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, 4906, 3)}, 107 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4906, 5)}, 108 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4906, 52)}, 109 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4906, 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, 4906, 3)}, 124 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4906, 5)}, 125 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4906, 52)}, 126 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4906, 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, 4906, 3)}, 142 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4906, 5)}, 143 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4906, 52)}, 144 {codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4906, 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 int soc15_query_video_codecs(struct amdgpu_device *adev, bool encode, 157 const struct amdgpu_video_codecs **codecs) 158 { 159 if (adev->ip_versions[VCE_HWIP][0]) { 160 switch (adev->ip_versions[VCE_HWIP][0]) { 161 case IP_VERSION(4, 0, 0): 162 case IP_VERSION(4, 1, 0): 163 if (encode) 164 *codecs = &vega_video_codecs_encode; 165 else 166 *codecs = &vega_video_codecs_decode; 167 return 0; 168 default: 169 return -EINVAL; 170 } 171 } else { 172 switch (adev->ip_versions[UVD_HWIP][0]) { 173 case IP_VERSION(1, 0, 0): 174 case IP_VERSION(1, 0, 1): 175 if (encode) 176 *codecs = &vega_video_codecs_encode; 177 else 178 *codecs = &rv_video_codecs_decode; 179 return 0; 180 case IP_VERSION(2, 5, 0): 181 case IP_VERSION(2, 6, 0): 182 case IP_VERSION(2, 2, 0): 183 if (encode) 184 *codecs = &vega_video_codecs_encode; 185 else 186 *codecs = &rn_video_codecs_decode; 187 return 0; 188 default: 189 return -EINVAL; 190 } 191 } 192 } 193 194 /* 195 * Indirect registers accessor 196 */ 197 static u32 soc15_pcie_rreg(struct amdgpu_device *adev, u32 reg) 198 { 199 unsigned long address, data; 200 address = adev->nbio.funcs->get_pcie_index_offset(adev); 201 data = adev->nbio.funcs->get_pcie_data_offset(adev); 202 203 return amdgpu_device_indirect_rreg(adev, address, data, reg); 204 } 205 206 static void soc15_pcie_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 207 { 208 unsigned long address, data; 209 210 address = adev->nbio.funcs->get_pcie_index_offset(adev); 211 data = adev->nbio.funcs->get_pcie_data_offset(adev); 212 213 amdgpu_device_indirect_wreg(adev, address, data, reg, v); 214 } 215 216 static u64 soc15_pcie_rreg64(struct amdgpu_device *adev, u32 reg) 217 { 218 unsigned long address, data; 219 address = adev->nbio.funcs->get_pcie_index_offset(adev); 220 data = adev->nbio.funcs->get_pcie_data_offset(adev); 221 222 return amdgpu_device_indirect_rreg64(adev, address, data, reg); 223 } 224 225 static void soc15_pcie_wreg64(struct amdgpu_device *adev, u32 reg, u64 v) 226 { 227 unsigned long address, data; 228 229 address = adev->nbio.funcs->get_pcie_index_offset(adev); 230 data = adev->nbio.funcs->get_pcie_data_offset(adev); 231 232 amdgpu_device_indirect_wreg64(adev, address, data, reg, v); 233 } 234 235 static u32 soc15_uvd_ctx_rreg(struct amdgpu_device *adev, u32 reg) 236 { 237 unsigned long flags, address, data; 238 u32 r; 239 240 address = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_INDEX); 241 data = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_DATA); 242 243 spin_lock_irqsave(&adev->uvd_ctx_idx_lock, flags); 244 WREG32(address, ((reg) & 0x1ff)); 245 r = RREG32(data); 246 spin_unlock_irqrestore(&adev->uvd_ctx_idx_lock, flags); 247 return r; 248 } 249 250 static void soc15_uvd_ctx_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 251 { 252 unsigned long flags, address, data; 253 254 address = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_INDEX); 255 data = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_DATA); 256 257 spin_lock_irqsave(&adev->uvd_ctx_idx_lock, flags); 258 WREG32(address, ((reg) & 0x1ff)); 259 WREG32(data, (v)); 260 spin_unlock_irqrestore(&adev->uvd_ctx_idx_lock, flags); 261 } 262 263 static u32 soc15_didt_rreg(struct amdgpu_device *adev, u32 reg) 264 { 265 unsigned long flags, address, data; 266 u32 r; 267 268 address = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_INDEX); 269 data = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_DATA); 270 271 spin_lock_irqsave(&adev->didt_idx_lock, flags); 272 WREG32(address, (reg)); 273 r = RREG32(data); 274 spin_unlock_irqrestore(&adev->didt_idx_lock, flags); 275 return r; 276 } 277 278 static void soc15_didt_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 279 { 280 unsigned long flags, address, data; 281 282 address = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_INDEX); 283 data = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_DATA); 284 285 spin_lock_irqsave(&adev->didt_idx_lock, flags); 286 WREG32(address, (reg)); 287 WREG32(data, (v)); 288 spin_unlock_irqrestore(&adev->didt_idx_lock, flags); 289 } 290 291 static u32 soc15_gc_cac_rreg(struct amdgpu_device *adev, u32 reg) 292 { 293 unsigned long flags; 294 u32 r; 295 296 spin_lock_irqsave(&adev->gc_cac_idx_lock, flags); 297 WREG32_SOC15(GC, 0, mmGC_CAC_IND_INDEX, (reg)); 298 r = RREG32_SOC15(GC, 0, mmGC_CAC_IND_DATA); 299 