case 1:
6270 case 2:
6271 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6272 ring = &adev->gfx.compute_ring[i];
6273 /* Per-queue interrupt is supported for MEC starting from VI.
6274 * The interrupt can only be enabled/disabled per pipe instead of per queue.
6275 */
6276 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6277 amdgpu_fence_process(ring);
6278 }
6279 break;
6280 }
6281 return 0;
6282}
6283
6284static void gfx_v9_0_fault(struct amdgpu_device *adev,
6285 struct amdgpu_iv_entry *entry)
6286{
6287 u8 me_id, pipe_id, queue_id;
6288 struct amdgpu_ring *ring;
6289 int i;
6290
6291 me_id = (entry->ring_id & 0x0c) >> 2;
6292 pipe_id = (entry->ring_id & 0x03) >> 0;
6293 queue_id = (entry->ring_id & 0x70) >> 4;
6294
6295 switch (me_id) {
6296 case 0:
6297 drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6298 break;
6299 case 1:
6300 case 2:
6301 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6302 ring = &adev->gfx.compute_ring[i];
6303 if (ring->me == me_id && ring->pipe == pipe_id &&
6304 ring->queue == queue_id)
6305 drm_sched_fault(&ring->sched);
6306 }
6307 break;
6308 }
6309}
6310
6311static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
6312 struct amdgpu_irq_src *source,
6313 struct amdgpu_iv_entry *entry)
6314{
6315 DRM_ERROR("Illegal register access in command stream\n");
6316 gfx_v9_0_fault(adev, entry);
6317 return 0;
6318}
6319
6320static int gfx_v9_0_bad_op_irq(struct amdgpu_device *adev,
6321 struct amdgpu_irq_src *source,
6322 struct amdgpu_iv_entry *entry)
6323{
6324 DRM_ERROR("Illegal opcode in command stream\n");
6325 gfx_v9_0_fault(adev, entry);
6326 return 0;
6327}
6328
6329static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
6330 struct amdgpu_irq_src *source,
6331 struct amdgpu_iv_entry *entry)
6332{
6333 DRM_ERROR("Illegal instruction in command stream\n");
6334 gfx_v9_0_fault(adev, entry);
6335 return 0;
6336}
6337
6338
6339static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = {
6340 { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT),
6341 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
6342 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT)
6343 },
6344 { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT),
6345 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
6346 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT)
6347 },
6348 { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
6349 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1),
6350 0, 0
6351 },
6352 { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
6353 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2),
6354 0, 0
6355 },
6356 { "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT),
6357 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
6358 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT)
6359 },
6360 { "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
6361 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT),
6362 0, 0
6363 },
6364 { "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
6365 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT),
6366 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT)
6367 },
6368 { "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT),
6369 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT),
6370 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT)
6371 },
6372 { "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT),
6373 SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1),
6374 0, 0
6375 },
6376 { "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT),
6377 SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1),
6378 0, 0
6379 },
6380 { "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT),
6381 SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1),
6382 0, 0
6383 },
6384 { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6385 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
6386 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED)
6387 },
6388 { "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6389 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED),
6390 0, 0
6391 },
6392 { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6393 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
6394 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED)
6395 },
6396 { "GDS_OA_PHY_PHY_CMD_RAM_MEM",
6397 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6398 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
6399 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED)
6400 },
6401 { "GDS_OA_PHY_PHY_DATA_RAM_MEM",
6402 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6403 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED),
6404 0, 0
6405 },
6406 { "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM",
6407 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6408 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
6409 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED)
6410 },
6411 { "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM",
6412 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6413 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
6414 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED)
6415 },
6416 { "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM",
6417 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6418 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
6419 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED)
6420 },
6421 { "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM",
6422 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6423 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
6424 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED)
6425 },
6426 { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT),
6427 SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT),
6428 0, 0
6429 },
6430 { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6431 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
6432 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT)
6433 },
6434 { "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6435 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT),
6436 0, 0
6437 },
6438 { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6439 SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT),
6440 0, 0
6441 },
6442 { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6443 SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT),
6444 0, 0
6445 },
6446 { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6447 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT),
6448 0, 0
6449 },
6450 { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6451 SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT),
6452 0, 0
6453 },
6454 { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6455 SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT),
6456 0, 0
6457 },
6458 { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6459 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
6460 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT)
6461 },
6462 { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6463 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
6464 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT)
6465 },
6466 { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6467 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
6468 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT)
6469 },
6470 { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6471 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
6472 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT)
6473 },
6474 { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6475 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
6476 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT)
6477 },
6478 { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6479 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT),
6480 0, 0
6481 },
6482 { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6483 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT),
6484 0, 0
6485 },
6486 { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6487 SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT),
6488 0, 0
6489 },
6490 { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6491 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT),
6492 0, 0
6493 },
6494 { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6495 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT),
6496 0, 0
6497 },
6498 { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6499 SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT),
