FILE
gfx_v9_4_3.c
File preview with syntax highlighting.
linux-6.19.14/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
1/* 2 * Copyright 2022 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23#include <linux/firmware.h> 24 25#include "amdgpu.h" 26#include "amdgpu_gfx.h" 27#include "soc15.h" 28#include "soc15d.h" 29#include "soc15_common.h" 30#include "vega10_enum.h" 31 32#include "v9_structs.h" 33 34#include "ivsrcid/gfx/irqsrcs_gfx_9_0.h" 35 36#include "gc/gc_9_4_3_offset.h" 37#include "gc/gc_9_4_3_sh_mask.h" 38 39#include "gfx_v9_4_3.h" 40#include "gfx_v9_4_3_cleaner_shader.h" 41#include "amdgpu_xcp.h" 42#include "amdgpu_aca.h" 43 44MODULE_FIRMWARE("amdgpu/gc_9_4_3_mec.bin"); 45MODULE_FIRMWARE("amdgpu/gc_9_4_4_mec.bin"); 46MODULE_FIRMWARE("amdgpu/gc_9_5_0_mec.bin"); 47MODULE_FIRMWARE("amdgpu/gc_9_4_3_rlc.bin"); 48MODULE_FIRMWARE("amdgpu/gc_9_4_4_rlc.bin"); 49MODULE_FIRMWARE("amdgpu/gc_9_5_0_rlc.bin"); 50MODULE_FIRMWARE("amdgpu/gc_9_4_3_sjt_mec.bin"); 51MODULE_FIRMWARE("amdgpu/gc_9_4_4_sjt_mec.bin"); 52 53#define GFX9_MEC_HPD_SIZE 4096 54#define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L 55 56#define GOLDEN_GB_ADDR_CONFIG 0x2a114042 57#define CP_HQD_PERSISTENT_STATE_DEFAULT 0xbe05301 58 59#define XCC_REG_RANGE_0_LOW 0x2000 /* XCC gfxdec0 lower Bound */ 60#define XCC_REG_RANGE_0_HIGH 0x3400 /* XCC gfxdec0 upper Bound */ 61#define XCC_REG_RANGE_1_LOW 0xA000 /* XCC gfxdec1 lower Bound */ 62#define XCC_REG_RANGE_1_HIGH 0x10000 /* XCC gfxdec1 upper Bound */ 63 64#define NORMALIZE_XCC_REG_OFFSET(offset) \ 65 (offset & 0xFFFF) 66 67static const struct amdgpu_hwip_reg_entry gc_reg_list_9_4_3[] = { 68 SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS), 69 SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS2), 70 SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT1), 71 SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT2), 72 SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_STALLED_STAT1), 73 SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_STALLED_STAT1), 74 SOC15_REG_ENTRY_STR(GC, 0, regCP_BUSY_STAT), 75 SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_BUSY_STAT), 76 SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_BUSY_STAT), 77 SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_STATUS), 78 SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_ERROR), 79 SOC15_REG_ENTRY_STR(GC, 0, regCPF_UTCL1_STATUS), 80 SOC15_REG_ENTRY_STR(GC, 0, regCPC_UTCL1_STATUS), 81 SOC15_REG_ENTRY_STR(GC, 0, regCPG_UTCL1_STATUS), 82 SOC15_REG_ENTRY_STR(GC, 0, regGDS_PROTECTION_FAULT), 83 SOC15_REG_ENTRY_STR(GC, 0, regGDS_VM_PROTECTION_FAULT), 84 SOC15_REG_ENTRY_STR(GC, 0, regRLC_UTCL1_STATUS), 85 SOC15_REG_ENTRY_STR(GC, 0, regRMI_UTCL1_STATUS), 86 SOC15_REG_ENTRY_STR(GC, 0, regSQC_DCACHE_UTCL1_STATUS), 87 SOC15_REG_ENTRY_STR(GC, 0, regSQC_ICACHE_UTCL1_STATUS), 88 SOC15_REG_ENTRY_STR(GC, 0, regSQ_UTCL1_STATUS), 89 SOC15_REG_ENTRY_STR(GC, 0, regTCP_UTCL1_STATUS), 90 SOC15_REG_ENTRY_STR(GC, 0, regWD_UTCL1_STATUS), 91 SOC15_REG_ENTRY_STR(GC, 0, regVM_L2_PROTECTION_FAULT_CNTL), 92 SOC15_REG_ENTRY_STR(GC, 0, regVM_L2_PROTECTION_FAULT_STATUS), 93 SOC15_REG_ENTRY_STR(GC, 0, regCP_DEBUG), 94 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_CNTL), 95 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC1_INSTR_PNTR), 96 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC2_INSTR_PNTR), 97 SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_STATUS), 98 SOC15_REG_ENTRY_STR(GC, 0, regRLC_STAT), 99 SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_COMMAND), 100 SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_MESSAGE), 101 SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_ARGUMENT_1), 102 SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_ARGUMENT_2), 103 SOC15_REG_ENTRY_STR(GC, 0, regSMU_RLC_RESPONSE), 104 SOC15_REG_ENTRY_STR(GC, 0, regRLC_SAFE_MODE), 105 SOC15_REG_ENTRY_STR(GC, 0, regRLC_SMU_SAFE_MODE), 106 SOC15_REG_ENTRY_STR(GC, 0, regRLC_INT_STAT), 107 SOC15_REG_ENTRY_STR(GC, 0, regRLC_GPM_GENERAL_6), 108 /* SE status registers */ 109 SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE0), 110 SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE1), 111 SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE2), 112 SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE3) 113}; 114 115static const struct amdgpu_hwip_reg_entry gc_cp_reg_list_9_4_3[] = { 116 /* compute queue registers */ 117 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_VMID), 118 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_ACTIVE), 119 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PERSISTENT_STATE), 120 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PIPE_PRIORITY), 121 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_QUEUE_PRIORITY), 122 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_QUANTUM), 123 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_BASE), 124 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_BASE_HI), 125 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_RPTR), 126 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR), 127 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI), 128 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL), 129 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_CONTROL), 130 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_BASE_ADDR), 131 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_BASE_ADDR_HI), 132 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_RPTR), 133 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_CONTROL), 134 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_DEQUEUE_REQUEST), 135 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_BASE_ADDR), 136 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI), 137 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_CONTROL), 138 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_RPTR), 139 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_WPTR), 140 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_EVENTS), 141 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_BASE_ADDR_LO), 142 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_BASE_ADDR_HI), 143 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_CONTROL), 144 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CNTL_STACK_OFFSET), 145 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CNTL_STACK_SIZE), 146 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_WG_STATE_OFFSET), 147 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_SIZE), 148 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_GDS_RESOURCE_STATE), 149 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_ERROR), 150 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_WPTR_MEM), 151 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_LO), 152 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_HI), 153 SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_GFX_STATUS), 154 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 155 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 156 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 157 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 158 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 159 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 160 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 161 SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP), 162}; 163 164struct amdgpu_gfx_ras gfx_v9_4_3_ras; 165 166static void gfx_v9_4_3_set_ring_funcs(struct amdgpu_device *adev); 167static void gfx_v9_4_3_set_irq_funcs(struct amdgpu_device *adev); 168static void gfx_v9_4_3_set_gds_init(struct amdgpu_device *adev); 169static void gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device *adev); 170static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev, 171 struct amdgpu_cu_info *cu_info); 172static void gfx_v9_4_3_xcc_set_safe_mode(struct amdgpu_device *adev, int xcc_id); 173static void gfx_v9_4_3_xcc_unset_safe_mode(struct amdgpu_device *adev, int xcc_id); 174 175static void gfx_v9_4_3_kiq_set_resources(struct amdgpu_ring *kiq_ring, 176 uint64_t queue_mask) 177{ 178 struct amdgpu_device *adev = kiq_ring->adev; 179 u64 shader_mc_addr; 180 181 /* Cleaner shader MC address */ 182 shader_mc_addr = adev->gfx.cleaner_shader_gpu_addr >> 8; 183 184 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6)); 185 amdgpu_ring_write(kiq_ring, 186 PACKET3_SET_RESOURCES_VMID_MASK(0) | 187 /* vmid_mask:0* queue_type:0 (KIQ) */ 188 PACKET3_SET_RESOURCES_QUEUE_TYPE(0)); 189 amdgpu_ring_write(kiq_ring, 190 lower_32_bits(queue_mask)); /* queue mask lo */ 191 amdgpu_ring_write(kiq_ring, 192 upper_32_bits(queue_mask)); /* queue mask hi */ 193 amdgpu_ring_write(kiq_ring, lower_32_bits(shader_mc_addr)); /* cleaner shader addr lo */ 194 amdgpu_ring_write(kiq_ring, upper_32_bits(shader_mc_addr)); /* cleaner shader addr hi */ 195 amdgpu_ring_write(kiq_ring, 0); /* oac mask */ 196 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */ 197} 198 199static void gfx_v9_4_3_kiq_map_queues(struct amdgpu_ring *kiq_ring, 200 struct amdgpu_ring *ring) 201{ 202 struct amdgpu_device *adev = kiq_ring->adev; 203 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj); 204 uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); 205 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; 206 207 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5)); 208 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/ 209 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 210 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */ 211 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */ 212 PACKET3_MAP_QUEUES_QUEUE(ring->queue) | 213 PACKET3_MAP_QUEUES_PIPE(ring->pipe) | 214 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) | 215 /*queue_type: normal compute queue */ 216 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | 217 /* alloc format: all_on_one_pipe */ 218 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | 219 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) | 220 /* num_queues: must be 1 */ 221 PACKET3_MAP_QUEUES_NUM_QUEUES(1)); 222 amdgpu_ring_write(kiq_ring, 223 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index)); 224 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr)); 225 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr)); 226 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr)); 227 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)); 228} 229 230static void gfx_v9_4_3_kiq_unmap_queues(struct amdgpu_ring *kiq_ring, 231 struct amdgpu_ring *ring, 232 enum amdgpu_unmap_queues_action action, 233 u64 gpu_addr, u64 seq) 234{ 235 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; 236 237 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4)); 238 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 239 PACKET3_UNMAP_QUEUES_ACTION(action) | 240 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) | 241 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) | 242 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1)); 243 amdgpu_ring_write(kiq_ring, 244 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index)); 245 246 if (action == PREEMPT_QUEUES_NO_UNMAP) { 247 amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr)); 248 amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr)); 249 amdgpu_ring_write(kiq_ring, seq); 250 } else { 251 amdgpu_ring_write(kiq_ring, 0); 252 amdgpu_ring_write(kiq_ring, 0); 253 amdgpu_ring_write(kiq_ring, 0); 254 } 255} 256 257static void gfx_v9_4_3_kiq_query_status(struct amdgpu_ring *kiq_ring, 258 struct amdgpu_ring *ring, 259 u64 addr, 260 u64 seq) 261{ 262 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; 263 264 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5)); 265 amdgpu_ring_write(kiq_ring, 266 PACKET3_QUERY_STATUS_CONTEXT_ID(0) | 267 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) | 268 PACKET3_QUERY_STATUS_COMMAND(2)); 269 /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 270 amdgpu_ring_write(kiq_ring, 271 PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) | 272 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel)); 273 amdgpu_ring_write(kiq_ring, lower_32_bits(addr)); 274 amdgpu_ring_write(kiq_ring, upper_32_bits(addr)); 275 amdgpu_ring_write(kiq_ring, lower_32_bits(seq)); 276 amdgpu_ring_write(kiq_ring, upper_32_bits(seq)); 277} 278 279static void gfx_v9_4_3_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring, 280 uint16_t pasid, uint32_t flush_type, 281 bool all_hub) 282{ 283 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0)); 284 amdgpu_ring_write(kiq_ring, 285 PACKET3_INVALIDATE_TLBS_DST_SEL(1) | 286 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) | 287 PACKET3_INVALIDATE_TLBS_PASID(pasid) | 288 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type)); 289} 290 291static void gfx_v9_4_3_kiq_reset_hw_queue(struct amdgpu_ring *kiq_ring, uint32_t queue_type, 292 uint32_t me_id, uint32_t pipe_id, uint32_t queue_id, 293 uint32_t xcc_id, uint32_t vmid) 294{ 295 struct amdgpu_device *adev = kiq_ring->adev; 296 unsigned i; 297 298 /* enter save mode */ 299 amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id); 300 mutex_lock(&adev->srbm_mutex); 301 soc15_grbm_select(adev, me_id, pipe_id, queue_id, 0, xcc_id); 302 303 if (queue_type == AMDGPU_RING_TYPE_COMPUTE) { 304 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 0x2); 305 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regSPI_COMPUTE_QUEUE_RESET, 0x1); 306 /* wait till dequeue take effects */ 307 for (i = 0; i < adev->usec_timeout; i++) { 308 if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) 309 break; 310 udelay(1); 311 } 312 if (i >= adev->usec_timeout) 313 dev_err(adev->dev, "fail to wait on hqd deactive\n"); 314 } else { 315 dev_err(adev->dev, "reset queue_type(%d) not supported\n\n", queue_type); 316 } 317 318 soc15_grbm_select(adev, 0, 0, 0, 0, 0); 319 mutex_unlock(&adev->srbm_mutex); 320 /* exit safe mode */ 321 amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id); 322} 323 324static const struct kiq_pm4_funcs gfx_v9_4_3_kiq_pm4_funcs = { 325 .kiq_set_resources = gfx_v9_4_3_kiq_set_resources, 326 .kiq_map_queues = gfx_v9_4_3_kiq_map_queues, 327 .kiq_unmap_queues = gfx_v9_4_3_kiq_unmap_queues, 328 .kiq_query_status = gfx_v9_4_3_kiq_query_status, 329 .kiq_invalidate_tlbs = gfx_v9_4_3_kiq_invalidate_tlbs, 330 .kiq_reset_hw_queue = gfx_v9_4_3_kiq_reset_hw_queue, 331 .set_resources_size = 8, 332 .map_queues_size = 7, 333 .unmap_queues_size = 6, 334 .query_status_size = 7, 335 .invalidate_tlbs_size = 2, 336}; 337 338static void gfx_v9_4_3_set_kiq_pm4_funcs(struct amdgpu_device *adev) 339{ 340 int i, num_xcc; 341 342 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 343 for (i = 0; i < num_xcc; i++) 344 adev->gfx.kiq[i].pmf = &gfx_v9_4_3_kiq_pm4_funcs; 345} 346 347static void gfx_v9_4_3_init_golden_registers(struct amdgpu_device *adev) 348{ 349 int i, num_xcc, dev_inst; 350 351 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 352 for (i = 0; i < num_xcc; i++) { 353 dev_inst = GET_INST(GC, i); 354 355 WREG32_SOC15(GC, dev_inst, regGB_ADDR_CONFIG, 356 GOLDEN_GB_ADDR_CONFIG); 357 WREG32_FIELD15_PREREG(GC, dev_inst, TCP_UTCL1_CNTL2, SPARE, 0x1); 358 } 359} 360 361static uint32_t gfx_v9_4_3_normalize_xcc_reg_offset(uint32_t reg) 362{ 363 uint32_t normalized_reg = NORMALIZE_XCC_REG_OFFSET(reg); 364 365 /* If it is an XCC reg, normalize the reg to keep 366 lower 16 bits in local xcc */ 367 368 if (((normalized_reg >= XCC_REG_RANGE_0_LOW) && (normalized_reg < XCC_REG_RANGE_0_HIGH)) || 369 ((normalized_reg >= XCC_REG_RANGE_1_LOW) && (normalized_reg < XCC_REG_RANGE_1_HIGH))) 370 return normalized_reg; 371 else 372 return reg; 373} 374 375static void gfx_v9_4_3_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel, 376 bool wc, uint32_t reg, uint32_t val) 377{ 378 reg = gfx_v9_4_3_normalize_xcc_reg_offset(reg); 379 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 380 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) | 381 WRITE_DATA_DST_SEL(0) | 382 (wc ? WR_CONFIRM : 0)); 383 amdgpu_ring_write(ring, reg); 384 amdgpu_ring_write(ring, 0); 385 amdgpu_ring_write(ring, val); 386} 387 388static void gfx_v9_4_3_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel, 389 int mem_space, int opt, uint32_t addr0, 390 uint32_t addr1, uint32_t ref, uint32_t mask, 391 uint32_t inv) 392{ 393 /* Only do the normalization on regspace */ 394 if (mem_space == 0) { 395 addr0 = gfx_v9_4_3_normalize_xcc_reg_offset(addr0); 396 addr1 = gfx_v9_4_3_normalize_xcc_reg_offset(addr1); 397 } 398 399 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 400 amdgpu_ring_write(ring, 401 /* memory (1) or register (0) */ 402 (WAIT_REG_MEM_MEM_SPACE(mem_space) | 403 WAIT_REG_MEM_OPERATION(opt) | /* wait */ 404 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 405 WAIT_REG_MEM_ENGINE(eng_sel))); 406 407 if (mem_space) 408 BUG_ON(addr0 & 0x3); /* Dword align */ 409 amdgpu_ring_write(ring, addr0); 410 amdgpu_ring_write(ring, addr1); 411 amdgpu_ring_write(ring, ref); 412 amdgpu_ring_write(ring, mask); 413 amdgpu_ring_write(ring, inv); /* poll interval */ 414} 415 416static int gfx_v9_4_3_ring_test_ring(struct amdgpu_ring *ring) 417{ 418 uint32_t scratch_reg0_offset, xcc_offset; 419 struct amdgpu_device *adev = ring->adev; 420 uint32_t tmp = 0; 421 unsigned i; 422 int r; 423 424 /* Use register offset which is local to XCC in the packet */ 425 xcc_offset = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0); 426 scratch_reg0_offset = SOC15_REG_OFFSET(GC, GET_INST(GC, ring->xcc_id), regSCRATCH_REG0); 427 WREG32(scratch_reg0_offset, 0xCAFEDEAD); 428 tmp = RREG32(scratch_reg0_offset); 429 430 r = amdgpu_ring_alloc(ring, 3); 431 if (r) 432 return r; 433 434 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); 435 amdgpu_ring_write(ring, xcc_offset - PACKET3_SET_UCONFIG_REG_START); 436 amdgpu_ring_write(ring, 0xDEADBEEF); 437 amdgpu_ring_commit(ring); 438 439 for (i = 0; i < adev->usec_timeout; i++) { 440 tmp = RREG32(scratch_reg0_offset); 441 if (tmp == 0xDEADBEEF) 442 break; 443 udelay(1); 444 } 445 446 if (i >= adev->usec_timeout) 447 r = -ETIMEDOUT; 448 return r; 449} 450 451static int gfx_v9_4_3_ring_test_ib(struct amdgpu_ring *ring, long timeout) 452{ 453 struct amdgpu_device *adev = ring->adev; 454 struct amdgpu_ib ib; 455 struct dma_fence *f = NULL; 456 457 unsigned index; 458 uint64_t gpu_addr; 459 uint32_t tmp; 460 long r; 461 462 r = amdgpu_device_wb_get(adev, &index); 463 if (r) 464 return r; 465 466 gpu_addr = adev->wb.gpu_addr + (index * 4); 467 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD); 468 memset(&ib, 0, sizeof(ib)); 469 470 r = amdgpu_ib_get(adev, NULL, 20, AMDGPU_IB_POOL_DIRECT, &ib); 471 if (r) 472 goto err1; 473 474 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3); 475 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM; 476 ib.ptr[2] = lower_32_bits(gpu_addr); 477 ib.ptr[3] = upper_32_bits(gpu_addr); 478 ib.ptr[4] = 0xDEADBEEF; 479 ib.length_dw = 5; 480 481 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 482 if (r) 483 goto err2; 484 485 r = dma_fence_wait_timeout(f, false, timeout); 486 if (r == 0) { 487 r = -ETIMEDOUT; 488 goto err2; 489 } else if (r < 0) { 490 goto err2; 491 } 492 493 tmp = adev->wb.wb[index]; 494 if (tmp == 0xDEADBEEF) 495 r = 0; 496 else 497 r = -EINVAL; 498 499err2: 500 amdgpu_ib_free(&ib, NULL); 501 dma_fence_put(f); 502err1: 503 amdgpu_device_wb_free(adev, index); 504 return r; 505} 506 507 508/* This value might differs per partition */ 509static uint64_t gfx_v9_4_3_get_gpu_clock_counter(struct amdgpu_device *adev) 510{ 511 uint64_t clock; 512 513 mutex_lock(&adev->gfx.gpu_clock_mutex); 514 WREG32_SOC15(GC, GET_INST(GC, 0), regRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 515 clock = (uint64_t)RREG32_SOC15(GC, GET_INST(GC, 0), regRLC_GPU_CLOCK_COUNT_LSB) | 516 ((uint64_t)RREG32_SOC15(GC, GET_INST(GC, 0), regRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 517 mutex_unlock(&adev->gfx.gpu_clock_mutex); 518 519 return clock; 520} 521 522static void gfx_v9_4_3_free_microcode(struct amdgpu_device *adev) 523{ 524 amdgpu_ucode_release(&adev->gfx.pfp_fw); 525 amdgpu_ucode_release(&adev->gfx.me_fw); 526 amdgpu_ucode_release(&adev->gfx.ce_fw); 527 amdgpu_ucode_release(&adev->gfx.rlc_fw); 528 amdgpu_ucode_release(&adev->gfx.mec_fw); 529 amdgpu_ucode_release(&adev->gfx.mec2_fw); 530 531 kfree(adev->gfx.rlc.register_list_format); 532} 533 534static int gfx_v9_4_3_init_rlc_microcode(struct amdgpu_device *adev, 535 const char *chip_name) 536{ 537 int err; 538 const struct rlc_firmware_header_v2_0 *rlc_hdr; 539 uint16_t version_major; 540 uint16_t version_minor; 541 542 543 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, 544 AMDGPU_UCODE_REQUIRED, 545 "amdgpu/%s_rlc.bin", chip_name); 546 if (err) 547 goto out; 548 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; 549 550 version_major = le16_to_cpu(rlc_hdr->header.header_version_major); 551 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor); 552 err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor); 553out: 554 if (err) 555 amdgpu_ucode_release(&adev->gfx.rlc_fw); 556 557 return err; 558} 559 560static int gfx_v9_4_3_init_cp_compute_microcode(struct amdgpu_device *adev, 561 const char *chip_name) 562{ 563 int err; 564 565 if (amdgpu_sriov_vf(adev)) { 566 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, 567 AMDGPU_UCODE_REQUIRED, 568 "amdgpu/%s_sjt_mec.bin", chip_name); 569 570 if (err) 571 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, 572 AMDGPU_UCODE_REQUIRED, 573 "amdgpu/%s_mec.bin", chip_name); 574 } else 575 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, 576 AMDGPU_UCODE_REQUIRED, 577 "amdgpu/%s_mec.bin", chip_name); 578 if (err) 579 goto out; 580 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1); 581 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT); 582 583 adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version; 584 adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version; 585 586out: 587 if (err) 588 amdgpu_ucode_release(&adev->gfx.mec_fw); 589 return err; 590} 591 592static int gfx_v9_4_3_init_microcode(struct amdgpu_device *adev) 593{ 594 char ucode_prefix[15]; 595 int r; 596 597 amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix)); 598 599 r = gfx_v9_4_3_init_rlc_microcode(adev, ucode_prefix); 600 if (r) 601 return r; 602 603 r = gfx_v9_4_3_init_cp_compute_microcode(adev, ucode_prefix); 