spin_unlock_irqrestore(&adev->gc_cac_idx_lock, flags); 300 return r; 301 } 302 303 static void soc15_gc_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 304 { 305 unsigned long flags; 306 307 spin_lock_irqsave(&adev->gc_cac_idx_lock, flags); 308 WREG32_SOC15(GC, 0, mmGC_CAC_IND_INDEX, (reg)); 309 WREG32_SOC15(GC, 0, mmGC_CAC_IND_DATA, (v)); 310 spin_unlock_irqrestore(&adev->gc_cac_idx_lock, flags); 311 } 312 313 static u32 soc15_se_cac_rreg(struct amdgpu_device *adev, u32 reg) 314 { 315 unsigned long flags; 316 u32 r; 317 318 spin_lock_irqsave(&adev->se_cac_idx_lock, flags); 319 WREG32_SOC15(GC, 0, mmSE_CAC_IND_INDEX, (reg)); 320 r = RREG32_SOC15(GC, 0, mmSE_CAC_IND_DATA); 321 spin_unlock_irqrestore(&adev->se_cac_idx_lock, flags); 322 return r; 323 } 324 325 static void soc15_se_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v) 326 { 327 unsigned long flags; 328 329 spin_lock_irqsave(&adev->se_cac_idx_lock, flags); 330 WREG32_SOC15(GC, 0, mmSE_CAC_IND_INDEX, (reg)); 331 WREG32_SOC15(GC, 0, mmSE_CAC_IND_DATA, (v)); 332 spin_unlock_irqrestore(&adev->se_cac_idx_lock, flags); 333 } 334 335 static u32 soc15_get_config_memsize(struct amdgpu_device *adev) 336 { 337 return adev->nbio.funcs->get_memsize(adev); 338 } 339 340 static u32 soc15_get_xclk(struct amdgpu_device *adev) 341 { 342 u32 reference_clock = adev->clock.spll.reference_freq; 343 344 if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(12, 0, 0) || 345 adev->ip_versions[MP1_HWIP][0] == IP_VERSION(12, 0, 1)) 346 return 10000; 347 if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(10, 0, 0) || 348 adev->ip_versions[MP1_HWIP][0] == IP_VERSION(10, 0, 1)) 349 return reference_clock / 4; 350 351 return reference_clock; 352 } 353 354 355 void soc15_grbm_select(struct amdgpu_device *adev, 356 u32 me, u32 pipe, u32 queue, u32 vmid) 357 { 358 u32 grbm_gfx_cntl = 0; 359 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, PIPEID, pipe); 360 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, MEID, me); 361 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, VMID, vmid); 362 grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, QUEUEID, queue); 363 364 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_CNTL, grbm_gfx_cntl); 365 } 366 367 static void soc15_vga_set_state(struct amdgpu_device *adev, bool state) 368 { 369 /* todo */ 370 } 371 372 static bool soc15_read_disabled_bios(struct amdgpu_device *adev) 373 { 374 /* todo */ 375 return false; 376 } 377 378 static struct soc15_allowed_register_entry soc15_allowed_read_registers[] = { 379 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS)}, 380 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS2)}, 381 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE0)}, 382 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE1)}, 383 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE2)}, 384 { SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE3)}, 385 { SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_STATUS_REG)}, 386 { SOC15_REG_ENTRY(SDMA1, 0, mmSDMA1_STATUS_REG)}, 387 { SOC15_REG_ENTRY(GC, 0, mmCP_STAT)}, 388 { SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT1)}, 389 { SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT2)}, 390 { SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT3)}, 391 { SOC15_REG_ENTRY(GC, 0, mmCP_CPF_BUSY_STAT)}, 392 { SOC15_REG_ENTRY(GC, 0, mmCP_CPF_STALLED_STAT1)}, 393 { SOC15_REG_ENTRY(GC, 0, mmCP_CPF_STATUS)}, 394 { SOC15_REG_ENTRY(GC, 0, mmCP_CPC_BUSY_STAT)}, 395 { SOC15_REG_ENTRY(GC, 0, mmCP_CPC_STALLED_STAT1)}, 396 { SOC15_REG_ENTRY(GC, 0, mmCP_CPC_STATUS)}, 397 { SOC15_REG_ENTRY(GC, 0, mmGB_ADDR_CONFIG)}, 398 { SOC15_REG_ENTRY(GC, 0, mmDB_DEBUG2)}, 399 }; 400 401 static uint32_t soc15_read_indexed_register(struct amdgpu_device *adev, u32 se_num, 402 u32 sh_num, u32 reg_offset) 403 { 404 uint32_t val; 405 406 mutex_lock(&adev->grbm_idx_mutex); 407 if (se_num != 0xffffffff || sh_num != 0xffffffff) 408 amdgpu_gfx_select_se_sh(adev, se_num, sh_num, 0xffffffff); 409 410 val = RREG32(reg_offset); 411 412 if (se_num != 0xffffffff || sh_num != 0xffffffff) 413 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 414 mutex_unlock(&adev->grbm_idx_mutex); 415 return val; 416 } 417 418 static uint32_t soc15_get_register_value(struct amdgpu_device *adev, 419 bool indexed, u32 se_num, 420 u32 sh_num, u32 reg_offset) 421 { 422 if (indexed) { 423 return soc15_read_indexed_register(adev, se_num, sh_num, reg_offset); 424 } else { 425 if (reg_offset == SOC15_REG_OFFSET(GC, 0, mmGB_ADDR_CONFIG)) 426 return adev->gfx.config.gb_addr_config; 427 else if (reg_offset == SOC15_REG_OFFSET(GC, 0, mmDB_DEBUG2)) 428 return adev->gfx.config.db_debug2; 429 return RREG32(reg_offset); 430 } 431 } 432 433 static int soc15_read_register(struct amdgpu_device *adev, u32 se_num, 434 u32 sh_num, u32 reg_offset, u32 *value) 435 { 436 uint32_t i; 437 struct soc15_allowed_register_entry *en; 438 439 *value = 0; 440 for (i = 0; i < ARRAY_SIZE(soc15_allowed_read_registers); i++) { 441 en = &soc15_allowed_read_registers[i]; 442 if (adev->reg_offset[en->hwip][en->inst] && 443 reg_offset != (adev->reg_offset[en->hwip][en->inst][en->seg] 444 + en->reg_offset)) 445 continue; 446 447 *value = soc15_get_register_value(adev, 448 soc15_allowed_read_registers[i].grbm_indexed, 449 se_num, sh_num, reg_offset); 450 return 0; 451 } 452 return -EINVAL; 453 } 454 455 456 /** 457 * soc15_program_register_sequence - program an array of registers. 