6500 0, 0
6501 },
6502 { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6503 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT),
6504 0, 0
6505 },
6506 { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6507 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT),
6508 0, 0
6509 },
6510 { "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6511 SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT),
6512 0, 0
6513 },
6514 { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6515 SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT),
6516 0, 0
6517 },
6518 { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6519 SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT),
6520 0, 0
6521 },
6522 { "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6523 SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT),
6524 0, 0
6525 },
6526 { "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6527 SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT),
6528 0, 0
6529 },
6530 { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT),
6531 SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT),
6532 0, 0
6533 },
6534 { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6535 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
6536 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT)
6537 },
6538 { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6539 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
6540 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT)
6541 },
6542 { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6543 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT),
6544 0, 0
6545 },
6546 { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6547 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
6548 0, 0
6549 },
6550 { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6551 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT),
6552 0, 0
6553 },
6554 { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6555 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
6556 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT)
6557 },
6558 { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6559 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
6560 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT)
6561 },
6562 { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6563 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
6564 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT)
6565 },
6566 { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6567 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
6568 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT)
6569 },
6570 { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6571 SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT),
6572 0, 0
6573 },
6574 { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6575 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
6576 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT)
6577 },
6578 { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6579 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
6580 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT)
6581 },
6582 { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6583 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
6584 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT)
6585 },
6586 { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6587 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
6588 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT)
6589 },
6590 { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6591 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
6592 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT)
6593 },
6594 { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6595 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
6596 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT)
6597 },
6598 { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6599 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
6600 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT)
6601 },
6602 { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6603 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
6604 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT)
6605 },
6606 { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6607 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
6608 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT)
6609 },
6610 { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6611 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
6612 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT)
6613 },
6614 { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6615 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
6616 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT)
6617 },
6618 { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6619 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
6620 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT)
6621 },
6622 { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6623 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
6624 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT)
6625 },
6626 { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6627 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
6628 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT)
6629 },
6630 { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6631 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
6632 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT)
6633 },
6634 { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6635 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
6636 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT)
6637 },
6638 { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6639 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
6640 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT)
6641 },
6642 { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6643 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT),
6644 0, 0
6645 },
6646 { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6647 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT),
6648 0, 0
6649 },
6650 { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6651 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT),
6652 0, 0
6653 },
6654 { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6655 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT),
6656 0, 0
6657 },
6658 { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6659 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT),
6660 0, 0
6661 },
6662 { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6663 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
6664 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT)
6665 },
6666 { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6667 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
6668 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT)
6669 },
6670 { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6671 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
6672 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT)
6673 },
6674 { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6675 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
6676 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT)
6677 },
6678 { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6679 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
6680 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT)
6681 },
6682 { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6683 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT),
6684 0, 0
6685 },
6686 { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6687 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT),
6688 0, 0
6689 },
6690 { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6691 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT),
6692 0, 0
6693 },
6694 { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6695 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT),
6696 0, 0
6697 },
6698 { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6699 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT),
6700 0, 0
6701 },
6702 { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6703 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
6704 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT)
6705 },
6706 { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6707 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