604 if (r) 605 return r; 606 607 return r; 608} 609 610static void gfx_v9_4_3_mec_fini(struct amdgpu_device *adev) 611{ 612 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL); 613 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL); 614} 615 616static int gfx_v9_4_3_mec_init(struct amdgpu_device *adev) 617{ 618 int r, i, num_xcc; 619 u32 *hpd; 620 const __le32 *fw_data; 621 unsigned fw_size; 622 u32 *fw; 623 size_t mec_hpd_size; 624 625 const struct gfx_firmware_header_v1_0 *mec_hdr; 626 627 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 628 for (i = 0; i < num_xcc; i++) 629 bitmap_zero(adev->gfx.mec_bitmap[i].queue_bitmap, 630 AMDGPU_MAX_COMPUTE_QUEUES); 631 632 /* take ownership of the relevant compute queues */ 633 amdgpu_gfx_compute_queue_acquire(adev); 634 mec_hpd_size = 635 adev->gfx.num_compute_rings * num_xcc * GFX9_MEC_HPD_SIZE; 636 if (mec_hpd_size) { 637 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, 638 AMDGPU_GEM_DOMAIN_VRAM | 639 AMDGPU_GEM_DOMAIN_GTT, 640 &adev->gfx.mec.hpd_eop_obj, 641 &adev->gfx.mec.hpd_eop_gpu_addr, 642 (void **)&hpd); 643 if (r) { 644 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r); 645 gfx_v9_4_3_mec_fini(adev); 646 return r; 647 } 648 649 if (amdgpu_emu_mode == 1) { 650 for (i = 0; i < mec_hpd_size / 4; i++) { 651 memset((void *)(hpd + i), 0, 4); 652 if (i % 50 == 0) 653 msleep(1); 654 } 655 } else { 656 memset(hpd, 0, mec_hpd_size); 657 } 658 659 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj); 660 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj); 661 } 662 663 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 664 665 fw_data = (const __le32 *) 666 (adev->gfx.mec_fw->data + 667 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)); 668 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes); 669 670 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes, 671 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, 672 &adev->gfx.mec.mec_fw_obj, 673 &adev->gfx.mec.mec_fw_gpu_addr, 674 (void **)&fw); 675 if (r) { 676 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r); 677 gfx_v9_4_3_mec_fini(adev); 678 return r; 679 } 680 681 memcpy(fw, fw_data, fw_size); 682 683 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj); 684 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj); 685 686 return 0; 687} 688 689static void gfx_v9_4_3_xcc_select_se_sh(struct amdgpu_device *adev, u32 se_num, 690 u32 sh_num, u32 instance, int xcc_id) 691{ 692 u32 data; 693 694 if (instance == 0xffffffff) 695 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, 696 INSTANCE_BROADCAST_WRITES, 1); 697 else 698 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, 699 INSTANCE_INDEX, instance); 700 701 if (se_num == 0xffffffff) 702 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, 703 SE_BROADCAST_WRITES, 1); 704 else 705 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num); 706 707 if (sh_num == 0xffffffff) 708 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, 709 SH_BROADCAST_WRITES, 1); 710 else 711 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num); 712 713 WREG32_SOC15_RLC_SHADOW_EX(reg, GC, GET_INST(GC, xcc_id), regGRBM_GFX_INDEX, data); 714} 715 716static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t address) 717{ 718 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSQ_IND_INDEX, 719 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 720 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 721 (address << SQ_IND_INDEX__INDEX__SHIFT) | 722 (SQ_IND_INDEX__FORCE_READ_MASK)); 723 return RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_IND_DATA); 724} 725 726static void wave_read_regs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, 727 uint32_t wave, uint32_t thread, 728 uint32_t regno, uint32_t num, uint32_t *out) 729{ 730 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSQ_IND_INDEX, 731 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 732 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 733 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 734 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 735 (SQ_IND_INDEX__FORCE_READ_MASK) | 736 (SQ_IND_INDEX__AUTO_INCR_MASK)); 737 while (num--) 738 *(out++) = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_IND_DATA); 739} 740 741static void gfx_v9_4_3_read_wave_data(struct amdgpu_device *adev, 742 uint32_t xcc_id, uint32_t simd, uint32_t wave, 743 uint32_t *dst, int *no_fields) 744{ 745 /* type 1 wave data */ 746 dst[(*no_fields)++] = 1; 747 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_STATUS); 748 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_PC_LO); 749 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_PC_HI); 750 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_EXEC_LO); 751 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_EXEC_HI); 752 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_HW_ID); 753 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_INST_DW0); 754 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_INST_DW1); 755 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_GPR_ALLOC); 756 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_LDS_ALLOC); 757 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_TRAPSTS); 758 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_IB_STS); 759 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_IB_DBG0); 760 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_M0); 761 dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_MODE); 762} 763 764static void gfx_v9_4_3_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, 765 uint32_t wave, uint32_t start, 766 uint32_t size, uint32_t *dst) 767{ 768 wave_read_regs(adev, xcc_id, simd, wave, 0, 769 start + SQIND_WAVE_SGPRS_OFFSET, size, dst); 770} 771 772static void gfx_v9_4_3_read_wave_vgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, 773 uint32_t wave, uint32_t thread, 774 uint32_t start, uint32_t size, 775 uint32_t *dst) 776{ 777 wave_read_regs(adev, xcc_id, simd, wave, thread, 778 start + SQIND_WAVE_VGPRS_OFFSET, size, dst); 779} 780 781static void gfx_v9_4_3_select_me_pipe_q(struct amdgpu_device *adev, 782 u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id) 783{ 784 soc15_grbm_select(adev, me, pipe, q, vm, GET_INST(GC, xcc_id)); 785} 786 787static int gfx_v9_4_3_get_xccs_per_xcp(struct amdgpu_device *adev) 788{ 789 u32 xcp_ctl; 790 791 /* Value is expected to be the same on all, fetch from first instance */ 792 xcp_ctl = RREG32_SOC15(GC, GET_INST(GC, 0), regCP_HYP_XCP_CTL); 793 794 return REG_GET_FIELD(xcp_ctl, CP_HYP_XCP_CTL, NUM_XCC_IN_XCP); 795} 796 797static int gfx_v9_4_3_switch_compute_partition(struct amdgpu_device *adev, 798 int num_xccs_per_xcp) 799{ 800 int ret, i, num_xcc; 801 u32 tmp = 0; 802 803 if (adev->psp.funcs) { 804 ret = psp_spatial_partition(&adev->psp, 805 NUM_XCC(adev->gfx.xcc_mask) / 806 num_xccs_per_xcp); 807 if (ret) 808 return ret; 809 } else { 810 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 811 812 for (i = 0; i < num_xcc; i++) { 813 tmp = REG_SET_FIELD(tmp, CP_HYP_XCP_CTL, NUM_XCC_IN_XCP, 814 num_xccs_per_xcp); 815 tmp = REG_SET_FIELD(tmp, CP_HYP_XCP_CTL, VIRTUAL_XCC_ID, 816 i % num_xccs_per_xcp); 817 WREG32_SOC15(GC, GET_INST(GC, i), regCP_HYP_XCP_CTL, 818 tmp); 819 } 820 ret = 0; 821 } 822 823 adev->gfx.num_xcc_per_xcp = num_xccs_per_xcp; 824 825 return ret; 826} 827 828static int gfx_v9_4_3_ih_to_xcc_inst(struct amdgpu_device *adev, int ih_node) 829{ 830 int xcc; 831 832 xcc = hweight8(adev->gfx.xcc_mask & GENMASK(ih_node / 2, 0)); 833 if (!xcc) { 834 dev_err(adev->dev, "Couldn't find xcc mapping from IH node"); 835 return -EINVAL; 836 } 837 838 return xcc - 1; 839} 840 841static const struct amdgpu_gfx_funcs gfx_v9_4_3_gfx_funcs = { 842 .get_gpu_clock_counter = &gfx_v9_4_3_get_gpu_clock_counter, 843 .select_se_sh = &gfx_v9_4_3_xcc_select_se_sh, 844 .read_wave_data = &gfx_v9_4_3_read_wave_data, 845 .read_wave_sgprs = &gfx_v9_4_3_read_wave_sgprs, 846 .read_wave_vgprs = &gfx_v9_4_3_read_wave_vgprs, 847 .select_me_pipe_q = &gfx_v9_4_3_select_me_pipe_q, 848 .switch_partition_mode = &gfx_v9_4_3_switch_compute_partition, 849 .ih_node_to_logical_xcc = &gfx_v9_4_3_ih_to_xcc_inst, 850 .get_xccs_per_xcp = &gfx_v9_4_3_get_xccs_per_xcp, 851}; 852 853static int gfx_v9_4_3_aca_bank_parser(struct aca_handle *handle, 854 struct aca_bank *bank, enum aca_smu_type type, 855 void *data) 856{ 857 struct aca_bank_info info; 858 u64 misc0; 859 u32 instlo; 860 int ret; 861 862 ret = aca_bank_info_decode(bank, &info); 863 if (ret) 864 return ret; 865 866 /* NOTE: overwrite info.die_id with xcd id for gfx */ 867 instlo = ACA_REG__IPID__INSTANCEIDLO(bank->regs[ACA_REG_IDX_IPID]); 868 instlo &= GENMASK(31, 1); 869 info.die_id = instlo == mmSMNAID_XCD0_MCA_SMU ? 0 : 1; 870 871 misc0 = bank->regs[ACA_REG_IDX_MISC0]; 872 873 switch (type) { 874 case ACA_SMU_TYPE_UE: 875 bank->aca_err_type = ACA_ERROR_TYPE_UE; 876 ret = aca_error_cache_log_bank_error(handle, &info, bank->aca_err_type, 1ULL); 877 break; 878 case ACA_SMU_TYPE_CE: 879 bank->aca_err_type = ACA_ERROR_TYPE_CE; 880 ret = aca_error_cache_log_bank_error(handle, &info, bank->aca_err_type, 881 ACA_REG__MISC0__ERRCNT(misc0)); 882 break; 883 default: 884 return -EINVAL; 885 } 886 887 return ret; 888} 889 890static bool gfx_v9_4_3_aca_bank_is_valid(struct aca_handle *handle, struct aca_bank *bank, 891 enum aca_smu_type type, void *data) 892{ 893 u32 instlo; 894 895 instlo = ACA_REG__IPID__INSTANCEIDLO(bank->regs[ACA_REG_IDX_IPID]); 896 instlo &= GENMASK(31, 1); 897 switch (instlo) { 898 case mmSMNAID_XCD0_MCA_SMU: 899 case mmSMNAID_XCD1_MCA_SMU: 900 case mmSMNXCD_XCD0_MCA_SMU: 901 return true; 902 default: 903 break; 904 } 905 906 return false; 907} 908 909static const struct aca_bank_ops gfx_v9_4_3_aca_bank_ops = { 910 .aca_bank_parser = gfx_v9_4_3_aca_bank_parser, 911 .aca_bank_is_valid = gfx_v9_4_3_aca_bank_is_valid, 912}; 913 914static const struct aca_info gfx_v9_4_3_aca_info = { 915 .hwip = ACA_HWIP_TYPE_SMU, 916 .mask = ACA_ERROR_UE_MASK | ACA_ERROR_CE_MASK, 917 .bank_ops = &gfx_v9_4_3_aca_bank_ops, 918}; 919 920static int gfx_v9_4_3_gpu_early_init(struct amdgpu_device *adev) 921{ 922 adev->gfx.funcs = &gfx_v9_4_3_gfx_funcs; 923 adev->gfx.ras = &gfx_v9_4_3_ras; 924 925 adev->gfx.config.max_hw_contexts = 8; 926 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 927 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 928 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 929 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 930 adev->gfx.config.gb_addr_config = GOLDEN_GB_ADDR_CONFIG; 931 932 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 << 933 REG_GET_FIELD( 934 adev->gfx.config.gb_addr_config, 935 GB_ADDR_CONFIG, 936 NUM_PIPES); 937 938 adev->gfx.config.max_tile_pipes = 939 adev->gfx.config.gb_addr_config_fields.num_pipes; 940 941 adev->gfx.config.gb_addr_config_fields.num_banks = 1 << 942 REG_GET_FIELD( 943 adev->gfx.config.gb_addr_config, 944 GB_ADDR_CONFIG, 945 NUM_BANKS); 946 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 << 947 REG_GET_FIELD( 948 adev->gfx.config.gb_addr_config, 949 GB_ADDR_CONFIG, 950 MAX_COMPRESSED_FRAGS); 951 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 << 952 REG_GET_FIELD( 953 adev->gfx.config.gb_addr_config, 954 GB_ADDR_CONFIG, 955 NUM_RB_PER_SE); 956 adev->gfx.config.gb_addr_config_fields.num_se = 1 << 957 REG_GET_FIELD( 958 adev->gfx.config.gb_addr_config, 959 GB_ADDR_CONFIG, 960 NUM_SHADER_ENGINES); 961 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 + 962 REG_GET_FIELD( 963 adev->gfx.config.gb_addr_config, 964 GB_ADDR_CONFIG, 965 PIPE_INTERLEAVE_SIZE)); 966 967 return 0; 968} 969 970static int gfx_v9_4_3_compute_ring_init(struct amdgpu_device *adev, int ring_id, 971 int xcc_id, int mec, int pipe, int queue) 972{ 973 unsigned irq_type; 974 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id]; 975 unsigned int hw_prio; 976 uint32_t xcc_doorbell_start; 977 978 ring = &adev->gfx.compute_ring[xcc_id * adev->gfx.num_compute_rings + 979 ring_id]; 980 981 /* mec0 is me1 */ 982 ring->xcc_id = xcc_id; 983 ring->me = mec + 1; 984 ring->pipe = pipe; 985 ring->queue = queue; 986 987 ring->ring_obj = NULL; 988 ring->use_doorbell = true; 989 xcc_doorbell_start = adev->doorbell_index.mec_ring0 + 990 xcc_id * adev->doorbell_index.xcc_doorbell_range; 991 ring->doorbell_index = (xcc_doorbell_start + ring_id) << 1; 992 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + 993 (ring_id + xcc_id * adev->gfx.num_compute_rings) * 994 GFX9_MEC_HPD_SIZE; 995 ring->vm_hub = AMDGPU_GFXHUB(xcc_id); 996 sprintf(ring->name, "comp_%d.%d.%d.%d", 997 ring->xcc_id, ring->me, ring->pipe, ring->queue); 998 999 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP 1000 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec) 1001 + ring->pipe; 1002 hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ? 1003 AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL; 1004 /* type-2 packets are deprecated on MEC, use type-3 instead */ 1005 return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type, 1006 hw_prio, NULL); 1007} 1008 1009static void gfx_v9_4_3_alloc_ip_dump(struct amdgpu_device *adev) 1010{ 1011 uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9_4_3); 1012 uint32_t *ptr, num_xcc, inst; 1013 1014 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1015 1016 ptr = kcalloc(reg_count * num_xcc, sizeof(uint32_t), GFP_KERNEL); 1017 if (!ptr) { 1018 DRM_ERROR("Failed to allocate memory for GFX IP Dump\n"); 1019 adev->gfx.ip_dump_core = NULL; 1020 } else { 1021 adev->gfx.ip_dump_core = ptr; 1022 } 1023 1024 /* Allocate memory for compute queue registers for all the instances */ 1025 reg_count = ARRAY_SIZE(gc_cp_reg_list_9_4_3); 1026 inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec * 1027 adev->gfx.mec.num_queue_per_pipe; 1028 1029 ptr = kcalloc(reg_count * inst * num_xcc, sizeof(uint32_t), GFP_KERNEL); 1030 if (!ptr) { 1031 DRM_ERROR("Failed to allocate memory for Compute Queues IP Dump\n"); 1032 adev->gfx.ip_dump_compute_queues = NULL; 1033 } else { 1034 adev->gfx.ip_dump_compute_queues = ptr; 1035 } 1036} 1037 1038static int gfx_v9_4_3_sw_init(struct amdgpu_ip_block *ip_block) 1039{ 1040 int i, j, k, r, ring_id, xcc_id, num_xcc; 1041 struct amdgpu_device *adev = ip_block->adev; 1042 1043 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { 1044 case IP_VERSION(9, 4, 3): 1045 case IP_VERSION(9, 4, 4): 1046 adev->gfx.cleaner_shader_ptr = gfx_9_4_3_cleaner_shader_hex; 1047 adev->gfx.cleaner_shader_size = sizeof(gfx_9_4_3_cleaner_shader_hex); 1048 if (adev->gfx.mec_fw_version >= 153) { 1049 adev->gfx.enable_cleaner_shader = true; 1050 r = amdgpu_gfx_cleaner_shader_sw_init(adev, adev->gfx.cleaner_shader_size); 1051 if (r) { 1052 adev->gfx.enable_cleaner_shader = false; 1053 dev_err(adev->dev, "Failed to initialize cleaner shader\n"); 1054 } 1055 } 1056 break; 1057 default: 1058 adev->gfx.enable_cleaner_shader = false; 1059 break; 1060 } 1061 1062 adev->gfx.mec.num_mec = 2; 1063 adev->gfx.mec.num_pipe_per_mec = 4; 1064 adev->gfx.mec.num_queue_per_pipe = 8; 1065 1066 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1067 1068 /* EOP Event */ 1069 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq); 1070 if (r) 1071 return r; 1072 1073 /* Bad opcode Event */ 1074 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, 1075 GFX_9_0__SRCID__CP_BAD_OPCODE_ERROR, 1076 &adev->gfx.bad_op_irq); 1077 if (r) 1078 return r; 1079 1080 /* Privileged reg */ 1081 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT, 1082 &adev->gfx.priv_reg_irq); 1083 if (r) 1084 return r; 1085 1086 /* Privileged inst */ 1087 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT, 1088 &adev->gfx.priv_inst_irq); 1089 if (r) 1090 return r; 1091 1092 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE; 1093 1094 r = adev->gfx.rlc.funcs->init(adev); 1095 if (r) { 1096 DRM_ERROR("Failed to init rlc BOs!\n"); 1097 return r; 1098 } 1099 1100 r = gfx_v9_4_3_mec_init(adev); 1101 if (r) { 1102 DRM_ERROR("Failed to init MEC BOs!\n"); 1103 return r; 1104 } 1105 1106 /* set up the compute queues - allocate horizontally across pipes */ 1107 for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { 1108 ring_id = 0; 1109 for (i = 0; i < adev->gfx.mec.num_mec; ++i) { 1110 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) { 1111 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; 1112 k++) { 1113 if (!amdgpu_gfx_is_mec_queue_enabled( 1114 adev, xcc_id, i, k, j)) 1115 continue; 1116 1117 r = gfx_v9_4_3_compute_ring_init(adev, 1118 ring_id, 1119 xcc_id, 1120 i, k, j); 1121 if (r) 1122 return r; 1123 1124 ring_id++; 1125 } 1126 } 1127 } 1128 1129 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE, xcc_id); 1130 if (r) { 1131 DRM_ERROR("Failed to init KIQ BOs!\n"); 1132 return r; 1133 } 1134 1135 r = amdgpu_gfx_kiq_init_ring(adev, xcc_id); 1136 if (r) 1137 return r; 1138 1139 /* create MQD for all compute queues as wel as KIQ for SRIOV case */ 1140 r = amdgpu_gfx_mqd_sw_init(adev, 1141 sizeof(struct v9_mqd_allocation), xcc_id); 1142 if (r) 1143 return r; 1144 } 1145 1146 adev->gfx.compute_supported_reset = 1147 amdgpu_get_soft_full_reset_mask(&adev->gfx.compute_ring[0]); 1148 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { 1149 case IP_VERSION(9, 4, 3): 1150 case IP_VERSION(9, 4, 4): 1151 if ((adev->gfx.mec_fw_version >= 155) && 1152 !amdgpu_sriov_vf(adev) && 1153 !adev->debug_disable_gpu_ring_reset) { 1154 adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; 1155 adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_PIPE; 1156 } 1157 break; 1158 case IP_VERSION(9, 5, 0): 1159 if ((adev->gfx.mec_fw_version >= 21) && 1160 !amdgpu_sriov_vf(adev) && 1161 !adev->debug_disable_gpu_ring_reset) { 1162 adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; 1163 adev->gfx.compute_supported_reset |= AMDGPU_RESET_TYPE_PER_PIPE; 1164 } 1165 break; 1166 default: 1167 break; 1168 } 1169 r = gfx_v9_4_3_gpu_early_init(adev); 1170 if (r) 1171 return r; 1172 1173 r = amdgpu_gfx_ras_sw_init(adev); 1174 if (r) 1175 return r; 1176 1177 r = amdgpu_gfx_sysfs_init(adev); 1178 if (r) 1179 return r; 1180 1181 gfx_v9_4_3_alloc_ip_dump(adev); 1182 1183 return 0; 1184} 1185 1186static int gfx_v9_4_3_sw_fini(struct amdgpu_ip_block *ip_block) 1187{ 1188 int i, num_xcc; 1189 struct amdgpu_device *adev = ip_block->adev; 1190 1191 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1192 for (i = 0; i < adev->gfx.num_compute_rings * num_xcc; i++) 1193 amdgpu_ring_fini(&adev->gfx.compute_ring[i]); 1194 1195 for (i = 0; i < num_xcc; i++) { 1196 amdgpu_gfx_mqd_sw_fini(adev, i); 1197 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[i].ring); 1198 amdgpu_gfx_kiq_fini(adev, i); 1199 } 1200 1201 amdgpu_gfx_cleaner_shader_sw_fini(adev); 1202 1203 gfx_v9_4_3_mec_fini(adev); 1204 amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj); 1205 gfx_v9_4_3_free_microcode(adev); 1206 amdgpu_gfx_sysfs_fini(adev); 1207 1208 kfree(adev->gfx.ip_dump_core); 1209 kfree(adev->gfx.ip_dump_compute_queues); 1210 1211 return 0; 1212} 1213 1214#define DEFAULT_SH_MEM_BASES (0x6000) 1215static void gfx_v9_4_3_xcc_init_compute_vmid(struct amdgpu_device *adev, 1216 int xcc_id) 1217{ 1218 int i; 1219 uint32_t sh_mem_config; 1220 uint32_t sh_mem_bases; 1221 uint32_t data; 1222 1223 /* 1224 * Configure apertures: 1225 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB) 1226 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB) 1227 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB) 1228 */ 1229 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16); 1230 1231 sh_mem_config = SH_MEM_ADDRESS_MODE_64 | 1232 SH_MEM_ALIGNMENT_MODE_UNALIGNED << 1233 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT; 1234 1235 mutex_lock(&adev->srbm_mutex); 1236 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 1237 soc15_grbm_select(adev, 0, 0, 0, i, GET_INST(GC, xcc_id)); 1238 /* CP and shaders */ 1239 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSH_MEM_CONFIG, sh_mem_config); 1240 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSH_MEM_BASES, sh_mem_bases); 1241 1242 /* Enable trap for each kfd vmid. */ 1243 data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSPI_GDBG_PER_VMID_CNTL); 1244 data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1); 1245 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSPI_GDBG_PER_VMID_CNTL, data); 1246 } 1247 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 1248 mutex_unlock(&adev->srbm_mutex); 1249 1250 /* 1251 * Initialize all compute VMIDs to have no GDS, GWS, or OA 1252 * access. These should be enabled by FW for target VMIDs. 1253 */ 1254 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 1255 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_BASE, 2 * i, 0); 1256 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_SIZE, 2 * i, 0); 1257 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_GWS_VMID0, i, 0); 1258 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_OA_VMID0, i, 0); 1259 } 1260} 1261 1262static void gfx_v9_4_3_xcc_init_gds_vmid(struct amdgpu_device *adev, int xcc_id) 1263{ 1264 int vmid; 1265 1266 /* 1267 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA 1268 * access. Compute VMIDs should be enabled by FW for target VMIDs, 1269 * the driver can enable them for graphics. VMID0 should maintain 1270 * access so that HWS firmware can save/restore entries. 1271 */ 1272 for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) { 1273 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_BASE, 2 * vmid, 0); 1274 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_SIZE, 2 * vmid, 0); 1275 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_GWS_VMID0, vmid, 0); 1276 WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_OA_VMID0, vmid, 0); 1277 } 1278} 1279 1280/* For ASICs that needs xnack chain and MEC version supports, set SG_CONFIG1 1281 * DISABLE_XNACK_CHECK_IN_RETRY_DISABLE bit and inform KFD to set xnack_chain 1282 * bit in SET_RESOURCES 1283 */ 1284static void gfx_v9_4_3_xcc_init_sq(struct amdgpu_device *adev, int xcc_id) 1285{ 1286 uint32_t data; 1287 1288 if (!