1 /* 2 * Copyright 2016 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 #include <linux/firmware.h> 24 #include <linux/slab.h> 25 #include <linux/module.h> 26 #include <linux/pci.h> 27 28 #include <drm/amdgpu_drm.h> 29 30 #include "amdgpu.h" 31 #include "amdgpu_atombios.h" 32 #include "amdgpu_ih.h" 33 #include "amdgpu_uvd.h" 34 #include "amdgpu_vce.h" 35 #include "amdgpu_ucode.h" 36 #include "amdgpu_psp.h" 37 #include "atom.h" 38 #include "amd_pcie.h" 39 40 #include "uvd/uvd_7_0_offset.h" 41 #include "gc/gc_9_0_offset.h" 42 #include "gc/gc_9_0_sh_mask.h" 43 #include "sdma0/sdma0_4_0_offset.h" 44 #include "sdma1/sdma1_4_0_offset.h" 45 #include "nbio/nbio_7_0_default.h" 46 #include "nbio/nbio_7_0_offset.h" 47 #include "nbio/nbio_7_0_sh_mask.h" 48 #include "nbio/nbio_7_0_smn.h" 49 #include "mp/mp_9_0_offset.h" 50 51 #include "soc15.h" 52 #include "soc15_common.h" 53 #include "gfx_v9_0.h" 54 #include "gmc_v9_0.h" 55 #include "gfxhub_v1_0.h" 56 #include "mmhub_v1_0.h" 57 #include "df_v1_7.h" 58 #include "df_v3_6.h" 59 #include "nbio_v6_1.h" 60 #include "nbio_v7_0.h" 61 #include "nbio_v7_4.h" 62 #include "hdp_v4_0.h" 63 #include "vega10_ih.h" 64 #include "vega20_ih.h" 65 #include "navi10_ih.h" 66 #include "sdma_v4_0.h" 67 #include "uvd_v7_0.h" 68 #include "vce_v4_0.h" 69 #include "vcn_v1_0.h" 70 #include "vcn_v2_0.h" 71 #include "jpeg_v2_0.h" 72 #include "vcn_v2_5.h" 73 #include "jpeg_v2_5.h" 74 #include "smuio_v9_0.h" 75 #include "smuio_v11_0.h" 76 #include "smuio_v13_0.h" 77 #include "amdgpu_vkms.h" 78 #include "mxgpu_ai.h" 79 #include "amdgpu_ras.h" 80 #include "amdgpu_xgmi.h" 81 #include <uapi/linux/kfd_ioctl.h> 82 83 #define mmMP0_MISC_CGTT_CTRL0 0x01b9 84 #define mmMP0_MISC_CGTT_CTRL0_BASE_IDX 0 85 #define mmMP0_MISC_LIGHT_SLEEP_CTRL 0x01ba 86 #define mmMP0_MISC_LIGHT_SLEEP_CTRL_BASE_IDX 0 87 88 static const struct amd_ip_funcs soc15_common_ip_funcs; 89 90 /* Vega, Raven, Arcturus */ 91 static const struct amdgpu_video_codec_info vega_video_codecs_encode_array[] = 92 { 93 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 2304, 0)}, 94 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 4096, 2304, 0)}, 95 }; 96 97 static const struct amdgpu_video_codecs vega_video_codecs_encode = 98 { 99 .codec_count = ARRAY_SIZE(vega_video_codecs_encode_array), 100 .codec_array = vega_video_codecs_encode_array, 101 }; 102 103 /* Vega */ 104 static const struct amdgpu_video_codec_info vega_video_codecs_decode_array[] = 105 { 106 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2, 4096, 4096, 3)}, 107 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4096, 5)}, 108 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)}, 109 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4096, 4)}, 110 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 4096, 4096, 186)}, 111 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)}, 112 }; 113 114 static const struct amdgpu_video_codecs vega_video_codecs_decode = 115 { 116 .codec_count = ARRAY_SIZE(vega_video_codecs_decode_array), 117 .codec_array = vega_video_codecs_decode_array, 118 }; 119 120 /* Raven */ 121 static const struct amdgpu_video_codec_info rv_video_codecs_decode_array[] = 122 { 123 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2, 4096, 4096, 3)}, 124 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4096, 5)}, 125 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)}, 126 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4096, 4)}, 127 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 4096, 4096, 186)}, 128 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)}, 129 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9, 4096, 4096, 0)}, 130 }; 131 132 static const struct amdgpu_video_codecs rv_video_codecs_decode = 133 { 134 .codec_count = ARRAY_SIZE(rv_video_codecs_decode_array), 135 .codec_array = rv_video_codecs_decode_array, 136 }; 137 138 /* Renoir, Arcturus */ 139 static const struct amdgpu_video_codec_info rn_video_codecs_decode_array[] = 140 { 141 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2, 4096, 4096, 3)}, 142 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4096, 5)}, 143 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)}, 144 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4096, 4)}, 145 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 8192, 4352, 186)}, 146 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)}, 147 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9, 8192, 4352, 0)}, 148 }; 149 150 static const struct amdgpu_video_codecs rn_video_codecs_decode = 151 { 152 .codec_count = ARRAY_SIZE(rn_video_codecs_decode_array), 153 .codec_array = rn_video_codecs_decode_array, 154 }; 155 156 static const struct amdgpu_video_codec_info vcn_4_0_3_video_codecs_decode_array[] = { 157 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)}, 158 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 8192, 4352, 186)}, 159 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)}, 160 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9, 8192, 4352, 0)}, 161 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1, 8192, 4352, 0)}, 162 }; 163 164 static const struct amdgpu_video_codecs vcn_4_0_3_video_codecs_decode = { 165 .codec_count = ARRAY_SIZE(vcn_4_0_3_video_codecs_decode_array), 166 .codec_array = vcn_4_0_3_video_codecs_decode_array, 167 }; 168 169 static const struct amdgpu_video_codecs vcn_4_0_3_video_codecs_encode = { 170 .codec_count = 0, 171 .codec_array = NULL, 172 }; 173 174 static int soc15_query_video_codecs(struct amdgpu_device *adev, bool encode, 175 const struct amdgpu_video_codecs **codecs) 176 { 177 if (adev->ip_versions[VCE_HWIP][0]) { 178 switch (adev->ip_versions[VCE_HWIP][0]) { 179 case IP_VERSION(4, 0, 0): 180 case IP_VERSION(4, 1, 0): 181 if (encode) 182 *codecs = &vega_video_codecs_encode; 183 else 184 *codecs = &vega_video_codecs_decode; 185 return 0; 186 default: 187 return -EINVAL; 188 } 189 } else { 190 switch (adev->ip_versions[UVD_HWIP][0]) { 