6708 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT)
6709 },
6710 { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6711 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
6712 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT)
6713 },
6714 { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6715 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
6716 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT)
6717 },
6718 { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6719 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
6720 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT)
6721 },
6722 { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6723 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
6724 0, 0
6725 },
6726 { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6727 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
6728 0, 0
6729 },
6730 { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6731 SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT),
6732 0, 0
6733 },
6734 { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6735 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
6736 0, 0
6737 },
6738 { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6739 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT),
6740 0, 0
6741 },
6742 { "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6743 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
6744 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT)
6745 },
6746 { "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6747 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
6748 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT)
6749 },
6750 { "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6751 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
6752 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT)
6753 },
6754 { "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6755 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
6756 0, 0
6757 },
6758 { "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6759 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
6760 0, 0
6761 },
6762 { "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6763 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
6764 0, 0
6765 },
6766 { "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6767 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
6768 0, 0
6769 },
6770 { "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6771 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
6772 0, 0
6773 },
6774 { "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6775 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
6776 0, 0
6777 }
6778};
6779
6780static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
6781 void *inject_if, uint32_t instance_mask)
6782{
6783 struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
6784 int ret;
6785 struct ta_ras_trigger_error_input block_info = { 0 };
6786
6787 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6788 return -EINVAL;
6789
6790 if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks))
6791 return -EINVAL;
6792
6793 if (!ras_gfx_subblocks[info->head.sub_block_index].name)
6794 return -EPERM;
6795
6796 if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type &
6797 info->head.type)) {
6798 DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n",
6799 ras_gfx_subblocks[info->head.sub_block_index].name,
6800 info->head.type);
6801 return -EPERM;
6802 }
6803
6804 if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type &
6805 info->head.type)) {
6806 DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n",
6807 ras_gfx_subblocks[info->head.sub_block_index].name,
6808 info->head.type);
6809 return -EPERM;
6810 }
6811
6812 block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
6813 block_info.sub_block_index =
6814 ras_gfx_subblocks[info->head.sub_block_index].ta_subblock;
6815 block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
6816 block_info.address = info->address;
6817 block_info.value = info->value;
6818
6819 mutex_lock(&adev->grbm_idx_mutex);
6820 ret = psp_ras_trigger_error(&adev->psp, &block_info, instance_mask);
6821 mutex_unlock(&adev->grbm_idx_mutex);
6822
6823 return ret;
6824}
6825
6826static const char * const vml2_mems[] = {
6827 "UTC_VML2_BANK_CACHE_0_BIGK_MEM0",
6828 "UTC_VML2_BANK_CACHE_0_BIGK_MEM1",
6829 "UTC_VML2_BANK_CACHE_0_4K_MEM0",
6830 "UTC_VML2_BANK_CACHE_0_4K_MEM1",
6831 "UTC_VML2_BANK_CACHE_1_BIGK_MEM0",
6832 "UTC_VML2_BANK_CACHE_1_BIGK_MEM1",
6833 "UTC_VML2_BANK_CACHE_1_4K_MEM0",
6834 "UTC_VML2_BANK_CACHE_1_4K_MEM1",
6835 "UTC_VML2_BANK_CACHE_2_BIGK_MEM0",
6836 "UTC_VML2_BANK_CACHE_2_BIGK_MEM1",
6837 "UTC_VML2_BANK_CACHE_2_4K_MEM0",
6838 "UTC_VML2_BANK_CACHE_2_4K_MEM1",
6839 "UTC_VML2_BANK_CACHE_3_BIGK_MEM0",
6840 "UTC_VML2_BANK_CACHE_3_BIGK_MEM1",
6841 "UTC_VML2_BANK_CACHE_3_4K_MEM0",
6842 "UTC_VML2_BANK_CACHE_3_4K_MEM1",
6843};
6844
6845static const char * const vml2_walker_mems[] = {
6846 "UTC_VML2_CACHE_PDE0_MEM0",
6847 "UTC_VML2_CACHE_PDE0_MEM1",
6848 "UTC_VML2_CACHE_PDE1_MEM0",
6849 "UTC_VML2_CACHE_PDE1_MEM1",
6850 "UTC_VML2_CACHE_PDE2_MEM0",
6851 "UTC_VML2_CACHE_PDE2_MEM1",
6852 "UTC_VML2_RDIF_LOG_FIFO",
6853};
6854
6855static const char * const atc_l2_cache_2m_mems[] = {
6856 "UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM",
6857 "UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM",
6858 "UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM",
6859 "UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM",
6860};
6861
6862static const char *atc_l2_cache_4k_mems[] = {
6863 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0",
6864 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1",
6865 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2",
6866 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3",
6867 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4",
6868 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5",
6869 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6",
6870 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7",
6871 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0",
6872 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1",
6873 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2",
6874 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3",
6875 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4",
6876 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5",
6877 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6",
6878 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7",
6879 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0",
6880 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1",
6881 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2",
6882 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3",
6883 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4",
6884 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5",
6885 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6",
6886 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7",
6887 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0",
6888 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1",
6889 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2",
6890 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3",
6891 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4",
6892 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5",
6893 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6",
6894 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7",
6895};
6896
6897static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev,
6898 struct ras_err_data *err_data)
6899{
6900 uint32_t i, data;
6901 uint32_t sec_count, ded_count;
6902
6903 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6904 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6905 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6906 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6907 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6908 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6909 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6910 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6911
6912 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6913 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6914 data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6915
6916 sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT);
6917 if (sec_count) {
6918 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6919 "SEC %d\n", i, vml2_mems[i], sec_count);