(adev->gmc.xnack_flags & AMDGPU_GMC_XNACK_FLAG_CHAIN)) 1289 return; 1290 1291 data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_CONFIG1); 1292 data = REG_SET_FIELD(data, SQ_CONFIG1, DISABLE_XNACK_CHECK_IN_RETRY_DISABLE, 1); 1293 WREG32_SOC15(GC, xcc_id, regSQ_CONFIG1, data); 1294} 1295 1296static void gfx_v9_4_3_xcc_constants_init(struct amdgpu_device *adev, 1297 int xcc_id) 1298{ 1299 u32 tmp; 1300 int i; 1301 1302 /* XXX SH_MEM regs */ 1303 /* where to put LDS, scratch, GPUVM in FSA64 space */ 1304 mutex_lock(&adev->srbm_mutex); 1305 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB(0)].num_ids; i++) { 1306 soc15_grbm_select(adev, 0, 0, 0, i, GET_INST(GC, xcc_id)); 1307 /* CP and shaders */ 1308 if (i == 0) { 1309 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, 1310 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 1311 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE, 1312 !!adev->gmc.noretry); 1313 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), 1314 regSH_MEM_CONFIG, tmp); 1315 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), 1316 regSH_MEM_BASES, 0); 1317 } else { 1318 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, 1319 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 1320 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE, 1321 !!adev->gmc.noretry); 1322 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), 1323 regSH_MEM_CONFIG, tmp); 1324 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE, 1325 (adev->gmc.private_aperture_start >> 1326 48)); 1327 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE, 1328 (adev->gmc.shared_aperture_start >> 1329 48)); 1330 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), 1331 regSH_MEM_BASES, tmp); 1332 } 1333 } 1334 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, 0)); 1335 1336 mutex_unlock(&adev->srbm_mutex); 1337 1338 gfx_v9_4_3_xcc_init_compute_vmid(adev, xcc_id); 1339 gfx_v9_4_3_xcc_init_gds_vmid(adev, xcc_id); 1340 gfx_v9_4_3_xcc_init_sq(adev, xcc_id); 1341} 1342 1343static void gfx_v9_4_3_constants_init(struct amdgpu_device *adev) 1344{ 1345 int i, num_xcc; 1346 1347 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1348 1349 gfx_v9_4_3_get_cu_info(adev, &adev->gfx.cu_info); 1350 adev->gfx.config.db_debug2 = 1351 RREG32_SOC15(GC, GET_INST(GC, 0), regDB_DEBUG2); 1352 1353 switch (amdgpu_ip_version(adev, GC_HWIP, 0)) { 1354 /* ToDo: GC 9.4.4 */ 1355 case IP_VERSION(9, 4, 3): 1356 if (adev->gfx.mec_fw_version >= 184 && 1357 (amdgpu_sriov_reg_access_sq_config(adev) || 1358 !amdgpu_sriov_vf(adev))) 1359 adev->gmc.xnack_flags |= AMDGPU_GMC_XNACK_FLAG_CHAIN; 1360 break; 1361 case IP_VERSION(9, 5, 0): 1362 if (adev->gfx.mec_fw_version >= 23) 1363 adev->gmc.xnack_flags |= AMDGPU_GMC_XNACK_FLAG_CHAIN; 1364 break; 1365 default: 1366 break; 1367 } 1368 1369 for (i = 0; i < num_xcc; i++) 1370 gfx_v9_4_3_xcc_constants_init(adev, i); 1371} 1372 1373static void 1374gfx_v9_4_3_xcc_enable_save_restore_machine(struct amdgpu_device *adev, 1375 int xcc_id) 1376{ 1377 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), RLC_SRM_CNTL, SRM_ENABLE, 1); 1378} 1379 1380static void gfx_v9_4_3_xcc_init_pg(struct amdgpu_device *adev, int xcc_id) 1381{ 1382 /* 1383 * Rlc save restore list is workable since v2_1. 1384 */ 1385 gfx_v9_4_3_xcc_enable_save_restore_machine(adev, xcc_id); 1386} 1387 1388static void gfx_v9_4_3_xcc_disable_gpa_mode(struct amdgpu_device *adev, int xcc_id) 1389{ 1390 uint32_t data; 1391 1392 data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCPC_PSP_DEBUG); 1393 data |= CPC_PSP_DEBUG__UTCL2IUGPAOVERRIDE_MASK; 1394 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCPC_PSP_DEBUG, data); 1395} 1396 1397static bool gfx_v9_4_3_is_rlc_enabled(struct amdgpu_device *adev) 1398{ 1399 uint32_t rlc_setting; 1400 1401 /* if RLC is not enabled, do nothing */ 1402 rlc_setting = RREG32_SOC15(GC, GET_INST(GC, 0), regRLC_CNTL); 1403 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) 1404 return false; 1405 1406 return true; 1407} 1408 1409static void gfx_v9_4_3_xcc_set_safe_mode(struct amdgpu_device *adev, int xcc_id) 1410{ 1411 uint32_t data; 1412 unsigned i; 1413 1414 data = RLC_SAFE_MODE__CMD_MASK; 1415 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); 1416 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SAFE_MODE, data); 1417 1418 /* wait for RLC_SAFE_MODE */ 1419 for (i = 0; i < adev->usec_timeout; i++) { 1420 if (!REG_GET_FIELD(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) 1421 break; 1422 udelay(1); 1423 } 1424} 1425 1426static void gfx_v9_4_3_xcc_unset_safe_mode(struct amdgpu_device *adev, 1427 int xcc_id) 1428{ 1429 uint32_t data; 1430 1431 data = RLC_SAFE_MODE__CMD_MASK; 1432 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SAFE_MODE, data); 1433} 1434 1435static void gfx_v9_4_3_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev) 1436{ 1437 int xcc_id, num_xcc; 1438 struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl; 1439 1440 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1441 for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { 1442 reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl[GET_INST(GC, xcc_id)]; 1443 reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG0); 1444 reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG1); 1445 reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG2); 1446 reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG3); 1447 reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regGRBM_GFX_CNTL); 1448 reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regGRBM_GFX_INDEX); 1449 reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regRLC_SPARE_INT); 1450 } 1451 adev->gfx.rlc.rlcg_reg_access_supported = true; 1452} 1453 1454static int gfx_v9_4_3_rlc_init(struct amdgpu_device *adev) 1455{ 1456 /* init spm vmid with 0xf */ 1457 if (adev->gfx.rlc.funcs->update_spm_vmid) 1458 adev->gfx.rlc.funcs->update_spm_vmid(adev, NULL, 0xf); 1459 1460 return 0; 1461} 1462 1463static void gfx_v9_4_3_xcc_wait_for_rlc_serdes(struct amdgpu_device *adev, 1464 int xcc_id) 1465{ 1466 u32 i, j, k; 1467 u32 mask; 1468 1469 mutex_lock(&adev->grbm_idx_mutex); 1470 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1471 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1472 gfx_v9_4_3_xcc_select_se_sh(adev, i, j, 0xffffffff, 1473 xcc_id); 1474 for (k = 0; k < adev->usec_timeout; k++) { 1475 if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SERDES_CU_MASTER_BUSY) == 0) 1476 break; 1477 udelay(1); 1478 } 1479 if (k == adev->usec_timeout) { 1480 gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 1481 0xffffffff, 1482 0xffffffff, xcc_id); 1483 mutex_unlock(&adev->grbm_idx_mutex); 1484 DRM_INFO("Timeout wait for RLC serdes %u,%u\n", 1485 i, j); 1486 return; 1487 } 1488 } 1489 } 1490 gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 1491 xcc_id); 1492 mutex_unlock(&adev->grbm_idx_mutex); 1493 1494 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK | 1495 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK | 1496 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK | 1497 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK; 1498 for (k = 0; k < adev->usec_timeout; k++) { 1499 if ((RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0) 1500 break; 1501 udelay(1); 1502 } 1503} 1504 1505static void gfx_v9_4_3_xcc_enable_gui_idle_interrupt(struct amdgpu_device *adev, 1506 bool enable, int xcc_id) 1507{ 1508 u32 tmp; 1509 1510 /* These interrupts should be enabled to drive DS clock */ 1511 1512 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_INT_CNTL_RING0); 1513 1514 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0); 1515 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0); 1516 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0); 1517 1518 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_INT_CNTL_RING0, tmp); 1519} 1520 1521static void gfx_v9_4_3_xcc_rlc_stop(struct amdgpu_device *adev, int xcc_id) 1522{ 1523 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), RLC_CNTL, 1524 RLC_ENABLE_F32, 0); 1525 gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, false, xcc_id); 1526 gfx_v9_4_3_xcc_wait_for_rlc_serdes(adev, xcc_id); 1527} 1528 1529static void gfx_v9_4_3_rlc_stop(struct amdgpu_device *adev) 1530{ 1531 int i, num_xcc; 1532 1533 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1534 for (i = 0; i < num_xcc; i++) 1535 gfx_v9_4_3_xcc_rlc_stop(adev, i); 1536} 1537 1538static void gfx_v9_4_3_xcc_rlc_reset(struct amdgpu_device *adev, int xcc_id) 1539{ 1540 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), GRBM_SOFT_RESET, 1541 SOFT_RESET_RLC, 1); 1542 udelay(50); 1543 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), GRBM_SOFT_RESET, 1544 SOFT_RESET_RLC, 0); 1545 udelay(50); 1546} 1547 1548static void gfx_v9_4_3_rlc_reset(struct amdgpu_device *adev) 1549{ 1550 int i, num_xcc; 1551 1552 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1553 for (i = 0; i < num_xcc; i++) 1554 gfx_v9_4_3_xcc_rlc_reset(adev, i); 1555} 1556 1557static void gfx_v9_4_3_xcc_rlc_start(struct amdgpu_device *adev, int xcc_id) 1558{ 1559 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), RLC_CNTL, 1560 RLC_ENABLE_F32, 1); 1561 udelay(50); 1562 1563 /* carrizo do enable cp interrupt after cp inited */ 1564 if (!(adev->flags & AMD_IS_APU)) { 1565 gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, true, xcc_id); 1566 udelay(50); 1567 } 1568} 1569 1570static void gfx_v9_4_3_rlc_start(struct amdgpu_device *adev) 1571{ 1572#ifdef AMDGPU_RLC_DEBUG_RETRY 1573 u32 rlc_ucode_ver; 1574#endif 1575 int i, num_xcc; 1576 1577 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1578 for (i = 0; i < num_xcc; i++) { 1579 gfx_v9_4_3_xcc_rlc_start(adev, i); 1580#ifdef AMDGPU_RLC_DEBUG_RETRY 1581 /* RLC_GPM_GENERAL_6 : RLC Ucode version */ 1582 rlc_ucode_ver = RREG32_SOC15(GC, GET_INST(GC, i), regRLC_GPM_GENERAL_6); 1583 if (rlc_ucode_ver == 0x108) { 1584 dev_info(adev->dev, 1585 "Using rlc debug ucode. regRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n", 1586 rlc_ucode_ver, adev->gfx.rlc_fw_version); 1587 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles, 1588 * default is 0x9C4 to create a 100us interval */ 1589 WREG32_SOC15(GC, GET_INST(GC, i), regRLC_GPM_TIMER_INT_3, 0x9C4); 1590 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr 1591 * to disable the page fault retry interrupts, default is 1592 * 0x100 (256) */ 1593 WREG32_SOC15(GC, GET_INST(GC, i), regRLC_GPM_GENERAL_12, 0x100); 1594 } 1595#endif 1596 } 1597} 1598 1599static int gfx_v9_4_3_xcc_rlc_load_microcode(struct amdgpu_device *adev, 1600 int xcc_id) 1601{ 1602 const struct rlc_firmware_header_v2_0 *hdr; 1603 const __le32 *fw_data; 1604 unsigned i, fw_size; 1605 1606 if (!adev->gfx.rlc_fw) 1607 return -EINVAL; 1608 1609 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; 1610 amdgpu_ucode_print_rlc_hdr(&hdr->header); 1611 1612 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data + 1613 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 1614 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 1615 1616 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_GPM_UCODE_ADDR, 1617 RLCG_UCODE_LOADING_START_ADDRESS); 1618 for (i = 0; i < fw_size; i++) { 1619 if (amdgpu_emu_mode == 1 && i % 100 == 0) { 1620 dev_info(adev->dev, "Write RLC ucode data %u DWs\n", i); 1621 msleep(1); 1622 } 1623 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++)); 1624 } 1625 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version); 1626 1627 return 0; 1628} 1629 1630static int gfx_v9_4_3_xcc_rlc_resume(struct amdgpu_device *adev, int xcc_id) 1631{ 1632 int r; 1633 1634 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { 1635 gfx_v9_4_3_xcc_rlc_stop(adev, xcc_id); 1636 /* legacy rlc firmware loading */ 1637 r = gfx_v9_4_3_xcc_rlc_load_microcode(adev, xcc_id); 1638 if (r) 1639 return r; 1640 gfx_v9_4_3_xcc_rlc_start(adev, xcc_id); 1641 } 1642 1643 amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id); 1644 /* disable CG */ 1645 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, 0); 1646 gfx_v9_4_3_xcc_init_pg(adev, xcc_id); 1647 amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id); 1648 1649 return 0; 1650} 1651 1652static int gfx_v9_4_3_rlc_resume(struct amdgpu_device *adev) 1653{ 1654 int r, i, num_xcc; 1655 1656 if (amdgpu_sriov_vf(adev)) 1657 return 0; 1658 1659 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 1660 for (i = 0; i < num_xcc; i++) { 1661 r = gfx_v9_4_3_xcc_rlc_resume(adev, i); 1662 if (r) 1663 return r; 1664 } 1665 1666 return 0; 1667} 1668 1669static void gfx_v9_4_3_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, 1670 unsigned vmid) 1671{ 1672 u32 reg, pre_data, data; 1673 1674 reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL); 1675 if (amdgpu_sriov_is_pp_one_vf(adev) && !amdgpu_sriov_runtime(adev)) 1676 pre_data = RREG32_NO_KIQ(reg); 1677 else 1678 pre_data = RREG32(reg); 1679 1680 data = pre_data & (~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK); 1681 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT; 1682 1683 if (pre_data != data) { 1684 if (amdgpu_sriov_is_pp_one_vf(adev) && !amdgpu_sriov_runtime(adev)) { 1685 WREG32_SOC15_NO_KIQ(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data); 1686 } else 1687 WREG32_SOC15(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data); 1688 } 1689} 1690 1691static const struct soc15_reg_rlcg rlcg_access_gc_9_4_3[] = { 1692 {SOC15_REG_ENTRY(GC, 0, regGRBM_GFX_INDEX)}, 1693 {SOC15_REG_ENTRY(GC, 0, regSQ_IND_INDEX)}, 1694}; 1695 1696static bool gfx_v9_4_3_check_rlcg_range(struct amdgpu_device *adev, 1697 uint32_t offset, 1698 struct soc15_reg_rlcg *entries, int arr_size) 1699{ 1700 int i, inst; 1701 uint32_t reg; 1702 1703 if (!entries) 1704 return false; 1705 1706 for (i = 0; i < arr_size; i++) { 1707 const struct soc15_reg_rlcg *entry; 1708 1709 entry = &entries[i]; 1710 inst = adev->ip_map.logical_to_dev_inst ? 1711 adev->ip_map.logical_to_dev_inst( 1712 adev, entry->hwip, entry->instance) : 1713 entry->instance; 1714 reg = adev->reg_offset[entry->hwip][inst][entry->segment] + 1715 entry->reg; 1716 if (offset == reg) 1717 return true; 1718 } 1719 1720 return false; 1721} 1722 1723static bool gfx_v9_4_3_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset) 1724{ 1725 return gfx_v9_4_3_check_rlcg_range(adev, offset, 1726 (void *)rlcg_access_gc_9_4_3, 1727 ARRAY_SIZE(rlcg_access_gc_9_4_3)); 1728} 1729 1730static void gfx_v9_4_3_xcc_cp_compute_enable(struct amdgpu_device *adev, 1731 bool enable, int xcc_id) 1732{ 1733 if (enable) { 1734 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MEC_CNTL, 0); 1735 } else { 1736 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MEC_CNTL, 1737 (CP_MEC_CNTL__MEC_INVALIDATE_ICACHE_MASK | 1738 CP_MEC_CNTL__MEC_ME1_PIPE0_RESET_MASK | 1739 CP_MEC_CNTL__MEC_ME1_PIPE1_RESET_MASK | 1740 CP_MEC_CNTL__MEC_ME1_PIPE2_RESET_MASK | 1741 CP_MEC_CNTL__MEC_ME1_PIPE3_RESET_MASK | 1742 CP_MEC_CNTL__MEC_ME2_PIPE0_RESET_MASK | 1743 CP_MEC_CNTL__MEC_ME2_PIPE1_RESET_MASK | 1744 CP_MEC_CNTL__MEC_ME1_HALT_MASK | 1745 CP_MEC_CNTL__MEC_ME2_HALT_MASK)); 1746 adev->gfx.kiq[xcc_id].ring.sched.ready = false; 1747 } 1748 udelay(50); 1749} 1750 1751static int gfx_v9_4_3_xcc_cp_compute_load_microcode(struct amdgpu_device *adev, 1752 int xcc_id) 1753{ 1754 const struct gfx_firmware_header_v1_0 *mec_hdr; 1755 const __le32 *fw_data; 1756 unsigned i; 1757 u32 tmp; 1758 u32 mec_ucode_addr_offset; 1759 u32 mec_ucode_data_offset; 1760 1761 if (!adev->gfx.mec_fw) 1762 return -EINVAL; 1763 1764 gfx_v9_4_3_xcc_cp_compute_enable(adev, false, xcc_id); 1765 1766 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 1767 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header); 1768 1769 fw_data = (const __le32 *) 1770 (adev->gfx.mec_fw->data + 1771 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)); 1772 tmp = 0; 1773 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0); 1774 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0); 1775 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_IC_BASE_CNTL, tmp); 1776 1777 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_IC_BASE_LO, 1778 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000); 1779 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_IC_BASE_HI, 1780 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr)); 1781 1782 mec_ucode_addr_offset = 1783 SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_MEC_ME1_UCODE_ADDR); 1784 mec_ucode_data_offset = 1785 SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_MEC_ME1_UCODE_DATA); 1786 1787 /* MEC1 */ 1788 WREG32(mec_ucode_addr_offset, mec_hdr->jt_offset); 1789 for (i = 0; i < mec_hdr->jt_size; i++) 1790 WREG32(mec_ucode_data_offset, 1791 le32_to_cpup(fw_data + mec_hdr->jt_offset + i)); 1792 1793 WREG32(mec_ucode_addr_offset, adev->gfx.mec_fw_version); 1794 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */ 1795 1796 return 0; 1797} 1798 1799/* KIQ functions */ 1800static void gfx_v9_4_3_xcc_kiq_setting(struct amdgpu_ring *ring, int xcc_id) 1801{ 1802 uint32_t tmp; 1803 struct amdgpu_device *adev = ring->adev; 1804 1805 /* tell RLC which is KIQ queue */ 1806 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS); 1807 tmp &= 0xffffff00; 1808 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue); 1809 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp | 0x80); 1810} 1811 1812static void gfx_v9_4_3_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd) 1813{ 1814 struct amdgpu_device *adev = ring->adev; 1815 1816 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 1817 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) { 1818 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH; 1819 mqd->cp_hqd_queue_priority = 1820 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM; 1821 } 1822 } 1823} 1824 1825static int gfx_v9_4_3_xcc_mqd_init(struct amdgpu_ring *ring, int xcc_id) 1826{ 1827 struct amdgpu_device *adev = ring->adev; 1828 struct v9_mqd *mqd = ring->mqd_ptr; 1829 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr; 1830 uint32_t tmp; 1831 1832 mqd->header = 0xC0310800; 1833 mqd->compute_pipelinestat_enable = 0x00000001; 1834 mqd->compute_static_thread_mgmt_se0 = 0xffffffff; 1835 mqd->compute_static_thread_mgmt_se1 = 0xffffffff; 1836 mqd->compute_static_thread_mgmt_se2 = 0xffffffff; 1837 mqd->compute_static_thread_mgmt_se3 = 0xffffffff; 1838 mqd->compute_misc_reserved = 0x00000003; 1839 1840 mqd->dynamic_cu_mask_addr_lo = 1841 lower_32_bits(ring->mqd_gpu_addr 1842 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask)); 1843 mqd->dynamic_cu_mask_addr_hi = 1844 upper_32_bits(ring->mqd_gpu_addr 1845 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask)); 1846 1847 eop_base_addr = ring->eop_gpu_addr >> 8; 1848 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr; 1849 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr); 1850 1851 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 1852 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_CONTROL); 1853 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE, 1854 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1)); 1855 1856 mqd->cp_hqd_eop_control = tmp; 1857 1858 /* enable doorbell? */ 1859 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL); 1860 1861 if (ring->use_doorbell) { 1862 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 1863 DOORBELL_OFFSET, ring->doorbell_index); 1864 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 1865 DOORBELL_EN, 1); 1866 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 1867 DOORBELL_SOURCE, 0); 1868 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 1869 DOORBELL_HIT, 0); 1870 if (amdgpu_sriov_multi_vf_mode(adev)) 1871 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 1872 DOORBELL_MODE, 1); 1873 } else { 1874 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 1875 DOORBELL_EN, 0); 1876 } 1877 1878 mqd->cp_hqd_pq_doorbell_control = tmp; 1879 1880 /* disable the queue if it's active */ 1881 ring->wptr = 0; 1882 mqd->cp_hqd_dequeue_request = 0; 1883 mqd->cp_hqd_pq_rptr = 0; 1884 mqd->cp_hqd_pq_wptr_lo = 0; 1885 mqd->cp_hqd_pq_wptr_hi = 0; 1886 1887 /* set the pointer to the MQD */ 1888 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc; 1889 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr); 1890 1891 /* set MQD vmid to 0 */ 1892 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MQD_CONTROL); 1893 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0); 1894 mqd->cp_mqd_control = tmp; 1895 1896 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */ 1897 hqd_gpu_addr = ring->gpu_addr >> 8; 1898 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr; 1899 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr); 1900 1901 /* set up the HQD, this is similar to CP_RB0_CNTL */ 1902 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_CONTROL); 1903 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, 1904 (order_base_2(ring->ring_size / 4) - 1)); 1905 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, 1906 ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8)); 1907#ifdef __BIG_ENDIAN 1908 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); 1909#endif 1910 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0); 1911 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0); 1912 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1); 1913 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1); 1914 mqd->cp_hqd_pq_control = tmp; 1915 1916 /* set the wb address whether it's enabled or not */ 1917 wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4); 1918 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc; 1919 mqd->cp_hqd_pq_rptr_report_addr_hi = 1920 upper_32_bits(wb_gpu_addr) & 0xffff; 1921 1922 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */ 1923 wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4); 1924 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc; 1925 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff; 1926 1927 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 1928 ring->wptr = 0; 1929 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR); 1930 1931 /* set the vmid for the queue */ 1932 mqd->cp_hqd_vmid = 0; 1933 1934 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE); 1935 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53); 1936 mqd->cp_hqd_persistent_state = tmp; 1937 1938 /* set MIN_IB_AVAIL_SIZE */ 1939 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_IB_CONTROL); 1940 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3); 1941 mqd->cp_hqd_ib_control = tmp; 1942 1943 /* set static priority for a queue/ring */ 1944 gfx_v9_4_3_mqd_set_priority(ring, mqd); 1945 mqd->cp_hqd_quantum = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_QUANTUM); 1946 1947 /* map_queues packet doesn't need activate the queue, 1948 * so only kiq need set this field. 