458 * 459 * @adev: amdgpu_device pointer 460 * @regs: pointer to the register array 461 * @array_size: size of the register array 462 * 463 * Programs an array or registers with and and or masks. 464 * This is a helper for setting golden registers. 465 */ 466 467 void soc15_program_register_sequence(struct amdgpu_device *adev, 468 const struct soc15_reg_golden *regs, 469 const u32 array_size) 470 { 471 const struct soc15_reg_golden *entry; 472 u32 tmp, reg; 473 int i; 474 475 for (i = 0; i < array_size; ++i) { 476 entry = ®s[i]; 477 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg; 478 479 if (entry->and_mask == 0xffffffff) { 480 tmp = entry->or_mask; 481 } else { 482 tmp = (entry->hwip == GC_HWIP) ? 483 RREG32_SOC15_IP(GC, reg) : RREG32(reg); 484 485 tmp &= ~(entry->and_mask); 486 tmp |= (entry->or_mask & entry->and_mask); 487 } 488 489 if (reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3) || 490 reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_ENHANCE) || 491 reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_ENHANCE_1) || 492 reg == SOC15_REG_OFFSET(GC, 0, mmSH_MEM_CONFIG)) 493 WREG32_RLC(reg, tmp); 494 else 495 (entry->hwip == GC_HWIP) ? 496 WREG32_SOC15_IP(GC, reg, tmp) : WREG32(reg, tmp); 497 498 } 499 500 } 501 502 static int soc15_asic_baco_reset(struct amdgpu_device *adev) 503 { 504 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 505 int ret = 0; 506 507 /* avoid NBIF got stuck when do RAS recovery in BACO reset */ 508 if (ras && adev->ras_enabled) 509 adev->nbio.funcs->enable_doorbell_interrupt(adev, false); 510 511 ret = amdgpu_dpm_baco_reset(adev); 512 if (ret) 513 return ret; 514 515 /* re-enable doorbell interrupt after BACO exit */ 516 if (ras && adev->ras_enabled) 517 adev->nbio.funcs->enable_doorbell_interrupt(adev, true); 518 519 return 0; 520 } 521 522 static enum amd_reset_method 523 soc15_asic_reset_method(struct amdgpu_device *adev) 524 { 525 bool baco_reset = false; 526 bool connected_to_cpu = false; 527 struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); 528 529 if (adev->gmc.xgmi.supported && adev->gmc.xgmi.connected_to_cpu) 530 connected_to_cpu = true; 531 532 if (amdgpu_reset_method == AMD_RESET_METHOD_MODE1 || 533 amdgpu_reset_method == AMD_RESET_METHOD_MODE2 || 534 amdgpu_reset_method == AMD_RESET_METHOD_BACO || 535 amdgpu_reset_method == AMD_RESET_METHOD_PCI) { 536 /* If connected to cpu, driver only support mode2 */ 537 if (connected_to_cpu) 538 return AMD_RESET_METHOD_MODE2; 539 return amdgpu_reset_method; 540 } 541 542 if (amdgpu_reset_method != -1) 543 dev_warn(adev->dev, "Specified reset method:%d isn't supported, using AUTO instead.\n", 544 amdgpu_reset_method); 545 546 switch (adev->ip_versions[MP1_HWIP][0]) { 547 case IP_VERSION(10, 0, 0): 548 case IP_VERSION(10, 0, 1): 549 case IP_VERSION(12, 0, 0): 550 case IP_VERSION(12, 0, 1): 551 return AMD_RESET_METHOD_MODE2; 552 case IP_VERSION(9, 0, 0): 553 case IP_VERSION(11, 0, 2): 554 if (adev->asic_type == CHIP_VEGA20) { 555 if (adev->psp.sos.fw_version >= 0x80067) 556 baco_reset = amdgpu_dpm_is_baco_supported(adev); 557 /* 558 * 1. PMFW version > 0x284300: all cases use baco 559 * 2. PMFW version <= 0x284300: only sGPU w/o RAS use baco 560 */ 561 if (ras && adev->ras_enabled && 562 adev->pm.fw_version <= 0x283400) 563 baco_reset = false; 564 } else { 565 baco_reset = amdgpu_dpm_is_baco_supported(adev); 566 } 567 break; 568 case IP_VERSION(13, 0, 2): 569 /* 570 * 1.connected to cpu: driver issue mode2 reset 571 * 2.discret gpu: driver issue mode1 reset 572 */ 573 if (connected_to_cpu) 574 return AMD_RESET_METHOD_MODE2; 575 break; 576 default: 577 break; 578 } 579 580 if (baco_reset) 581 return AMD_RESET_METHOD_BACO; 582 else 583 return AMD_RESET_METHOD_MODE1; 584 } 585 586 static int soc15_asic_reset(struct amdgpu_device *adev) 587 { 588 /* original raven doesn't have full asic reset */ 589 if ((adev->apu_flags & AMD_APU_IS_RAVEN) || 590 (adev->apu_flags & AMD_APU_IS_RAVEN2)) 591 return 0; 592 593 switch (soc15_asic_reset_method(adev)) { 594 case AMD_RESET_METHOD_PCI: 595 dev_info(adev->dev, "PCI reset\n"); 596 return amdgpu_device_pci_reset(adev); 597 case AMD_RESET_METHOD_BACO: 598 dev_info(adev->dev, "BACO reset\n"); 599 return soc15_asic_baco_reset(adev); 600 case AMD_RESET_METHOD_MODE2: 601 dev_info(adev->dev, "MODE2 reset\n"); 602 return amdgpu_dpm_mode2_reset(adev); 603 default: 604 dev_info(adev->dev, "MODE1 reset\n"); 605 return amdgpu_device_mode1_reset(adev); 606 } 607 } 608 609 static bool soc15_supports_baco(struct amdgpu_device *adev) 610 { 611 switch (adev->ip_versions[MP1_HWIP][0]) { 612 case IP_VERSION(9, 0, 0): 613 case IP_VERSION(11, 0, 2): 614 if (adev->asic_type == CHIP_VEGA20) { 615 if (adev->psp.sos.fw_version >= 0x80067) 616 return amdgpu_dpm_is_baco_supported(adev); 617 return false; 618 } else { 619 return amdgpu_dpm_is_baco_supported(adev); 620 } 621 break; 622 default: 623 return false; 624 } 625 } 626 627 /*static int soc15_set_uvd_clock(struct amdgpu_device *adev, u32 clock, 628 u32 cntl_reg, u32 status_reg) 629 { 630 return 0; 631 }*/ 632 633 static int soc15_set_uvd_clocks(struct amdgpu_device *adev, u32 vclk, u32 dclk) 634 { 635 /*int r; 636 637 r = soc15_set_uvd_clock(adev, vclk, ixCG_VCLK_CNTL, ixCG_VCLK_STATUS); 638 if (r) 639 return r; 640 641 r = soc15_set_uvd_clock(adev, dclk, ixCG_DCLK_CNTL, ixCG_DCLK_STATUS); 642 */ 643 return 0; 644 } 645 646 static int soc15_set_vce_clocks(struct amdgpu_device *adev, u32 evclk, u32 ecclk) 647 { 648 /* todo */ 649 650 return 0; 651 } 652 653 static void soc15_pcie_gen3_enable(struct amdgpu_device *adev) 654 { 655 if (pci_is_root_bus(adev->pdev->bus)) 656 return; 657 658 if (amdgpu_pcie_gen2 == 0) 659 return; 660 661 if (adev->flags & AMD_IS_APU) 662 return; 663 664 if (!