6920 err_data->ce_count += sec_count;
6921 }
6922
6923 ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT);
6924 if (ded_count) {
6925 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6926 "DED %d\n", i, vml2_mems[i], ded_count);
6927 err_data->ue_count += ded_count;
6928 }
6929 }
6930
6931 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6932 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6933 data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6934
6935 sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6936 SEC_COUNT);
6937 if (sec_count) {
6938 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6939 "SEC %d\n", i, vml2_walker_mems[i], sec_count);
6940 err_data->ce_count += sec_count;
6941 }
6942
6943 ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6944 DED_COUNT);
6945 if (ded_count) {
6946 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6947 "DED %d\n", i, vml2_walker_mems[i], ded_count);
6948 err_data->ue_count += ded_count;
6949 }
6950 }
6951
6952 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6953 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6954 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6955
6956 sec_count = (data & 0x00006000L) >> 0xd;
6957 if (sec_count) {
6958 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6959 "SEC %d\n", i, atc_l2_cache_2m_mems[i],
6960 sec_count);
6961 err_data->ce_count += sec_count;
6962 }
6963 }
6964
6965 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6966 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6967 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6968
6969 sec_count = (data & 0x00006000L) >> 0xd;
6970 if (sec_count) {
6971 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6972 "SEC %d\n", i, atc_l2_cache_4k_mems[i],
6973 sec_count);
6974 err_data->ce_count += sec_count;
6975 }
6976
6977 ded_count = (data & 0x00018000L) >> 0xf;
6978 if (ded_count) {
6979 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6980 "DED %d\n", i, atc_l2_cache_4k_mems[i],
6981 ded_count);
6982 err_data->ue_count += ded_count;
6983 }
6984 }
6985
6986 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6987 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6988 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6989 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6990
6991 return 0;
6992}
6993
6994static int gfx_v9_0_ras_error_count(struct amdgpu_device *adev,
6995 const struct soc15_reg_entry *reg,
6996 uint32_t se_id, uint32_t inst_id, uint32_t value,
6997 uint32_t *sec_count, uint32_t *ded_count)
6998{
6999 uint32_t i;
7000 uint32_t sec_cnt, ded_cnt;
7001
7002 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) {
7003 if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset ||
7004 gfx_v9_0_ras_fields[i].seg != reg->seg ||
7005 gfx_v9_0_ras_fields[i].inst != reg->inst)
7006 continue;
7007
7008 sec_cnt = (value &
7009 gfx_v9_0_ras_fields[i].sec_count_mask) >>
7010 gfx_v9_0_ras_fields[i].sec_count_shift;
7011 if (sec_cnt) {
7012 dev_info(adev->dev, "GFX SubBlock %s, "
7013 "Instance[%d][%d], SEC %d\n",
7014 gfx_v9_0_ras_fields[i].name,
7015 se_id, inst_id,
7016 sec_cnt);
7017 *sec_count += sec_cnt;
7018 }
7019
7020 ded_cnt = (value &
7021 gfx_v9_0_ras_fields[i].ded_count_mask) >>
7022 gfx_v9_0_ras_fields[i].ded_count_shift;
7023 if (ded_cnt) {
7024 dev_info(adev->dev, "GFX SubBlock %s, "
7025 "Instance[%d][%d], DED %d\n",
7026 gfx_v9_0_ras_fields[i].name,
7027 se_id, inst_id,
7028 ded_cnt);
7029 *ded_count += ded_cnt;
7030 }
7031 }
7032
7033 return 0;
7034}
7035
7036static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev)
7037{
7038 int i, j, k;
7039
7040 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
7041 return;
7042
7043 /* read back registers to clear the counters */
7044 mutex_lock(&adev->grbm_idx_mutex);
7045 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
7046 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
7047 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
7048 amdgpu_gfx_select_se_sh(adev, j, 0x0, k, 0);
7049 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
7050 }
7051 }
7052 }
7053 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
7054 mutex_unlock(&adev->grbm_idx_mutex);
7055
7056 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
7057 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
7058 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
7059 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
7060 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
7061 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
7062 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
7063 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
7064
7065 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
7066 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
7067 RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
7068 }
7069
7070 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
7071 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
7072 RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
7073 }
7074
7075 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
7076 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
7077 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
7078 }
7079
7080 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
7081 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
7082 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
7083 }
7084
7085 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
7086 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
7087 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
7088 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
7089}
7090
7091static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
7092 void *ras_error_status)
7093{
7094 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
7095 uint32_t sec_count = 0, ded_count = 0;
7096 uint32_t i, j, k;
7097 uint32_t reg_value;
7098
7099 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
7100 return;
7101
7102 err_data->ue_count = 0;
7103 err_data->ce_count = 0;
7104
7105 mutex_lock(&adev->grbm_idx_mutex);
7106
7107 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
7108 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
7109 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
7110 amdgpu_gfx_select_se_sh(adev, j, 0, k, 0);
7111 reg_value =
7112 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
7113 if (reg_value)
7114 gfx_v9_0_ras_error_count(adev,
7115 &gfx_v9_0_edc_counter_regs[i],
7116 j, k, reg_value,
7117 &sec_count, &ded_count);
7118 }
7119 }
7120 }
7121
7122 err_data->ce_count += sec_count;
7123 err_data->ue_count += ded_count;
7124
7125 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
7126 mutex_unlock(&adev->grbm_idx_mutex);
7127
7128 gfx_v9_0_query_utc_edc_status(adev, err_data);
7129}
7130
7131static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring)
7132{
7133 const unsigned int cp_coher_cntl =
7134 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) |
7135 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) |
7136 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) |
7137 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) |
7138 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1);
7139
7140 /* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */
7141 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
7142 amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */
7143 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */
7144 amdgpu_ring_write(ring, 0xffffff); /* CP_COHER_SIZE_HI */
7145 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
7146 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */
7147 amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
7148}
7149
7150static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
7151 uint32_t pipe, bool enable)
7152{
7153 struct amdgpu_device *adev = ring->adev;
7154 uint32_t val;
7155 uint32_t wcl_cs_reg;
7156
7157 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
7158 val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT;
7159
7160 switch (pipe) {
7161 case 0:
7162 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0);
7163 break;
7164 case 1:
7165 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1);
7166 break;
7167 case 2:
7168 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2);
7169 break;
7170 case 3:
7171 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3);
7172 break;
7173 default:
7174 DRM_DEBUG("invalid pipe %d\n", pipe);
7175 return;
7176 }
7177
7178 amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
7179
7180}