191 case IP_VERSION(1, 0, 0): 192 case IP_VERSION(1, 0, 1): 193 if (encode) 194 *codecs = &vega_video_codecs_encode; 195 else 196 *codecs = &rv_video_codecs_decode; 197 return 0; 198 case IP_VERSION(2, 5, 0): 199 case IP_VERSION(2, 6, 0): 200 case IP_VERSION(2, 2, 0): 201 if (encode) 202 *codecs = &vega_video_codecs_encode; 203 else 204 *codecs = &rn_video_codecs_decode; 205 return 0; 206 case IP_VERSION(4, 0, 3): 207 if (encode) 208 *codecs = &vcn_4_0_3_video_codecs_encode; 209 else 210 *codecs = &vcn_4_0_3_video_codecs_decode; 211 return 0; 212 default: 213 return -EINVAL; 214 } 215 } 216 } 217 218 static u32 soc15_uvd_ctx_rreg(struct amdgpu_device *adev, u32 reg) 219 { 220 unsigned long flags, address, data; 221 u32 r; 222 223 address = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_INDEX); 224 data = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_DATA); 225 226 spin_lock_irqsave(&adev->uvd_ctx_idx_lock, flags); 227 WREG32(address, ((reg) & 0x1ff)); 228 r = RREG32(data); 229 spin_unlock_irqrestore(&adev->uvd_ctx_idx_lock, flags); 230 return r; 231 } 232 233 static void soc15_uvd_ctx_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 234 { 235 unsigned long flags, address, data; 236 237 address = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_INDEX); 238 data = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_DATA); 239 240 spin_lock_irqsave(&adev->uvd_ctx_idx_lock, flags); 241 WREG32(address, ((reg) & 0x1ff)); 242 WREG32(data, (v)); 243 spin_unlock_irqrestore(&adev->uvd_ctx_idx_lock, flags); 244 } 245 246 static u32 soc15_didt_rreg(struct amdgpu_device *adev, u32 reg) 247 { 248 unsigned long flags, address, data; 249 u32 r; 250 251 address = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_INDEX); 252 data = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_DATA); 253 254 spin_lock_irqsave(&adev->didt_idx_lock, flags); 255 WREG32(address, (reg)); 256 r = RREG32(data); 257 spin_unlock_irqrestore(&adev->didt_idx_lock, flags); 258 return r; 259 } 260 261 static void soc15_didt_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 262 { 263 unsigned long flags, address, data; 264 265 address = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_INDEX); 266 data = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_DATA); 267 268 spin_lock_irqsave(&adev->didt_idx_lock, flags); 269 WREG32(address, (reg)); 270 WREG32(data, (v)); 271 spin_unlock_irqrestore(&adev->didt_idx_lock, flags); 272 } 273 274 static u32 soc15_gc_cac_rreg(struct amdgpu_device *adev, u32 reg) 275 { 276 unsigned long flags; 277 u32 r; 278 279 spin_lock_irqsave(&adev->gc_cac_idx_lock, flags); 280 WREG32_SOC15(GC, 0, mmGC_CAC_IND_INDEX, (reg)); 281 r = RREG32_SOC15(GC, 0, mmGC_CAC_IND_DATA); 282 spin_unlock_irqrestore(&adev->gc_cac_idx_lock, flags); 283 return r; 284 } 285 286 static void soc15_gc_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 287 { 288 unsigned long flags; 289 290 spin_lock_irqsave(&adev->gc_cac_idx_lock, flags); 291 WREG32_SOC15(GC, 0, mmGC_CAC_IND_INDEX, (reg)); 292 WREG32_SOC15(GC, 0, mmGC_CAC_IND_DATA, (v)); 293 spin_unlock_irqrestore(&adev->gc_cac_idx_lock, flags); 294 } 295 296 static u32 soc15_se_cac_rreg(struct amdgpu_device *adev, u32 reg) 297 { 298 unsigned long flags; 299 u32 r; 300 301 spin_lock_irqsave(&adev->se_cac_idx_lock, flags); 302 WREG32_SOC15(GC, 0, mmSE_CAC_IND_INDEX, (reg)); 303 r = RREG32_SOC15(GC, 0, mmSE_CAC_IND_DATA); 304 spin_unlock_irqrestore(&adev->se_cac_idx_lock, flags); 305 return r; 306 } 307 308 static void soc15_se_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 309 { 310 unsigned long flags; 311 312 spin_lock_irqsave(&adev->se_cac_idx_lock, flags); 313 WREG32_SOC15(GC, 0, mmSE_CAC_IND_INDEX, (reg)); 314 WREG32_SOC15(GC, 0, mmSE_CAC_IND_DATA, (v)); 315 spin_unlock_irqrestore(&adev->se_cac_idx_lock, flags); 316 } 317 318 static u32 soc15_get_config_memsize(struct amdgpu_device *adev) 319 { 320 return adev->nbio.funcs->get_memsize(adev); 321 } 322 323 static u32 soc15_get_xclk(struct amdgpu_device *adev) 324 { 325 u32 reference_clock = adev->clock.spll.reference_freq; 326 327 if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(12, 0, 0) || 328 adev->ip_versions[MP1_HWIP][0] == IP_VERSION(12, 0, 1) || 329 adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 6)) 330 return 10000; 331 if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(10, 0, 0) || 332 adev->ip_versions[MP1_HWIP][0] == IP_VERSION(10, 0, 1)) 333 return reference_clock / 4; 334 335 return reference_clock; 336 } 337 338 339 void soc15_grbm_select(struct amdgpu_device *adev, 340 u32 me, u32 pipe, u32 queue, u32 vmid, int xcc_id) 341 { 342 u32 grbm_gfx_cntl = 0; 343 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, PIPEID, pipe); 344 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, MEID, me); 345 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, VMID, vmid); 346 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, QUEUEID, queue); 347 348 WREG32_SOC15_RLC_SHADOW(GC, xcc_id, mmGRBM_GFX_CNTL, grbm_gfx_cntl); 349 } 350 351 static bool soc15_read_disabled_bios(struct amdgpu_device *adev) 352 { 353 /* todo */ 354 return false; 355 } 356 357 static struct soc15_allowed_register_entry soc15_allowed_read_registers[] = { 358 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS)}, 359 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS2)}, 360 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE0)}, 361 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE1)}, 362 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE2)}, 363 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE3)}, 364 { SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_STATUS_REG)}, 365 { SOC15_REG_ENTRY(SDMA1, 0, mmSDMA1_STATUS_REG)}, 366 { SOC15_REG_ENTRY(GC, 0, mmCP_STAT)}, 367 { SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT1)}, 368 { SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT2)}, 369 { SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT3)}, 