1949 */ 1950 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) 1951 mqd->cp_hqd_active = 1; 1952 1953 return 0; 1954} 1955 1956static int gfx_v9_4_3_xcc_kiq_init_register(struct amdgpu_ring *ring, 1957 int xcc_id) 1958{ 1959 struct amdgpu_device *adev = ring->adev; 1960 struct v9_mqd *mqd = ring->mqd_ptr; 1961 int j; 1962 1963 /* disable wptr polling */ 1964 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), CP_PQ_WPTR_POLL_CNTL, EN, 0); 1965 1966 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_BASE_ADDR, 1967 mqd->cp_hqd_eop_base_addr_lo); 1968 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_BASE_ADDR_HI, 1969 mqd->cp_hqd_eop_base_addr_hi); 1970 1971 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 1972 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_CONTROL, 1973 mqd->cp_hqd_eop_control); 1974 1975 /* enable doorbell? */ 1976 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 1977 mqd->cp_hqd_pq_doorbell_control); 1978 1979 /* disable the queue if it's active */ 1980 if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1) { 1981 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 1); 1982 for (j = 0; j < adev->usec_timeout; j++) { 1983 if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) 1984 break; 1985 udelay(1); 1986 } 1987 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 1988 mqd->cp_hqd_dequeue_request); 1989 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR, 1990 mqd->cp_hqd_pq_rptr); 1991 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO, 1992 mqd->cp_hqd_pq_wptr_lo); 1993 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI, 1994 mqd->cp_hqd_pq_wptr_hi); 1995 } 1996 1997 /* set the pointer to the MQD */ 1998 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MQD_BASE_ADDR, 1999 mqd->cp_mqd_base_addr_lo); 2000 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MQD_BASE_ADDR_HI, 2001 mqd->cp_mqd_base_addr_hi); 2002 2003 /* set MQD vmid to 0 */ 2004 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MQD_CONTROL, 2005 mqd->cp_mqd_control); 2006 2007 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */ 2008 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_BASE, 2009 mqd->cp_hqd_pq_base_lo); 2010 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_BASE_HI, 2011 mqd->cp_hqd_pq_base_hi); 2012 2013 /* set up the HQD, this is similar to CP_RB0_CNTL */ 2014 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_CONTROL, 2015 mqd->cp_hqd_pq_control); 2016 2017 /* set the wb address whether it's enabled or not */ 2018 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR_REPORT_ADDR, 2019 mqd->cp_hqd_pq_rptr_report_addr_lo); 2020 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR_REPORT_ADDR_HI, 2021 mqd->cp_hqd_pq_rptr_report_addr_hi); 2022 2023 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */ 2024 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_POLL_ADDR, 2025 mqd->cp_hqd_pq_wptr_poll_addr_lo); 2026 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_POLL_ADDR_HI, 2027 mqd->cp_hqd_pq_wptr_poll_addr_hi); 2028 2029 /* enable the doorbell if requested */ 2030 if (ring->use_doorbell) { 2031 WREG32_SOC15( 2032 GC, GET_INST(GC, xcc_id), 2033 regCP_MEC_DOORBELL_RANGE_LOWER, 2034 ((adev->doorbell_index.kiq + 2035 xcc_id * adev->doorbell_index.xcc_doorbell_range) * 2036 2) << 2); 2037 WREG32_SOC15( 2038 GC, GET_INST(GC, xcc_id), 2039 regCP_MEC_DOORBELL_RANGE_UPPER, 2040 ((adev->doorbell_index.userqueue_end + 2041 xcc_id * adev->doorbell_index.xcc_doorbell_range) * 2042 2) << 2); 2043 } 2044 2045 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 2046 mqd->cp_hqd_pq_doorbell_control); 2047 2048 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 2049 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO, 2050 mqd->cp_hqd_pq_wptr_lo); 2051 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI, 2052 mqd->cp_hqd_pq_wptr_hi); 2053 2054 /* set the vmid for the queue */ 2055 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_VMID, mqd->cp_hqd_vmid); 2056 2057 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE, 2058 mqd->cp_hqd_persistent_state); 2059 2060 /* activate the queue */ 2061 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE, 2062 mqd->cp_hqd_active); 2063 2064 if (ring->use_doorbell) 2065 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), CP_PQ_STATUS, DOORBELL_ENABLE, 1); 2066 2067 return 0; 2068} 2069 2070static int gfx_v9_4_3_xcc_q_fini_register(struct amdgpu_ring *ring, 2071 int xcc_id) 2072{ 2073 struct amdgpu_device *adev = ring->adev; 2074 int j; 2075 2076 /* disable the queue if it's active */ 2077 if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1) { 2078 2079 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 1); 2080 2081 for (j = 0; j < adev->usec_timeout; j++) { 2082 if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) 2083 break; 2084 udelay(1); 2085 } 2086 2087 if (j == AMDGPU_MAX_USEC_TIMEOUT) { 2088 DRM_DEBUG("%s dequeue request failed.\n", ring->name); 2089 2090 /* Manual disable if dequeue request times out */ 2091 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE, 0); 2092 } 2093 2094 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 2095 0); 2096 } 2097 2098 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_IQ_TIMER, 0); 2099 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_IB_CONTROL, 0); 2100 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE, CP_HQD_PERSISTENT_STATE_DEFAULT); 2101 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000); 2102 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 0); 2103 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR, 0); 2104 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI, 0); 2105 WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO, 0); 2106 2107 return 0; 2108} 2109 2110static int gfx_v9_4_3_xcc_kiq_init_queue(struct amdgpu_ring *ring, int xcc_id) 2111{ 2112 struct amdgpu_device *adev = ring->adev; 2113 struct v9_mqd *mqd = ring->mqd_ptr; 2114 struct v9_mqd *tmp_mqd; 2115 2116 gfx_v9_4_3_xcc_kiq_setting(ring, xcc_id); 2117 2118 /* GPU could be in bad state during probe, driver trigger the reset 2119 * after load the SMU, in this case , the mqd is not be initialized. 2120 * driver need to re-init the mqd. 2121 * check mqd->cp_hqd_pq_control since this value should not be 0 2122 */ 2123 tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[xcc_id].mqd_backup; 2124 if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control) { 2125 /* for GPU_RESET case , reset MQD to a clean status */ 2126 if (adev->gfx.kiq[xcc_id].mqd_backup) 2127 memcpy(mqd, adev->gfx.kiq[xcc_id].mqd_backup, sizeof(struct v9_mqd_allocation)); 2128 2129 /* reset ring buffer */ 2130 ring->wptr = 0; 2131 amdgpu_ring_clear_ring(ring); 2132 mutex_lock(&adev->srbm_mutex); 2133 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, xcc_id)); 2134 gfx_v9_4_3_xcc_kiq_init_register(ring, xcc_id); 2135 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 2136 mutex_unlock(&adev->srbm_mutex); 2137 } else { 2138 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation)); 2139 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF; 2140 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF; 2141 mutex_lock(&adev->srbm_mutex); 2142 if (amdgpu_sriov_vf(adev) && adev->in_suspend) 2143 amdgpu_ring_clear_ring(ring); 2144 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, xcc_id)); 2145 gfx_v9_4_3_xcc_mqd_init(ring, xcc_id); 2146 gfx_v9_4_3_xcc_kiq_init_register(ring, xcc_id); 2147 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 2148 mutex_unlock(&adev->srbm_mutex); 2149 2150 if (adev->gfx.kiq[xcc_id].mqd_backup) 2151 memcpy(adev->gfx.kiq[xcc_id].mqd_backup, mqd, sizeof(struct v9_mqd_allocation)); 2152 } 2153 2154 return 0; 2155} 2156 2157static void gfx_v9_4_3_xcc_kcq_init_queue(struct amdgpu_ring *ring, int xcc_id, 2158 bool restore) 2159{ 2160 struct amdgpu_device *adev = ring->adev; 2161 struct v9_mqd *mqd = ring->mqd_ptr; 2162 int mqd_idx = ring - &adev->gfx.compute_ring[0]; 2163 struct v9_mqd *tmp_mqd; 2164 2165 /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control 2166 * is not be initialized before 2167 */ 2168 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx]; 2169 2170 if (!restore && (!tmp_mqd->cp_hqd_pq_control || 2171 (!amdgpu_in_reset(adev) && !adev->in_suspend))) { 2172 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation)); 2173 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF; 2174 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF; 2175 mutex_lock(&adev->srbm_mutex); 2176 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, xcc_id)); 2177 gfx_v9_4_3_xcc_mqd_init(ring, xcc_id); 2178 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 2179 mutex_unlock(&adev->srbm_mutex); 2180 2181 if (adev->gfx.mec.mqd_backup[mqd_idx]) 2182 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation)); 2183 } else { 2184 /* restore MQD to a clean status */ 2185 if (adev->gfx.mec.mqd_backup[mqd_idx]) 2186 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation)); 2187 /* reset ring buffer */ 2188 ring->wptr = 0; 2189 atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], 0); 2190 amdgpu_ring_clear_ring(ring); 2191 } 2192} 2193 2194static int gfx_v9_4_3_xcc_kcq_fini_register(struct amdgpu_device *adev, int xcc_id) 2195{ 2196 struct amdgpu_ring *ring; 2197 int j; 2198 2199 for (j = 0; j < adev->gfx.num_compute_rings; j++) { 2200 ring = &adev->gfx.compute_ring[j + xcc_id * adev->gfx.num_compute_rings]; 2201 if (!amdgpu_in_reset(adev) && !adev->in_suspend) { 2202 mutex_lock(&adev->srbm_mutex); 2203 soc15_grbm_select(adev, ring->me, 2204 ring->pipe, 2205 ring->queue, 0, GET_INST(GC, xcc_id)); 2206 gfx_v9_4_3_xcc_q_fini_register(ring, xcc_id); 2207 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 2208 mutex_unlock(&adev->srbm_mutex); 2209 } 2210 } 2211 2212 return 0; 2213} 2214 2215static int gfx_v9_4_3_xcc_kiq_resume(struct amdgpu_device *adev, int xcc_id) 2216{ 2217 gfx_v9_4_3_xcc_kiq_init_queue(&adev->gfx.kiq[xcc_id].ring, xcc_id); 2218 return 0; 2219} 2220 2221static int gfx_v9_4_3_xcc_kcq_resume(struct amdgpu_device *adev, int xcc_id) 2222{ 2223 struct amdgpu_ring *ring; 2224 int i; 2225 2226 gfx_v9_4_3_xcc_cp_compute_enable(adev, true, xcc_id); 2227 2228 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 2229 ring = &adev->gfx.compute_ring[i + xcc_id * 2230 adev->gfx.num_compute_rings]; 2231 2232 gfx_v9_4_3_xcc_kcq_init_queue(ring, xcc_id, false); 2233 } 2234 2235 return amdgpu_gfx_enable_kcq(adev, xcc_id); 2236} 2237 2238static int gfx_v9_4_3_xcc_cp_resume(struct amdgpu_device *adev, int xcc_id) 2239{ 2240 struct amdgpu_ring *ring; 2241 int r, j; 2242 2243 gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, false, xcc_id); 2244 2245 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { 2246 gfx_v9_4_3_xcc_disable_gpa_mode(adev, xcc_id); 2247 2248 r = gfx_v9_4_3_xcc_cp_compute_load_microcode(adev, xcc_id); 2249 if (r) 2250 return r; 2251 } else { 2252 gfx_v9_4_3_xcc_cp_compute_enable(adev, false, xcc_id); 2253 } 2254 2255 r = gfx_v9_4_3_xcc_kiq_resume(adev, xcc_id); 2256 if (r) 2257 return r; 2258 2259 r = gfx_v9_4_3_xcc_kcq_resume(adev, xcc_id); 2260 if (r) 2261 return r; 2262 2263 for (j = 0; j < adev->gfx.num_compute_rings; j++) { 2264 ring = &adev->gfx.compute_ring 2265 [j + xcc_id * adev->gfx.num_compute_rings]; 2266 r = amdgpu_ring_test_helper(ring); 2267 if (r) 2268 return r; 2269 } 2270 2271 gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, true, xcc_id); 2272 2273 return 0; 2274} 2275 2276static int gfx_v9_4_3_cp_resume(struct amdgpu_device *adev) 2277{ 2278 int r = 0, i, num_xcc, num_xcp, num_xcc_per_xcp; 2279 2280 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 2281 if (amdgpu_sriov_vf(adev)) { 2282 enum amdgpu_gfx_partition mode; 2283 2284 mode = amdgpu_xcp_query_partition_mode(adev->xcp_mgr, 2285 AMDGPU_XCP_FL_NONE); 2286 if (mode == AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) 2287 return -EINVAL; 2288 num_xcc_per_xcp = gfx_v9_4_3_get_xccs_per_xcp(adev); 2289 adev->gfx.num_xcc_per_xcp = num_xcc_per_xcp; 2290 num_xcp = num_xcc / num_xcc_per_xcp; 2291 r = amdgpu_xcp_init(adev->xcp_mgr, num_xcp, mode); 2292 2293 } else { 2294 if (adev->in_suspend) 2295 amdgpu_xcp_restore_partition_mode(adev->xcp_mgr); 2296 else if (amdgpu_xcp_query_partition_mode(adev->xcp_mgr, 2297 AMDGPU_XCP_FL_NONE) == 2298 AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) 2299 r = amdgpu_xcp_switch_partition_mode( 2300 adev->xcp_mgr, amdgpu_user_partt_mode); 2301 } 2302 if (r) 2303 return r; 2304 2305 for (i = 0; i < num_xcc; i++) { 2306 r = gfx_v9_4_3_xcc_cp_resume(adev, i); 2307 if (r) 2308 return r; 2309 } 2310 2311 return 0; 2312} 2313 2314static void gfx_v9_4_3_xcc_fini(struct amdgpu_device *adev, int xcc_id) 2315{ 2316 if (amdgpu_gfx_disable_kcq(adev, xcc_id)) 2317 DRM_ERROR("XCD %d KCQ disable failed\n", xcc_id); 2318 2319 if (amdgpu_sriov_vf(adev)) { 2320 /* must disable polling for SRIOV when hw finished, otherwise 2321 * CPC engine may still keep fetching WB address which is already 2322 * invalid after sw finished and trigger DMAR reading error in 2323 * hypervisor side. 2324 */ 2325 WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), CP_PQ_WPTR_POLL_CNTL, EN, 0); 2326 return; 2327 } 2328 2329 /* Use deinitialize sequence from CAIL when unbinding device 2330 * from driver, otherwise KIQ is hanging when binding back 2331 */ 2332 if (!amdgpu_in_reset(adev) && !adev->in_suspend) { 2333 mutex_lock(&adev->srbm_mutex); 2334 soc15_grbm_select(adev, adev->gfx.kiq[xcc_id].ring.me, 2335 adev->gfx.kiq[xcc_id].ring.pipe, 2336 adev->gfx.kiq[xcc_id].ring.queue, 0, 2337 GET_INST(GC, xcc_id)); 2338 gfx_v9_4_3_xcc_q_fini_register(&adev->gfx.kiq[xcc_id].ring, 2339 xcc_id); 2340 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 2341 mutex_unlock(&adev->srbm_mutex); 2342 } 2343 2344 gfx_v9_4_3_xcc_kcq_fini_register(adev, xcc_id); 2345 gfx_v9_4_3_xcc_cp_compute_enable(adev, false, xcc_id); 2346} 2347 2348static int gfx_v9_4_3_hw_init(struct amdgpu_ip_block *ip_block) 2349{ 2350 int r; 2351 struct amdgpu_device *adev = ip_block->adev; 2352 2353 amdgpu_gfx_cleaner_shader_init(adev, adev->gfx.cleaner_shader_size, 2354 adev->gfx.cleaner_shader_ptr); 2355 2356 if (!amdgpu_sriov_vf(adev)) 2357 gfx_v9_4_3_init_golden_registers(adev); 2358 2359 gfx_v9_4_3_constants_init(adev); 2360 2361 r = adev->gfx.rlc.funcs->resume(adev); 2362 if (r) 2363 return r; 2364 2365 r = gfx_v9_4_3_cp_resume(adev); 2366 if (r) 2367 return r; 2368 2369 return r; 2370} 2371 2372static int gfx_v9_4_3_hw_fini(struct amdgpu_ip_block *ip_block) 2373{ 2374 struct amdgpu_device *adev = ip_block->adev; 2375 int i, num_xcc; 2376 2377 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); 2378 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); 2379 amdgpu_irq_put(adev, &adev->gfx.bad_op_irq, 0); 2380 2381 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 2382 for (i = 0; i < num_xcc; i++) { 2383 gfx_v9_4_3_xcc_fini(adev, i); 2384 } 2385 2386 return 0; 2387} 2388 2389static int gfx_v9_4_3_suspend(struct amdgpu_ip_block *ip_block) 2390{ 2391 return gfx_v9_4_3_hw_fini(ip_block); 2392} 2393 2394static int gfx_v9_4_3_resume(struct amdgpu_ip_block *ip_block) 2395{ 2396 return gfx_v9_4_3_hw_init(ip_block); 2397} 2398 2399static bool gfx_v9_4_3_is_idle(struct amdgpu_ip_block *ip_block) 2400{ 2401 struct amdgpu_device *adev = ip_block->adev; 2402 int i, num_xcc; 2403 2404 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 2405 for (i = 0; i < num_xcc; i++) { 2406 if (REG_GET_FIELD(RREG32_SOC15(GC, GET_INST(GC, i), regGRBM_STATUS), 2407 GRBM_STATUS, GUI_ACTIVE)) 2408 return false; 2409 } 2410 return true; 2411} 2412 2413static int gfx_v9_4_3_wait_for_idle(struct amdgpu_ip_block *ip_block) 2414{ 2415 unsigned i; 2416 struct amdgpu_device *adev = ip_block->adev; 2417 2418 for (i = 0; i < adev->usec_timeout; i++) { 2419 if (gfx_v9_4_3_is_idle(ip_block)) 2420 return 0; 2421 udelay(1); 2422 } 2423 return -ETIMEDOUT; 2424} 2425 2426static int gfx_v9_4_3_soft_reset(struct amdgpu_ip_block *ip_block) 2427{ 2428 u32 grbm_soft_reset = 0; 2429 u32 tmp; 2430 struct amdgpu_device *adev = ip_block->adev; 2431 2432 /* GRBM_STATUS */ 2433 tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_STATUS); 2434 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | 2435 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | 2436 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | 2437 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | 2438 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | 2439 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) { 2440 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 2441 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 2442 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 2443 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1); 2444 } 2445 2446 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) { 2447 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 2448 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 2449 } 2450 2451 /* GRBM_STATUS2 */ 2452 tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_STATUS2); 2453 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY)) 2454 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 2455 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 2456 2457 2458 if (grbm_soft_reset) { 2459 /* stop the rlc */ 2460 adev->gfx.rlc.funcs->stop(adev); 2461 2462 /* Disable MEC parsing/prefetching */ 2463 gfx_v9_4_3_xcc_cp_compute_enable(adev, false, 0); 2464 2465 tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET); 2466 tmp |= grbm_soft_reset; 2467 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 2468 WREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET, tmp); 2469 tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET); 2470 2471 udelay(50); 2472 2473 tmp &= ~grbm_soft_reset; 2474 WREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET, tmp); 2475 tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET); 2476 2477 /* Wait a little for things to settle down */ 2478 udelay(50); 2479 } 2480 return 0; 2481} 2482 2483static void gfx_v9_4_3_ring_emit_gds_switch(struct amdgpu_ring *ring, 2484 uint32_t vmid, 2485 uint32_t gds_base, uint32_t gds_size, 2486 uint32_t gws_base, uint32_t gws_size, 2487 uint32_t oa_base, uint32_t oa_size) 2488{ 2489 struct amdgpu_device *adev = ring->adev; 2490 2491 /* GDS Base */ 2492 gfx_v9_4_3_write_data_to_reg(ring, 0, false, 2493 SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_VMID0_BASE) + 2 * vmid, 2494 gds_base); 2495 2496 /* GDS Size */ 2497 gfx_v9_4_3_write_data_to_reg(ring, 0, false, 2498 SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_VMID0_SIZE) + 2 * vmid, 2499 gds_size); 2500 2501 /* GWS */ 2502 gfx_v9_4_3_write_data_to_reg(ring, 0, false, 2503 SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_GWS_VMID0) + vmid, 2504 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base); 2505 2506 /* OA */ 2507 gfx_v9_4_3_write_data_to_reg(ring, 0, false, 2508 SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_OA_VMID0) + vmid, 2509 (1 << (oa_size + oa_base)) - (1 << oa_base)); 2510} 2511 2512static int gfx_v9_4_3_early_init(struct amdgpu_ip_block *ip_block) 2513{ 2514 struct amdgpu_device *adev = ip_block->adev; 2515 2516 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), 2517 AMDGPU_MAX_COMPUTE_RINGS); 2518 gfx_v9_4_3_set_kiq_pm4_funcs(adev); 2519 gfx_v9_4_3_set_ring_funcs(adev); 2520 gfx_v9_4_3_set_irq_funcs(adev); 2521 gfx_v9_4_3_set_gds_init(adev); 2522 gfx_v9_4_3_set_rlc_funcs(adev); 2523 2524 /* init rlcg reg access ctrl */ 2525 gfx_v9_4_3_init_rlcg_reg_access_ctrl(adev); 2526 2527 return gfx_v9_4_3_init_microcode(adev); 2528} 2529 2530static int gfx_v9_4_3_late_init(struct amdgpu_ip_block *ip_block) 2531{ 2532 struct amdgpu_device *adev = ip_block->adev; 2533 int r; 2534 2535 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); 2536 if (r) 2537 return r; 2538 2539 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); 2540 if (r) 2541 return r; 2542 2543 r = amdgpu_irq_get(adev, &adev->gfx.bad_op_irq, 0); 2544 if (r) 2545 return r; 2546 2547 if (adev->gfx.ras && 2548 adev->gfx.ras->enable_watchdog_timer) 2549 adev->gfx.ras->enable_watchdog_timer(adev); 2550 2551 return 0; 2552} 2553 2554static void gfx_v9_4_3_xcc_update_sram_fgcg(struct amdgpu_device *adev, 2555 bool enable, int xcc_id) 2556{ 2557 uint32_t def, data; 2558 2559 if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_FGCG)) 2560 return; 2561 2562 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), 2563 regRLC_CGTT_MGCG_OVERRIDE); 2564 2565 if (enable) 2566 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK; 2567 else 2568 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK; 2569 2570 if (def != data) 2571 WREG32_SOC15(GC, GET_INST(GC, xcc_id), 2572 regRLC_CGTT_MGCG_OVERRIDE, data); 2573 2574} 2575 2576static void gfx_v9_4_3_xcc_update_repeater_fgcg(struct amdgpu_device *adev, 2577 bool enable, int xcc_id) 2578{ 2579 uint32_t def, data; 2580 2581 if (!