7181static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
7182{
7183 struct amdgpu_device *adev = ring->adev;
7184 uint32_t val;
7185 int i;
7186
7187
7188 /* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
7189 * number of gfx waves. Setting 5 bit will make sure gfx only gets
7190 * around 25% of gpu resources.
7191 */
7192 val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
7193 amdgpu_ring_emit_wreg(ring,
7194 SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX),
7195 val);
7196
7197 /* Restrict waves for normal/low priority compute queues as well
7198 * to get best QoS for high priority compute jobs.
7199 *
7200 * amdgpu controls only 1st ME(0-3 CS pipes).
7201 */
7202 for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
7203 if (i != ring->pipe)
7204 gfx_v9_0_emit_wave_limit_cs(ring, i, enable);
7205
7206 }
7207}
7208
7209static void gfx_v9_ring_insert_nop(struct amdgpu_ring *ring, uint32_t num_nop)
7210{
7211 /* Header itself is a NOP packet */
7212 if (num_nop == 1) {
7213 amdgpu_ring_write(ring, ring->funcs->nop);
7214 return;
7215 }
7216
7217 /* Max HW optimization till 0x3ffe, followed by remaining one NOP at a time*/
7218 amdgpu_ring_write(ring, PACKET3(PACKET3_NOP, min(num_nop - 2, 0x3ffe)));
7219
7220 /* Header is at index 0, followed by num_nops - 1 NOP packet's */
7221 amdgpu_ring_insert_nop(ring, num_nop - 1);
7222}
7223
7224static int gfx_v9_0_reset_kgq(struct amdgpu_ring *ring, unsigned int vmid)
7225{
7226 struct amdgpu_device *adev = ring->adev;
7227 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
7228 struct amdgpu_ring *kiq_ring = &kiq->ring;
7229 unsigned long flags;
7230 u32 tmp;
7231 int r;
7232
7233 if (amdgpu_sriov_vf(adev))
7234 return -EINVAL;
7235
7236 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
7237 return -EINVAL;
7238
7239 spin_lock_irqsave(&kiq->ring_lock, flags);
7240
7241 if (amdgpu_ring_alloc(kiq_ring, 5)) {
7242 spin_unlock_irqrestore(&kiq->ring_lock, flags);
7243 return -ENOMEM;
7244 }
7245
7246 tmp = REG_SET_FIELD(0, CP_VMID_RESET, RESET_REQUEST, 1 << vmid);
7247 gfx_v9_0_ring_emit_wreg(kiq_ring,
7248 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_RESET), tmp);
7249 amdgpu_ring_commit(kiq_ring);
7250
7251 spin_unlock_irqrestore(&kiq->ring_lock, flags);
7252
7253 r = amdgpu_ring_test_ring(kiq_ring);
7254 if (r)
7255 return r;
7256
7257 if (amdgpu_ring_alloc(ring, 7 + 7 + 5))
7258 return -ENOMEM;
7259 gfx_v9_0_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
7260 ring->fence_drv.sync_seq, AMDGPU_FENCE_FLAG_EXEC);
7261 gfx_v9_0_ring_emit_reg_wait(ring,
7262 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_RESET), 0, 0xffff);
7263 gfx_v9_0_ring_emit_wreg(ring,
7264 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_RESET), 0);
7265
7266 return amdgpu_ring_test_ring(ring);
7267}
7268
7269static int gfx_v9_0_reset_kcq(struct amdgpu_ring *ring,
7270 unsigned int vmid)
7271{
7272 struct amdgpu_device *adev = ring->adev;
7273 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
7274 struct amdgpu_ring *kiq_ring = &kiq->ring;
7275 unsigned long flags;
7276 int i, r;
7277
7278 if (amdgpu_sriov_vf(adev))
7279 return -EINVAL;
7280
7281 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
7282 return -EINVAL;
7283
7284 spin_lock_irqsave(&kiq->ring_lock, flags);
7285
7286 if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
7287 spin_unlock_irqrestore(&kiq->ring_lock, flags);
7288 return -ENOMEM;
7289 }
7290
7291 kiq->pmf->kiq_unmap_queues(kiq_ring, ring, RESET_QUEUES,
7292 0, 0);
7293 amdgpu_ring_commit(kiq_ring);
7294
7295 spin_unlock_irqrestore(&kiq->ring_lock, flags);
7296
7297 r = amdgpu_ring_test_ring(kiq_ring);
7298 if (r)
7299 return r;
7300
7301 /* make sure dequeue is complete*/
7302 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
7303 mutex_lock(&adev->srbm_mutex);
7304 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
7305 for (i = 0; i < adev->usec_timeout; i++) {
7306 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
7307 break;
7308 udelay(1);
7309 }
7310 if (i >= adev->usec_timeout)
7311 r = -ETIMEDOUT;
7312 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
7313 mutex_unlock(&adev->srbm_mutex);
7314 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
7315 if (r) {
7316 dev_err(adev->dev, "fail to wait on hqd deactive\n");
7317 return r;
7318 }
7319
7320 r = amdgpu_bo_reserve(ring->mqd_obj, false);
7321 if (unlikely(r != 0)){
7322 dev_err(adev->dev, "fail to resv mqd_obj\n");
7323 return r;
7324 }
7325 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
7326 if (!r) {
7327 r = gfx_v9_0_kcq_init_queue(ring, true);
7328 amdgpu_bo_kunmap(ring->mqd_obj);
7329 ring->mqd_ptr = NULL;
7330 }
7331 amdgpu_bo_unreserve(ring->mqd_obj);
7332 if (r) {
7333 dev_err(adev->dev, "fail to unresv mqd_obj\n");
7334 return r;
7335 }
7336 spin_lock_irqsave(&kiq->ring_lock, flags);
7337 r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size);
7338 if (r) {
7339 spin_unlock_irqrestore(&kiq->ring_lock, flags);
7340 return -ENOMEM;
7341 }
7342 kiq->pmf->kiq_map_queues(kiq_ring, ring);
7343 amdgpu_ring_commit(kiq_ring);
7344 spin_unlock_irqrestore(&kiq->ring_lock, flags);
7345 r = amdgpu_ring_test_ring(kiq_ring);
7346 if (r) {
7347 DRM_ERROR("fail to remap queue\n");
7348 return r;
7349 }
7350 return amdgpu_ring_test_ring(ring);
7351}
7352
7353static void gfx_v9_ip_print(struct amdgpu_ip_block *ip_block, struct drm_printer *p)
7354{
7355 struct amdgpu_device *adev = ip_block->adev;
7356 uint32_t i, j, k, reg, index = 0;
7357 uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9);
7358
7359 if (!adev->gfx.ip_dump_core)
7360 return;
7361
7362 for (i = 0; i < reg_count; i++)
7363 drm_printf(p, "%-50s \t 0x%08x\n",
7364 gc_reg_list_9[i].reg_name,
7365 adev->gfx.ip_dump_core[i]);
7366
7367 /* print compute queue registers for all instances */
7368 if (!adev->gfx.ip_dump_compute_queues)
7369 return;
7370
7371 reg_count = ARRAY_SIZE(gc_cp_reg_list_9);
7372 drm_printf(p, "\nnum_mec: %d num_pipe: %d num_queue: %d\n",
7373 adev->gfx.mec.num_mec,
7374 adev->gfx.mec.num_pipe_per_mec,
7375 adev->gfx.mec.num_queue_per_pipe);
7376
7377 for (i = 0; i < adev->gfx.mec.num_mec; i++) {
7378 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
7379 for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) {
7380 drm_printf(p, "\nmec %d, pipe %d, queue %d\n", i, j, k);
7381 for (reg = 0; reg < reg_count; reg++) {
7382 drm_printf(p, "%-50s \t 0x%08x\n",
7383 gc_cp_reg_list_9[reg].reg_name,
7384 adev->gfx.ip_dump_compute_queues[index + reg]);
7385 }
7386 index += reg_count;
7387 }
7388 }
7389 }
7390
7391}
7392
7393static void gfx_v9_ip_dump(struct amdgpu_ip_block *ip_block)
7394{
7395 struct amdgpu_device *adev = ip_block->adev;
7396 uint32_t i, j, k, reg, index = 0;
7397 uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9);
7398
7399 if (!adev->gfx.ip_dump_core || !adev->gfx.num_gfx_rings)
7400 return;
7401
7402 amdgpu_gfx_off_ctrl(adev, false);
7403 for (i = 0; i < reg_count; i++)
7404 adev->gfx.ip_dump_core[i] = RREG32(SOC15_REG_ENTRY_OFFSET(gc_reg_list_9[i]));
7405 amdgpu_gfx_off_ctrl(adev, true);
7406
7407 /* dump compute queue registers for all instances */
7408 if (!adev->gfx.ip_dump_compute_queues)
7409 return;
7410
7411 reg_count = ARRAY_SIZE(gc_cp_reg_list_9);
7412 amdgpu_gfx_off_ctrl(adev, false);
7413 mutex_lock(&adev->srbm_mutex);
7414 for (i = 0; i < adev->gfx.mec.num_mec; i++) {
7415 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
7416 for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) {
7417 /* ME0 is for GFX so start from 1 for CP */
7418 soc15_grbm_select(adev, 1 + i, j, k, 0, 0);
7419
7420 for (reg = 0; reg < reg_count; reg++) {
7421 adev->gfx.ip_dump_compute_queues[index + reg] =
7422 RREG32(SOC15_REG_ENTRY_OFFSET(
7423 gc_cp_reg_list_9[reg]));
7424 }
7425 index += reg_count;
7426 }
7427 }
7428 }
7429 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
7430 mutex_unlock(&adev->srbm_mutex);
7431 amdgpu_gfx_off_ctrl(adev, true);
7432
7433}
7434
7435static void gfx_v9_0_ring_emit_cleaner_shader(struct amdgpu_ring *ring)
7436{
7437 /* Emit the cleaner shader */
7438 amdgpu_ring_write(ring, PACKET3(PACKET3_RUN_CLEANER_SHADER, 0));
7439 amdgpu_ring_write(ring, 0); /* RESERVED field, programmed to zero */
7440}
7441
7442static void gfx_v9_0_ring_begin_use_compute(struct amdgpu_ring *ring)
7443{
7444 struct amdgpu_device *adev = ring->adev;
7445
7446 amdgpu_gfx_enforce_isolation_ring_begin_use(ring);
7447
7448 /* Raven and PCO APUs seem to have stability issues
7449 * with compute and gfxoff and gfx pg. Disable gfx pg during
7450 * submission and allow again afterwards.