370 { SOC15_REG_ENTRY(GC, 0, mmCP_CPF_BUSY_STAT)}, 371 { SOC15_REG_ENTRY(GC, 0, mmCP_CPF_STALLED_STAT1)}, 372 { SOC15_REG_ENTRY(GC, 0, mmCP_CPF_STATUS)}, 373 { SOC15_REG_ENTRY(GC, 0, mmCP_CPC_BUSY_STAT)}, 374 { SOC15_REG_ENTRY(GC, 0, mmCP_CPC_STALLED_STAT1)}, 375 { SOC15_REG_ENTRY(GC, 0, mmCP_CPC_STATUS)}, 376 { SOC15_REG_ENTRY(GC, 0, mmGB_ADDR_CONFIG)}, 377 { SOC15_REG_ENTRY(GC, 0, mmDB_DEBUG2)}, 378 }; 379 380 static uint32_t soc15_read_indexed_register(struct amdgpu_device *adev, u32 se_num, 381 u32 sh_num, u32 reg_offset) 382 { 383 uint32_t val; 384 385 mutex_lock(&adev->grbm_idx_mutex); 386 if (se_num != 0xffffffff || sh_num != 0xffffffff) 387 amdgpu_gfx_select_se_sh(adev, se_num, sh_num, 0xffffffff, 0); 388 389 val = RREG32(reg_offset); 390 391 if (se_num != 0xffffffff || sh_num != 0xffffffff) 392 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 393 mutex_unlock(&adev->grbm_idx_mutex); 394 return val; 395 } 396 397 static uint32_t soc15_get_register_value(struct amdgpu_device *adev, 398 bool indexed, u32 se_num, 399 u32 sh_num, u32 reg_offset) 400 { 401 if (indexed) { 402 return soc15_read_indexed_register(adev, se_num, sh_num, reg_offset); 403 } else { 404 if (reg_offset == SOC15_REG_OFFSET(GC, 0, mmGB_ADDR_CONFIG)) 405 return adev->gfx.config.gb_addr_config; 406 else if (reg_offset == SOC15_REG_OFFSET(GC, 0, mmDB_DEBUG2)) 407 return adev->gfx.config.db_debug2; 408 return RREG32(reg_offset); 409 } 410 } 411 412 static int soc15_read_register(struct amdgpu_device *adev, u32 se_num, 413 u32 sh_num, u32 reg_offset, u32 *value) 414 { 415 uint32_t i; 416 struct soc15_allowed_register_entry *en; 417 418 *value = 0; 419 for (i = 0; i < ARRAY_SIZE(soc15_allowed_read_registers); i++) { 420 en = &soc15_allowed_read_registers[i]; 421 if (!adev->reg_offset[en->hwip][en->inst]) 422 continue; 423 else if (reg_offset != (adev->reg_offset[en->hwip][en->inst][en->seg] 424 + en->reg_offset)) 425 continue; 426 427 *value = soc15_get_register_value(adev, 428 soc15_allowed_read_registers[i].grbm_indexed, 429 se_num, sh_num, reg_offset); 430 return 0; 431 } 432 return -EINVAL; 433 } 434 435 436 /** 437 * soc15_program_register_sequence - program an array of registers. 438 * 439 * @adev: amdgpu_device pointer 440 * @regs: pointer to the register array 441 * @array_size: size of the register array 442 * 443 * Programs an array or registers with and and or masks. 444 * This is a helper for setting golden registers. 445 */ 446 447 void soc15_program_register_sequence(struct amdgpu_device *adev, 448 const struct soc15_reg_golden *regs, 449 const u32 array_size) 450 { 451 const struct soc15_reg_golden *entry; 452 u32 tmp, reg; 453 int i; 454 455 for (i = 0; i < array_size; ++i) { 456 entry = ®s[i]; 457 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg; 458 459 if (entry->and_mask == 0xffffffff) { 460 tmp = entry->or_mask; 461 } else { 462 tmp = (entry->hwip == GC_HWIP) ? 463 RREG32_SOC15_IP(GC, reg) : RREG32(reg); 464 465 tmp &= ~(entry->and_mask); 466 tmp |= (entry->or_mask & entry->and_mask); 467 } 468 469 if (reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3) || 470 reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_ENHANCE) || 471 reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_ENHANCE_1) || 472 reg == SOC15_REG_OFFSET(GC, 0, mmSH_MEM_CONFIG)) 473 WREG32_RLC(reg, tmp); 474 else 475 (entry->hwip == GC_HWIP) ? 476 WREG32_SOC15_IP(GC, reg, tmp) : WREG32(reg, tmp); 477 478 } 479 480 } 481 482 static int soc15_asic_baco_reset(struct amdgpu_device *adev) 483 { 484 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 485 int ret = 0; 486 487 /* avoid NBIF got stuck when do RAS recovery in BACO reset */ 488 if (ras && adev->ras_enabled) 489 adev->nbio.funcs->enable_doorbell_interrupt(adev, false); 490 491 ret = amdgpu_dpm_baco_reset(adev); 492 if (ret) 493 return ret; 494 495 /* re-enable doorbell interrupt after BACO exit */ 496 if (ras && adev->ras_enabled) 497 adev->nbio.funcs->enable_doorbell_interrupt(adev, true); 498 499 return 0; 500 } 501 502 static enum amd_reset_method 503 soc15_asic_reset_method(struct amdgpu_device *adev) 504 { 505 bool baco_reset = false; 506 bool connected_to_cpu = false; 507 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 508 509 if (adev->gmc.xgmi.supported && adev->gmc.xgmi.connected_to_cpu) 510 connected_to_cpu = true; 511 512 if (amdgpu_reset_method == AMD_RESET_METHOD_MODE1 || 513 amdgpu_reset_method == AMD_RESET_METHOD_MODE2 || 514 amdgpu_reset_method == AMD_RESET_METHOD_BACO || 515 amdgpu_reset_method == AMD_RESET_METHOD_PCI) { 516 /* If connected to cpu, driver only support mode2 */ 517 if (connected_to_cpu) 518 return AMD_RESET_METHOD_MODE2; 519 return amdgpu_reset_method; 520 } 521 522 if (amdgpu_reset_method != -1) 523 dev_warn(adev->dev, "Specified reset method:%d isn't supported, using AUTO instead.\n", 524 amdgpu_reset_method); 525 526 switch (adev->ip_versions[MP1_HWIP][0]) { 527 case IP_VERSION(10, 0, 0): 528 case IP_VERSION(10, 0, 1): 529 case IP_VERSION(12, 0, 0): 530 case IP_VERSION(12, 0, 1): 531 return AMD_RESET_METHOD_MODE2; 532 case IP_VERSION(9, 0, 0): 533 case IP_VERSION(11, 0, 2): 534 if (adev->asic_type == CHIP_VEGA20) { 535 if (adev->psp.sos.fw_version >= 0x80067) 536 baco_reset = amdgpu_dpm_is_baco_supported(adev); 537 /* 538 * 1. PMFW version > 0x284300: all cases use baco 539 * 2. PMFW version <= 0x284300: only sGPU w/o RAS use baco 540 */ 541 if (ras && adev->ras_enabled && 542 adev->pm.fw_version <= 0x283400) 543 baco_reset = false; 544 } else { 545 baco_reset = amdgpu_dpm_is_baco_supported(adev); 546 } 547 break; 548 case IP_VERSION(13, 0, 2): 549 /* 550 * 1.connected to cpu: driver issue mode2 reset 551 * 2.discret gpu: driver issue mode1 reset 552 */ 553 if (connected_to_cpu) 554 return AMD_RESET_METHOD_MODE2; 555 break; 556 case IP_VERSION(13, 0, 6): 557 /* Use gpu_recovery param to target a reset method. 558 * Enable triggering of GPU reset only if specified 559 * by module parameter. 