(adev->cg_flags & AMD_CG_SUPPORT_REPEATER_FGCG)) 2582 return; 2583 2584 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), 2585 regRLC_CGTT_MGCG_OVERRIDE); 2586 2587 if (enable) 2588 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_REP_FGCG_OVERRIDE_MASK; 2589 else 2590 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_REP_FGCG_OVERRIDE_MASK; 2591 2592 if (def != data) 2593 WREG32_SOC15(GC, GET_INST(GC, xcc_id), 2594 regRLC_CGTT_MGCG_OVERRIDE, data); 2595} 2596 2597static void 2598gfx_v9_4_3_xcc_update_medium_grain_clock_gating(struct amdgpu_device *adev, 2599 bool enable, int xcc_id) 2600{ 2601 uint32_t data, def; 2602 2603 /* It is disabled by HW by default */ 2604 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 2605 /* 1 - RLC_CGTT_MGCG_OVERRIDE */ 2606 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE); 2607 2608 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK | 2609 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK | 2610 RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK | 2611 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK); 2612 2613 if (def != data) 2614 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data); 2615 2616 /* MGLS is a global flag to control all MGLS in GFX */ 2617 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 2618 /* 2 - RLC memory Light sleep */ 2619 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) { 2620 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL); 2621 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 2622 if (def != data) 2623 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL, data); 2624 } 2625 /* 3 - CP memory Light sleep */ 2626 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 2627 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL); 2628 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 2629 if (def != data) 2630 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL, data); 2631 } 2632 } 2633 } else { 2634 /* 1 - MGCG_OVERRIDE */ 2635 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE); 2636 2637 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK | 2638 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK | 2639 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK | 2640 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK); 2641 2642 if (def != data) 2643 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data); 2644 2645 /* 2 - disable MGLS in RLC */ 2646 data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL); 2647 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) { 2648 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 2649 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL, data); 2650 } 2651 2652 /* 3 - disable MGLS in CP */ 2653 data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL); 2654 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 2655 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 2656 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL, data); 2657 } 2658 } 2659 2660} 2661 2662static void 2663gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(struct amdgpu_device *adev, 2664 bool enable, int xcc_id) 2665{ 2666 uint32_t def, data; 2667 2668 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 2669 2670 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE); 2671 /* unset CGCG override */ 2672 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK; 2673 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) 2674 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK; 2675 else 2676 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK; 2677 /* update CGCG and CGLS override bits */ 2678 if (def != data) 2679 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data); 2680 2681 /* CGCG Hysteresis: 400us */ 2682 def = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL); 2683 2684 data = (0x2710 2685 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 2686 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; 2687 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) 2688 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) | 2689 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 2690 if (def != data) 2691 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, data); 2692 2693 /* set IDLE_POLL_COUNT(0x33450100)*/ 2694 def = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_RB_WPTR_POLL_CNTL); 2695 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) | 2696 (0x3345 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 2697 if (def != data) 2698 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_RB_WPTR_POLL_CNTL, data); 2699 } else { 2700 def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL); 2701 /* reset CGCG/CGLS bits */ 2702 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 2703 /* disable cgcg and cgls in FSM */ 2704 if (def != data) 2705 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, data); 2706 } 2707 2708} 2709 2710static int gfx_v9_4_3_xcc_update_gfx_clock_gating(struct amdgpu_device *adev, 2711 bool enable, int xcc_id) 2712{ 2713 amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id); 2714 2715 if (enable) { 2716 /* FGCG */ 2717 gfx_v9_4_3_xcc_update_sram_fgcg(adev, enable, xcc_id); 2718 gfx_v9_4_3_xcc_update_repeater_fgcg(adev, enable, xcc_id); 2719 2720 /* CGCG/CGLS should be enabled after MGCG/MGLS 2721 * === MGCG + MGLS === 2722 */ 2723 gfx_v9_4_3_xcc_update_medium_grain_clock_gating(adev, enable, 2724 xcc_id); 2725 /* === CGCG + CGLS === */ 2726 gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(adev, enable, 2727 xcc_id); 2728 } else { 2729 /* CGCG/CGLS should be disabled before MGCG/MGLS 2730 * === CGCG + CGLS === 2731 */ 2732 gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(adev, enable, 2733 xcc_id); 2734 /* === MGCG + MGLS === */ 2735 gfx_v9_4_3_xcc_update_medium_grain_clock_gating(adev, enable, 2736 xcc_id); 2737 2738 /* FGCG */ 2739 gfx_v9_4_3_xcc_update_sram_fgcg(adev, enable, xcc_id); 2740 gfx_v9_4_3_xcc_update_repeater_fgcg(adev, enable, xcc_id); 2741 } 2742 2743 amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id); 2744 2745 return 0; 2746} 2747 2748static const struct amdgpu_rlc_funcs gfx_v9_4_3_rlc_funcs = { 2749 .is_rlc_enabled = gfx_v9_4_3_is_rlc_enabled, 2750 .set_safe_mode = gfx_v9_4_3_xcc_set_safe_mode, 2751 .unset_safe_mode = gfx_v9_4_3_xcc_unset_safe_mode, 2752 .init = gfx_v9_4_3_rlc_init, 2753 .resume = gfx_v9_4_3_rlc_resume, 2754 .stop = gfx_v9_4_3_rlc_stop, 2755 .reset = gfx_v9_4_3_rlc_reset, 2756 .start = gfx_v9_4_3_rlc_start, 2757 .update_spm_vmid = gfx_v9_4_3_update_spm_vmid, 2758 .is_rlcg_access_range = gfx_v9_4_3_is_rlcg_access_range, 2759}; 2760 2761static int gfx_v9_4_3_set_powergating_state(struct amdgpu_ip_block *ip_block, 2762 enum amd_powergating_state state) 2763{ 2764 return 0; 2765} 2766 2767static int gfx_v9_4_3_set_clockgating_state(struct amdgpu_ip_block *ip_block, 2768 enum amd_clockgating_state state) 2769{ 2770 struct amdgpu_device *adev = ip_block->adev; 2771 int i, num_xcc; 2772 2773 if (amdgpu_sriov_vf(adev)) 2774 return 0; 2775 2776 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 2777 for (i = 0; i < num_xcc; i++) 2778 gfx_v9_4_3_xcc_update_gfx_clock_gating( 2779 adev, state == AMD_CG_STATE_GATE, i); 2780 2781 return 0; 2782} 2783 2784static void gfx_v9_4_3_get_clockgating_state(struct amdgpu_ip_block *ip_block, u64 *flags) 2785{ 2786 struct amdgpu_device *adev = ip_block->adev; 2787 int data; 2788 2789 if (amdgpu_sriov_vf(adev)) 2790 *flags = 0; 2791 2792 /* AMD_CG_SUPPORT_GFX_MGCG */ 2793 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_CGTT_MGCG_OVERRIDE)); 2794 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK)) 2795 *flags |= AMD_CG_SUPPORT_GFX_MGCG; 2796 2797 /* AMD_CG_SUPPORT_GFX_CGCG */ 2798 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_CGCG_CGLS_CTRL)); 2799 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK) 2800 *flags |= AMD_CG_SUPPORT_GFX_CGCG; 2801 2802 /* AMD_CG_SUPPORT_GFX_CGLS */ 2803 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK) 2804 *flags |= AMD_CG_SUPPORT_GFX_CGLS; 2805 2806 /* AMD_CG_SUPPORT_GFX_RLC_LS */ 2807 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_MEM_SLP_CNTL)); 2808 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) 2809 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS; 2810 2811 /* AMD_CG_SUPPORT_GFX_CP_LS */ 2812 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regCP_MEM_SLP_CNTL)); 2813 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) 2814 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS; 2815} 2816 2817static void gfx_v9_4_3_ring_emit_hdp_flush(struct amdgpu_ring *ring) 2818{ 2819 struct amdgpu_device *adev = ring->adev; 2820 u32 ref_and_mask, reg_mem_engine; 2821 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg; 2822 2823 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 2824 switch (ring->me) { 2825 case 1: 2826 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe; 2827 break; 2828 case 2: 2829 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe; 2830 break; 2831 default: 2832 return; 2833 } 2834 reg_mem_engine = 0; 2835 } else { 2836 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0; 2837 reg_mem_engine = 1; /* pfp */ 2838 } 2839 2840 gfx_v9_4_3_wait_reg_mem(ring, reg_mem_engine, 0, 1, 2841 adev->nbio.funcs->get_hdp_flush_req_offset(adev), 2842 adev->nbio.funcs->get_hdp_flush_done_offset(adev), 2843 ref_and_mask, ref_and_mask, 0x20); 2844} 2845 2846static void gfx_v9_4_3_ring_emit_ib_compute(struct amdgpu_ring *ring, 2847 struct amdgpu_job *job, 2848 struct amdgpu_ib *ib, 2849 uint32_t flags) 2850{ 2851 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 2852 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); 2853 2854 /* Currently, there is a high possibility to get wave ID mismatch 2855 * between ME and GDS, leading to a hw deadlock, because ME generates 2856 * different wave IDs than the GDS expects. This situation happens 2857 * randomly when at least 5 compute pipes use GDS ordered append. 2858 * The wave IDs generated by ME are also wrong after suspend/resume. 2859 * Those are probably bugs somewhere else in the kernel driver. 2860 * 2861 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and 2862 * GDS to 0 for this ring (me/pipe). 2863 */ 2864 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { 2865 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 2866 amdgpu_ring_write(ring, regGDS_COMPUTE_MAX_WAVE_ID); 2867 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); 2868 } 2869 2870 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); 2871 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */ 2872 amdgpu_ring_write(ring, 2873#ifdef __BIG_ENDIAN 2874 (2 << 0) | 2875#endif 2876 lower_32_bits(ib->gpu_addr)); 2877 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 2878 amdgpu_ring_write(ring, control); 2879} 2880 2881static void gfx_v9_4_3_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 2882 u64 seq, unsigned flags) 2883{ 2884 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 2885 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 2886 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY; 2887 2888 /* RELEASE_MEM - flush caches, send int */ 2889 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6)); 2890 amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN | 2891 EOP_TC_NC_ACTION_EN) : 2892 (EOP_TCL1_ACTION_EN | 2893 EOP_TC_ACTION_EN | 2894 EOP_TC_WB_ACTION_EN | 2895 EOP_TC_MD_ACTION_EN)) | 2896 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 2897 EVENT_INDEX(5))); 2898 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 2899 2900 /* 2901 * the address should be Qword aligned if 64bit write, Dword 2902 * aligned if only send 32bit data low (discard data high) 2903 */ 2904 if (write64bit) 2905 BUG_ON(addr & 0x7); 2906 else 2907 BUG_ON(addr & 0x3); 2908 amdgpu_ring_write(ring, lower_32_bits(addr)); 2909 amdgpu_ring_write(ring, upper_32_bits(addr)); 2910 amdgpu_ring_write(ring, lower_32_bits(seq)); 2911 amdgpu_ring_write(ring, upper_32_bits(seq)); 2912 amdgpu_ring_write(ring, 0); 2913} 2914 2915static void gfx_v9_4_3_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 2916{ 2917 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 2918 uint32_t seq = ring->fence_drv.sync_seq; 2919 uint64_t addr = ring->fence_drv.gpu_addr; 2920 2921 gfx_v9_4_3_wait_reg_mem(ring, usepfp, 1, 0, 2922 lower_32_bits(addr), upper_32_bits(addr), 2923 seq, 0xffffffff, 4); 2924} 2925 2926static void gfx_v9_4_3_ring_emit_vm_flush(struct amdgpu_ring *ring, 2927 unsigned vmid, uint64_t pd_addr) 2928{ 2929 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 2930} 2931 2932static u64 gfx_v9_4_3_ring_get_rptr_compute(struct amdgpu_ring *ring) 2933{ 2934 return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 hardware is 32bit rptr */ 2935} 2936 2937static u64 gfx_v9_4_3_ring_get_wptr_compute(struct amdgpu_ring *ring) 2938{ 2939 u64 wptr; 2940 2941 /* XXX check if swapping is necessary on BE */ 2942 if (ring->use_doorbell) 2943 wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]); 2944 else 2945 BUG(); 2946 return wptr; 2947} 2948 2949static void gfx_v9_4_3_ring_set_wptr_compute(struct amdgpu_ring *ring) 2950{ 2951 struct amdgpu_device *adev = ring->adev; 2952 2953 /* XXX check if swapping is necessary on BE */ 2954 if (ring->use_doorbell) { 2955 atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], ring->wptr); 2956 WDOORBELL64(ring->doorbell_index, ring->wptr); 2957 } else { 2958 BUG(); /* only DOORBELL method supported on gfx9 now */ 2959 } 2960} 2961 2962static void gfx_v9_4_3_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr, 2963 u64 seq, unsigned int flags) 2964{ 2965 struct amdgpu_device *adev = ring->adev; 2966 2967 /* we only allocate 32bit for each seq wb address */ 2968 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 2969 2970 /* write fence seq to the "addr" */ 2971 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 2972 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 2973 WRITE_DATA_DST_SEL(5) | WR_CONFIRM)); 2974 amdgpu_ring_write(ring, lower_32_bits(addr)); 2975 amdgpu_ring_write(ring, upper_32_bits(addr)); 2976 amdgpu_ring_write(ring, lower_32_bits(seq)); 2977 2978 if (flags & AMDGPU_FENCE_FLAG_INT) { 2979 /* set register to trigger INT */ 2980 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 2981 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 2982 WRITE_DATA_DST_SEL(0) | WR_CONFIRM)); 2983 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regCPC_INT_STATUS)); 2984 amdgpu_ring_write(ring, 0); 2985 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */ 2986 } 2987} 2988 2989static void gfx_v9_4_3_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg, 2990 uint32_t reg_val_offs) 2991{ 2992 struct amdgpu_device *adev = ring->adev; 2993 2994 reg = gfx_v9_4_3_normalize_xcc_reg_offset(reg); 2995 2996 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); 2997 amdgpu_ring_write(ring, 0 | /* src: register*/ 2998 (5 << 8) | /* dst: memory */ 2999 (1 << 20)); /* write confirm */ 3000 amdgpu_ring_write(ring, reg); 3001 amdgpu_ring_write(ring, 0); 3002 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + 3003 reg_val_offs * 4)); 3004 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + 3005 reg_val_offs * 4)); 3006} 3007 3008static void gfx_v9_4_3_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, 3009 uint32_t val) 3010{ 3011 uint32_t cmd = 0; 3012 3013 reg = gfx_v9_4_3_normalize_xcc_reg_offset(reg); 3014 3015 switch (ring->funcs->type) { 3016 case AMDGPU_RING_TYPE_GFX: 3017 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM; 3018 break; 3019 case AMDGPU_RING_TYPE_KIQ: 3020 cmd = (1 << 16); /* no inc addr */ 3021 break; 3022 default: 3023 cmd = WR_CONFIRM; 3024 break; 3025 } 3026 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 3027 amdgpu_ring_write(ring, cmd); 3028 amdgpu_ring_write(ring, reg); 3029 amdgpu_ring_write(ring, 0); 3030 amdgpu_ring_write(ring, val); 3031} 3032 3033static void gfx_v9_4_3_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg, 3034 uint32_t val, uint32_t mask) 3035{ 3036 gfx_v9_4_3_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20); 3037} 3038 3039static void gfx_v9_4_3_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, 3040 uint32_t reg0, uint32_t reg1, 3041 uint32_t ref, uint32_t mask) 3042{ 3043 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1, 3044 ref, mask); 3045} 3046 3047static void gfx_v9_4_3_ring_soft_recovery(struct amdgpu_ring *ring, 3048 unsigned vmid) 3049{ 3050 struct amdgpu_device *adev = ring->adev; 3051 uint32_t value = 0; 3052 3053 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); 3054 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); 3055 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); 3056 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); 3057 amdgpu_gfx_rlc_enter_safe_mode(adev, ring->xcc_id); 3058 WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regSQ_CMD, value); 3059 amdgpu_gfx_rlc_exit_safe_mode(adev, ring->xcc_id); 3060} 3061 3062static void gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3063 struct amdgpu_device *adev, int me, int pipe, 3064 enum amdgpu_interrupt_state state, int xcc_id) 3065{ 3066 u32 mec_int_cntl, mec_int_cntl_reg; 3067 3068 /* 3069 * amdgpu controls only the first MEC. That's why this function only 3070 * handles the setting of interrupts for this specific MEC. All other 3071 * pipes' interrupts are set by amdkfd. 3072 */ 3073 3074 if (me == 1) { 3075 switch (pipe) { 3076 case 0: 3077 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE0_INT_CNTL); 3078 break; 3079 case 1: 3080 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE1_INT_CNTL); 3081 break; 3082 case 2: 3083 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE2_INT_CNTL); 3084 break; 3085 case 3: 3086 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE3_INT_CNTL); 3087 break; 3088 default: 3089 DRM_DEBUG("invalid pipe %d\n", pipe); 3090 return; 3091 } 3092 } else { 3093 DRM_DEBUG("invalid me %d\n", me); 3094 return; 3095 } 3096 3097 switch (state) { 3098 case AMDGPU_IRQ_STATE_DISABLE: 3099 mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, xcc_id); 3100 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 3101 TIME_STAMP_INT_ENABLE, 0); 3102 WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, xcc_id); 3103 break; 3104 case AMDGPU_IRQ_STATE_ENABLE: 3105 mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, xcc_id); 3106 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 3107 TIME_STAMP_INT_ENABLE, 1); 3108 WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, xcc_id); 3109 break; 3110 default: 3111 break; 3112 } 3113} 3114 3115static u32 gfx_v9_4_3_get_cpc_int_cntl(struct amdgpu_device *adev, 3116 int xcc_id, int me, int pipe) 3117{ 3118 /* 3119 * amdgpu controls only the first MEC. That's why this function only 3120 * handles the setting of interrupts for this specific MEC. All other 3121 * pipes' interrupts are set by amdkfd. 3122 */ 3123 if (me != 1) 3124 return 0; 3125 3126 switch (pipe) { 3127 case 0: 3128 return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE0_INT_CNTL); 3129 case 1: 3130 return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE1_INT_CNTL); 3131 case 2: 3132 return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE2_INT_CNTL); 3133 case 3: 3134 return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE3_INT_CNTL); 3135 default: 3136 return 0; 3137 } 3138} 3139 3140static int gfx_v9_4_3_set_priv_reg_fault_state(struct amdgpu_device *adev, 3141 struct amdgpu_irq_src *source, 3142 unsigned type, 3143 enum amdgpu_interrupt_state state) 3144{ 3145 u32 mec_int_cntl_reg, mec_int_cntl; 3146 int i, j, k, num_xcc; 3147 3148 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 3149 switch (state) { 3150 case AMDGPU_IRQ_STATE_DISABLE: 3151 case AMDGPU_IRQ_STATE_ENABLE: 3152 for (i = 0; i < num_xcc; i++) { 3153 WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0, 3154 PRIV_REG_INT_ENABLE, 3155 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 3156 for (j = 0; j < adev->gfx.mec.num_mec; j++) { 3157 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { 3158 /* MECs start at 1 */ 3159 mec_int_cntl_reg = gfx_v9_4_3_get_cpc_int_cntl(adev, i, j + 1, k); 3160 3161 if (mec_int_cntl_reg) { 3162 mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, i); 3163 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 3164 PRIV_REG_INT_ENABLE, 3165 state == AMDGPU_IRQ_STATE_ENABLE ? 3166 1 : 0); 3167 WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, i); 3168 } 3169 } 3170 } 3171 } 3172 break; 3173 default: 3174 break; 3175 } 3176 3177 return 0; 3178} 3179 3180static int gfx_v9_4_3_set_bad_op_fault_state(struct amdgpu_device *adev, 3181 struct amdgpu_irq_src *source, 3182 unsigned type, 3183 enum amdgpu_interrupt_state state) 3184{ 3185 u32 mec_int_cntl_reg, mec_int_cntl; 3186 int i, j, k, num_xcc; 3187 3188 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 3189 switch (state) { 3190 case AMDGPU_IRQ_STATE_DISABLE: 3191 case AMDGPU_IRQ_STATE_ENABLE: 3192 for (i = 0; i < num_xcc; i++) { 3193 WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0, 3194 OPCODE_ERROR_INT_ENABLE, 3195 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 3196 for (j = 0; j < adev->gfx.mec.num_mec; j++) { 3197 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { 3198 /* MECs start at 1 */ 3199 mec_int_cntl_reg = gfx_v9_4_3_get_cpc_int_cntl(adev, i, j + 1, k); 3200 3201 if (mec_int_cntl_reg) { 3202 mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, i); 3203 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 3204 OPCODE_ERROR_INT_ENABLE, 3205 state == AMDGPU_IRQ_STATE_ENABLE ? 3206 1 : 0); 3207 WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, i); 3208 } 3209 } 3210 } 3211 } 3212 break; 3213 default: 3214 break; 3215 } 3216 3217 return 0; 3218} 3219 3220static int gfx_v9_4_3_set_priv_inst_fault_state(struct amdgpu_device *adev, 3221 struct amdgpu_irq_src *source, 3222 unsigned type, 3223 enum amdgpu_interrupt_state state) 3224{ 3225 int i, num_xcc; 3226 3227 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 3228 switch (state) { 3229 case AMDGPU_IRQ_STATE_DISABLE: 3230 case AMDGPU_IRQ_STATE_ENABLE: 3231 for (i = 0; i < num_xcc; i++) 3232 WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0, 3233 PRIV_INSTR_INT_ENABLE, 3234 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 3235 break; 3236 default: 3237 break; 3238 } 3239 3240 return 0; 3241} 3242 3243static int gfx_v9_4_3_set_eop_interrupt_state(struct amdgpu_device *adev, 3244 struct amdgpu_irq_src *src, 3245 unsigned type, 3246 enum amdgpu_interrupt_state state) 3247{ 3248 int i, num_xcc; 3249 3250 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 3251 for (i = 0; i < num_xcc; i++) { 3252 switch (type) { 3253 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 3254 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3255 adev, 1, 0, state, i); 3256 break; 3257 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 3258 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3259 adev, 1, 1, state, i); 3260 break; 3261 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP: 3262 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3263 adev, 1, 2, state, i); 3264 break; 3265 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP: 3266 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3267 adev, 1, 3, state, i); 3268 break; 3269 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP: 3270 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3271 adev, 2, 0, state, i); 3272 break; 3273 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP: 3274 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3275 adev, 2, 1, state, i); 3276 break; 3277 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP: 3278 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3279 adev, 2, 2, state, i); 3280 break; 3281 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP: 3282 gfx_v9_4_3_xcc_set_compute_eop_interrupt_state( 3283 adev, 2, 3, state, i); 3284 break; 3285 default: 3286 break; 3287 } 3288 } 3289 3290 return 0; 3291} 3292 3293static int gfx_v9_4_3_eop_irq(struct amdgpu_device *adev, 3294 struct amdgpu_irq_src *source, 3295 struct amdgpu_iv_entry *entry) 3296{ 3297 int i, xcc_id; 3298 u8 me_id, pipe_id, queue_id; 3299 struct amdgpu_ring *ring; 3300 3301 DRM_DEBUG("IH: CP EOP\n"); 3302 me_id = (entry->ring_id & 0x0c) >> 2; 3303 pipe_id = (entry->ring_id & 0x03) >> 0; 3304 queue_id = (entry->ring_id & 0x70) >> 4; 3305 3306 xcc_id = gfx_v9_4_3_ih_to_xcc_inst(adev, entry->node_id); 3307 3308 if (xcc_id == -EINVAL) 3309 return -EINVAL; 3310 3311 switch (me_id) { 3312 case 0: 3313 case 1: 3314 case 2: 3315 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3316 ring = &adev->gfx.compute_ring 3317 [i + 3318 xcc_id * adev->gfx.num_compute_rings]; 3319 /* Per-queue interrupt is supported for MEC starting from VI. 3320 * The interrupt can only be enabled/disabled per pipe instead of per queue. 