7451 */
7452 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 1, 0))
7453 gfx_v9_0_set_powergating_state(adev, AMD_PG_STATE_UNGATE);
7454}
7455
7456static void gfx_v9_0_ring_end_use_compute(struct amdgpu_ring *ring)
7457{
7458 struct amdgpu_device *adev = ring->adev;
7459
7460 /* Raven and PCO APUs seem to have stability issues
7461 * with compute and gfxoff and gfx pg. Disable gfx pg during
7462 * submission and allow again afterwards.
7463 */
7464 if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 1, 0))
7465 gfx_v9_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
7466
7467 amdgpu_gfx_enforce_isolation_ring_end_use(ring);
7468}
7469
7470static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
7471 .name = "gfx_v9_0",
7472 .early_init = gfx_v9_0_early_init,
7473 .late_init = gfx_v9_0_late_init,
7474 .sw_init = gfx_v9_0_sw_init,
7475 .sw_fini = gfx_v9_0_sw_fini,
7476 .hw_init = gfx_v9_0_hw_init,
7477 .hw_fini = gfx_v9_0_hw_fini,
7478 .suspend = gfx_v9_0_suspend,
7479 .resume = gfx_v9_0_resume,
7480 .is_idle = gfx_v9_0_is_idle,
7481 .wait_for_idle = gfx_v9_0_wait_for_idle,
7482 .soft_reset = gfx_v9_0_soft_reset,
7483 .set_clockgating_state = gfx_v9_0_set_clockgating_state,
7484 .set_powergating_state = gfx_v9_0_set_powergating_state,
7485 .get_clockgating_state = gfx_v9_0_get_clockgating_state,
7486 .dump_ip_state = gfx_v9_ip_dump,
7487 .print_ip_state = gfx_v9_ip_print,
7488};
7489
7490static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
7491 .type = AMDGPU_RING_TYPE_GFX,
7492 .align_mask = 0xff,
7493 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
7494 .support_64bit_ptrs = true,
7495 .secure_submission_supported = true,
7496 .get_rptr = gfx_v9_0_ring_get_rptr_gfx,
7497 .get_wptr = gfx_v9_0_ring_get_wptr_gfx,
7498 .set_wptr = gfx_v9_0_ring_set_wptr_gfx,
7499 .emit_frame_size = /* totally 242 maximum if 16 IBs */
7500 5 + /* COND_EXEC */
7501 7 + /* PIPELINE_SYNC */
7502 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
7503 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
7504 2 + /* VM_FLUSH */
7505 8 + /* FENCE for VM_FLUSH */
7506 20 + /* GDS switch */
7507 4 + /* double SWITCH_BUFFER,
7508 the first COND_EXEC jump to the place just
7509 prior to this double SWITCH_BUFFER */
7510 5 + /* COND_EXEC */
7511 7 + /* HDP_flush */
7512 4 + /* VGT_flush */
7513 14 + /* CE_META */
7514 31 + /* DE_META */
7515 3 + /* CNTX_CTRL */
7516 5 + /* HDP_INVL */
7517 8 + 8 + /* FENCE x2 */
7518 2 + /* SWITCH_BUFFER */
7519 7 + /* gfx_v9_0_emit_mem_sync */
7520 2, /* gfx_v9_0_ring_emit_cleaner_shader */
7521 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
7522 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
7523 .emit_fence = gfx_v9_0_ring_emit_fence,
7524 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
7525 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
7526 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
7527 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
7528 .test_ring = gfx_v9_0_ring_test_ring,
7529 .insert_nop = gfx_v9_ring_insert_nop,
7530 .pad_ib = amdgpu_ring_generic_pad_ib,
7531 .emit_switch_buffer = gfx_v9_ring_emit_sb,
7532 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
7533 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
7534 .preempt_ib = gfx_v9_0_ring_preempt_ib,
7535 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
7536 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7537 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7538 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7539 .soft_recovery = gfx_v9_0_ring_soft_recovery,
7540 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
7541 .reset = gfx_v9_0_reset_kgq,
7542 .emit_cleaner_shader = gfx_v9_0_ring_emit_cleaner_shader,
7543 .begin_use = amdgpu_gfx_enforce_isolation_ring_begin_use,
7544 .end_use = amdgpu_gfx_enforce_isolation_ring_end_use,
7545};
7546
7547static const struct amdgpu_ring_funcs gfx_v9_0_sw_ring_funcs_gfx = {
7548 .type = AMDGPU_RING_TYPE_GFX,
7549 .align_mask = 0xff,
7550 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
7551 .support_64bit_ptrs = true,
7552 .secure_submission_supported = true,
7553 .get_rptr = amdgpu_sw_ring_get_rptr_gfx,
7554 .get_wptr = amdgpu_sw_ring_get_wptr_gfx,
7555 .set_wptr = amdgpu_sw_ring_set_wptr_gfx,
7556 .emit_frame_size = /* totally 242 maximum if 16 IBs */
7557 5 + /* COND_EXEC */
7558 7 + /* PIPELINE_SYNC */
7559 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
7560 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
7561 2 + /* VM_FLUSH */
7562 8 + /* FENCE for VM_FLUSH */
7563 20 + /* GDS switch */
7564 4 + /* double SWITCH_BUFFER,
7565 * the first COND_EXEC jump to the place just
7566 * prior to this double SWITCH_BUFFER
7567 */
7568 5 + /* COND_EXEC */
7569 7 + /* HDP_flush */
7570 4 + /* VGT_flush */
7571 14 + /* CE_META */
7572 31 + /* DE_META */
7573 3 + /* CNTX_CTRL */
7574 5 + /* HDP_INVL */
7575 8 + 8 + /* FENCE x2 */
7576 2 + /* SWITCH_BUFFER */
7577 7 + /* gfx_v9_0_emit_mem_sync */
7578 2, /* gfx_v9_0_ring_emit_cleaner_shader */
7579 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
7580 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
7581 .emit_fence = gfx_v9_0_ring_emit_fence,
7582 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
7583 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
7584 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
7585 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
7586 .test_ring = gfx_v9_0_ring_test_ring,
7587 .test_ib = gfx_v9_0_ring_test_ib,
7588 .insert_nop = gfx_v9_ring_insert_nop,
7589 .pad_ib = amdgpu_ring_generic_pad_ib,
7590 .emit_switch_buffer = gfx_v9_ring_emit_sb,
7591 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