(adev->pm.pcie_gen_mask & (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | 665 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3))) 666 return; 667 668 /* todo */ 669 } 670 671 static void soc15_program_aspm(struct amdgpu_device *adev) 672 { 673 if (!amdgpu_device_should_use_aspm(adev)) 674 return; 675 676 if (!(adev->flags & AMD_IS_APU) && 677 (adev->nbio.funcs->program_aspm)) 678 adev->nbio.funcs->program_aspm(adev); 679 } 680 681 static void soc15_enable_doorbell_aperture(struct amdgpu_device *adev, 682 bool enable) 683 { 684 adev->nbio.funcs->enable_doorbell_aperture(adev, enable); 685 adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, enable); 686 } 687 688 const struct amdgpu_ip_block_version vega10_common_ip_block = 689 { 690 .type = AMD_IP_BLOCK_TYPE_COMMON, 691 .major = 2, 692 .minor = 0, 693 .rev = 0, 694 .funcs = &soc15_common_ip_funcs, 695 }; 696 697 static uint32_t soc15_get_rev_id(struct amdgpu_device *adev) 698 { 699 return adev->nbio.funcs->get_rev_id(adev); 700 } 701 702 static void soc15_reg_base_init(struct amdgpu_device *adev) 703 { 704 int r; 705 706 /* Set IP register base before any HW register access */ 707 switch (adev->asic_type) { 708 case CHIP_VEGA10: 709 case CHIP_VEGA12: 710 case CHIP_RAVEN: 711 vega10_reg_base_init(adev); 712 break; 713 case CHIP_RENOIR: 714 /* It's safe to do ip discovery here for Renoir, 715 * it doesn't support SRIOV. */ 716 if (amdgpu_discovery) { 717 r = amdgpu_discovery_reg_base_init(adev); 718 if (r == 0) 719 break; 720 DRM_WARN("failed to init reg base from ip discovery table, " 721 "fallback to legacy init method\n"); 722 } 723 vega10_reg_base_init(adev); 724 break; 725 case CHIP_VEGA20: 726 vega20_reg_base_init(adev); 727 break; 728 case CHIP_ARCTURUS: 729 arct_reg_base_init(adev); 730 break; 731 case CHIP_ALDEBARAN: 732 aldebaran_reg_base_init(adev); 733 break; 734 default: 735 DRM_ERROR("Unsupported asic type: %d!\n", adev->asic_type); 736 break; 737 } 738 } 739 740 void soc15_set_virt_ops(struct amdgpu_device *adev) 741 { 742 adev->virt.ops = &xgpu_ai_virt_ops; 743 744 /* init soc15 reg base early enough so we can 745 * request request full access for sriov before 746 * set_ip_blocks. */ 747 soc15_reg_base_init(adev); 748 } 749 750 static bool soc15_need_full_reset(struct amdgpu_device *adev) 751 { 752 /* change this when we implement soft reset */ 753 return true; 754 } 755 756 static void soc15_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, 757 uint64_t *count1) 758 { 759 uint32_t perfctr = 0; 760 uint64_t cnt0_of, cnt1_of; 761 int tmp; 762 763 /* This reports 0 on APUs, so return to avoid writing/reading registers 764 * that may or may not be different from their GPU counterparts 765 */ 766 if (adev->flags & AMD_IS_APU) 767 return; 768 769 /* Set the 2 events that we wish to watch, defined above */ 770 /* Reg 40 is # received msgs */ 771 /* Reg 104 is # of posted requests sent */ 772 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40); 773 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104); 774 775 /* Write to enable desired perf counters */ 776 WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK, perfctr); 777 /* Zero out and enable the perf counters 778 * Write 0x5: 779 * Bit 0 = Start all counters(1) 780 * Bit 2 = Global counter reset enable(1) 781 */ 782 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005); 783 784 msleep(1000); 785 786 /* Load the shadow and disable the perf counters 787 * Write 0x2: 788 * Bit 0 = Stop counters(0) 789 * Bit 1 = Load the shadow counters(1) 790 */ 791 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002); 792 793 /* Read register values to get any >32bit overflow */ 794 tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK); 795 cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER); 796 cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER); 797 798 /* Get the values and add the overflow */ 799 *count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32); 800 *count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32); 801 } 802 803 static void vega20_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0, 804 uint64_t *count1) 805 { 806 uint32_t perfctr = 0; 807 uint64_t cnt0_of, cnt1_of; 808 int tmp; 809 810 /* This reports 0 on APUs, so return to avoid writing/reading registers 811 * that may or may not be different from their GPU counterparts 812 */ 813 if (adev->flags & AMD_IS_APU) 814 return; 815 816 /* Set the 2 events that