560 */ 561 if (amdgpu_gpu_recovery == 4 || amdgpu_gpu_recovery == 5) 562 return AMD_RESET_METHOD_MODE2; 563 else if (!(adev->flags & AMD_IS_APU)) 564 return AMD_RESET_METHOD_MODE1; 565 else 566 return AMD_RESET_METHOD_MODE2; 567 default: 568 break; 569 } 570 571 if (baco_reset) 572 return AMD_RESET_METHOD_BACO; 573 else 574 return AMD_RESET_METHOD_MODE1; 575 } 576 577 static bool soc15_need_reset_on_resume(struct amdgpu_device *adev) 578 { 579 u32 sol_reg; 580 581 sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81); 582 583 /* Will reset for the following suspend abort cases. 584 * 1) Only reset limit on APU side, dGPU hasn't checked yet. 585 * 2) S3 suspend abort and TOS already launched. 586 */ 587 if (adev->flags & AMD_IS_APU && adev->in_s3 && 588 !adev->suspend_complete && 589 sol_reg) 590 return true; 591 592 return false; 593 } 594 595 static int soc15_asic_reset(struct amdgpu_device *adev) 596 { 597 /* original raven doesn't have full asic reset */ 598 /* On the latest Raven, the GPU reset can be performed 599 * successfully. So now, temporarily enable it for the 600 * S3 suspend abort case. 601 */ 602 if (((adev->apu_flags & AMD_APU_IS_RAVEN) || 603 (adev->apu_flags & AMD_APU_IS_RAVEN2)) && 604 !soc15_need_reset_on_resume(adev)) 605 return 0; 606 607 switch (soc15_asic_reset_method(adev)) { 608 case AMD_RESET_METHOD_PCI: 609 dev_info(adev->dev, "PCI reset\n"); 610 return amdgpu_device_pci_reset(adev); 611 case AMD_RESET_METHOD_BACO: 612 dev_info(adev->dev, "BACO reset\n"); 613 return soc15_asic_baco_reset(adev); 614 case AMD_RESET_METHOD_MODE2: 615 dev_info(adev->dev, "MODE2 reset\n"); 616 return amdgpu_dpm_mode2_reset(adev); 617 default: 618 dev_info(adev->dev, "MODE1 reset\n"); 619 return amdgpu_device_mode1_reset(adev); 620 } 621 } 622 623 static bool soc15_supports_baco(struct amdgpu_device *adev) 624 { 625 switch (adev->ip_versions[MP1_HWIP][0]) { 626 case IP_VERSION(9, 0, 0): 627 case IP_VERSION(11, 0, 2): 628 if (adev->asic_type == CHIP_VEGA20) { 629 if (adev->psp.sos.fw_version >= 0x80067) 630 return amdgpu_dpm_is_baco_supported(adev); 631 return false; 632 } else { 633 return amdgpu_dpm_is_baco_supported(adev); 634 } 635 break; 636 default: 637 return false; 638 } 639 } 640 641 /*static int soc15_set_uvd_clock(struct amdgpu_device *adev, u32 clock, 642 u32 cntl_reg, u32 status_reg) 643 { 644 return 0; 645 }*/ 646 647 static int soc15_set_uvd_clocks(struct amdgpu_device *adev, u32 vclk, u32 dclk) 648 { 649 /*int r; 650 651 r = soc15_set_uvd_clock(adev, vclk, ixCG_VCLK_CNTL, ixCG_VCLK_STATUS); 652 if (r) 653 return r; 654 655 r = soc15_set_uvd_clock(adev, dclk, ixCG_DCLK_CNTL, ixCG_DCLK_STATUS); 656 */ 657 return 0; 658 } 659 660 static int soc15_set_vce_clocks(struct amdgpu_device *adev, u32 evclk, u32 ecclk) 661 { 662 /* todo */ 663 664 return 0; 665 } 666 667 static void soc15_program_aspm(struct amdgpu_device *adev) 668 { 669 if (!amdgpu_device_should_use_aspm(adev)) 670 return; 671 672 if (!(adev->flags & AMD_IS_APU) && 673 (adev->nbio.funcs->program_aspm)) 674 adev->nbio.funcs->program_aspm(adev); 675 } 676 677 const struct amdgpu_ip_block_version vega10_common_ip_block = 678 { 679 .type = AMD_IP_BLOCK_TYPE_COMMON, 680 .major = 2, 681 .minor = 0, 682 .rev = 0, 683 .funcs = &soc15_common_ip_funcs, 684 }; 685 686 static void soc15_reg_base_init(struct amdgpu_device *adev) 687 { 688 /* Set IP register base before any HW register access */ 689 switch (adev->asic_type) { 690 case CHIP_VEGA10: 691 case CHIP_VEGA12: 692 case CHIP_RAVEN: 693 case CHIP_RENOIR: 694 vega10_reg_base_init(adev); 695 break; 696 case CHIP_VEGA20: 697 vega20_reg_base_init(adev); 698 break; 699 case CHIP_ARCTURUS: 700 arct_reg_base_init(adev); 701 break; 702 case CHIP_ALDEBARAN: 703 aldebaran_reg_base_init(adev); 704 break; 705 default: 706 DRM_ERROR("Unsupported asic type: %d!\n", adev->asic_type); 707 break; 708 } 709 } 710 711 void soc15_set_virt_ops(struct amdgpu_device *adev) 712 { 713 adev->virt.ops = &xgpu_ai_virt_ops; 714 715 /* init soc15 reg base early enough so we can 716 * request request full access for sriov before 717 * set_ip_blocks. */ 718 soc15_reg_base_init(adev); 719 } 720 721 static bool soc15_need_full_reset(struct amdgpu_device *adev) 722 { 723 /* change this when we implement soft reset */ 724 return true; 725 } 726 727 static void soc15_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, 728 uint64_t *count1) 729 { 730 uint32_t perfctr = 0; 731 uint64_t cnt0_of, cnt1_of; 732 int tmp; 733 734 /* This reports 0 on APUs, so return to avoid writing/reading registers 735 * that may or may not be different from their GPU counterparts 736 */ 737 if (adev->flags & AMD_IS_APU) 738 return; 739 740 /* Set the 2 events that we wish to watch, defined above */ 741 /* Reg 40 is # received msgs */ 742 /* Reg 104 is # of posted requests sent */ 743 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40); 744 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104); 745 746 /* Write to enable desired perf counters */ 747 WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK, perfctr); 748 /* Zero out and enable the perf counters 749 * Write 0x5: 750 * Bit 0 = Start all counters(1) 751 * Bit 2 = Global counter reset enable(1) 752 */ 753 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005); 754 755 msleep(1000); 756 757 /* Load the shadow and disable the perf counters 758 * Write 0x2: 759 * Bit 0 = Stop counters(0) 760 * Bit 1 = Load the shadow counters(1) 761 */ 762 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002); 763 764 /* Read register values to get any >32bit overflow */ 765 tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK); 766 cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER); 767 cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER); 768 769 /* Get the values and add the overflow */ 770 *count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32); 771 *count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32); 