7592 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
7593 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
7594 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7595 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7596 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7597 .soft_recovery = gfx_v9_0_ring_soft_recovery,
7598 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
7599 .patch_cntl = gfx_v9_0_ring_patch_cntl,
7600 .patch_de = gfx_v9_0_ring_patch_de_meta,
7601 .patch_ce = gfx_v9_0_ring_patch_ce_meta,
7602 .emit_cleaner_shader = gfx_v9_0_ring_emit_cleaner_shader,
7603 .begin_use = amdgpu_gfx_enforce_isolation_ring_begin_use,
7604 .end_use = amdgpu_gfx_enforce_isolation_ring_end_use,
7605};
7606
7607static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
7608 .type = AMDGPU_RING_TYPE_COMPUTE,
7609 .align_mask = 0xff,
7610 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
7611 .support_64bit_ptrs = true,
7612 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
7613 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
7614 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
7615 .emit_frame_size =
7616 20 + /* gfx_v9_0_ring_emit_gds_switch */
7617 7 + /* gfx_v9_0_ring_emit_hdp_flush */
7618 5 + /* hdp invalidate */
7619 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
7620 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
7621 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
7622 8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
7623 7 + /* gfx_v9_0_emit_mem_sync */
7624 5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
7625 15 + /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
7626 2, /* gfx_v9_0_ring_emit_cleaner_shader */
7627 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
7628 .emit_ib = gfx_v9_0_ring_emit_ib_compute,
7629 .emit_fence = gfx_v9_0_ring_emit_fence,
7630 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
7631 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
7632 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
7633 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
7634 .test_ring = gfx_v9_0_ring_test_ring,
7635 .test_ib = gfx_v9_0_ring_test_ib,
7636 .insert_nop = gfx_v9_ring_insert_nop,
7637 .pad_ib = amdgpu_ring_generic_pad_ib,
7638 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7639 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7640 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7641 .soft_recovery = gfx_v9_0_ring_soft_recovery,
7642 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
7643 .emit_wave_limit = gfx_v9_0_emit_wave_limit,
7644 .reset = gfx_v9_0_reset_kcq,
7645 .emit_cleaner_shader = gfx_v9_0_ring_emit_cleaner_shader,
7646 .begin_use = gfx_v9_0_ring_begin_use_compute,
7647 .end_use = gfx_v9_0_ring_end_use_compute,
7648};
7649
7650static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
7651 .type = AMDGPU_RING_TYPE_KIQ,
7652 .align_mask = 0xff,
7653 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
7654 .support_64bit_ptrs = true,
7655 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
7656 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
7657 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
7658 .emit_frame_size =
7659 20 + /* gfx_v9_0_ring_emit_gds_switch */
7660 7 + /* gfx_v9_0_ring_emit_hdp_flush */
7661 5 + /* hdp invalidate */
7662 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
7663 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
7664 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
7665 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7666 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
7667 .emit_fence = gfx_v9_0_ring_emit_fence_kiq,
7668 .test_ring = gfx_v9_0_ring_test_ring,
7669 .insert_nop = amdgpu_ring_insert_nop,
7670 .pad_ib = amdgpu_ring_generic_pad_ib,
7671 .emit_rreg = gfx_v9_0_ring_emit_rreg,
7672 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7673 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7674 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7675};
7676
7677static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
7678{
7679 int i;
7680
7681 adev->gfx.kiq[0].ring.funcs = &gfx_v9_0_ring_funcs_kiq;
7682
7683 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7684 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
7685
7686 if (adev->gfx.mcbp && adev->gfx.num_gfx_rings) {
7687 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
7688 adev->gfx.sw_gfx_ring[i].funcs = &gfx_v9_0_sw_ring_funcs_gfx;
7689 }
7690
7691 for (i = 0; i < adev->gfx.num_compute_rings; i++)
7692 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
7693}
7694
7695static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
7696 .set = gfx_v9_0_set_eop_interrupt_state,
7697 .process = gfx_v9_0_eop_irq,
7698};
7699
7700static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
7701 .set = gfx_v9_0_set_priv_reg_fault_state,
7702 .process = gfx_v9_0_priv_reg_irq,
7703};
7704
7705static const struct amdgpu_irq_src_funcs gfx_v9_0_bad_op_irq_funcs = {
7706 .set = gfx_v9_0_set_bad_op_fault_state,
7707 .process = gfx_v9_0_bad_op_irq,
7708};
7709
7710static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
7711 .set = gfx_v9_0_set_priv_inst_fault_state,
7712 .process = gfx_v9_0_priv_inst_irq,
7713};
7714
7715static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
7716 .set = gfx_v9_0_set_cp_ecc_error_state,
7717 .process = amdgpu_gfx_cp_ecc_error_irq,
7718};
7719
7720
7721static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
7722{
7723 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7724 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
7725
7726 adev->gfx.priv_reg_irq.num_types = 1;
7727 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
7728