we wish to watch, defined above */ 817 /* Reg 40 is # received msgs */ 818 /* Reg 108 is # of posted requests sent on VG20 */ 819 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK3, 820 EVENT0_SEL, 40); 821 perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK3, 822 EVENT1_SEL, 108); 823 824 /* Write to enable desired perf counters */ 825 WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK3, perfctr); 826 /* Zero out and enable the perf counters 827 * Write 0x5: 828 * Bit 0 = Start all counters(1) 829 * Bit 2 = Global counter reset enable(1) 830 */ 831 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005); 832 833 msleep(1000); 834 835 /* Load the shadow and disable the perf counters 836 * Write 0x2: 837 * Bit 0 = Stop counters(0) 838 * Bit 1 = Load the shadow counters(1) 839 */ 840 WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002); 841 842 /* Read register values to get any >32bit overflow */ 843 tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK3); 844 cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK3, COUNTER0_UPPER); 845 cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK3, COUNTER1_UPPER); 846 847 /* Get the values and add the overflow */ 848 *count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK3) | (cnt0_of << 32); 849 *count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK3) | (cnt1_of << 32); 850 } 851 852 static bool soc15_need_reset_on_init(struct amdgpu_device *adev) 853 { 854 u32 sol_reg; 855 856 /* Just return false for soc15 GPUs. Reset does not seem to 857 * be necessary. 858 */ 859 if (!amdgpu_passthrough(adev)) 860 return false; 861 862 if (adev->flags & AMD_IS_APU) 863 return false; 864 865 /* Check sOS sign of life register to confirm sys driver and sOS 866 * are already been loaded. 867 */ 868 sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81); 869 if (sol_reg) 870 return true; 871 872 return false; 873 } 874 875 static uint64_t soc15_get_pcie_replay_count(struct amdgpu_device *adev) 876 { 877 uint64_t nak_r, nak_g; 878 879 /* Get the number of NAKs received and generated */ 880 nak_r = RREG32_PCIE(smnPCIE_RX_NUM_NAK); 881 nak_g = RREG32_PCIE(smnPCIE_RX_NUM_NAK_GENERATED); 882 883 /* Add the total number of NAKs, i.e the number of replays */ 884 return (nak_r + nak_g); 885 } 886 887 static void soc15_pre_asic_init(struct amdgpu_device *adev) 888 { 889 gmc_v9_0_restore_registers(adev); 890 } 891 892 static const struct amdgpu_asic_funcs soc15_asic_funcs = 893 { 894 .read_disabled_bios = &soc15_read_disabled_bios, 895 .read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom, 896 .read_register = &soc15_read_register, 897 .reset = &soc15_asic_reset, 898 .reset_method = &soc15_asic_reset_method, 899 .set_vga_state = &soc15_vga_set_state, 900 .get_xclk = &soc15_get_xclk, 901 .set_uvd_clocks = &soc15_set_uvd_clocks, 902 .set_vce_clocks = &soc15_set_vce_clocks, 903 .get_config_memsize = &soc15_get_config_memsize, 904 .need_full_reset = &soc15_need_full_reset, 905 .init_doorbell_index = &vega10_doorbell_index_init, 906 .get_pcie_usage = &soc15_get_pcie_usage, 907 .need_reset_on_init = &soc15_need_reset_on_init, 908 .get_pcie_replay_count = &soc15_get_pcie_replay_count, 909 .supports_baco = &soc15_supports_baco, 910 .pre_asic_init = &soc15_pre_asic_init, 911 .query_video_codecs = &soc15_query_video_codecs, 912 }; 913 914 static const struct amdgpu_asic_funcs vega20_asic_funcs = 915 { 916 .read_disabled_bios = &soc15_read_disabled_bios, 917 .read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom, 918 .read_register = &soc15_read_register, 919 .reset = &soc15_asic_reset, 920 .reset_method = &soc15_asic_reset_method, 921 .set_vga_state = &soc15_vga_set_state, 922 .get_xclk = &soc15_get_xclk, 923 .set_uvd_clocks = &soc15_set_uvd_clocks, 924 .set_vce_clocks = &soc15_set_vce_clocks, 925 .get_config_memsize = &soc15_get_config_memsize, 926 .need_full_reset = &soc15_need_full_reset, 927 .init_doorbell_index = &vega20_doorbell_index_init, 928 .get_pcie_usage = &vega20_get_pcie_usage, 929 .need_reset_on_init = &soc15_need_reset_on_init, 930 .get_pcie_replay_count = &soc15_get_pcie_replay_count, 931 .supports_baco = &soc15_supports_baco, 932 .pre_asic_init = &soc15_pre_asic_init, 933 .query_video_codecs = &soc15_query_video_codecs, 934 }; 935 936 static int soc15_common_early_init(void *handle) 937 { 938 #define MMIO_REG_HOLE_OFFSET (0x80000 - PAGE_SIZE) 939 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 940 941 if (!amdgpu_sriov_vf(adev)) { 942 adev->rmmio_remap.reg_offset = MMIO_REG_HOLE_OFFSET; 943 adev->rmmio_remap.bus_addr = adev->rmmio_base + MMIO_REG_HOLE_OFFSET; 944 } 945 adev->smc_rreg = NULL; 946 adev->smc_wreg = NULL; 947 adev->pcie_rreg = &soc15_pcie_rreg; 948 adev->pcie_wreg = &soc15_pcie_wreg; 949 adev->pcie_rreg64 = &soc15_pcie_rreg64; 950 adev->pcie_wreg64 = &soc15_pcie_wreg64; 951 adev->uvd_ctx_rreg = &soc15_uvd_ctx_rreg; 952 adev->uvd_ctx_wreg = &soc15_uvd_ctx_wreg; 953 adev->didt_rreg = &soc15_didt_rreg; 954 adev->didt_wreg = &soc15_didt_wreg; 955 adev->gc_cac_rreg = &soc15_gc_cac_rreg; 956 adev->gc_cac_wreg = &soc15_gc_cac_wreg; 957 adev->se_cac_rreg = &soc15_se_cac_rreg; 958 adev->se_cac_wreg = &soc15_se_cac_wreg; 959 960 adev->rev_id = soc15_get_rev_id(adev); 961 adev->external_rev_id = 0xFF; 962 /* TODO: split the GC and PG flags based on the relevant IP version for which 963 * they are relevant. 