772 } 773 774 static void vega20_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, 775 uint64_t *count1) 776 { 777 uint32_t perfctr = 0; 778 uint64_t cnt0_of, cnt1_of; 779 int tmp; 780 781 /* This reports 0 on APUs, so return to avoid writing/reading registers 782 * that may or may not be different from their GPU counterparts 783 */ 784 if (adev->flags & AMD_IS_APU) 785 return; 786 787 /* Set the 2 events that we wish to watch, defined above */ 788 /* Reg 40 is # received msgs */ 789 /* Reg 108 is # of posted requests sent on VG20 */ 790 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK3, 791 EVENT0_SEL, 40); 792 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK3, 793 EVENT1_SEL, 108); 794 795 /* Write to enable desired perf counters */ 796 WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK3, perfctr); 797 /* Zero out and enable the perf counters 798 * Write 0x5: 799 * Bit 0 = Start all counters(1) 800 * Bit 2 = Global counter reset enable(1) 801 */ 802 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005); 803 804 msleep(1000); 805 806 /* Load the shadow and disable the perf counters 807 * Write 0x2: 808 * Bit 0 = Stop counters(0) 809 * Bit 1 = Load the shadow counters(1) 810 */ 811 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002); 812 813 /* Read register values to get any >32bit overflow */ 814 tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK3); 815 cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK3, COUNTER0_UPPER); 816 cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK3, COUNTER1_UPPER); 817 818 /* Get the values and add the overflow */ 819 *count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK3) | (cnt0_of << 32); 820 *count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK3) | (cnt1_of << 32); 821 } 822 823 static bool soc15_need_reset_on_init(struct amdgpu_device *adev) 824 { 825 u32 sol_reg; 826 827 /* CP hangs in IGT reloading test on RN, reset to WA */ 828 if (adev->asic_type == CHIP_RENOIR) 829 return true; 830 831 /* Just return false for soc15 GPUs. Reset does not seem to 832 * be necessary. 833 */ 834 if (!amdgpu_passthrough(adev)) 835 return false; 836 837 if (adev->flags & AMD_IS_APU) 838 return false; 839 840 /* Check sOS sign of life register to confirm sys driver and sOS 841 * are already been loaded. 842 */ 843 sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81); 844 if (sol_reg) 845 return true; 846 847 return false; 848 } 849 850 static uint64_t soc15_get_pcie_replay_count(struct amdgpu_device *adev) 851 { 852 uint64_t nak_r, nak_g; 853 854 /* Get the number of NAKs received and generated */ 855 nak_r = RREG32_PCIE(smnPCIE_RX_NUM_NAK); 856 nak_g = RREG32_PCIE(smnPCIE_RX_NUM_NAK_GENERATED); 857 858 /* Add the total number of NAKs, i.e the number of replays */ 859 return (nak_r + nak_g); 860 } 861 862 static void soc15_pre_asic_init(struct amdgpu_device *adev) 863 { 864 gmc_v9_0_restore_registers(adev); 865 } 866 867 static const struct amdgpu_asic_funcs soc15_asic_funcs = 868 { 869 .read_disabled_bios = &soc15_read_disabled_bios, 870 .read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom, 871 .read_register = &soc15_read_register, 872 .reset = &soc15_asic_reset, 873 .reset_method = &soc15_asic_reset_method, 874 .get_xclk = &soc15_get_xclk, 875 .set_uvd_clocks = &soc15_set_uvd_clocks, 876 .set_vce_clocks = &soc15_set_vce_clocks, 877 .get_config_memsize = &soc15_get_config_memsize, 878 .need_full_reset = &soc15_need_full_reset, 879 .init_doorbell_index = &vega10_doorbell_index_init, 880 .get_pcie_usage = &soc15_get_pcie_usage, 881 .need_reset_on_init = &soc15_need_reset_on_init, 882 .get_pcie_replay_count = &soc15_get_pcie_replay_count, 883 .supports_baco = &soc15_supports_baco, 884 .pre_asic_init = &soc15_pre_asic_init, 885 .query_video_codecs = &soc15_query_video_codecs, 886 }; 887 888 static const struct amdgpu_asic_funcs vega20_asic_funcs = 889 { 890 .read_disabled_bios = &soc15_read_disabled_bios, 891 .read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom, 892 .read_register = &soc15_read_register, 893 .reset = &soc15_asic_reset, 894 .reset_method = &soc15_asic_reset_method, 895 .get_xclk = &soc15_get_xclk, 896 .set_uvd_clocks = &soc15_set_uvd_clocks, 897 .set_vce_clocks = &soc15_set_vce_clocks, 898 .get_config_memsize = &soc15_get_config_memsize, 899 .need_full_reset = &soc15_need_full_reset, 900 .init_doorbell_index = &vega20_doorbell_index_init, 901 .get_pcie_usage = &vega20_get_pcie_usage, 902 .need_reset_on_init = &soc15_need_reset_on_init, 903 .get_pcie_replay_count = &soc15_get_pcie_replay_count, 904 .supports_baco = &soc15_supports_baco, 905 .pre_asic_init = &soc15_pre_asic_init, 906 .query_video_codecs = &soc15_query_video_codecs, 907 }; 908 909 static const struct amdgpu_asic_funcs aqua_vanjaram_asic_funcs = 910 { 911 .read_disabled_bios = &soc15_read_disabled_bios, 912 .read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom, 913 .read_register = &soc15_read_register, 914 .reset = &soc15_asic_reset, 915 .reset_method = &soc15_asic_reset_method, 916 .get_xclk = &soc15_get_xclk, 917 .set_uvd_clocks = &soc15_set_uvd_clocks, 918 .set_vce_clocks = &soc15_set_vce_clocks, 919 .get_config_memsize = &soc15_get_config_memsize, 920 .need_full_reset = &soc15_need_full_reset, 921 .init_doorbell_index = &aqua_vanjaram_doorbell_index_init, 922 .get_pcie_usage = &amdgpu_nbio_get_pcie_usage, 923 .need_reset_on_init = &soc15_need_reset_on_init, 924 .get_pcie_replay_count = &amdgpu_nbio_get_pcie_replay_count, 925 .supports_baco = &soc15_supports_baco, 926 .pre_asic_init = &soc15_pre_asic_init, 927 .query_video_codecs = &soc15_query_video_codecs, 928 .encode_ext_smn_addressing = &aqua_vanjaram_encode_ext_smn_addressing, 929 }; 930 931 static int soc15_common_early_init(void *handle) 932 { 933 #define MMIO_REG_HOLE_OFFSET (0x80000 - PAGE_SIZE) 934 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 935 936 if (!amdgpu_sriov_vf(adev)) { 937 adev->rmmio_remap.reg_offset = MMIO_REG_HOLE_OFFSET; 938 adev->rmmio_remap.bus_addr = adev->rmmio_base + MMIO_REG_HOLE_OFFSET; 939 } 940 adev->smc_rreg = NULL; 941 adev->smc_wreg = NULL; 942 adev->pcie_rreg = &amdgpu_device_indirect_rreg; 943 adev->pcie_wreg = &amdgpu_device_indirect_wreg; 944 adev->pcie_rreg_ext = &amdgpu_device_indirect_rreg_ext; 945 adev->pcie_wreg_ext = &amdgpu_device_indirect_wreg_ext; 946 adev->pcie_rreg64 = &amdgpu_device_indirect_rreg64; 947 adev->pcie_wreg64 = &amdgpu_device_indirect_wreg64; 948 adev->uvd_ctx_rreg = &soc15_uvd_ctx_rreg; 949 adev->uvd_ctx_wreg = &soc15_uvd_ctx_wreg; 950 adev->didt_rreg = &soc15_didt_rreg; 951 adev->didt_wreg = &soc15_didt_wreg; 952 adev->gc_cac_rreg = &soc15_gc_cac_rreg; 953 adev->gc_cac_wreg = &soc15_gc_cac_wreg; 954 adev->se_cac_rreg = &soc15_se_cac_rreg; 955 adev->se_cac_wreg = &soc15_se_cac_wreg; 956 957 adev->rev_id = amdgpu_device_get_rev_id(adev); 958 adev->external_rev_id = 0xFF; 959 /* TODO: split the GC and PG flags based on the relevant IP version for which 960 * they are relevant. 961 */ 962 switch (adev->ip_versions[GC_HWIP][0]) { 963 case IP_VERSION(9, 0, 1): 964 adev->asic_funcs = &soc15_asic_funcs; 965 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 966 AMD_CG_SUPPORT_GFX_MGLS | 967 AMD_CG_SUPPORT_GFX_RLC_LS | 968 AMD_CG_SUPPORT_GFX_CP_LS | 969 AMD_CG_SUPPORT_GFX_3D_CGCG | 970 AMD_CG_SUPPORT_GFX_3D_CGLS | 971 AMD_CG_SUPPORT_GFX_CGCG | 972 AMD_CG_SUPPORT_GFX_CGLS | 973 AMD_CG_SUPPORT_BIF_MGCG | 974 AMD_CG_SUPPORT_BIF_LS | 975 AMD_CG_SUPPORT_HDP_LS | 976 AMD_CG_SUPPORT_DRM_MGCG | 977 AMD_CG_SUPPORT_DRM_LS | 978 AMD_CG_SUPPORT_ROM_MGCG | 979 AMD_CG_SUPPORT_DF_MGCG | 980 AMD_CG_SUPPORT_SDMA_MGCG | 981 AMD_CG_SUPPORT_SDMA_LS | 982 AMD_CG_SUPPORT_MC_MGCG | 983 AMD_CG_SUPPORT_MC_LS; 984 adev->pg_flags = 0; 985 adev->external_rev_id = 0x1; 986 break; 987 case IP_VERSION(9, 2, 1): 988 adev->asic_funcs = &soc15_asic_funcs; 989 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 990 AMD_CG_SUPPORT_GFX_MGLS | 991 AMD_CG_SUPPORT_GFX_CGCG | 992 AMD_CG_SUPPORT_GFX_CGLS | 993 AMD_CG_SUPPORT_GFX_3D_CGCG | 994 AMD_CG_SUPPORT_GFX_3D_CGLS | 995 AMD_CG_SUPPORT_GFX_CP_LS | 996 AMD_CG_SUPPORT_MC_LS | 997 AMD_CG_SUPPORT_MC_MGCG | 998 AMD_CG_SUPPORT_SDMA_MGCG | 999 AMD_CG_SUPPORT_SDMA_LS | 1000 AMD_CG_SUPPORT_BIF_MGCG | 1001 AMD_CG_SUPPORT_BIF_LS | 1002 AMD_CG_SUPPORT_HDP_MGCG | 1003 AMD_CG_SUPPORT_HDP_LS | 1004 AMD_CG_SUPPORT_ROM_MGCG | 1005 AMD_CG_SUPPORT_VCE_MGCG | 1006 AMD_CG_SUPPORT_UVD_MGCG; 1007 adev->pg_flags = 0; 1008 adev->external_rev_id = adev->rev_id + 0x14; 1009 break; 1010 case IP_VERSION(9, 4, 0): 1011 adev->asic_funcs = &vega20_asic_funcs; 1012 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1013 AMD_CG_SUPPORT_GFX_MGLS | 1014 AMD_CG_SUPPORT_GFX_CGCG | 1015 AMD_CG_SUPPORT_GFX_CGLS | 1016 AMD_CG_SUPPORT_GFX_3D_CGCG | 1017 AMD_CG_SUPPORT_GFX_3D_CGLS | 1018 AMD_CG_SUPPORT_GFX_CP_LS | 1019 AMD_CG_SUPPORT_MC_LS | 1020 AMD_CG_SUPPORT_MC_MGCG | 1021 AMD_CG_SUPPORT_SDMA_MGCG | 1022 AMD_CG_SUPPORT_SDMA_LS | 1023 AMD_CG_SUPPORT_BIF_MGCG | 1024 AMD_CG_SUPPORT_BIF_LS | 1025 AMD_CG_SUPPORT_HDP_MGCG | 1026 AMD_CG_SUPPORT_HDP_LS | 1027 AMD_CG_SUPPORT_ROM_MGCG | 1028 AMD_CG_SUPPORT_VCE_MGCG | 1029 AMD_CG_SUPPORT_UVD_MGCG; 1030 adev->pg_flags = 0; 1031 adev->external_rev_id = adev->rev_id + 0x28; 1032 break; 1033 case IP_VERSION(9, 1, 0): 1034 case IP_VERSION(9, 2, 2): 1035 adev->asic_funcs = &soc15_asic_funcs; 1036 1037 if (adev->rev_id >= 0x8) 1038 adev->apu_flags |= AMD_APU_IS_RAVEN2; 1039 1040 if (adev->apu_flags & AMD_APU_IS_RAVEN2) 1041 adev->external_rev_id = adev->rev_id + 0x79; 1042 else if (adev->apu_flags & AMD_APU_IS_PICASSO) 1043 adev->external_rev_id = adev->rev_id + 0x41; 1044 else if (adev->rev_id == 1) 1045 adev->external_rev_id = adev->rev_id + 0x20; 1046 else 1047 adev->external_rev_id = adev->rev_id + 0x01; 1048 1049 if (adev->apu_flags & AMD_APU_IS_RAVEN2) { 1050 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1051 AMD_CG_SUPPORT_GFX_MGLS | 1052 AMD_CG_SUPPORT_GFX_CP_LS | 1053 AMD_CG_SUPPORT_GFX_3D_CGCG | 1054 AMD_CG_SUPPORT_GFX_3D_CGLS | 1055 AMD_CG_SUPPORT_GFX_CGCG | 1056 AMD_CG_SUPPORT_GFX_CGLS | 1057 AMD_CG_SUPPORT_BIF_LS | 1058 AMD_CG_SUPPORT_HDP_LS | 1059 AMD_CG_SUPPORT_MC_MGCG | 1060 AMD_CG_SUPPORT_MC_LS | 1061 AMD_CG_SUPPORT_SDMA_MGCG | 1062 AMD_CG_SUPPORT_SDMA_LS | 1063 AMD_CG_SUPPORT_VCN_MGCG; 1064 1065 adev->pg_flags = AMD_PG_SUPPORT_SDMA | AMD_PG_SUPPORT_VCN; 1066 } else if (adev->apu_flags & AMD_APU_IS_PICASSO) { 1067 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1068 AMD_CG_SUPPORT_GFX_MGLS | 1069 AMD_CG_SUPPORT_GFX_CP_LS | 1070 AMD_CG_SUPPORT_GFX_3D_CGLS | 1071 AMD_CG_SUPPORT_GFX_CGCG | 1072 AMD_CG_SUPPORT_GFX_CGLS | 1073 AMD_CG_SUPPORT_BIF_LS | 1074 AMD_CG_SUPPORT_HDP_LS | 1075 AMD_CG_SUPPORT_MC_MGCG | 1076 AMD_CG_SUPPORT_MC_LS | 1077 AMD_CG_SUPPORT_SDMA_MGCG | 1078 AMD_CG_SUPPORT_SDMA_LS | 1079 AMD_CG_SUPPORT_VCN_MGCG; 1080 1081 /* 1082 * MMHUB PG needs to be disabled for Picasso for 1083 * stability reasons. 1084 */ 1085 adev->pg_flags = AMD_PG_SUPPORT_SDMA | 1086 AMD_PG_SUPPORT_VCN; 1087 } else { 1088 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1089 AMD_CG_SUPPORT_GFX_MGLS | 1090 AMD_CG_SUPPORT_GFX_RLC_LS | 1091 AMD_CG_SUPPORT_GFX_CP_LS | 1092 AMD_CG_SUPPORT_GFX_3D_CGLS | 1093 AMD_CG_SUPPORT_GFX_CGCG | 1094 AMD_CG_SUPPORT_GFX_CGLS | 1095 AMD_CG_SUPPORT_BIF_MGCG | 1096 AMD_CG_SUPPORT_BIF_LS | 1097 AMD_CG_SUPPORT_HDP_MGCG | 1098 AMD_CG_SUPPORT_HDP_LS | 1099 AMD_CG_SUPPORT_DRM_MGCG | 1100 AMD_CG_SUPPORT_DRM_LS | 1101 AMD_CG_SUPPORT_MC_MGCG | 1102 AMD_CG_SUPPORT_MC_LS | 1103 AMD_CG_SUPPORT_SDMA_MGCG | 1104 AMD_CG_SUPPORT_SDMA_LS | 1105 AMD_CG_SUPPORT_VCN_MGCG; 1106 1107 adev->pg_flags = AMD_PG_SUPPORT_SDMA | AMD_PG_SUPPORT_VCN; 1108 } 1109 break; 1110 case IP_VERSION(9, 4, 1): 1111 adev->asic_funcs = &vega20_asic_funcs; 1112 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1113 AMD_CG_SUPPORT_GFX_MGLS | 1114 AMD_CG_SUPPORT_GFX_CGCG | 1115 AMD_CG_SUPPORT_GFX_CGLS | 1116 AMD_CG_SUPPORT_GFX_CP_LS | 1117 AMD_CG_SUPPORT_HDP_MGCG | 1118 AMD_CG_SUPPORT_HDP_LS | 1119 AMD_CG_SUPPORT_SDMA_MGCG | 1120 AMD_CG_SUPPORT_SDMA_LS | 1121 AMD_CG_SUPPORT_MC_MGCG | 1122 AMD_CG_SUPPORT_MC_LS | 1123 AMD_CG_SUPPORT_IH_CG | 1124 AMD_CG_SUPPORT_VCN_MGCG | 1125 AMD_CG_SUPPORT_JPEG_MGCG; 1126 adev->pg_flags = AMD_PG_SUPPORT_VCN | AMD_PG_SUPPORT_VCN_DPG; 1127 adev->external_rev_id = adev->rev_id + 0x32; 1128 break; 1129 case IP_VERSION(9, 3, 0): 1130 adev->asic_funcs = &soc15_asic_funcs; 1131 1132 if (adev->apu_flags & AMD_APU_IS_RENOIR) 1133 adev->external_rev_id = adev->rev_id + 0x91; 1134 else 1135 adev->external_rev_id = adev->rev_id + 0xa1; 1136 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1137 AMD_CG_SUPPORT_GFX_MGLS | 1138 AMD_CG_SUPPORT_GFX_3D_CGCG | 1139 AMD_CG_SUPPORT_GFX_3D_CGLS | 1140 AMD_CG_SUPPORT_GFX_CGCG | 1141 AMD_CG_SUPPORT_GFX_CGLS | 1142 AMD_CG_SUPPORT_GFX_CP_LS | 1143 AMD_CG_SUPPORT_MC_MGCG | 1144 AMD_CG_SUPPORT_MC_LS | 1145 AMD_CG_SUPPORT_SDMA_MGCG | 1146 AMD_CG_SUPPORT_SDMA_LS | 1147 AMD_CG_SUPPORT_BIF_LS | 1148 AMD_CG_SUPPORT_HDP_LS | 1149 AMD_CG_SUPPORT_VCN_MGCG | 1150 AMD_CG_SUPPORT_JPEG_MGCG | 1151 AMD_CG_SUPPORT_IH_CG | 1152 AMD_CG_SUPPORT_ATHUB_LS | 1153 AMD_CG_SUPPORT_ATHUB_MGCG | 1154 AMD_CG_SUPPORT_DF_MGCG; 1155 adev->pg_flags = AMD_PG_SUPPORT_SDMA | 1156 AMD_PG_SUPPORT_VCN | 1157 AMD_PG_SUPPORT_JPEG | 1158 AMD_PG_SUPPORT_VCN_DPG; 1159 break; 1160 case IP_VERSION(9, 4, 2): 1161 adev->asic_funcs = &vega20_asic_funcs; 1162 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1163 AMD_CG_SUPPORT_GFX_MGLS | 1164 AMD_CG_SUPPORT_GFX_CP_LS | 1165 AMD_CG_SUPPORT_HDP_LS | 1166 AMD_CG_SUPPORT_SDMA_MGCG | 1167 AMD_CG_SUPPORT_SDMA_LS | 1168 AMD_CG_SUPPORT_IH_CG | 1169 AMD_CG_SUPPORT_VCN_MGCG | AMD_CG_SUPPORT_JPEG_MGCG; 1170 adev->pg_flags = AMD_PG_SUPPORT_VCN_DPG; 1171 adev->external_rev_id = adev->rev_id + 0x3c; 1172 break; 1173 case IP_VERSION(9, 4, 3): 1174 adev->asic_funcs = &aqua_vanjaram_asic_funcs; 1175 adev->cg_flags = 1176 AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_CGCG | 1177 AMD_CG_SUPPORT_GFX_CGLS | AMD_CG_SUPPORT_SDMA_MGCG | 1178 AMD_CG_SUPPORT_GFX_FGCG | AMD_CG_SUPPORT_REPEATER_FGCG | 1179 AMD_CG_SUPPORT_VCN_MGCG | AMD_CG_SUPPORT_JPEG_MGCG | 1180 AMD_CG_SUPPORT_IH_CG; 1181 adev->pg_flags = 1182 AMD_PG_SUPPORT_VCN | 1183 AMD_PG_SUPPORT_VCN_DPG | 1184 AMD_PG_SUPPORT_JPEG; 1185 adev->external_rev_id = adev->rev_id + 0x46; 1186 /* GC 9.4.3 uses MMIO register region hole at a different offset */ 1187 if (!amdgpu_sriov_vf(adev)) { 1188 adev->rmmio_remap.reg_offset = 0x1A000; 1189 adev->rmmio_remap.bus_addr = adev->rmmio_base + 0x1A000; 1190 } 1191 break; 1192 default: 1193 /* FIXME: not supported yet */ 1194 return -EINVAL; 1195 } 1196 1197 if (amdgpu_sriov_vf(adev)) { 1198 amdgpu_virt_init_setting(adev); 1199 xgpu_ai_mailbox_set_irq_funcs(adev); 1200 } 1201 1202 return 0; 1203 } 1204 1205 static int soc15_common_late_init(void *handle) 1206 { 1207 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1208 1209 if (amdgpu_sriov_vf(adev)) 1210 xgpu_ai_mailbox_get_irq(adev); 1211 1212 /* Enable selfring doorbell aperture late because doorbell BAR 1213 * aperture will change if resize BAR successfully in gmc sw_init. 1214 */ 1215 adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, true); 1216 1217 return 0; 1218 } 1219 1220 static int soc15_common_sw_init(void *handle) 1221 { 1222 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1223 1224 if (amdgpu_sriov_vf(adev)) 1225 xgpu_ai_mailbox_add_irq_id(adev); 1226 1227 if (adev->df.funcs && 1228 adev->df.funcs->sw_init) 1229 adev->df.funcs->sw_init(adev); 1230 1231 return 0; 1232 } 1233 1234 static int soc15_common_sw_fini(void *handle) 1235 { 1236 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1237 1238 if (adev->df.funcs && 1239 adev->df.funcs->sw_fini) 1240 adev->df.funcs->sw_fini(adev); 1241 return 0; 1242 } 1243 1244 static void soc15_sdma_doorbell_range_init(struct amdgpu_device *adev) 1245 { 1246 int i; 1247 1248 /* sdma doorbell range is programed by hypervisor */ 1249 if (!amdgpu_sriov_vf(adev)) { 1250 for (i = 0; i < adev->sdma.num_instances; i++) { 1251 adev->nbio.funcs->sdma_doorbell_range(adev, i, 1252 true, adev->doorbell_index.sdma_engine[i] << 1, 1253 adev->doorbell_index.sdma_doorbell_range); 1254 } 1255 } 1256 } 1257 1258 static int soc15_common_hw_init(void *handle) 1259 { 1260 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1261 1262 /* enable aspm */ 1263 soc15_program_aspm(adev); 1264 /* setup nbio registers */ 1265 adev->nbio.funcs->init_registers(adev); 1266 /* remap HDP registers to a hole in mmio space, 1267 * for the purpose of expose those registers 1268 * to process space 1269 */ 1270 if (adev->nbio.funcs->remap_hdp_registers && !amdgpu_sriov_vf(adev)) 1271 adev->nbio.funcs->remap_hdp_registers(adev); 1272 1273 /* enable the doorbell aperture */ 1274 adev->nbio.funcs->enable_doorbell_aperture(adev, true); 1275 1276 /* HW doorbell routing policy: doorbell writing not 1277 * in SDMA/IH/MM/ACV range will be routed to CP. So 1278 * we need to init SDMA doorbell range prior 1279 * to CP ip block init and ring test. IH already 1280 * happens before CP. 1281 */ 1282 soc15_sdma_doorbell_range_init(adev); 1283 1284 return 0; 1285 } 1286 1287 static int soc15_common_hw_fini(void *handle) 1288 { 1289 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1290 1291 /* Disable the doorbell aperture and selfring doorbell aperture 1292 * separately in hw_fini because soc15_enable_doorbell_aperture 1293 * has been removed and there is no need to delay disabling 1294 * selfring doorbell. 