3321 */ 3322 3323 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id)) 3324 amdgpu_fence_process(ring); 3325 } 3326 break; 3327 } 3328 return 0; 3329} 3330 3331static void gfx_v9_4_3_fault(struct amdgpu_device *adev, 3332 struct amdgpu_iv_entry *entry) 3333{ 3334 u8 me_id, pipe_id, queue_id; 3335 struct amdgpu_ring *ring; 3336 int i, xcc_id; 3337 3338 me_id = (entry->ring_id & 0x0c) >> 2; 3339 pipe_id = (entry->ring_id & 0x03) >> 0; 3340 queue_id = (entry->ring_id & 0x70) >> 4; 3341 3342 xcc_id = gfx_v9_4_3_ih_to_xcc_inst(adev, entry->node_id); 3343 3344 if (xcc_id == -EINVAL) 3345 return; 3346 3347 switch (me_id) { 3348 case 0: 3349 case 1: 3350 case 2: 3351 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3352 ring = &adev->gfx.compute_ring 3353 [i + 3354 xcc_id * adev->gfx.num_compute_rings]; 3355 if (ring->me == me_id && ring->pipe == pipe_id && 3356 ring->queue == queue_id) 3357 drm_sched_fault(&ring->sched); 3358 } 3359 break; 3360 } 3361} 3362 3363static int gfx_v9_4_3_priv_reg_irq(struct amdgpu_device *adev, 3364 struct amdgpu_irq_src *source, 3365 struct amdgpu_iv_entry *entry) 3366{ 3367 DRM_ERROR("Illegal register access in command stream\n"); 3368 gfx_v9_4_3_fault(adev, entry); 3369 return 0; 3370} 3371 3372static int gfx_v9_4_3_bad_op_irq(struct amdgpu_device *adev, 3373 struct amdgpu_irq_src *source, 3374 struct amdgpu_iv_entry *entry) 3375{ 3376 DRM_ERROR("Illegal opcode in command stream\n"); 3377 gfx_v9_4_3_fault(adev, entry); 3378 return 0; 3379} 3380 3381static int gfx_v9_4_3_priv_inst_irq(struct amdgpu_device *adev, 3382 struct amdgpu_irq_src *source, 3383 struct amdgpu_iv_entry *entry) 3384{ 3385 DRM_ERROR("Illegal instruction in command stream\n"); 3386 gfx_v9_4_3_fault(adev, entry); 3387 return 0; 3388} 3389 3390static void gfx_v9_4_3_emit_mem_sync(struct amdgpu_ring *ring) 3391{ 3392 const unsigned int cp_coher_cntl = 3393 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) | 3394 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) | 3395 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) | 3396 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) | 3397 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1); 3398 3399 /* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */ 3400 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5)); 3401 amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */ 3402 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 3403 amdgpu_ring_write(ring, 0xffffff); /* CP_COHER_SIZE_HI */ 3404 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 3405 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */ 3406 amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */ 3407} 3408 3409static void gfx_v9_4_3_emit_wave_limit_cs(struct amdgpu_ring *ring, 3410 uint32_t pipe, bool enable) 3411{ 3412 struct amdgpu_device *adev = ring->adev; 3413 uint32_t val; 3414 uint32_t wcl_cs_reg; 3415 3416 /* regSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */ 3417 val = enable ? 0x1 : 0x7f; 3418 3419 switch (pipe) { 3420 case 0: 3421 wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS0); 3422 break; 3423 case 1: 3424 wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS1); 3425 break; 3426 case 2: 3427 wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS2); 3428 break; 3429 case 3: 3430 wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS3); 3431 break; 3432 default: 3433 DRM_DEBUG("invalid pipe %d\n", pipe); 3434 return; 3435 } 3436 3437 amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val); 3438 3439} 3440static void gfx_v9_4_3_emit_wave_limit(struct amdgpu_ring *ring, bool enable) 3441{ 3442 struct amdgpu_device *adev = ring->adev; 3443 uint32_t val; 3444 int i; 3445 3446 /* regSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit 3447 * number of gfx waves. Setting 5 bit will make sure gfx only gets 3448 * around 25% of gpu resources. 3449 */ 3450 val = enable ? 0x1f : 0x07ffffff; 3451 amdgpu_ring_emit_wreg(ring, 3452 SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_GFX), 3453 val); 3454 3455 /* Restrict waves for normal/low priority compute queues as well 3456 * to get best QoS for high priority compute jobs. 3457 * 3458 * amdgpu controls only 1st ME(0-3 CS pipes). 3459 */ 3460 for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) { 3461 if (i != ring->pipe) 3462 gfx_v9_4_3_emit_wave_limit_cs(ring, i, enable); 3463 3464 } 3465} 3466 3467static int gfx_v9_4_3_unmap_done(struct amdgpu_device *adev, uint32_t me, 3468 uint32_t pipe, uint32_t queue, 3469 uint32_t xcc_id) 3470{ 3471 int i, r; 3472 /* make sure dequeue is complete*/ 3473 gfx_v9_4_3_xcc_set_safe_mode(adev, xcc_id); 3474 mutex_lock(&adev->srbm_mutex); 3475 soc15_grbm_select(adev, me, pipe, queue, 0, GET_INST(GC, xcc_id)); 3476 for (i = 0; i < adev->usec_timeout; i++) { 3477 if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1)) 3478 break; 3479 udelay(1); 3480 } 3481 if (i >= adev->usec_timeout) 3482 r = -ETIMEDOUT; 3483 else 3484 r = 0; 3485 soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id)); 3486 mutex_unlock(&adev->srbm_mutex); 3487 gfx_v9_4_3_xcc_unset_safe_mode(adev, xcc_id); 3488 3489 return r; 3490 3491} 3492 3493static bool gfx_v9_4_3_pipe_reset_support(struct amdgpu_device *adev) 3494{ 3495 if (!!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_PIPE)) 3496 return true; 3497 else 3498 dev_warn_once(adev->dev, "Please use the latest MEC version to see whether support pipe reset\n"); 3499 3500 return false; 3501} 3502 3503static int gfx_v9_4_3_reset_hw_pipe(struct amdgpu_ring *ring) 3504{ 3505 struct amdgpu_device *adev = ring->adev; 3506 uint32_t reset_pipe, clean_pipe; 3507 int r; 3508 3509 if (!gfx_v9_4_3_pipe_reset_support(adev)) 3510 return -EINVAL; 3511 3512 gfx_v9_4_3_xcc_set_safe_mode(adev, ring->xcc_id); 3513 mutex_lock(&adev->srbm_mutex); 3514 3515 reset_pipe = RREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL); 3516 clean_pipe = reset_pipe; 3517 3518 if (ring->me == 1) { 3519 switch (ring->pipe) { 3520 case 0: 3521 reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, 3522 MEC_ME1_PIPE0_RESET, 1); 3523 break; 3524 case 1: 3525 reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, 3526 MEC_ME1_PIPE1_RESET, 1); 3527 break; 3528 case 2: 3529 reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, 3530 MEC_ME1_PIPE2_RESET, 1); 3531 break; 3532 case 3: 3533 reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, 3534 MEC_ME1_PIPE3_RESET, 1); 3535 break; 3536 default: 3537 break; 3538 } 3539 } else { 3540 if (ring->pipe) 3541 reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, 3542 MEC_ME2_PIPE1_RESET, 1); 3543 else 3544 reset_pipe = REG_SET_FIELD(reset_pipe, CP_MEC_CNTL, 3545 MEC_ME2_PIPE0_RESET, 1); 3546 } 3547 3548 WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL, reset_pipe); 3549 WREG32_SOC15(GC, GET_INST(GC, ring->xcc_id), regCP_MEC_CNTL, clean_pipe); 3550 mutex_unlock(&adev->srbm_mutex); 3551 gfx_v9_4_3_xcc_unset_safe_mode(adev, ring->xcc_id); 3552 3553 r = gfx_v9_4_3_unmap_done(adev, ring->me, ring->pipe, ring->queue, ring->xcc_id); 3554 return r; 3555} 3556 3557static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring, 3558 unsigned int vmid, 3559 struct amdgpu_fence *timedout_fence) 3560{ 3561 struct amdgpu_device *adev = ring->adev; 3562 struct amdgpu_kiq *kiq = &adev->gfx.kiq[ring->xcc_id]; 3563 struct amdgpu_ring *kiq_ring = &kiq->ring; 3564 int reset_mode = AMDGPU_RESET_TYPE_PER_QUEUE; 3565 unsigned long flags; 3566 int r; 3567 3568 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) 3569 return -EINVAL; 3570 3571 amdgpu_ring_reset_helper_begin(ring, timedout_fence); 3572 3573 spin_lock_irqsave(&kiq->ring_lock, flags); 3574 3575 if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) { 3576 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3577 return -ENOMEM; 3578 } 3579 3580 kiq->pmf->kiq_unmap_queues(kiq_ring, ring, RESET_QUEUES, 3581 0, 0); 3582 amdgpu_ring_commit(kiq_ring); 3583 3584 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3585 3586 r = amdgpu_ring_test_ring(kiq_ring); 3587 if (r) { 3588 dev_err(adev->dev, "kiq ring test failed after ring: %s queue reset\n", 3589 ring->name); 3590 goto pipe_reset; 3591 } 3592 3593 r = gfx_v9_4_3_unmap_done(adev, ring->me, ring->pipe, ring->queue, ring->xcc_id); 3594 if (r) 3595 dev_err(adev->dev, "fail to wait on hqd deactive and will try pipe reset\n"); 3596 3597pipe_reset: 3598 if (r) { 3599 if (!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_PIPE)) 3600 return -EOPNOTSUPP; 3601 r = gfx_v9_4_3_reset_hw_pipe(ring); 3602 reset_mode = AMDGPU_RESET_TYPE_PER_PIPE; 3603 dev_info(adev->dev, "ring: %s pipe reset :%s\n", ring->name, 3604 r ? "failed" : "successfully"); 3605 if (r) 3606 return r; 3607 } 3608 3609 gfx_v9_4_3_xcc_kcq_init_queue(ring, ring->xcc_id, true); 3610 3611 spin_lock_irqsave(&kiq->ring_lock, flags); 3612 r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size); 3613 if (r) { 3614 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3615 return -ENOMEM; 3616 } 3617 kiq->pmf->kiq_map_queues(kiq_ring, ring); 3618 amdgpu_ring_commit(kiq_ring); 3619 r = amdgpu_ring_test_ring(kiq_ring); 3620 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3621 if (r) { 3622 if (reset_mode == AMDGPU_RESET_TYPE_PER_QUEUE) 3623 goto pipe_reset; 3624 3625 dev_err(adev->dev, "fail to remap queue\n"); 3626 return r; 3627 } 3628 3629 if (reset_mode == AMDGPU_RESET_TYPE_PER_QUEUE) { 3630 r = amdgpu_ring_test_ring(ring); 3631 if (r) 3632 goto pipe_reset; 3633 } 3634 3635 3636 return amdgpu_ring_reset_helper_end(ring, timedout_fence); 3637} 3638 3639enum amdgpu_gfx_cp_ras_mem_id { 3640 AMDGPU_GFX_CP_MEM1 = 1, 3641 AMDGPU_GFX_CP_MEM2, 3642 AMDGPU_GFX_CP_MEM3, 3643 AMDGPU_GFX_CP_MEM4, 3644 AMDGPU_GFX_CP_MEM5, 3645}; 3646 3647enum amdgpu_gfx_gcea_ras_mem_id { 3648 AMDGPU_GFX_GCEA_IOWR_CMDMEM = 4, 3649 AMDGPU_GFX_GCEA_IORD_CMDMEM, 3650 AMDGPU_GFX_GCEA_GMIWR_CMDMEM, 3651 AMDGPU_GFX_GCEA_GMIRD_CMDMEM, 3652 AMDGPU_GFX_GCEA_DRAMWR_CMDMEM, 3653 AMDGPU_GFX_GCEA_DRAMRD_CMDMEM, 3654 AMDGPU_GFX_GCEA_MAM_DMEM0, 3655 AMDGPU_GFX_GCEA_MAM_DMEM1, 3656 AMDGPU_GFX_GCEA_MAM_DMEM2, 3657 AMDGPU_GFX_GCEA_MAM_DMEM3, 3658 AMDGPU_GFX_GCEA_MAM_AMEM0, 3659 AMDGPU_GFX_GCEA_MAM_AMEM1, 3660 AMDGPU_GFX_GCEA_MAM_AMEM2, 3661 AMDGPU_GFX_GCEA_MAM_AMEM3, 3662 AMDGPU_GFX_GCEA_MAM_AFLUSH_BUFFER, 3663 AMDGPU_GFX_GCEA_WRET_TAGMEM, 3664 AMDGPU_GFX_GCEA_RRET_TAGMEM, 3665 AMDGPU_GFX_GCEA_IOWR_DATAMEM, 3666 AMDGPU_GFX_GCEA_GMIWR_DATAMEM, 3667 AMDGPU_GFX_GCEA_DRAM_DATAMEM, 3668}; 3669 3670enum amdgpu_gfx_gc_cane_ras_mem_id { 3671 AMDGPU_GFX_GC_CANE_MEM0 = 0, 3672}; 3673 3674enum amdgpu_gfx_gcutcl2_ras_mem_id { 3675 AMDGPU_GFX_GCUTCL2_MEM2P512X95 = 160, 3676}; 3677 3678enum amdgpu_gfx_gds_ras_mem_id { 3679 AMDGPU_GFX_GDS_MEM0 = 0, 3680}; 3681 3682enum amdgpu_gfx_lds_ras_mem_id { 3683 AMDGPU_GFX_LDS_BANK0 = 0, 3684 AMDGPU_GFX_LDS_BANK1, 3685 AMDGPU_GFX_LDS_BANK2, 3686 AMDGPU_GFX_LDS_BANK3, 3687 AMDGPU_GFX_LDS_BANK4, 3688 AMDGPU_GFX_LDS_BANK5, 3689 AMDGPU_GFX_LDS_BANK6, 3690 AMDGPU_GFX_LDS_BANK7, 3691 AMDGPU_GFX_LDS_BANK8, 3692 AMDGPU_GFX_LDS_BANK9, 3693 AMDGPU_GFX_LDS_BANK10, 3694 AMDGPU_GFX_LDS_BANK11, 3695 AMDGPU_GFX_LDS_BANK12, 3696 AMDGPU_GFX_LDS_BANK13, 3697 AMDGPU_GFX_LDS_BANK14, 3698 AMDGPU_GFX_LDS_BANK15, 3699 AMDGPU_GFX_LDS_BANK16, 3700 AMDGPU_GFX_LDS_BANK17, 3701 AMDGPU_GFX_LDS_BANK18, 3702 AMDGPU_GFX_LDS_BANK19, 3703 AMDGPU_GFX_LDS_BANK20, 3704 AMDGPU_GFX_LDS_BANK21, 3705 AMDGPU_GFX_LDS_BANK22, 3706 AMDGPU_GFX_LDS_BANK23, 3707 AMDGPU_GFX_LDS_BANK24, 3708 AMDGPU_GFX_LDS_BANK25, 3709 AMDGPU_GFX_LDS_BANK26, 3710 AMDGPU_GFX_LDS_BANK27, 3711 AMDGPU_GFX_LDS_BANK28, 3712 AMDGPU_GFX_LDS_BANK29, 3713 AMDGPU_GFX_LDS_BANK30, 3714 AMDGPU_GFX_LDS_BANK31, 3715 AMDGPU_GFX_LDS_SP_BUFFER_A, 3716 AMDGPU_GFX_LDS_SP_BUFFER_B, 3717}; 3718 3719enum amdgpu_gfx_rlc_ras_mem_id { 3720 AMDGPU_GFX_RLC_GPMF32 = 1, 3721 AMDGPU_GFX_RLC_RLCVF32, 3722 AMDGPU_GFX_RLC_SCRATCH, 3723 AMDGPU_GFX_RLC_SRM_ARAM, 3724 AMDGPU_GFX_RLC_SRM_DRAM, 3725 AMDGPU_GFX_RLC_TCTAG, 3726 AMDGPU_GFX_RLC_SPM_SE, 3727 AMDGPU_GFX_RLC_SPM_GRBMT, 3728}; 3729 3730enum amdgpu_gfx_sp_ras_mem_id { 3731 AMDGPU_GFX_SP_SIMDID0 = 0, 3732}; 3733 3734enum amdgpu_gfx_spi_ras_mem_id { 3735 AMDGPU_GFX_SPI_MEM0 = 0, 3736 AMDGPU_GFX_SPI_MEM1, 3737 AMDGPU_GFX_SPI_MEM2, 3738 AMDGPU_GFX_SPI_MEM3, 3739}; 3740 3741enum amdgpu_gfx_sqc_ras_mem_id { 3742 AMDGPU_GFX_SQC_INST_CACHE_A = 100, 3743 AMDGPU_GFX_SQC_INST_CACHE_B = 101, 3744 AMDGPU_GFX_SQC_INST_CACHE_TAG_A = 102, 3745 AMDGPU_GFX_SQC_INST_CACHE_TAG_B = 103, 3746 AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_A = 104, 3747 AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_B = 105, 3748 AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_A = 106, 3749 AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_B = 107, 3750 AMDGPU_GFX_SQC_DATA_CACHE_A = 200, 3751 AMDGPU_GFX_SQC_DATA_CACHE_B = 201, 3752 AMDGPU_GFX_SQC_DATA_CACHE_TAG_A = 202, 3753 AMDGPU_GFX_SQC_DATA_CACHE_TAG_B = 203, 3754 AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_A = 204, 3755 AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_B = 205, 3756 AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_A = 206, 3757 AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_B = 207, 3758 AMDGPU_GFX_SQC_DIRTY_BIT_A = 208, 3759 AMDGPU_GFX_SQC_DIRTY_BIT_B = 209, 3760 AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU0 = 210, 3761 AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU1 = 211, 3762 AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_A = 212, 3763 AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_B = 213, 3764 AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_INST_CACHE = 108, 3765}; 3766 3767enum amdgpu_gfx_sq_ras_mem_id { 3768 AMDGPU_GFX_SQ_SGPR_MEM0 = 0, 3769 AMDGPU_GFX_SQ_SGPR_MEM1, 3770 AMDGPU_GFX_SQ_SGPR_MEM2, 3771 AMDGPU_GFX_SQ_SGPR_MEM3, 3772}; 3773 3774enum amdgpu_gfx_ta_ras_mem_id { 3775 AMDGPU_GFX_TA_FS_AFIFO_RAM_LO = 1, 3776 AMDGPU_GFX_TA_FS_AFIFO_RAM_HI, 3777 AMDGPU_GFX_TA_FS_CFIFO_RAM, 3778 AMDGPU_GFX_TA_FSX_LFIFO, 3779 AMDGPU_GFX_TA_FS_DFIFO_RAM, 3780}; 3781 3782enum amdgpu_gfx_tcc_ras_mem_id { 3783 AMDGPU_GFX_TCC_MEM1 = 1, 3784}; 3785 3786enum amdgpu_gfx_tca_ras_mem_id { 3787 AMDGPU_GFX_TCA_MEM1 = 1, 3788}; 3789 3790enum amdgpu_gfx_tci_ras_mem_id { 3791 AMDGPU_GFX_TCIW_MEM = 1, 3792}; 3793 3794enum amdgpu_gfx_tcp_ras_mem_id { 3795 AMDGPU_GFX_TCP_LFIFO0 = 1, 3796 AMDGPU_GFX_TCP_SET0BANK0_RAM, 3797 AMDGPU_GFX_TCP_SET0BANK1_RAM, 3798 AMDGPU_GFX_TCP_SET0BANK2_RAM, 3799 AMDGPU_GFX_TCP_SET0BANK3_RAM, 3800 AMDGPU_GFX_TCP_SET1BANK0_RAM, 3801 AMDGPU_GFX_TCP_SET1BANK1_RAM, 3802 AMDGPU_GFX_TCP_SET1BANK2_RAM, 3803 AMDGPU_GFX_TCP_SET1BANK3_RAM, 3804 AMDGPU_GFX_TCP_SET2BANK0_RAM, 3805 AMDGPU_GFX_TCP_SET2BANK1_RAM, 3806 AMDGPU_GFX_TCP_SET2BANK2_RAM, 3807 AMDGPU_GFX_TCP_SET2BANK3_RAM, 3808 AMDGPU_GFX_TCP_SET3BANK0_RAM, 3809 AMDGPU_GFX_TCP_SET3BANK1_RAM, 3810 AMDGPU_GFX_TCP_SET3BANK2_RAM, 3811 AMDGPU_GFX_TCP_SET3BANK3_RAM, 3812 AMDGPU_GFX_TCP_VM_FIFO, 3813 AMDGPU_GFX_TCP_DB_TAGRAM0, 3814 AMDGPU_GFX_TCP_DB_TAGRAM1, 3815 AMDGPU_GFX_TCP_DB_TAGRAM2, 3816 AMDGPU_GFX_TCP_DB_TAGRAM3, 3817 AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE0, 3818 AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE1, 3819 AMDGPU_GFX_TCP_CMD_FIFO, 3820}; 3821 3822enum amdgpu_gfx_td_ras_mem_id { 3823 AMDGPU_GFX_TD_UTD_CS_FIFO_MEM = 1, 3824 AMDGPU_GFX_TD_UTD_SS_FIFO_LO_MEM, 3825 AMDGPU_GFX_TD_UTD_SS_FIFO_HI_MEM, 3826}; 3827 3828enum amdgpu_gfx_tcx_ras_mem_id { 3829 AMDGPU_GFX_TCX_FIFOD0 = 0, 3830 AMDGPU_GFX_TCX_FIFOD1, 3831 AMDGPU_GFX_TCX_FIFOD2, 3832 AMDGPU_GFX_TCX_FIFOD3, 3833 AMDGPU_GFX_TCX_FIFOD4, 3834 AMDGPU_GFX_TCX_FIFOD5, 3835 AMDGPU_GFX_TCX_FIFOD6, 3836 AMDGPU_GFX_TCX_FIFOD7, 3837 AMDGPU_GFX_TCX_FIFOB0, 3838 AMDGPU_GFX_TCX_FIFOB1, 3839 AMDGPU_GFX_TCX_FIFOB2, 3840 AMDGPU_GFX_TCX_FIFOB3, 3841 AMDGPU_GFX_TCX_FIFOB4, 3842 AMDGPU_GFX_TCX_FIFOB5, 3843 AMDGPU_GFX_TCX_FIFOB6, 3844 AMDGPU_GFX_TCX_FIFOB7, 3845 AMDGPU_GFX_TCX_FIFOA0, 3846 AMDGPU_GFX_TCX_FIFOA1, 3847 AMDGPU_GFX_TCX_FIFOA2, 3848 AMDGPU_GFX_TCX_FIFOA3, 3849 AMDGPU_GFX_TCX_FIFOA4, 3850 AMDGPU_GFX_TCX_FIFOA5, 3851 AMDGPU_GFX_TCX_FIFOA6, 3852 AMDGPU_GFX_TCX_FIFOA7, 3853 AMDGPU_GFX_TCX_CFIFO0, 3854 AMDGPU_GFX_TCX_CFIFO1, 3855 AMDGPU_GFX_TCX_CFIFO2, 3856 AMDGPU_GFX_TCX_CFIFO3, 3857 AMDGPU_GFX_TCX_CFIFO4, 3858 AMDGPU_GFX_TCX_CFIFO5, 3859 AMDGPU_GFX_TCX_CFIFO6, 3860 AMDGPU_GFX_TCX_CFIFO7, 3861 AMDGPU_GFX_TCX_FIFO_ACKB0, 3862 AMDGPU_GFX_TCX_FIFO_ACKB1, 3863 AMDGPU_GFX_TCX_FIFO_ACKB2, 3864 AMDGPU_GFX_TCX_FIFO_ACKB3, 3865 AMDGPU_GFX_TCX_FIFO_ACKB4, 3866 AMDGPU_GFX_TCX_FIFO_ACKB5, 3867 AMDGPU_GFX_TCX_FIFO_ACKB6, 3868 AMDGPU_GFX_TCX_FIFO_ACKB7, 3869 AMDGPU_GFX_TCX_FIFO_ACKD0, 3870 AMDGPU_GFX_TCX_FIFO_ACKD1, 3871 AMDGPU_GFX_TCX_FIFO_ACKD2, 3872 AMDGPU_GFX_TCX_FIFO_ACKD3, 3873 AMDGPU_GFX_TCX_FIFO_ACKD4, 3874 AMDGPU_GFX_TCX_FIFO_ACKD5, 3875 AMDGPU_GFX_TCX_FIFO_ACKD6, 3876 AMDGPU_GFX_TCX_FIFO_ACKD7, 3877 AMDGPU_GFX_TCX_DST_FIFOA0, 3878 AMDGPU_GFX_TCX_DST_FIFOA1, 3879 AMDGPU_GFX_TCX_DST_FIFOA2, 3880 AMDGPU_GFX_TCX_DST_FIFOA3, 3881 AMDGPU_GFX_TCX_DST_FIFOA4, 3882 AMDGPU_GFX_TCX_DST_FIFOA5, 3883 AMDGPU_GFX_TCX_DST_FIFOA6, 3884 AMDGPU_GFX_TCX_DST_FIFOA7, 3885 AMDGPU_GFX_TCX_DST_FIFOB0, 3886 AMDGPU_GFX_TCX_DST_FIFOB1, 3887 AMDGPU_GFX_TCX_DST_FIFOB2, 3888 AMDGPU_GFX_TCX_DST_FIFOB3, 3889 AMDGPU_GFX_TCX_DST_FIFOB4, 3890 AMDGPU_GFX_TCX_DST_FIFOB5, 3891 AMDGPU_GFX_TCX_DST_FIFOB6, 3892 AMDGPU_GFX_TCX_DST_FIFOB7, 3893 AMDGPU_GFX_TCX_DST_FIFOD0, 3894 AMDGPU_GFX_TCX_DST_FIFOD1, 3895 AMDGPU_GFX_TCX_DST_FIFOD2, 3896 AMDGPU_GFX_TCX_DST_FIFOD3, 3897 AMDGPU_GFX_TCX_DST_FIFOD4, 3898 AMDGPU_GFX_TCX_DST_FIFOD5, 3899 AMDGPU_GFX_TCX_DST_FIFOD6, 3900 AMDGPU_GFX_TCX_DST_FIFOD7, 3901 AMDGPU_GFX_TCX_DST_FIFO_ACKB0, 3902 AMDGPU_GFX_TCX_DST_FIFO_ACKB1, 3903 AMDGPU_GFX_TCX_DST_FIFO_ACKB2, 3904 AMDGPU_GFX_TCX_DST_FIFO_ACKB3, 3905 AMDGPU_GFX_TCX_DST_FIFO_ACKB4, 3906 AMDGPU_GFX_TCX_DST_FIFO_ACKB5, 3907 AMDGPU_GFX_TCX_DST_FIFO_ACKB6, 3908 AMDGPU_GFX_TCX_DST_FIFO_ACKB7, 3909 AMDGPU_GFX_TCX_DST_FIFO_ACKD0, 3910 AMDGPU_GFX_TCX_DST_FIFO_ACKD1, 3911 AMDGPU_GFX_TCX_DST_FIFO_ACKD2, 3912 AMDGPU_GFX_TCX_DST_FIFO_ACKD3, 3913 AMDGPU_GFX_TCX_DST_FIFO_ACKD4, 3914 AMDGPU_GFX_TCX_DST_FIFO_ACKD5, 3915 AMDGPU_GFX_TCX_DST_FIFO_ACKD6, 3916 AMDGPU_GFX_TCX_DST_FIFO_ACKD7, 3917}; 3918 3919enum amdgpu_gfx_atc_l2_ras_mem_id { 3920 AMDGPU_GFX_ATC_L2_MEM0 = 0, 3921}; 3922 3923enum amdgpu_gfx_utcl2_ras_mem_id { 3924 AMDGPU_GFX_UTCL2_MEM0 = 0, 3925}; 3926 3927enum amdgpu_gfx_vml2_ras_mem_id { 3928 AMDGPU_GFX_VML2_MEM0 = 0, 3929}; 3930 3931enum amdgpu_gfx_vml2_walker_ras_mem_id { 3932 AMDGPU_GFX_VML2_WALKER_MEM0 = 0, 3933}; 3934 3935static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_cp_mem_list[] = { 3936 {AMDGPU_GFX_CP_MEM1, "CP_MEM1"}, 3937 {AMDGPU_GFX_CP_MEM2, "CP_MEM2"}, 3938 {AMDGPU_GFX_CP_MEM3, "CP_MEM3"}, 3939 {AMDGPU_GFX_CP_MEM4, "CP_MEM4"}, 3940 {AMDGPU_GFX_CP_MEM5, "CP_MEM5"}, 3941}; 3942 3943static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gcea_mem_list[] = { 3944 {AMDGPU_GFX_GCEA_IOWR_CMDMEM, "GCEA_IOWR_CMDMEM"}, 3945 {AMDGPU_GFX_GCEA_IORD_CMDMEM, "GCEA_IORD_CMDMEM"}, 3946 {AMDGPU_GFX_GCEA_GMIWR_CMDMEM, "GCEA_GMIWR_CMDMEM"}, 3947 {AMDGPU_GFX_GCEA_GMIRD_CMDMEM, "GCEA_GMIRD_CMDMEM"}, 3948 {AMDGPU_GFX_GCEA_DRAMWR_CMDMEM, "GCEA_DRAMWR_CMDMEM"}, 3949 {AMDGPU_GFX_GCEA_DRAMRD_CMDMEM, "GCEA_DRAMRD_CMDMEM"}, 3950 {AMDGPU_GFX_GCEA_MAM_DMEM0, "GCEA_MAM_DMEM0"}, 3951 {AMDGPU_GFX_GCEA_MAM_DMEM1, "GCEA_MAM_DMEM1"}, 3952 {AMDGPU_GFX_GCEA_MAM_DMEM2, "GCEA_MAM_DMEM2"}, 3953 {AMDGPU_GFX_GCEA_MAM_DMEM3, "GCEA_MAM_DMEM3"}, 3954 {AMDGPU_GFX_GCEA_MAM_AMEM0, "GCEA_MAM_AMEM0"}, 3955 {AMDGPU_GFX_GCEA_MAM_AMEM1, "GCEA_MAM_AMEM1"}, 3956 {AMDGPU_GFX_GCEA_MAM_AMEM2, "GCEA_MAM_AMEM2"}, 3957 {AMDGPU_GFX_GCEA_MAM_AMEM3, "GCEA_MAM_AMEM3"}, 3958 {AMDGPU_GFX_GCEA_MAM_AFLUSH_BUFFER, "GCEA_MAM_AFLUSH_BUFFER"}, 3959 {AMDGPU_GFX_GCEA_WRET_TAGMEM, "GCEA_WRET_TAGMEM"}, 3960 {AMDGPU_GFX_GCEA_RRET_TAGMEM, "GCEA_RRET_TAGMEM"}, 3961 {AMDGPU_GFX_GCEA_IOWR_DATAMEM, "GCEA_IOWR_DATAMEM"}, 3962 {AMDGPU_GFX_GCEA_GMIWR_DATAMEM, "GCEA_GMIWR_DATAMEM"}, 3963 {AMDGPU_GFX_GCEA_DRAM_DATAMEM, "GCEA_DRAM_DATAMEM"}, 3964}; 3965 3966static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gc_cane_mem_list[] = { 3967 {AMDGPU_GFX_GC_CANE_MEM0, "GC_CANE_MEM0"}, 3968}; 3969 3970static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gcutcl2_mem_list[] = { 3971 {AMDGPU_GFX_GCUTCL2_MEM2P512X95, "GCUTCL2_MEM2P512X95"}, 3972}; 3973 3974static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gds_mem_list[] = { 3975 {AMDGPU_GFX_GDS_MEM0, "GDS_MEM"}, 3976}; 3977 3978static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_lds_mem_list[] = { 3979 {AMDGPU_GFX_LDS_BANK0, "LDS_BANK0"}, 3980 {AMDGPU_GFX_LDS_BANK1, "LDS_BANK1"}, 3981 {AMDGPU_GFX_LDS_BANK2, "LDS_BANK2"}, 3982 {AMDGPU_GFX_LDS_BANK3, "LDS_BANK3"}, 3983 {AMDGPU_GFX_LDS_BANK4, "LDS_BANK4"}, 3984 {AMDGPU_GFX_LDS_BANK5, "LDS_BANK5"}, 3985 {AMDGPU_GFX_LDS_BANK6, "LDS_BANK6"}, 3986 {AMDGPU_GFX_LDS_BANK7, "LDS_BANK7"}, 3987 {AMDGPU_GFX_LDS_BANK8, "LDS_BANK8"}, 3988 {AMDGPU_GFX_LDS_BANK9, "LDS_BANK9"}, 3989 {AMDGPU_GFX_LDS_BANK10, "LDS_BANK10"}, 3990 {AMDGPU_GFX_LDS_BANK11, "LDS_BANK11"}, 3991 {AMDGPU_GFX_LDS_BANK12, "LDS_BANK12"}, 3992 {AMDGPU_GFX_LDS_BANK13, "LDS_BANK13"}, 3993 {AMDGPU_GFX_LDS_BANK14, "LDS_BANK14"}, 3994 {AMDGPU_GFX_LDS_BANK15, "LDS_BANK15"}, 3995 {AMDGPU_GFX_LDS_BANK16, "LDS_BANK16"}, 3996 {AMDGPU_GFX_LDS_BANK17, "LDS_BANK17"}, 3997 {AMDGPU_GFX_LDS_BANK18, "LDS_BANK18"}, 3998 {AMDGPU_GFX_LDS_BANK19, "LDS_BANK19"}, 3999 {AMDGPU_GFX_LDS_BANK20, "LDS_BANK20"}, 4000 {AMDGPU_GFX_LDS_BANK21, "LDS_BANK21"}, 4001 {AMDGPU_GFX_LDS_BANK22, "LDS_BANK22"}, 4002 {AMDGPU_GFX_LDS_BANK23, "LDS_BANK23"}, 4003 {AMDGPU_GFX_LDS_BANK24, "LDS_BANK24"}, 4004 {AMDGPU_GFX_LDS_BANK25, "LDS_BANK25"}, 4005 {AMDGPU_GFX_LDS_BANK26, "LDS_BANK26"}, 4006 {AMDGPU_GFX_LDS_BANK27, "LDS_BANK27"}, 4007 {AMDGPU_GFX_LDS_BANK28, "LDS_BANK28"}, 4008 {AMDGPU_GFX_LDS_BANK29, "LDS_BANK29"}, 4009 {AMDGPU_GFX_LDS_BANK30, "LDS_BANK30"}, 4010 {AMDGPU_GFX_LDS_BANK31, "LDS_BANK31"}, 4011 {AMDGPU_GFX_LDS_SP_BUFFER_A, "LDS_SP_BUFFER_A"}, 4012 {AMDGPU_GFX_LDS_SP_BUFFER_B, "LDS_SP_BUFFER_B"}, 4013}; 4014 4015static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_rlc_mem_list[] = { 4016 {AMDGPU_GFX_RLC_GPMF32, "RLC_GPMF32"}, 4017 {AMDGPU_GFX_RLC_RLCVF32, "RLC_RLCVF32"}, 4018 {AMDGPU_GFX_RLC_SCRATCH, "RLC_SCRATCH"}, 4019 {AMDGPU_GFX_RLC_SRM_ARAM, "RLC_SRM_ARAM"}, 4020 {AMDGPU_GFX_RLC_SRM_DRAM, "RLC_SRM_DRAM"}, 4021 {AMDGPU_GFX_RLC_TCTAG, "RLC_TCTAG"}, 4022 {AMDGPU_GFX_RLC_SPM_SE, "RLC_SPM_SE"}, 4023 {AMDGPU_GFX_RLC_SPM_GRBMT, "RLC_SPM_GRBMT"}, 4024}; 4025 4026static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_sp_mem_list[] = { 4027 {AMDGPU_GFX_SP_SIMDID0, "SP_SIMDID0"}, 4028}; 4029 4030static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_spi_mem_list[] = { 4031 {AMDGPU_GFX_SPI_MEM0, "SPI_MEM0"}, 4032 {AMDGPU_GFX_SPI_MEM1, "SPI_MEM1"}, 4033 {AMDGPU_GFX_SPI_MEM2, "SPI_MEM2"}, 4034 {AMDGPU_GFX_SPI_MEM3, "SPI_MEM3"}, 4035}; 4036 4037static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_sqc_mem_list[] = { 4038 {AMDGPU_GFX_SQC_INST_CACHE_A, "SQC_INST_CACHE_A"}, 4039 {AMDGPU_GFX_SQC_INST_CACHE_B, "SQC_INST_CACHE_B"}, 4040 {AMDGPU_GFX_SQC_INST_CACHE_TAG_A, "SQC_INST_CACHE_TAG_A"}, 4041 {AMDGPU_GFX_SQC_INST_CACHE_TAG_B, "SQC_INST_CACHE_TAG_B"}, 4042 {AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_A, "SQC_INST_CACHE_MISS_FIFO_A"}, 4043 {AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_B, "SQC_INST_CACHE_MISS_FIFO_B"}, 4044 {AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_A, "SQC_INST_CACHE_GATCL1_MISS_FIFO_A"}, 4045 {AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_B, "SQC_INST_CACHE_GATCL1_MISS_FIFO_B"}, 4046 {AMDGPU_GFX_SQC_DATA_CACHE_A, "SQC_DATA_CACHE_A"}, 4047 {AMDGPU_GFX_SQC_DATA_CACHE_B, "SQC_DATA_CACHE_B"}, 4048 {AMDGPU_GFX_SQC_DATA_CACHE_TAG_A, "SQC_DATA_CACHE_TAG_A"}, 4049 {AMDGPU_GFX_SQC_DATA_CACHE_TAG_B, "SQC_DATA_CACHE_TAG_B"}, 4050 {AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_A, "SQC_DATA_CACHE_MISS_FIFO_A"}, 4051 {AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_B, "SQC_DATA_CACHE_MISS_FIFO_B"}, 4052 {AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_A, "SQC_DATA_CACHE_HIT_FIFO_A"}, 4053 {AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_B, "SQC_DATA_CACHE_HIT_FIFO_B"}, 4054 {AMDGPU_GFX_SQC_DIRTY_BIT_A, "SQC_DIRTY_BIT_A"}, 4055 {AMDGPU_GFX_SQC_DIRTY_BIT_B, "SQC_DIRTY_BIT_B"}, 4056 {AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU0, "SQC_WRITE_DATA_BUFFER_CU0"}, 4057 {AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU1, "SQC_WRITE_DATA_BUFFER_CU1"}, 4058 {AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_A, "SQC_UTCL1_MISS_LFIFO_DATA_CACHE_A"}, 4059 {AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_B, "SQC_UTCL1_MISS_LFIFO_DATA_CACHE_B"}, 4060 {AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_INST_CACHE, "SQC_UTCL1_MISS_LFIFO_INST_CACHE"}, 4061}; 4062 4063static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_sq_mem_list[] = { 4064 {AMDGPU_GFX_SQ_SGPR_MEM0, "SQ_SGPR_MEM0"}, 4065 {AMDGPU_GFX_SQ_SGPR_MEM1, "SQ_SGPR_MEM1"}, 4066 {AMDGPU_GFX_SQ_SGPR_MEM2, "SQ_SGPR_MEM2"}, 4067 {AMDGPU_GFX_SQ_SGPR_MEM3, "SQ_SGPR_MEM3"}, 4068}; 4069 4070static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_ta_mem_list[] = { 4071 {AMDGPU_GFX_TA_FS_AFIFO_RAM_LO, "TA_FS_AFIFO_RAM_LO"}, 4072 {AMDGPU_GFX_TA_FS_AFIFO_RAM_HI, "TA_FS_AFIFO_RAM_HI"}, 4073 {AMDGPU_GFX_TA_FS_CFIFO_RAM, "TA_FS_CFIFO_RAM"}, 4074 {AMDGPU_GFX_TA_FSX_LFIFO, "TA_FSX_LFIFO"}, 4075 {AMDGPU_GFX_TA_FS_DFIFO_RAM, "TA_FS_DFIFO_RAM"}, 4076}; 4077 4078static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tcc_mem_list[] = { 4079 {AMDGPU_GFX_TCC_MEM1, "TCC_MEM1"}, 4080}; 4081 4082static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tca_mem_list[] = { 4083 {AMDGPU_GFX_TCA_MEM1, "TCA_MEM1"}, 4084}; 4085 4086static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tci_mem_list[] = { 4087 {AMDGPU_GFX_TCIW_MEM, "TCIW_MEM"}, 4088}; 4089 4090static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tcp_mem_list[] = { 4091 {AMDGPU_GFX_TCP_LFIFO0, "TCP_LFIFO0"}, 4092 {AMDGPU_GFX_TCP_SET0BANK0_RAM, "TCP_SET0BANK0_RAM"}, 4093 {AMDGPU_GFX_TCP_SET0BANK1_RAM, "TCP_SET0BANK1_RAM"}, 4094 {AMDGPU_GFX_TCP_SET0BANK2_RAM, "TCP_SET0BANK2_RAM"}, 4095 {AMDGPU_GFX_TCP_SET0BANK3_RAM, "TCP_SET0BANK3_RAM"}, 4096 {AMDGPU_GFX_TCP_SET1BANK0_RAM, "TCP_SET1BANK0_RAM"}, 4097 {AMDGPU_GFX_TCP_SET1BANK1_RAM, "TCP_SET1BANK1_RAM"}, 4098 {AMDGPU_GFX_TCP_SET1BANK2_RAM, "TCP_SET1BANK2_RAM"}, 4099 {AMDGPU_GFX_TCP_SET1BANK3_RAM, "TCP_SET1BANK3_RAM"}, 4100 {AMDGPU_GFX_TCP_SET2BANK0_RAM, "TCP_SET2BANK0_RAM"}, 4101 {AMDGPU_GFX_TCP_SET2BANK1_RAM, "TCP_SET2BANK1_RAM"}, 4102 {AMDGPU_GFX_TCP_SET2BANK2_RAM, "TCP_SET2BANK2_RAM"}, 4103 {AMDGPU_GFX_TCP_SET2BANK3_RAM, "TCP_SET2BANK3_RAM"}, 4104 {AMDGPU_GFX_TCP_SET3BANK0_RAM, "TCP_SET3BANK0_RAM"}, 4105 {AMDGPU_GFX_TCP_SET3BANK1_RAM, "TCP_SET3BANK1_RAM"}, 4106 {AMDGPU_GFX_TCP_SET3BANK2_RAM, "TCP_SET3BANK2_RAM"}, 4107 {AMDGPU_GFX_TCP_SET3BANK3_RAM, "TCP_SET3BANK3_RAM"}, 4108 {AMDGPU_GFX_TCP_VM_FIFO, "TCP_VM_FIFO"}, 4109 {AMDGPU_GFX_TCP_DB_TAGRAM0, "TCP_DB_TAGRAM0"}, 4110 {AMDGPU_GFX_TCP_DB_TAGRAM1, "TCP_DB_TAGRAM1"}, 4111 {AMDGPU_GFX_TCP_DB_TAGRAM2, "TCP_DB_TAGRAM2"}, 4112 {AMDGPU_GFX_TCP_DB_TAGRAM3, "TCP_DB_TAGRAM3"}, 4113 {AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE0, "TCP_UTCL1_LFIFO_PROBE0"}, 4114 {AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE1, "TCP_UTCL1_LFIFO_PROBE1"}, 4115 {AMDGPU_GFX_TCP_CMD_FIFO, "TCP_CMD_FIFO"}, 4116}; 4117 4118static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_td_mem_list[] = { 4119 {AMDGPU_GFX_TD_UTD_CS_FIFO_MEM, "TD_UTD_CS_FIFO_MEM"}, 4120 {AMDGPU_GFX_TD_UTD_SS_FIFO_LO_MEM, "TD_UTD_SS_FIFO_LO_MEM"}, 4121 {AMDGPU_GFX_TD_UTD_SS_FIFO_HI_MEM, "TD_UTD_SS_FIFO_HI_MEM"}, 4122}; 4123 4124static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tcx_mem_list[] = { 4125 {AMDGPU_GFX_TCX_FIFOD0, "TCX_FIFOD0"}, 4126 {AMDGPU_GFX_TCX_FIFOD1, "TCX_FIFOD1"}, 4127 {AMDGPU_GFX_TCX_FIFOD2, "TCX_FIFOD2"}, 4128 {AMDGPU_GFX_TCX_FIFOD3, "TCX_FIFOD3"}, 4129 {AMDGPU_GFX_TCX_FIFOD4, "TCX_FIFOD4"}, 4130 {AMDGPU_GFX_TCX_FIFOD5, "TCX_FIFOD5"}, 4131 {AMDGPU_GFX_TCX_FIFOD6, "TCX_FIFOD6"}, 4132 {AMDGPU_GFX_TCX_FIFOD7, "TCX_FIFOD7"}, 4133 {AMDGPU_GFX_TCX_FIFOB0, "TCX_FIFOB0"}, 4134 {AMDGPU_GFX_TCX_FIFOB1, "TCX_FIFOB1"}, 4135 {AMDGPU_GFX_TCX_FIFOB2, "TCX_FIFOB2"}, 4136 {AMDGPU_GFX_TCX_FIFOB3, "TCX_FIFOB3"}, 4137 {AMDGPU_GFX_TCX_FIFOB4, "TCX_FIFOB4"}, 4138 {AMDGPU_GFX_TCX_FIFOB5, "TCX_FIFOB5"}, 4139 {AMDGPU_GFX_TCX_FIFOB6, "TCX_FIFOB6"}, 4140 {AMDGPU_GFX_TCX_FIFOB7, "TCX_FIFOB7"}, 4141 {AMDGPU_GFX_TCX_FIFOA0, "TCX_FIFOA0"}, 4142 {AMDGPU_GFX_TCX_FIFOA1, "TCX_FIFOA1"}, 4143 {AMDGPU_GFX_TCX_FIFOA2, "TCX_FIFOA2"}, 4144 {AMDGPU_GFX_TCX_FIFOA3, "TCX_FIFOA3"}, 4145 {AMDGPU_GFX_TCX_FIFOA4, "TCX_FIFOA4"}, 4146 {AMDGPU_GFX_TCX_FIFOA5, "TCX_FIFOA5"}, 4147 {AMDGPU_GFX_TCX_FIFOA6, "TCX_FIFOA6"}, 4148 {AMDGPU_GFX_TCX_FIFOA7, "TCX_FIFOA7"}, 4149 {AMDGPU_GFX_TCX_CFIFO0, "TCX_CFIFO0"}, 4150 {AMDGPU_GFX_TCX_CFIFO1, "TCX_CFIFO1"}, 4151 {AMDGPU_GFX_TCX_CFIFO2, "TCX_CFIFO2"}, 4152 {AMDGPU_GFX_TCX_CFIFO3, "TCX_CFIFO3"}, 4153 {AMDGPU_GFX_TCX_CFIFO4, "TCX_CFIFO4"}, 4154 {AMDGPU_GFX_TCX_CFIFO5, "TCX_CFIFO5"}, 4155 {AMDGPU_GFX_TCX_CFIFO6, "TCX_CFIFO6"}, 4156 {AMDGPU_GFX_TCX_CFIFO7, "TCX_CFIFO7"}, 4157 {AMDGPU_GFX_TCX_FIFO_ACKB0, "TCX_FIFO_ACKB0"}, 4158 {AMDGPU_GFX_TCX_FIFO_ACKB1, "TCX_FIFO_ACKB1"}, 4159 {AMDGPU_GFX_TCX_FIFO_ACKB2, "TCX_FIFO_ACKB2"}, 4160 {AMDGPU_GFX_TCX_FIFO_ACKB3, "TCX_FIFO_ACKB3"}, 4161 {AMDGPU_GFX_TCX_FIFO_ACKB4, "TCX_FIFO_ACKB4"}, 4162 {AMDGPU_GFX_TCX_FIFO_ACKB5, "TCX_FIFO_ACKB5"}, 4163 {AMDGPU_GFX_TCX_FIFO_ACKB6, "TCX_FIFO_ACKB6"}, 4164 {AMDGPU_GFX_TCX_FIFO_ACKB7, "TCX_FIFO_ACKB7"}, 4165 {AMDGPU_GFX_TCX_FIFO_ACKD0, "TCX_FIFO_ACKD0"}, 4166 {AMDGPU_GFX_TCX_FIFO_ACKD1, "TCX_FIFO_ACKD1"}, 4167 {AMDGPU_GFX_TCX_FIFO_ACKD2, "TCX_FIFO_ACKD2"}, 4168 {AMDGPU_GFX_TCX_FIFO_ACKD3, "TCX_FIFO_ACKD3"}, 4169 {AMDGPU_GFX_TCX_FIFO_ACKD4, "TCX_FIFO_ACKD4"}, 4170 {AMDGPU_GFX_TCX_FIFO_ACKD5, "TCX_FIFO_ACKD5"}, 4171 {AMDGPU_GFX_TCX_FIFO_ACKD6, "TCX_FIFO_ACKD6"}, 4172 {AMDGPU_GFX_TCX_FIFO_ACKD7, "TCX_FIFO_ACKD7"}, 4173 {AMDGPU_GFX_TCX_DST_FIFOA0, "TCX_DST_FIFOA0"}, 4174 {AMDGPU_GFX_TCX_DST_FIFOA1, "TCX_DST_FIFOA1"}, 4175 {AMDGPU_GFX_TCX_DST_FIFOA2, "TCX_DST_FIFOA2"}, 4176 {AMDGPU_GFX_TCX_DST_FIFOA3, "TCX_DST_FIFOA3"}, 4177 {AMDGPU_GFX_TCX_DST_FIFOA4, "TCX_DST_FIFOA4"}, 4178 {AMDGPU_GFX_TCX_DST_FIFOA5, "TCX_DST_FIFOA5"}, 4179 {AMDGPU_GFX_TCX_DST_FIFOA6, "TCX_DST_FIFOA6"}, 4180 {AMDGPU_GFX_TCX_DST_FIFOA7, "TCX_DST_FIFOA7"}, 4181 {AMDGPU_GFX_TCX_DST_FIFOB0, "TCX_DST_FIFOB0"}, 4182 {AMDGPU_GFX_TCX_DST_FIFOB1, "TCX_DST_FIFOB1"}, 4183 {AMDGPU_GFX_TCX_DST_FIFOB2, "TCX_DST_FIFOB2"}, 4184 {AMDGPU_GFX_TCX_DST_FIFOB3, "TCX_DST_FIFOB3"}, 4185 {AMDGPU_GFX_TCX_DST_FIFOB4, "TCX_DST_FIFOB4"}, 4186 {AMDGPU_GFX_TCX_DST_FIFOB5, "TCX_DST_FIFOB5"}, 4187 {AMDGPU_GFX_TCX_DST_FIFOB6, "TCX_DST_FIFOB6"}, 4188 {AMDGPU_GFX_TCX_DST_FIFOB7, "TCX_DST_FIFOB7"}, 4189 {AMDGPU_GFX_TCX_DST_FIFOD0, "TCX_DST_FIFOD0"}, 4190 {AMDGPU_GFX_TCX_DST_FIFOD1, "TCX_DST_FIFOD1"}, 4191 {AMDGPU_GFX_TCX_DST_FIFOD2, "TCX_DST_FIFOD2"}, 4192 {AMDGPU_GFX_TCX_DST_FIFOD3, "TCX_DST_FIFOD3"}, 4193 {AMDGPU_GFX_TCX_DST_FIFOD4, "TCX_DST_FIFOD4"}, 4194 {AMDGPU_GFX_TCX_DST_FIFOD5, "TCX_DST_FIFOD5"}, 4195 {AMDGPU_GFX_TCX_DST_FIFOD6, "TCX_DST_FIFOD6"}, 4196 {AMDGPU_GFX_TCX_DST_FIFOD7, "TCX_DST_FIFOD7"}, 4197 {AMDGPU_GFX_TCX_DST_FIFO_ACKB0, "TCX_DST_FIFO_ACKB0"}, 4198 {AMDGPU_GFX_TCX_DST_FIFO_ACKB1, "TCX_DST_FIFO_ACKB1"}, 4199 {AMDGPU_GFX_TCX_DST_FIFO_ACKB2, "TCX_DST_FIFO_ACKB2"}, 4200 {AMDGPU_GFX_TCX_DST_FIFO_ACKB3, "TCX_DST_FIFO_ACKB3"}, 4201 {AMDGPU_GFX_TCX_DST_FIFO_ACKB4, "TCX_DST_FIFO_ACKB4"}, 4202 {AMDGPU_GFX_TCX_DST_FIFO_ACKB5, "TCX_DST_FIFO_ACKB5"}, 4203 {AMDGPU_GFX_TCX_DST_FIFO_ACKB6, "TCX_DST_FIFO_ACKB6"}, 4204 {AMDGPU_GFX_TCX_DST_FIFO_ACKB7, "TCX_DST_FIFO_ACKB7"}, 4205 {AMDGPU_GFX_TCX_DST_FIFO_ACKD0, "TCX_DST_FIFO_ACKD0"}, 4206 {AMDGPU_GFX_TCX_DST_FIFO_ACKD1, "TCX_DST_FIFO_ACKD1"}, 4207 {AMDGPU_GFX_TCX_DST_FIFO_ACKD2, "TCX_DST_FIFO_ACKD2"}, 4208 {AMDGPU_GFX_TCX_DST_FIFO_ACKD3, "TCX_DST_FIFO_ACKD3"}, 4209 {AMDGPU_GFX_TCX_DST_FIFO_ACKD4, "TCX_DST_FIFO_ACKD4"}, 4210 {AMDGPU_GFX_TCX_DST_FIFO_ACKD5, "TCX_DST_FIFO_ACKD5"}, 4211 {AMDGPU_GFX_TCX_DST_FIFO_ACKD6, "TCX_DST_FIFO_ACKD6"}, 4212 {AMDGPU_GFX_TCX_DST_FIFO_ACKD7, "TCX_DST_FIFO_ACKD7"}, 4213}; 4214 4215static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_atc_l2_mem_list[] = { 4216 {AMDGPU_GFX_ATC_L2_MEM, "ATC_L2_MEM"}, 4217}; 4218 4219static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_utcl2_mem_list[] = { 4220 {AMDGPU_GFX_UTCL2_MEM, "UTCL2_MEM"}, 4221}; 4222 4223static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_vml2_mem_list[] = { 4224 {AMDGPU_GFX_VML2_MEM, "VML2_MEM"}, 4225}; 4226 4227static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_vml2_walker_mem_list[] = { 4228 {AMDGPU_GFX_VML2_WALKER_MEM, "VML2_WALKER_MEM"}, 4229}; 4230 4231static const struct amdgpu_gfx_ras_mem_id_entry gfx_v9_4_3_ras_mem_list_array[AMDGPU_GFX_MEM_TYPE_NUM] = { 4232 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_cp_mem_list) 4233 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gcea_mem_list) 4234 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gc_cane_mem_list) 4235 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gcutcl2_mem_list) 4236 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gds_mem_list) 4237 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_lds_mem_list) 4238 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_rlc_mem_list) 4239 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_sp_mem_list) 4240 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_spi_mem_list) 4241 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_sqc_mem_list) 4242 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_sq_mem_list) 4243 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_ta_mem_list) 4244 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tcc_mem_list) 4245 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tca_mem_list) 4246 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tci_mem_list) 4247 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tcp_mem_list) 4248 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_td_mem_list) 4249 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tcx_mem_list) 4250 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_atc_l2_mem_list) 4251 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_utcl2_mem_list) 4252 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_vml2_mem_list) 4253 AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_vml2_walker_mem_list) 4254}; 4255 4256static const struct amdgpu_gfx_ras_reg_entry gfx_v9_4_3_ce_reg_list[] = { 4257 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regRLC_CE_ERR_STATUS_LOW, regRLC_CE_ERR_STATUS_HIGH), 4258 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "RLC"}, 4259 AMDGPU_GFX_RLC_MEM, 1}, 4260 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPC_CE_ERR_STATUS_LO, regCPC_CE_ERR_STATUS_HI), 4261 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPC"}, 4262 AMDGPU_GFX_CP_MEM, 1}, 4263 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPF_CE_ERR_STATUS_LO, regCPF_CE_ERR_STATUS_HI), 4264 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPF"}, 4265 AMDGPU_GFX_CP_MEM, 1}, 4266 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPG_CE_ERR_STATUS_LO, regCPG_CE_ERR_STATUS_HI), 4267 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPG"}, 4268 AMDGPU_GFX_CP_MEM, 1}, 4269 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regGDS_CE_ERR_STATUS_LO, regGDS_CE_ERR_STATUS_HI), 4270 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GDS"}, 4271 AMDGPU_GFX_GDS_MEM, 1}, 4272 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regGC_CANE_CE_ERR_STATUS_LO, regGC_CANE_CE_ERR_STATUS_HI), 4273 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CANE"}, 4274 AMDGPU_GFX_GC_CANE_MEM, 1}, 4275 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSPI_CE_ERR_STATUS_LO, regSPI_CE_ERR_STATUS_HI), 4276 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SPI"}, 4277 AMDGPU_GFX_SPI_MEM, 1}, 4278 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP0_CE_ERR_STATUS_LO, regSP0_CE_ERR_STATUS_HI), 4279 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP0"}, 4280 AMDGPU_GFX_SP_MEM, 4}, 4281 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP1_CE_ERR_STATUS_LO, regSP1_CE_ERR_STATUS_HI), 4282 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP1"}, 4283 AMDGPU_GFX_SP_MEM, 4}, 4284 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQ_CE_ERR_STATUS_LO, regSQ_CE_ERR_STATUS_HI), 4285 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQ"}, 4286 AMDGPU_GFX_SQ_MEM, 4}, 4287 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQC_CE_EDC_LO, regSQC_CE_EDC_HI), 4288 5, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQC"}, 4289 AMDGPU_GFX_SQC_MEM, 4}, 4290 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCX_CE_ERR_STATUS_LO, regTCX_CE_ERR_STATUS_HI), 4291 2, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCX"}, 4292 AMDGPU_GFX_TCX_MEM, 1}, 4293 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCC_CE_ERR_STATUS_LO, regTCC_CE_ERR_STATUS_HI), 4294 16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCC"}, 4295 AMDGPU_GFX_TCC_MEM, 1}, 4296 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTA_CE_EDC_LO, regTA_CE_EDC_HI), 4297 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TA"}, 4298 AMDGPU_GFX_TA_MEM, 4}, 4299 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCI_CE_EDC_LO_REG, regTCI_CE_EDC_HI_REG), 4300 27, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCI"}, 4301 AMDGPU_GFX_TCI_MEM, 1}, 4302 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCP_CE_EDC_LO_REG, regTCP_CE_EDC_HI_REG), 4303 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCP"}, 4304 AMDGPU_GFX_TCP_MEM, 4}, 4305 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTD_CE_EDC_LO, regTD_CE_EDC_HI), 4306 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TD"}, 4307 AMDGPU_GFX_TD_MEM, 4}, 4308 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regGCEA_CE_ERR_STATUS_LO, regGCEA_CE_ERR_STATUS_HI), 4309 16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GCEA"}, 4310 AMDGPU_GFX_GCEA_MEM, 1}, 4311 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regLDS_CE_ERR_STATUS_LO, regLDS_CE_ERR_STATUS_HI), 4312 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "LDS"}, 4313 AMDGPU_GFX_LDS_MEM, 4}, 4314}; 4315 4316static const struct amdgpu_gfx_ras_reg_entry gfx_v9_4_3_ue_reg_list[] = { 4317 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regRLC_UE_ERR_STATUS_LOW, regRLC_UE_ERR_STATUS_HIGH), 4318 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "RLC"}, 4319 AMDGPU_GFX_RLC_MEM, 1}, 4320 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPC_UE_ERR_STATUS_LO, regCPC_UE_ERR_STATUS_HI), 4321 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPC"}, 4322 AMDGPU_GFX_CP_MEM, 1}, 4323 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPF_UE_ERR_STATUS_LO, regCPF_UE_ERR_STATUS_HI), 4324 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPF"}, 4325 AMDGPU_GFX_CP_MEM, 1}, 4326 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPG_UE_ERR_STATUS_LO, regCPG_UE_ERR_STATUS_HI), 4327 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPG"}, 4328 AMDGPU_GFX_CP_MEM, 1}, 4329 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regGDS_UE_ERR_STATUS_LO, regGDS_UE_ERR_STATUS_HI), 4330 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GDS"}, 4331 AMDGPU_GFX_GDS_MEM, 1}, 4332 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regGC_CANE_UE_ERR_STATUS_LO, regGC_CANE_UE_ERR_STATUS_HI), 4333 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CANE"}, 4334 AMDGPU_GFX_GC_CANE_MEM, 1}, 4335 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSPI_UE_ERR_STATUS_LO, regSPI_UE_ERR_STATUS_HI), 4336 1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SPI"}, 4337 AMDGPU_GFX_SPI_MEM, 1}, 4338 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP0_UE_ERR_STATUS_LO, regSP0_UE_ERR_STATUS_HI), 4339 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP0"}, 4340 AMDGPU_GFX_SP_MEM, 4}, 4341 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP1_UE_ERR_STATUS_LO, regSP1_UE_ERR_STATUS_HI), 4342 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP1"}, 4343 AMDGPU_GFX_SP_MEM, 4}, 4344 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQ_UE_ERR_STATUS_LO, regSQ_UE_ERR_STATUS_HI), 4345 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQ"}, 4346 AMDGPU_GFX_SQ_MEM, 4}, 4347 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQC_UE_EDC_LO, regSQC_UE_EDC_HI), 4348 5, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQC"}, 4349 AMDGPU_GFX_SQC_MEM, 4}, 4350 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCX_UE_ERR_STATUS_LO, regTCX_UE_ERR_STATUS_HI), 4351 2, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCX"}, 4352 AMDGPU_GFX_TCX_MEM, 1}, 4353 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCC_UE_ERR_STATUS_LO, regTCC_UE_ERR_STATUS_HI), 4354 16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCC"}, 4355 AMDGPU_GFX_TCC_MEM, 1}, 4356 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTA_UE_EDC_LO, regTA_UE_EDC_HI), 4357 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TA"}, 4358 AMDGPU_GFX_TA_MEM, 4}, 4359 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCI_UE_EDC_LO_REG, regTCI_UE_EDC_HI_REG), 4360 27, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCI"}, 4361 AMDGPU_GFX_TCI_MEM, 1}, 4362 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCP_UE_EDC_LO_REG, regTCP_UE_EDC_HI_REG), 4363 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCP"}, 4364 AMDGPU_GFX_TCP_MEM, 4}, 4365 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTD_UE_EDC_LO, regTD_UE_EDC_HI), 4366 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TD"}, 4367 AMDGPU_GFX_TD_MEM, 4}, 4368 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCA_UE_ERR_STATUS_LO, regTCA_UE_ERR_STATUS_HI), 4369 2, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCA"}, 4370 AMDGPU_GFX_TCA_MEM, 1}, 4371 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regGCEA_UE_ERR_STATUS_LO, regGCEA_UE_ERR_STATUS_HI), 4372 16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GCEA"}, 4373 AMDGPU_GFX_GCEA_MEM, 1}, 4374 {{AMDGPU_RAS_REG_ENTRY(GC, 0, regLDS_UE_ERR_STATUS_LO, regLDS_UE_ERR_STATUS_HI), 4375 10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "LDS"}, 4376 AMDGPU_GFX_LDS_MEM, 4}, 4377}; 4378 4379static void gfx_v9_4_3_inst_query_ras_err_count(struct amdgpu_device *adev, 4380 void *ras_error_status, int xcc_id) 4381{ 4382 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 4383 unsigned long ce_count = 0, ue_count = 0; 4384 uint32_t i, j, k; 4385 4386 /* NOTE: convert xcc_id to physical XCD ID (XCD0 or XCD1) */ 4387 struct amdgpu_smuio_mcm_config_info mcm_info = { 4388 .socket_id = adev->smuio.funcs->get_socket_id(adev), 4389 .die_id = xcc_id & 0x01 ? 