964 */ 965 switch (adev->ip_versions[GC_HWIP][0]) { 966 case IP_VERSION(9, 0, 1): 967 adev->asic_funcs = &soc15_asic_funcs; 968 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 969 AMD_CG_SUPPORT_GFX_MGLS | 970 AMD_CG_SUPPORT_GFX_RLC_LS | 971 AMD_CG_SUPPORT_GFX_CP_LS | 972 AMD_CG_SUPPORT_GFX_3D_CGCG | 973 AMD_CG_SUPPORT_GFX_3D_CGLS | 974 AMD_CG_SUPPORT_GFX_CGCG | 975 AMD_CG_SUPPORT_GFX_CGLS | 976 AMD_CG_SUPPORT_BIF_MGCG | 977 AMD_CG_SUPPORT_BIF_LS | 978 AMD_CG_SUPPORT_HDP_LS | 979 AMD_CG_SUPPORT_DRM_MGCG | 980 AMD_CG_SUPPORT_DRM_LS | 981 AMD_CG_SUPPORT_ROM_MGCG | 982 AMD_CG_SUPPORT_DF_MGCG | 983 AMD_CG_SUPPORT_SDMA_MGCG | 984 AMD_CG_SUPPORT_SDMA_LS | 985 AMD_CG_SUPPORT_MC_MGCG | 986 AMD_CG_SUPPORT_MC_LS; 987 adev->pg_flags = 0; 988 adev->external_rev_id = 0x1; 989 break; 990 case IP_VERSION(9, 2, 1): 991 adev->asic_funcs = &soc15_asic_funcs; 992 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 993 AMD_CG_SUPPORT_GFX_MGLS | 994 AMD_CG_SUPPORT_GFX_CGCG | 995 AMD_CG_SUPPORT_GFX_CGLS | 996 AMD_CG_SUPPORT_GFX_3D_CGCG | 997 AMD_CG_SUPPORT_GFX_3D_CGLS | 998 AMD_CG_SUPPORT_GFX_CP_LS | 999 AMD_CG_SUPPORT_MC_LS | 1000 AMD_CG_SUPPORT_MC_MGCG | 1001 AMD_CG_SUPPORT_SDMA_MGCG | 1002 AMD_CG_SUPPORT_SDMA_LS | 1003 AMD_CG_SUPPORT_BIF_MGCG | 1004 AMD_CG_SUPPORT_BIF_LS | 1005 AMD_CG_SUPPORT_HDP_MGCG | 1006 AMD_CG_SUPPORT_HDP_LS | 1007 AMD_CG_SUPPORT_ROM_MGCG | 1008 AMD_CG_SUPPORT_VCE_MGCG | 1009 AMD_CG_SUPPORT_UVD_MGCG; 1010 adev->pg_flags = 0; 1011 adev->external_rev_id = adev->rev_id + 0x14; 1012 break; 1013 case IP_VERSION(9, 4, 0): 1014 adev->asic_funcs = &vega20_asic_funcs; 1015 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1016 AMD_CG_SUPPORT_GFX_MGLS | 1017 AMD_CG_SUPPORT_GFX_CGCG | 1018 AMD_CG_SUPPORT_GFX_CGLS | 1019 AMD_CG_SUPPORT_GFX_3D_CGCG | 1020 AMD_CG_SUPPORT_GFX_3D_CGLS | 1021 AMD_CG_SUPPORT_GFX_CP_LS | 1022 AMD_CG_SUPPORT_MC_LS | 1023 AMD_CG_SUPPORT_MC_MGCG | 1024 AMD_CG_SUPPORT_SDMA_MGCG | 1025 AMD_CG_SUPPORT_SDMA_LS | 1026 AMD_CG_SUPPORT_BIF_MGCG | 1027 AMD_CG_SUPPORT_BIF_LS | 1028 AMD_CG_SUPPORT_HDP_MGCG | 1029 AMD_CG_SUPPORT_HDP_LS | 1030 AMD_CG_SUPPORT_ROM_MGCG | 1031 AMD_CG_SUPPORT_VCE_MGCG | 1032 AMD_CG_SUPPORT_UVD_MGCG; 1033 adev->pg_flags = 0; 1034 adev->external_rev_id = adev->rev_id + 0x28; 1035 break; 1036 case IP_VERSION(9, 1, 0): 1037 case IP_VERSION(9, 2, 2): 1038 adev->asic_funcs = &soc15_asic_funcs; 1039 1040 if (adev->rev_id >= 0x8) 1041 adev->apu_flags |= AMD_APU_IS_RAVEN2; 1042 1043 if (adev->apu_flags & AMD_APU_IS_RAVEN2) 1044 adev->external_rev_id = adev->rev_id + 0x79; 1045 else if (adev->apu_flags & AMD_APU_IS_PICASSO) 1046 adev->external_rev_id = adev->rev_id + 0x41; 1047 else if (adev->rev_id == 1) 1048 adev->external_rev_id = adev->rev_id + 0x20; 1049 else 1050 adev->external_rev_id = adev->rev_id + 0x01; 1051 1052 if (adev->apu_flags & AMD_APU_IS_RAVEN2) { 1053 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1054 AMD_CG_SUPPORT_GFX_MGLS | 1055 AMD_CG_SUPPORT_GFX_CP_LS | 1056 AMD_CG_SUPPORT_GFX_3D_CGCG | 1057 AMD_CG_SUPPORT_GFX_3D_CGLS | 1058 AMD_CG_SUPPORT_GFX_CGCG | 1059 AMD_CG_SUPPORT_GFX_CGLS | 1060 AMD_CG_SUPPORT_BIF_LS | 1061 AMD_CG_SUPPORT_HDP_LS | 1062 AMD_CG_SUPPORT_MC_MGCG | 1063 AMD_CG_SUPPORT_MC_LS | 1064 AMD_CG_SUPPORT_SDMA_MGCG | 1065 AMD_CG_SUPPORT_SDMA_LS | 1066 AMD_CG_SUPPORT_VCN_MGCG; 1067 1068 adev->pg_flags = AMD_PG_SUPPORT_SDMA | AMD_PG_SUPPORT_VCN; 1069 } else if (adev->apu_flags & AMD_APU_IS_PICASSO) { 1070 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1071 AMD_CG_SUPPORT_GFX_MGLS | 1072 AMD_CG_SUPPORT_GFX_CP_LS | 1073 AMD_CG_SUPPORT_GFX_3D_CGLS | 1074 AMD_CG_SUPPORT_GFX_CGCG | 1075 AMD_CG_SUPPORT_GFX_CGLS | 1076 AMD_CG_SUPPORT_BIF_LS | 1077 AMD_CG_SUPPORT_HDP_LS | 1078 AMD_CG_SUPPORT_MC_MGCG | 1079 AMD_CG_SUPPORT_MC_LS | 1080 AMD_CG_SUPPORT_SDMA_MGCG | 1081 AMD_CG_SUPPORT_SDMA_LS | 1082 AMD_CG_SUPPORT_VCN_MGCG; 1083 1084 /* 1085 * MMHUB PG needs to be disabled for Picasso for 1086 * stability reasons. 1087 */ 1088 adev->pg_flags = AMD_PG_SUPPORT_SDMA | 1089 AMD_PG_SUPPORT_VCN; 1090 } else { 1091 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1092 AMD_CG_SUPPORT_GFX_MGLS | 1093 AMD_CG_SUPPORT_GFX_RLC_LS | 1094 AMD_CG_SUPPORT_GFX_CP_LS | 1095 AMD_CG_SUPPORT_GFX_3D_CGLS | 1096 AMD_CG_SUPPORT_GFX_CGCG | 1097 AMD_CG_SUPPORT_GFX_CGLS | 1098 AMD_CG_SUPPORT_BIF_MGCG | 1099 AMD_CG_SUPPORT_BIF_LS | 1100 AMD_CG_SUPPORT_HDP_MGCG | 1101 AMD_CG_SUPPORT_HDP_LS | 1102 AMD_CG_SUPPORT_DRM_MGCG | 1103 AMD_CG_SUPPORT_DRM_LS | 1104 AMD_CG_SUPPORT_MC_MGCG | 1105 AMD_CG_SUPPORT_MC_LS | 1106 AMD_CG_SUPPORT_SDMA_MGCG | 1107 AMD_CG_SUPPORT_SDMA_LS | 1108 AMD_CG_SUPPORT_VCN_MGCG; 1109 1110 adev->pg_flags = AMD_PG_SUPPORT_SDMA | AMD_PG_SUPPORT_VCN; 1111 } 1112 break; 1113 case IP_VERSION(9, 4, 1): 1114 adev->asic_funcs = &vega20_asic_funcs; 1115 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1116 AMD_CG_SUPPORT_GFX_MGLS | 1117 AMD_CG_SUPPORT_GFX_CGCG | 1118 AMD_CG_SUPPORT_GFX_CGLS | 1119 AMD_CG_SUPPORT_GFX_CP_LS | 1120 AMD_CG_SUPPORT_HDP_MGCG | 1121 AMD_CG_SUPPORT_HDP_LS | 1122 AMD_CG_SUPPORT_SDMA_MGCG | 1123 AMD_CG_SUPPORT_SDMA_LS | 1124 AMD_CG_SUPPORT_MC_MGCG | 1125 AMD_CG_SUPPORT_MC_LS | 1126 AMD_CG_SUPPORT_IH_CG | 1127 AMD_CG_SUPPORT_VCN_MGCG | 1128 AMD_CG_SUPPORT_JPEG_MGCG; 1129 adev->pg_flags = AMD_PG_SUPPORT_VCN | AMD_PG_SUPPORT_VCN_DPG; 1130 adev->external_rev_id = adev->rev_id + 0x32; 1131 break; 1132 case IP_VERSION(9, 3, 0): 1133 adev->asic_funcs = &soc15_asic_funcs; 1134 1135 if (adev->apu_flags & AMD_APU_IS_RENOIR) 1136 adev->external_rev_id = adev->rev_id + 0x91; 1137 else 1138 adev->external_rev_id = adev->rev_id + 0xa1; 1139 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1140 AMD_CG_SUPPORT_GFX_MGLS | 1141 AMD_CG_SUPPORT_GFX_3D_CGCG | 1142 AMD_CG_SUPPORT_GFX_3D_CGLS | 1143 AMD_CG_SUPPORT_GFX_CGCG | 1144 AMD_CG_SUPPORT_GFX_CGLS | 1145 AMD_CG_SUPPORT_GFX_CP_LS | 1146 AMD_CG_SUPPORT_MC_MGCG | 1147 AMD_CG_SUPPORT_MC_LS | 1148 AMD_CG_SUPPORT_SDMA_MGCG | 1149 AMD_CG_SUPPORT_SDMA_LS | 1150 AMD_CG_SUPPORT_BIF_LS | 1151 AMD_CG_SUPPORT_HDP_LS | 1152 AMD_CG_SUPPORT_VCN_MGCG | 1153 AMD_CG_SUPPORT_JPEG_MGCG | 1154 AMD_CG_SUPPORT_IH_CG | 1155 AMD_CG_SUPPORT_ATHUB_LS | 1156 AMD_CG_SUPPORT_ATHUB_MGCG | 1157 AMD_CG_SUPPORT_DF_MGCG; 1158 adev->pg_flags = AMD_PG_SUPPORT_SDMA | 1159 AMD_PG_SUPPORT_VCN | 1160 AMD_PG_SUPPORT_JPEG | 1161 AMD_PG_SUPPORT_VCN_DPG; 1162 break; 1163 case IP_VERSION(9, 4, 2): 1164 adev->asic_funcs = &vega20_asic_funcs; 1165 adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG | 1166 AMD_CG_SUPPORT_GFX_MGLS | 1167 AMD_CG_SUPPORT_GFX_CP_LS | 1168 AMD_CG_SUPPORT_HDP_LS | 1169 AMD_CG_SUPPORT_SDMA_MGCG | 1170 AMD_CG_SUPPORT_SDMA_LS | 1171 AMD_CG_SUPPORT_IH_CG | 1172 AMD_CG_SUPPORT_VCN_MGCG | AMD_CG_SUPPORT_JPEG_MGCG; 1173 adev->pg_flags = AMD_PG_SUPPORT_VCN_DPG; 1174 adev->external_rev_id = adev->rev_id + 0x3c; 1175 break; 1176 default: 1177 /* FIXME: not supported yet */ 1178 return -EINVAL; 1179 } 1180 1181 if (amdgpu_sriov_vf(adev)) { 1182 amdgpu_virt_init_setting(adev); 1183 xgpu_ai_mailbox_set_irq_funcs(adev); 1184 } 1185 1186 return 0; 1187 } 1188 1189 static int soc15_common_late_init(void *handle) 1190 { 1191 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1192 1193 if (amdgpu_sriov_vf(adev)) 1194 xgpu_ai_mailbox_get_irq(adev); 1195 1196 return 0; 1197 } 1198 1199 static int soc15_common_sw_init(void *handle) 1200 { 1201 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1202 1203 if (amdgpu_sriov_vf(adev)) 1204 xgpu_ai_mailbox_add_irq_id(adev); 1205 1206 if (adev->df.funcs && 1207 adev->df.funcs->sw_init) 1208 adev->df.funcs->sw_init(adev); 1209 1210 return 0; 1211 } 1212 1213 static int soc15_common_sw_fini(void *handle) 1214 { 1215 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1216 1217 if (adev->nbio.ras && adev->nbio.ras->ras_block.ras_fini) 1218 adev->nbio.ras->ras_block.ras_fini(adev); 1219 1220 if (adev->df.funcs && 1221 adev->df.funcs->sw_fini) 1222 adev->df.funcs->sw_fini(adev); 1223 return 0; 1224 } 1225 1226 static void soc15_doorbell_range_init(struct amdgpu_device *adev) 1227 { 1228 int i; 1229 struct amdgpu_ring *ring; 1230 1231 /* sdma/ih doorbell range are programed by hypervisor */ 1232 if (!amdgpu_sriov_vf(adev)) { 1233 for (i = 0; i < adev->sdma.num_instances; i++) { 1234 ring = &adev->sdma.instance[i].ring; 1235 adev->nbio.funcs->sdma_doorbell_range(adev, i, 1236 ring->use_doorbell, ring->doorbell_index, 1237 adev->doorbell_index.sdma_doorbell_range); 1238 } 1239 1240 adev->nbio.funcs->ih_doorbell_range(adev, adev->irq.ih.use_doorbell, 1241 adev->irq.ih.doorbell_index); 1242 } 1243 } 1244 1245 static int soc15_common_hw_init(void *handle) 1246 { 1247 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1248 1249 /* enable pcie gen2/3 link */ 1250 soc15_pcie_gen3_enable(adev); 1251 /* enable aspm */ 1252 soc15_program_aspm(adev); 1253 /* setup nbio registers */ 1254 adev->nbio.funcs->init_registers(adev); 1255 /* remap HDP registers to a hole in mmio space, 1256 * for the purpose of expose those registers 1257 * to process space 1258 */ 1259 if (adev->nbio.funcs->remap_hdp_registers && !amdgpu_sriov_vf(adev)) 1260 adev->nbio.funcs->remap_hdp_registers(adev); 1261 1262 /* enable the doorbell aperture */ 1263 soc15_enable_doorbell_aperture(adev, true); 1264 /* HW doorbell routing policy: doorbell writing not 1265 * in SDMA/IH/MM/ACV range will be routed to CP. So 1266 * we need to init SDMA/IH/MM/ACV doorbell range prior 1267 * to CP ip block init and ring test. 1268 */ 1269 soc15_doorbell_range_init(adev); 1270 1271 return 0; 1272 } 1273 1274 static int soc15_common_hw_fini(void *handle) 1275 { 1276 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1277 1278 /* disable the doorbell aperture */ 1279 soc15_enable_doorbell_aperture(adev, false); 1280 if (amdgpu_sriov_vf(adev)) 1281 xgpu_ai_mailbox_put_irq(adev); 1282 1283 if (adev->nbio.ras_if && 1284 amdgpu_ras_is_supported(adev, adev->nbio.ras_if->block)) { 1285 if (adev->nbio.ras && 1286 adev->nbio.ras->init_ras_controller_interrupt) 1287 amdgpu_irq_put(adev, &adev->nbio.ras_controller_irq, 0); 1288 if (adev->nbio.ras && 1289 adev->nbio.ras->init_ras_err_event_athub_interrupt) 1290 amdgpu_irq_put(adev, &adev->nbio.ras_err_event_athub_irq, 0); 1291 } 1292 1293 return 0; 1294 } 1295 1296 static int soc15_common_suspend(void *handle) 1297 { 1298 