1295 */ 1296 adev->nbio.funcs->enable_doorbell_aperture(adev, false); 1297 adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, false); 1298 1299 if (amdgpu_sriov_vf(adev)) 1300 xgpu_ai_mailbox_put_irq(adev); 1301 1302 if (adev->nbio.ras_if && 1303 amdgpu_ras_is_supported(adev, adev->nbio.ras_if->block)) { 1304 if (adev->nbio.ras && 1305 adev->nbio.ras->init_ras_controller_interrupt) 1306 amdgpu_irq_put(adev, &adev->nbio.ras_controller_irq, 0); 1307 if (adev->nbio.ras && 1308 adev->nbio.ras->init_ras_err_event_athub_interrupt) 1309 amdgpu_irq_put(adev, &adev->nbio.ras_err_event_athub_irq, 0); 1310 } 1311 1312 return 0; 1313 } 1314 1315 static int soc15_common_suspend(void *handle) 1316 { 1317 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1318 1319 return soc15_common_hw_fini(adev); 1320 } 1321 1322 static int soc15_common_resume(void *handle) 1323 { 1324 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1325 1326 if (soc15_need_reset_on_resume(adev)) { 1327 dev_info(adev->dev, "S3 suspend abort case, let's reset ASIC.\n"); 1328 soc15_asic_reset(adev); 1329 } 1330 return soc15_common_hw_init(adev); 1331 } 1332 1333 static bool soc15_common_is_idle(void *handle) 1334 { 1335 return true; 1336 } 1337 1338 static int soc15_common_wait_for_idle(void *handle) 1339 { 1340 return 0; 1341 } 1342 1343 static int soc15_common_soft_reset(void *handle) 1344 { 1345 return 0; 1346 } 1347 1348 static void soc15_update_drm_clock_gating(struct amdgpu_device *adev, bool enable) 1349 { 1350 uint32_t def, data; 1351 1352 def = data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0)); 1353 1354 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DRM_MGCG)) 1355 data &= ~(0x01000000 | 1356 0x02000000 | 1357 0x04000000 | 1358 0x08000000 | 1359 0x10000000 | 1360 0x20000000 | 1361 0x40000000 | 1362 0x80000000); 1363 else 1364 data |= (0x01000000 | 1365 0x02000000 | 1366 0x04000000 | 1367 0x08000000 | 1368 0x10000000 | 1369 0x20000000 | 1370 0x40000000 | 1371 0x80000000); 1372 1373 if (def != data) 1374 WREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0), data); 1375 } 1376 1377 static void soc15_update_drm_light_sleep(struct amdgpu_device *adev, bool enable) 1378 { 1379 uint32_t def, data; 1380 1381 def = data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL)); 1382 1383 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DRM_LS)) 1384 data |= 1; 1385 else 1386 data &= ~1; 1387 1388 if (def != data) 1389 WREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL), data); 1390 } 1391 1392 static int soc15_common_set_clockgating_state(void *handle, 1393 enum amd_clockgating_state state) 1394 { 1395 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1396 1397 if (amdgpu_sriov_vf(adev)) 1398 return 0; 1399 1400 switch (adev->ip_versions[NBIO_HWIP][0]) { 1401 case IP_VERSION(6, 1, 0): 1402 case IP_VERSION(6, 2, 0): 1403 case IP_VERSION(7, 4, 0): 1404 adev->nbio.funcs->update_medium_grain_clock_gating(adev, 1405 state == AMD_CG_STATE_GATE); 1406 adev->nbio.funcs->update_medium_grain_light_sleep(adev, 1407 state == AMD_CG_STATE_GATE); 1408 adev->hdp.funcs->update_clock_gating(adev, 1409 state == AMD_CG_STATE_GATE); 1410 soc15_update_drm_clock_gating(adev, 1411 state == AMD_CG_STATE_GATE); 1412 soc15_update_drm_light_sleep(adev, 1413 state == AMD_CG_STATE_GATE); 1414 adev->smuio.funcs->update_rom_clock_gating(adev, 1415 state == AMD_CG_STATE_GATE); 1416 adev->df.funcs->update_medium_grain_clock_gating(adev, 1417 state == AMD_CG_STATE_GATE); 1418 break; 1419 case IP_VERSION(7, 0, 0): 1420 case IP_VERSION(7, 0, 1): 1421 case IP_VERSION(2, 5, 0): 1422 adev->nbio.funcs->update_medium_grain_clock_gating(adev, 1423 state == AMD_CG_STATE_GATE); 1424 adev->nbio.funcs->update_medium_grain_light_sleep(adev, 1425 state == AMD_CG_STATE_GATE); 1426 adev->hdp.funcs->update_clock_gating(adev, 1427 state == AMD_CG_STATE_GATE); 1428 soc15_update_drm_clock_gating(adev, 1429 state == AMD_CG_STATE_GATE); 1430 soc15_update_drm_light_sleep(adev, 1431 state == AMD_CG_STATE_GATE); 1432 break; 1433 case IP_VERSION(7, 4, 1): 1434 case IP_VERSION(7, 4, 4): 1435 adev->hdp.funcs->update_clock_gating(adev, 1436 state == AMD_CG_STATE_GATE); 1437 break; 1438 default: 1439 break; 1440 } 1441 return 0; 1442 } 1443 1444 static void soc15_common_get_clockgating_state(void *handle, u64 *flags) 1445 { 1446 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1447 int data; 1448 1449 if (amdgpu_sriov_vf(adev)) 1450 *flags = 0; 1451 1452 if (adev->nbio.funcs && adev->nbio.funcs->get_clockgating_state) 1453 adev->nbio.funcs->get_clockgating_state(adev, flags); 1454 1455 if (adev->hdp.funcs && adev->hdp.funcs->get_clock_gating_state) 1456 adev->hdp.funcs->get_clock_gating_state(adev, flags); 1457 1458 if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(13, 0, 2)) { 1459 1460 /* AMD_CG_SUPPORT_DRM_MGCG */ 1461 data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0)); 1462 if (!(data & 0x01000000)) 1463 *flags |= AMD_CG_SUPPORT_DRM_MGCG; 1464 1465 /* AMD_CG_SUPPORT_DRM_LS */ 1466 data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL)); 1467 if (data & 0x1) 1468 *flags |= AMD_CG_SUPPORT_DRM_LS; 1469 } 1470 1471 /* AMD_CG_SUPPORT_ROM_MGCG */ 1472 if (adev->smuio.funcs && adev->smuio.funcs->get_clock_gating_state) 1473 adev->smuio.funcs->get_clock_gating_state(adev, flags); 1474 1475 if (adev->df.funcs && adev->df.funcs->get_clockgating_state) 1476 adev->df.funcs->get_clockgating_state(adev, flags); 1477 } 1478 1479 static int soc15_common_set_powergating_state(void *handle, 1480 enum amd_powergating_state state) 1481 { 1482 /* todo */ 1483 return 0; 1484 } 1485 1486 static const struct amd_ip_funcs soc15_common_ip_funcs = { 1487 .name = "soc15_common", 1488 .early_init = soc15_common_early_init, 1489 .late_init = soc15_common_late_init, 1490 .sw_init = soc15_common_sw_init, 1491 .sw_fini = soc15_common_sw_fini, 1492 .hw_init = soc15_common_hw_init, 1493 .hw_fini = soc15_common_hw_fini, 1494 .suspend = soc15_common_suspend, 1495 .resume = soc15_common_resume, 1496 .is_idle = soc15_common_is_idle, 1497 .wait_for_idle = soc15_common_wait_for_idle, 1498 .soft_reset = soc15_common_soft_reset, 1499 .set_clockgating_state = soc15_common_set_clockgating_state, 1500 .set_powergating_state = soc15_common_set_powergating_state, 1501 .get_clockgating_state= soc15_common_get_clockgating_state, 1502 }; 1503