1 : 0, 4390 }; 4391 4392 mutex_lock(&adev->grbm_idx_mutex); 4393 4394 for (i = 0; i < ARRAY_SIZE(gfx_v9_4_3_ce_reg_list); i++) { 4395 for (j = 0; j < gfx_v9_4_3_ce_reg_list[i].se_num; j++) { 4396 for (k = 0; k < gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst; k++) { 4397 /* no need to select if instance number is 1 */ 4398 if (gfx_v9_4_3_ce_reg_list[i].se_num > 1 || 4399 gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst > 1) 4400 gfx_v9_4_3_xcc_select_se_sh(adev, j, 0, k, xcc_id); 4401 4402 amdgpu_ras_inst_query_ras_error_count(adev, 4403 &(gfx_v9_4_3_ce_reg_list[i].reg_entry), 4404 1, 4405 gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ce_reg_list[i].mem_id_type].mem_id_ent, 4406 gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ce_reg_list[i].mem_id_type].size, 4407 GET_INST(GC, xcc_id), 4408 AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE, 4409 &ce_count); 4410 4411 amdgpu_ras_inst_query_ras_error_count(adev, 4412 &(gfx_v9_4_3_ue_reg_list[i].reg_entry), 4413 1, 4414 gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ue_reg_list[i].mem_id_type].mem_id_ent, 4415 gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ue_reg_list[i].mem_id_type].size, 4416 GET_INST(GC, xcc_id), 4417 AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE, 4418 &ue_count); 4419 } 4420 } 4421 } 4422 4423 /* handle extra register entries of UE */ 4424 for (; i < ARRAY_SIZE(gfx_v9_4_3_ue_reg_list); i++) { 4425 for (j = 0; j < gfx_v9_4_3_ue_reg_list[i].se_num; j++) { 4426 for (k = 0; k < gfx_v9_4_3_ue_reg_list[i].reg_entry.reg_inst; k++) { 4427 /* no need to select if instance number is 1 */ 4428 if (gfx_v9_4_3_ue_reg_list[i].se_num > 1 || 4429 gfx_v9_4_3_ue_reg_list[i].reg_entry.reg_inst > 1) 4430 gfx_v9_4_3_xcc_select_se_sh(adev, j, 0, k, xcc_id); 4431 4432 amdgpu_ras_inst_query_ras_error_count(adev, 4433 &(gfx_v9_4_3_ue_reg_list[i].reg_entry), 4434 1, 4435 gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ue_reg_list[i].mem_id_type].mem_id_ent, 4436 gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ue_reg_list[i].mem_id_type].size, 4437 GET_INST(GC, xcc_id), 4438 AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE, 4439 &ue_count); 4440 } 4441 } 4442 } 4443 4444 gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 4445 xcc_id); 4446 mutex_unlock(&adev->grbm_idx_mutex); 4447 4448 /* the caller should make sure initialize value of 4449 * err_data->ue_count and err_data->ce_count 4450 */ 4451 amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count); 4452 amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count); 4453} 4454 4455static void gfx_v9_4_3_inst_reset_ras_err_count(struct amdgpu_device *adev, 4456 void *ras_error_status, int xcc_id) 4457{ 4458 uint32_t i, j, k; 4459 4460 mutex_lock(&adev->grbm_idx_mutex); 4461 4462 for (i = 0; i < ARRAY_SIZE(gfx_v9_4_3_ce_reg_list); i++) { 4463 for (j = 0; j < gfx_v9_4_3_ce_reg_list[i].se_num; j++) { 4464 for (k = 0; k < gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst; k++) { 4465 /* no need to select if instance number is 1 */ 4466 if (gfx_v9_4_3_ce_reg_list[i].se_num > 1 || 4467 gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst > 1) 4468 gfx_v9_4_3_xcc_select_se_sh(adev, j, 0, k, xcc_id); 4469 4470 amdgpu_ras_inst_reset_ras_error_count(adev, 4471 &(gfx_v9_4_3_ce_reg_list[i].reg_entry), 4472 1, 4473 GET_INST(GC, xcc_id)); 4474 4475 amdgpu_ras_inst_reset_ras_error_count(adev, 4476 &(gfx_v9_4_3_ue_reg_list[i].reg_entry), 4477 1, 4478 GET_INST(GC, xcc_id)); 4479 } 4480 } 4481 } 4482 4483 /* handle extra register entries of UE */ 4484 for (; i < ARRAY_SIZE(gfx_v9_4_3_ue_reg_list); i++) { 4485 for (j = 0; j < gfx_v9_4_3_ue_reg_list[i].se_num; j++) { 4486 for (k = 0; k < gfx_v9_4_3_ue_reg_list[i].reg_entry.reg_inst; k++) { 4487 /* no need to select if instance number is 1 */ 4488 if (gfx_v9_4_3_ue_reg_list[i].se_num > 1 || 4489 gfx_v9_4_3_ue_reg_list[i].reg_entry.reg_inst > 1) 4490 gfx_v9_4_3_xcc_select_se_sh(adev, j, 0, k, xcc_id); 4491 4492 amdgpu_ras_inst_reset_ras_error_count(adev, 4493 &(gfx_v9_4_3_ue_reg_list[i].reg_entry), 4494 1, 4495 GET_INST(GC, xcc_id)); 4496 } 4497 } 4498 } 4499 4500 gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 4501 xcc_id); 4502 mutex_unlock(&adev->grbm_idx_mutex); 4503} 4504 4505static void gfx_v9_4_3_inst_enable_watchdog_timer(struct amdgpu_device *adev, 4506 void *ras_error_status, int xcc_id) 4507{ 4508 uint32_t i; 4509 uint32_t data; 4510 4511 if (amdgpu_sriov_vf(adev)) 4512 return; 4513 4514 data = RREG32_SOC15(GC, GET_INST(GC, 0), regSQ_TIMEOUT_CONFIG); 4515 data = REG_SET_FIELD(data, SQ_TIMEOUT_CONFIG, TIMEOUT_FATAL_DISABLE, 4516 amdgpu_watchdog_timer.timeout_fatal_disable ? 1 : 0); 4517 4518 if (amdgpu_watchdog_timer.timeout_fatal_disable && 4519 (amdgpu_watchdog_timer.period < 1 || 4520 amdgpu_watchdog_timer.period > 0x23)) { 4521 dev_warn(adev->dev, "Watchdog period range is 1 to 0x23\n"); 4522 amdgpu_watchdog_timer.period = 0x23; 4523 } 4524 data = REG_SET_FIELD(data, SQ_TIMEOUT_CONFIG, PERIOD_SEL, 4525 amdgpu_watchdog_timer.period); 4526 4527 mutex_lock(&adev->grbm_idx_mutex); 4528 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 4529 gfx_v9_4_3_xcc_select_se_sh(adev, i, 0xffffffff, 0xffffffff, xcc_id); 4530 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_TIMEOUT_CONFIG, data); 4531 } 4532 gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 4533 xcc_id); 4534 mutex_unlock(&adev->grbm_idx_mutex); 4535} 4536 4537static void gfx_v9_4_3_query_ras_error_count(struct amdgpu_device *adev, 4538 void *ras_error_status) 4539{ 4540 amdgpu_gfx_ras_error_func(adev, ras_error_status, 4541 gfx_v9_4_3_inst_query_ras_err_count); 4542} 4543 4544static void gfx_v9_4_3_reset_ras_error_count(struct amdgpu_device *adev) 4545{ 4546 amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_reset_ras_err_count); 4547} 4548 4549static void gfx_v9_4_3_enable_watchdog_timer(struct amdgpu_device *adev) 4550{ 4551 amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_enable_watchdog_timer); 4552} 4553 4554static void gfx_v9_4_3_ring_insert_nop(struct amdgpu_ring *ring, uint32_t num_nop) 4555{ 4556 /* Header itself is a NOP packet */ 4557 if (num_nop == 1) { 4558 amdgpu_ring_write(ring, ring->funcs->nop); 4559 return; 4560 } 4561 4562 /* Max HW optimization till 0x3ffe, followed by remaining one NOP at a time*/ 4563 amdgpu_ring_write(ring, PACKET3(PACKET3_NOP, min(num_nop - 2, 0x3ffe))); 4564 4565 /* Header is at index 0, followed by num_nops - 1 NOP packet's */ 4566 amdgpu_ring_insert_nop(ring, num_nop - 1); 4567} 4568 4569static void gfx_v9_4_3_ip_print(struct amdgpu_ip_block *ip_block, struct drm_printer *p) 4570{ 4571 struct amdgpu_device *adev = ip_block->adev; 4572 uint32_t i, j, k; 4573 uint32_t xcc_id, xcc_offset, inst_offset; 4574 uint32_t num_xcc, reg, num_inst; 4575 uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9_4_3); 4576 4577 if (!adev->gfx.ip_dump_core) 4578 return; 4579 4580 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 4581 drm_printf(p, "Number of Instances:%d\n", num_xcc); 4582 for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { 4583 xcc_offset = xcc_id * reg_count; 4584 drm_printf(p, "\nInstance id:%d\n", xcc_id); 4585 for (i = 0; i < reg_count; i++) 4586 drm_printf(p, "%-50s \t 0x%08x\n", 4587 gc_reg_list_9_4_3[i].reg_name, 4588 adev->gfx.ip_dump_core[xcc_offset + i]); 4589 } 4590 4591 /* print compute queue registers for all instances */ 4592 if (!adev->gfx.ip_dump_compute_queues) 4593 return; 4594 4595 num_inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec * 4596 adev->gfx.mec.num_queue_per_pipe; 4597 4598 reg_count = ARRAY_SIZE(gc_cp_reg_list_9_4_3); 4599 drm_printf(p, "\nnum_xcc: %d num_mec: %d num_pipe: %d num_queue: %d\n", 4600 num_xcc, 4601 adev->gfx.mec.num_mec, 4602 adev->gfx.mec.num_pipe_per_mec, 4603 adev->gfx.mec.num_queue_per_pipe); 4604 4605 for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { 4606 xcc_offset = xcc_id * reg_count * num_inst; 4607 inst_offset = 0; 4608 for (i = 0; i < adev->gfx.mec.num_mec; i++) { 4609 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) { 4610 for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) { 4611 drm_printf(p, 4612 "\nxcc:%d mec:%d, pipe:%d, queue:%d\n", 4613 xcc_id, i, j, k); 4614 for (reg = 0; reg < reg_count; reg++) { 4615 if (i && gc_cp_reg_list_9_4_3[reg].reg_offset == 4616 regCP_MEC_ME1_HEADER_DUMP) 4617 drm_printf(p, 4618 "%-50s \t 0x%08x\n", 4619 "regCP_MEC_ME2_HEADER_DUMP", 4620 adev->gfx.ip_dump_compute_queues 4621 [xcc_offset + inst_offset + 4622 reg]); 4623 else 4624 drm_printf(p, 4625 "%-50s \t 0x%08x\n", 4626 gc_cp_reg_list_9_4_3[reg].reg_name, 4627 adev->gfx.ip_dump_compute_queues 4628 [xcc_offset + inst_offset + 4629 reg]); 4630 } 4631 inst_offset += reg_count; 4632 } 4633 } 4634 } 4635 } 4636} 4637 4638static void gfx_v9_4_3_ip_dump(struct amdgpu_ip_block *ip_block) 4639{ 4640 struct amdgpu_device *adev = ip_block->adev; 4641 uint32_t i, j, k; 4642 uint32_t num_xcc, reg, num_inst; 4643 uint32_t xcc_id, xcc_offset, inst_offset; 4644 uint32_t reg_count = ARRAY_SIZE(gc_reg_list_9_4_3); 4645 4646 if (!adev->gfx.ip_dump_core) 4647 return; 4648 4649 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 4650 4651 for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { 4652 xcc_offset = xcc_id * reg_count; 4653 for (i = 0; i < reg_count; i++) 4654 adev->gfx.ip_dump_core[xcc_offset + i] = 4655 RREG32(SOC15_REG_ENTRY_OFFSET_INST(gc_reg_list_9_4_3[i], 4656 GET_INST(GC, xcc_id))); 4657 } 4658 4659 /* dump compute queue registers for all instances */ 4660 if (!adev->gfx.ip_dump_compute_queues) 4661 return; 4662 4663 num_inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec * 4664 adev->gfx.mec.num_queue_per_pipe; 4665 reg_count = ARRAY_SIZE(gc_cp_reg_list_9_4_3); 4666 mutex_lock(&adev->srbm_mutex); 4667 for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) { 4668 xcc_offset = xcc_id * reg_count * num_inst; 4669 inst_offset = 0; 4670 for (i = 0; i < adev->gfx.mec.num_mec; i++) { 4671 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) { 4672 for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) { 4673 /* ME0 is for GFX so start from 1 for CP */ 4674 soc15_grbm_select(adev, 1 + i, j, k, 0, 4675 GET_INST(GC, xcc_id)); 4676 4677 for (reg = 0; reg < reg_count; reg++) { 4678 if (i && gc_cp_reg_list_9_4_3[reg].reg_offset == 4679 regCP_MEC_ME1_HEADER_DUMP) 4680 adev->gfx.ip_dump_compute_queues 4681 [xcc_offset + 4682 inst_offset + reg] = 4683 RREG32(SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), 4684 regCP_MEC_ME2_HEADER_DUMP)); 4685 else 4686 adev->gfx.ip_dump_compute_queues 4687 [xcc_offset + 4688 inst_offset + reg] = 4689 RREG32(SOC15_REG_ENTRY_OFFSET_INST( 4690 gc_cp_reg_list_9_4_3[reg], 4691 GET_INST(GC, xcc_id))); 4692 } 4693 inst_offset += reg_count; 4694 } 4695 } 4696 } 4697 } 4698 soc15_grbm_select(adev, 0, 0, 0, 0, 0); 4699 mutex_unlock(&adev->srbm_mutex); 4700} 4701 4702static void gfx_v9_4_3_ring_emit_cleaner_shader(struct amdgpu_ring *ring) 4703{ 4704 /* Emit the cleaner shader */ 4705 amdgpu_ring_write(ring, PACKET3(PACKET3_RUN_CLEANER_SHADER, 0)); 4706 amdgpu_ring_write(ring, 0); /* RESERVED field, programmed to zero */ 4707} 4708 4709static const struct amd_ip_funcs gfx_v9_4_3_ip_funcs = { 4710 .name = "gfx_v9_4_3", 4711 .early_init = gfx_v9_4_3_early_init, 4712 .late_init = gfx_v9_4_3_late_init, 4713 .sw_init = gfx_v9_4_3_sw_init, 4714 .sw_fini = gfx_v9_4_3_sw_fini, 4715 .hw_init = gfx_v9_4_3_hw_init, 4716 .hw_fini = gfx_v9_4_3_hw_fini, 4717 .suspend = gfx_v9_4_3_suspend, 4718 .resume = gfx_v9_4_3_resume, 4719 .is_idle = gfx_v9_4_3_is_idle, 4720 .wait_for_idle = gfx_v9_4_3_wait_for_idle, 4721 .soft_reset = gfx_v9_4_3_soft_reset, 4722 .set_clockgating_state = gfx_v9_4_3_set_clockgating_state, 4723 .set_powergating_state = gfx_v9_4_3_set_powergating_state, 4724 .get_clockgating_state = gfx_v9_4_3_get_clockgating_state, 4725 .dump_ip_state = gfx_v9_4_3_ip_dump, 4726 .print_ip_state = gfx_v9_4_3_ip_print, 4727}; 4728 4729static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_compute = { 4730 .type = AMDGPU_RING_TYPE_COMPUTE, 4731 .align_mask = 0xff, 4732 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 4733 .support_64bit_ptrs = true, 4734 .get_rptr = gfx_v9_4_3_ring_get_rptr_compute, 4735 .get_wptr = gfx_v9_4_3_ring_get_wptr_compute, 4736 .set_wptr = gfx_v9_4_3_ring_set_wptr_compute, 4737 .emit_frame_size = 4738 20 + /* gfx_v9_4_3_ring_emit_gds_switch */ 4739 7 + /* gfx_v9_4_3_ring_emit_hdp_flush */ 4740 5 + /* hdp invalidate */ 4741 7 + /* gfx_v9_4_3_ring_emit_pipeline_sync */ 4742 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 4743 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 4744 2 + /* gfx_v9_4_3_ring_emit_vm_flush */ 4745 8 + 8 + 8 + /* gfx_v9_4_3_ring_emit_fence x3 for user fence, vm fence */ 4746 7 + /* gfx_v9_4_3_emit_mem_sync */ 4747 5 + /* gfx_v9_4_3_emit_wave_limit for updating regSPI_WCL_PIPE_PERCENT_GFX register */ 4748 15 + /* for updating 3 regSPI_WCL_PIPE_PERCENT_CS registers */ 4749 2, /* gfx_v9_4_3_ring_emit_cleaner_shader */ 4750 .emit_ib_size = 7, /* gfx_v9_4_3_ring_emit_ib_compute */ 4751 .emit_ib = gfx_v9_4_3_ring_emit_ib_compute, 4752 .emit_fence = gfx_v9_4_3_ring_emit_fence, 4753 .emit_pipeline_sync = gfx_v9_4_3_ring_emit_pipeline_sync, 4754 .emit_vm_flush = gfx_v9_4_3_ring_emit_vm_flush, 4755 .emit_gds_switch = gfx_v9_4_3_ring_emit_gds_switch, 4756 .emit_hdp_flush = gfx_v9_4_3_ring_emit_hdp_flush, 4757 .test_ring = gfx_v9_4_3_ring_test_ring, 4758 .test_ib = gfx_v9_4_3_ring_test_ib, 4759 .insert_nop = gfx_v9_4_3_ring_insert_nop, 4760 .pad_ib = amdgpu_ring_generic_pad_ib, 4761 .emit_wreg = gfx_v9_4_3_ring_emit_wreg, 4762 .emit_reg_wait = gfx_v9_4_3_ring_emit_reg_wait, 4763 .emit_reg_write_reg_wait = gfx_v9_4_3_ring_emit_reg_write_reg_wait, 4764 .soft_recovery = gfx_v9_4_3_ring_soft_recovery, 4765 .emit_mem_sync = gfx_v9_4_3_emit_mem_sync, 4766 .emit_wave_limit = gfx_v9_4_3_emit_wave_limit, 4767 .reset = gfx_v9_4_3_reset_kcq, 4768 .emit_cleaner_shader = gfx_v9_4_3_ring_emit_cleaner_shader, 4769 .begin_use = amdgpu_gfx_enforce_isolation_ring_begin_use, 4770 .end_use = amdgpu_gfx_enforce_isolation_ring_end_use, 4771}; 4772 4773static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_kiq = { 4774 .type = AMDGPU_RING_TYPE_KIQ, 4775 .align_mask = 0xff, 4776 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 4777 .support_64bit_ptrs = true, 4778 .get_rptr = gfx_v9_4_3_ring_get_rptr_compute, 4779 .get_wptr = gfx_v9_4_3_ring_get_wptr_compute, 4780 .set_wptr = gfx_v9_4_3_ring_set_wptr_compute, 4781 .emit_frame_size = 4782 20 + /* gfx_v9_4_3_ring_emit_gds_switch */ 4783 7 + /* gfx_v9_4_3_ring_emit_hdp_flush */ 4784 5 + /* hdp invalidate */ 4785 7 + /* gfx_v9_4_3_ring_emit_pipeline_sync */ 4786 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 4787 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 4788 2 + /* gfx_v9_4_3_ring_emit_vm_flush */ 4789 8 + 8 + 8, /* gfx_v9_4_3_ring_emit_fence_kiq x3 for user fence, vm fence */ 4790 .emit_ib_size = 7, /* gfx_v9_4_3_ring_emit_ib_compute */ 4791 .emit_fence = gfx_v9_4_3_ring_emit_fence_kiq, 4792 .test_ring = gfx_v9_4_3_ring_test_ring, 4793 .insert_nop = amdgpu_ring_insert_nop, 4794 .pad_ib = amdgpu_ring_generic_pad_ib, 4795 .emit_rreg = gfx_v9_4_3_ring_emit_rreg, 4796 .emit_wreg = gfx_v9_4_3_ring_emit_wreg, 4797 .emit_reg_wait = gfx_v9_4_3_ring_emit_reg_wait, 4798 .emit_reg_write_reg_wait = gfx_v9_4_3_ring_emit_reg_write_reg_wait, 4799 .emit_hdp_flush = gfx_v9_4_3_ring_emit_hdp_flush, 4800}; 4801 4802static void gfx_v9_4_3_set_ring_funcs(struct amdgpu_device *adev) 4803{ 4804 int i, j, num_xcc; 4805 4806 num_xcc = NUM_XCC(adev->gfx.xcc_mask); 4807 for (i = 0; i < num_xcc; i++) { 4808 adev->gfx.kiq[i].ring.funcs = &gfx_v9_4_3_ring_funcs_kiq; 4809 4810 for (j = 0; j < adev->gfx.num_compute_rings; j++) 4811 adev->gfx.compute_ring[j + i * adev->gfx.num_compute_rings].funcs 4812 = &gfx_v9_4_3_ring_funcs_compute; 4813 } 4814} 4815 4816static const struct amdgpu_irq_src_funcs gfx_v9_4_3_eop_irq_funcs = { 4817 .set = gfx_v9_4_3_set_eop_interrupt_state, 4818 .process = gfx_v9_4_3_eop_irq, 4819}; 4820 4821static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_reg_irq_funcs = { 4822 .set = gfx_v9_4_3_set_priv_reg_fault_state, 4823 .process = gfx_v9_4_3_priv_reg_irq, 4824}; 4825 4826static const struct amdgpu_irq_src_funcs gfx_v9_4_3_bad_op_irq_funcs = { 4827 .set = gfx_v9_4_3_set_bad_op_fault_state, 4828 .process = gfx_v9_4_3_bad_op_irq, 4829}; 4830 4831static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_inst_irq_funcs = { 4832 .set = gfx_v9_4_3_set_priv_inst_fault_state, 4833 .process = gfx_v9_4_3_priv_inst_irq, 4834}; 4835 4836static void gfx_v9_4_3_set_irq_funcs(struct amdgpu_device *adev) 4837{ 4838 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 4839 adev->gfx.eop_irq.funcs = &gfx_v9_4_3_eop_irq_funcs; 4840 4841 adev->gfx.priv_reg_irq.num_types = 1; 4842 adev->gfx.priv_reg_irq.funcs = &gfx_v9_4_3_priv_reg_irq_funcs; 4843 4844 adev->gfx.bad_op_irq.num_types = 1; 4845 adev->gfx.bad_op_irq.funcs = &gfx_v9_4_3_bad_op_irq_funcs; 4846 4847 adev->gfx.priv_inst_irq.num_types = 1; 4848 adev->gfx.priv_inst_irq.funcs = &gfx_v9_4_3_priv_inst_irq_funcs; 4849} 4850 4851static void gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device *adev) 4852{ 4853 adev->gfx.rlc.funcs = &gfx_v9_4_3_rlc_funcs; 4854} 4855 4856 4857static void gfx_v9_4_3_set_gds_init(struct amdgpu_device *adev) 4858{ 4859 /* 9.4.3 variants removed all the GDS internal memory, 4860 * only support GWS opcode in kernel, like barrier 4861 * semaphore.etc */ 4862 4863 /* init asic gds info */ 4864 adev->gds.gds_size = 0; 4865 adev->gds.gds_compute_max_wave_id = 0; 4866 adev->gds.gws_size = 64; 4867 adev->gds.oa_size = 16; 4868} 4869 4870static void gfx_v9_4_3_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 4871 u32 bitmap, int xcc_id) 4872{ 4873 u32 data; 4874 4875 if (!bitmap) 4876 return; 4877 4878 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 4879 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 4880 4881 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regGC_USER_SHADER_ARRAY_CONFIG, data); 4882} 4883 4884static u32 gfx_v9_4_3_get_cu_active_bitmap(struct amdgpu_device *adev, int xcc_id) 4885{ 4886 u32 data, mask; 4887 4888 data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCC_GC_SHADER_ARRAY_CONFIG); 4889 data |= RREG32_SOC15(GC, GET_INST(GC, xcc_id), regGC_USER_SHADER_ARRAY_CONFIG); 4890 4891 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 4892 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 4893 4894 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 4895 4896 return (~data) & mask; 4897} 4898 4899static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev, 4900 struct amdgpu_cu_info *cu_info) 4901{ 4902 int i, j, k, prev_counter, counter, xcc_id, active_cu_number = 0; 4903 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0, tmp; 4904 unsigned disable_masks[4 * 4]; 4905 bool is_symmetric_cus; 4906 4907 if (!adev || !cu_info) 4908 return -EINVAL; 4909 4910 /* 4911 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs 4912 */ 4913 if (adev->gfx.config.max_shader_engines * 4914 adev->gfx.config.max_sh_per_se > 16) 4915 return -EINVAL; 4916 4917 amdgpu_gfx_parse_disable_cu(disable_masks, 4918 adev->gfx.config.max_shader_engines, 4919 adev->gfx.config.max_sh_per_se); 4920 4921 mutex_lock(&adev->grbm_idx_mutex); 4922 for (xcc_id = 0; xcc_id < NUM_XCC(adev->gfx.xcc_mask); xcc_id++) { 4923 is_symmetric_cus = true; 4924 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 4925 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 4926 mask = 1; 4927 ao_bitmap = 0; 4928 counter = 0; 4929 gfx_v9_4_3_xcc_select_se_sh(adev, i, j, 0xffffffff, xcc_id); 4930 gfx_v9_4_3_set_user_cu_inactive_bitmap( 4931 adev, 4932 disable_masks[i * adev->gfx.config.max_sh_per_se + j], 4933 xcc_id); 4934 bitmap = gfx_v9_4_3_get_cu_active_bitmap(adev, xcc_id); 4935 4936 cu_info->bitmap[xcc_id][i][j] = bitmap; 4937 4938 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) { 4939 if (bitmap & mask) { 4940 if (counter < adev->gfx.config.max_cu_per_sh) 4941 ao_bitmap |= mask; 4942 counter++; 4943 } 4944 mask <<= 1; 4945 } 4946 active_cu_number += counter; 4947 if (i < 2 && j < 2) 4948 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 4949 cu_info->ao_cu_bitmap[i][j] = ao_bitmap; 4950 } 4951 if (i && is_symmetric_cus && prev_counter != counter) 4952 is_symmetric_cus = false; 4953 prev_counter = counter; 4954 } 4955 if (is_symmetric_cus) { 4956 tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_DEBUG); 4957 tmp = REG_SET_FIELD(tmp, CP_CPC_DEBUG, CPC_HARVESTING_RELAUNCH_DISABLE, 1); 4958 tmp = REG_SET_FIELD(tmp, CP_CPC_DEBUG, CPC_HARVESTING_DISPATCH_DISABLE, 1); 4959 WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_DEBUG, tmp); 4960 } 4961 gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 4962 xcc_id); 4963 } 4964 mutex_unlock(&adev->grbm_idx_mutex); 4965 4966 cu_info->number = active_cu_number; 4967 cu_info->ao_cu_mask = ao_cu_mask; 4968 cu_info->simd_per_cu = NUM_SIMD_PER_CU; 4969 4970 return 0; 4971} 4972 4973const struct amdgpu_ip_block_version gfx_v9_4_3_ip_block = { 4974 .type = AMD_IP_BLOCK_TYPE_GFX, 4975 .major = 9, 4976 .minor = 4, 4977 .rev = 3, 4978 .funcs = &gfx_v9_4_3_ip_funcs, 4979}; 4980 4981static int gfx_v9_4_3_xcp_resume(void *handle, uint32_t inst_mask) 4982{ 4983 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4984 uint32_t tmp_mask; 4985 int i, r; 4986 4987 /* TODO : Initialize golden regs */ 4988 /* gfx_v9_4_3_init_golden_registers(adev); */ 4989 4990 tmp_mask = inst_mask; 4991 for_each_inst(i, tmp_mask) 4992 gfx_v9_4_3_xcc_constants_init(adev, i); 4993 4994 if (!amdgpu_sriov_vf(adev)) { 4995 tmp_mask = inst_mask; 4996 for_each_inst(i, tmp_mask) { 4997 r = gfx_v9_4_3_xcc_rlc_resume(adev, i); 4998 if (r) 4999 return r; 5000 } 5001 } 5002 5003 tmp_mask = inst_mask; 5004 for_each_inst(i, tmp_mask) { 5005 r = gfx_v9_4_3_xcc_cp_resume(adev, i); 5006 if (r) 5007 return r; 5008 } 5009 5010 return 0; 5011} 5012 5013static int gfx_v9_4_3_xcp_suspend(void *handle, uint32_t inst_mask) 5014{ 5015 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5016 int i; 5017 5018 for_each_inst(i, inst_mask) 5019 gfx_v9_4_3_xcc_fini(adev, i); 5020 5021 return 0; 5022} 5023 5024struct amdgpu_xcp_ip_funcs gfx_v9_4_3_xcp_funcs = { 5025 .suspend = &gfx_v9_4_3_xcp_suspend, 5026 .resume = &gfx_v9_4_3_xcp_resume 5027}; 5028 5029struct amdgpu_ras_block_hw_ops gfx_v9_4_3_ras_ops = { 5030 .query_ras_error_count = &gfx_v9_4_3_query_ras_error_count, 5031 .reset_ras_error_count = &gfx_v9_4_3_reset_ras_error_count, 5032}; 5033 5034static int gfx_v9_4_3_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block) 5035{ 5036 int r; 5037 5038 r = amdgpu_ras_block_late_init(adev, ras_block); 5039 if (r) 5040 return r; 5041 5042 r = amdgpu_ras_bind_aca(adev, AMDGPU_RAS_BLOCK__GFX, 5043 &gfx_v9_4_3_aca_info, 5044 NULL); 5045 if (r) 5046 goto late_fini; 5047 5048 return 0; 5049 5050late_fini: 5051 amdgpu_ras_block_late_fini(adev, ras_block); 5052 5053 return r; 5054} 5055 5056struct amdgpu_gfx_ras gfx_v9_4_3_ras = { 5057 .ras_block = { 5058 .hw_ops = &gfx_v9_4_3_ras_ops, 5059 .ras_late_init = &gfx_v9_4_3_ras_late_init, 5060 }, 5061 .enable_watchdog_timer = &gfx_v9_4_3_enable_watchdog_timer, 5062};
~/
Read-only file preview.