7729 adev->gfx.bad_op_irq.num_types = 1;
7730 adev->gfx.bad_op_irq.funcs = &gfx_v9_0_bad_op_irq_funcs;
7731
7732 adev->gfx.priv_inst_irq.num_types = 1;
7733 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
7734
7735 adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
7736 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
7737}
7738
7739static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
7740{
7741 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
7742 case IP_VERSION(9, 0, 1):
7743 case IP_VERSION(9, 2, 1):
7744 case IP_VERSION(9, 4, 0):
7745 case IP_VERSION(9, 2, 2):
7746 case IP_VERSION(9, 1, 0):
7747 case IP_VERSION(9, 4, 1):
7748 case IP_VERSION(9, 3, 0):
7749 case IP_VERSION(9, 4, 2):
7750 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
7751 break;
7752 default:
7753 break;
7754 }
7755}
7756
7757static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
7758{
7759 /* init asci gds info */
7760 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
7761 case IP_VERSION(9, 0, 1):
7762 case IP_VERSION(9, 2, 1):
7763 case IP_VERSION(9, 4, 0):
7764 adev->gds.gds_size = 0x10000;
7765 break;
7766 case IP_VERSION(9, 2, 2):
7767 case IP_VERSION(9, 1, 0):
7768 case IP_VERSION(9, 4, 1):
7769 adev->gds.gds_size = 0x1000;
7770 break;
7771 case IP_VERSION(9, 4, 2):
7772 /* aldebaran removed all the GDS internal memory,
7773 * only support GWS opcode in kernel, like barrier
7774 * semaphore.etc */
7775 adev->gds.gds_size = 0;
7776 break;
7777 default:
7778 adev->gds.gds_size = 0x10000;
7779 break;
7780 }
7781
7782 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
7783 case IP_VERSION(9, 0, 1):
7784 case IP_VERSION(9, 4, 0):
7785 adev->gds.gds_compute_max_wave_id = 0x7ff;
7786 break;
7787 case IP_VERSION(9, 2, 1):
7788 adev->gds.gds_compute_max_wave_id = 0x27f;
7789 break;
7790 case IP_VERSION(9, 2, 2):
7791 case IP_VERSION(9, 1, 0):
7792 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
7793 adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */
7794 else
7795 adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */
7796 break;
7797 case IP_VERSION(9, 4, 1):
7798 adev->gds.gds_compute_max_wave_id = 0xfff;
7799 break;
7800 case IP_VERSION(9, 4, 2):
7801 /* deprecated for Aldebaran, no usage at all */
7802 adev->gds.gds_compute_max_wave_id = 0;
7803 break;
7804 default:
7805 /* this really depends on the chip */
7806 adev->gds.gds_compute_max_wave_id = 0x7ff;
7807 break;
7808 }
7809
7810 adev->gds.gws_size = 64;
7811 adev->gds.oa_size = 16;
7812}
7813
7814static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7815 u32 bitmap)
7816{
7817 u32 data;
7818
7819 if (!bitmap)
7820 return;
7821
7822 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7823 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7824
7825 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
7826}
7827
7828static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7829{
7830 u32 data, mask;
7831
7832 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
7833 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
7834
7835 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7836 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7837
7838 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7839
7840 return (~data) & mask;
7841}
7842
7843static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
7844 struct amdgpu_cu_info *cu_info)
7845{
7846 int i, j, k, counter, active_cu_number = 0;
7847 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7848 unsigned disable_masks[4 * 4];
7849
7850 if (!adev || !cu_info)
7851 return -EINVAL;
7852
7853 /*
7854 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
7855 */
7856 if (adev->gfx.config.max_shader_engines *
7857 adev->gfx.config.max_sh_per_se > 16)
7858 return -EINVAL;
7859
7860 amdgpu_gfx_parse_disable_cu(disable_masks,
7861 adev->gfx.config.max_shader_engines,
7862 adev->gfx.config.max_sh_per_se);
7863
7864 mutex_lock(&adev->grbm_idx_mutex);
7865 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7866 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7867 mask = 1;
7868 ao_bitmap = 0;
7869 counter = 0;
7870 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
7871 gfx_v9_0_set_user_cu_inactive_bitmap(
7872 adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]);
7873 bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
7874
7875 /*
7876 * The bitmap(and ao_cu_bitmap) in cu_info structure is
7877 * 4x4 size array, and it's usually suitable for Vega
7878 * ASICs which has 4*2 SE/SH layout.
7879 * But for Arcturus, SE/SH layout is changed to 8*1.
7880 * To mostly reduce the impact, we make it compatible
7881 * with current bitmap array as below:
7882 * SE4,SH0 --> bitmap[0][1]
7883 * SE5,SH0 --> bitmap[1][1]
7884 * SE6,SH0 --> bitmap[2][1]
7885 * SE7,SH0 --> bitmap[3][1]
7886 */
7887 cu_info->bitmap[0][i % 4][j + i / 4] = bitmap;
7888
7889 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7890 if (bitmap & mask) {
7891 if (counter < adev->gfx.config.max_cu_per_sh)
7892 ao_bitmap |= mask;
7893 counter ++;
7894 }
7895 mask <<= 1;
7896 }
7897 active_cu_number += counter;
7898 if (i < 2 && j < 2)
7899 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7900 cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap;
7901 }
7902 }
7903 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
7904 mutex_unlock(&adev->grbm_idx_mutex);
7905
7906 cu_info->number = active_cu_number;
7907 cu_info->ao_cu_mask = ao_cu_mask;
7908 cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7909
7910 return 0;
7911}
7912
7913const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
7914{
7915 .type = AMD_IP_BLOCK_TYPE_GFX,
7916 .major = 9,
7917 .minor = 0,
7918 .rev = 0,
7919 .funcs = &gfx_v9_0_ip_funcs,
7920};