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1299 1300 return soc15_common_hw_fini(adev); 1301 } 1302 1303 static int soc15_common_resume(void *handle) 1304 { 1305 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1306 1307 return soc15_common_hw_init(adev); 1308 } 1309 1310 static bool soc15_common_is_idle(void *handle) 1311 { 1312 return true; 1313 } 1314 1315 static int soc15_common_wait_for_idle(void *handle) 1316 { 1317 return 0; 1318 } 1319 1320 static int soc15_common_soft_reset(void *handle) 1321 { 1322 return 0; 1323 } 1324 1325 static void soc15_update_drm_clock_gating(struct amdgpu_device *adev, bool enable) 1326 { 1327 uint32_t def, data; 1328 1329 def = data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0)); 1330 1331 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DRM_MGCG)) 1332 data &= ~(0x01000000 | 1333 0x02000000 | 1334 0x04000000 | 1335 0x08000000 | 1336 0x10000000 | 1337 0x20000000 | 1338 0x40000000 | 1339 0x80000000); 1340 else 1341 data |= (0x01000000 | 1342 0x02000000 | 1343 0x04000000 | 1344 0x08000000 | 1345 0x10000000 | 1346 0x20000000 | 1347 0x40000000 | 1348 0x80000000); 1349 1350 if (def != data) 1351 WREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0), data); 1352 } 1353 1354 static void soc15_update_drm_light_sleep(struct amdgpu_device *adev, bool enable) 1355 { 1356 uint32_t def, data; 1357 1358 def = data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL)); 1359 1360 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DRM_LS)) 1361 data |= 1; 1362 else 1363 data &= ~1; 1364 1365 if (def != data) 1366 WREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL), data); 1367 } 1368 1369 static int soc15_common_set_clockgating_state(void *handle, 1370 enum amd_clockgating_state state) 1371 { 1372 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1373 1374 if (amdgpu_sriov_vf(adev)) 1375 return 0; 1376 1377 switch (adev->ip_versions[NBIO_HWIP][0]) { 1378 case IP_VERSION(6, 1, 0): 1379 case IP_VERSION(6, 2, 0): 1380 case IP_VERSION(7, 4, 0): 1381 adev->nbio.funcs->update_medium_grain_clock_gating(adev, 1382 state == AMD_CG_STATE_GATE); 1383 adev->nbio.funcs->update_medium_grain_light_sleep(adev, 1384 state == AMD_CG_STATE_GATE); 1385 adev->hdp.funcs->update_clock_gating(adev, 1386 state == AMD_CG_STATE_GATE); 1387 soc15_update_drm_clock_gating(adev, 1388 state == AMD_CG_STATE_GATE); 1389 soc15_update_drm_light_sleep(adev, 1390 state == AMD_CG_STATE_GATE); 1391 adev->smuio.funcs->update_rom_clock_gating(adev, 1392 state == AMD_CG_STATE_GATE); 1393 adev->df.funcs->update_medium_grain_clock_gating(adev, 1394 state == AMD_CG_STATE_GATE); 1395 break; 1396 case IP_VERSION(7, 0, 0): 1397 case IP_VERSION(7, 0, 1): 1398 case IP_VERSION(2, 5, 0): 1399 adev->nbio.funcs->update_medium_grain_clock_gating(adev, 1400 state == AMD_CG_STATE_GATE); 1401 adev->nbio.funcs->update_medium_grain_light_sleep(adev, 1402 state == AMD_CG_STATE_GATE); 1403 adev->hdp.funcs->update_clock_gating(adev, 1404 state == AMD_CG_STATE_GATE); 1405 soc15_update_drm_clock_gating(adev, 1406 state == AMD_CG_STATE_GATE); 1407 soc15_update_drm_light_sleep(adev, 1408 state == AMD_CG_STATE_GATE); 1409 break; 1410 case IP_VERSION(7, 4, 1): 1411 case IP_VERSION(7, 4, 4): 1412 adev->hdp.funcs->update_clock_gating(adev, 1413 state == AMD_CG_STATE_GATE); 1414 break; 1415 default: 1416 break; 1417 } 1418 return 0; 1419 } 1420 1421 static void soc15_common_get_clockgating_state(void *handle, u32 *flags) 1422 { 1423 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1424 int data; 1425 1426 if (amdgpu_sriov_vf(adev)) 1427 *flags = 0; 1428 1429 adev->nbio.funcs->get_clockgating_state(adev, flags); 1430 1431 adev->hdp.funcs->get_clock_gating_state(adev, flags); 1432 1433 if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(13, 0, 2)) { 1434 1435 /* AMD_CG_SUPPORT_DRM_MGCG */ 1436 data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0)); 1437 if (!(data & 0x01000000)) 1438 *flags |= AMD_CG_SUPPORT_DRM_MGCG; 1439 1440 /* AMD_CG_SUPPORT_DRM_LS */ 1441 data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL)); 1442 if (data & 0x1) 1443 *flags |= AMD_CG_SUPPORT_DRM_LS; 1444 } 1445 1446 /* AMD_CG_SUPPORT_ROM_MGCG */ 1447 adev->smuio.funcs->get_clock_gating_state(adev, flags); 1448 1449 adev->df.funcs->get_clockgating_state(adev, flags); 1450 } 1451 1452 static int soc15_common_set_powergating_state(void *handle, 1453 enum amd_powergating_state state) 1454 { 1455 /* todo */ 1456 return 0; 1457 } 1458 1459 static const struct amd_ip_funcs soc15_common_ip_funcs = { 1460 .name = "soc15_common", 1461 .early_init = soc15_common_early_init, 1462 .late_init = soc15_common_late_init, 1463 .sw_init = soc15_common_sw_init, 1464 .sw_fini = soc15_common_sw_fini, 1465 .hw_init = soc15_common_hw_init, 1466 .hw_fini = soc15_common_hw_fini, 1467 .suspend = soc15_common_suspend, 1468 .resume = soc15_common_resume, 1469 .is_idle = soc15_common_is_idle, 1470 .wait_for_idle = soc15_common_wait_for_idle, 1471 .soft_reset = soc15_common_soft_reset, 1472 .set_clockgating_state = soc15_common_set_clockgating_state, 1473 .set_powergating_state = soc15_common_set_powergating_state, 1474 .get_clockgating_state= soc15_common_get_clockgating_state, 1475 }; 1476