| 1 | /* |
| 2 | * Copyright 2023 Advanced Micro Devices, Inc. |
| 3 | * |
| 4 | * Permission is hereby granted, free of charge, to any person obtaining a |
| 5 | * copy of this software and associated documentation files (the "Software"), |
| 6 | * to deal in the Software without restriction, including without limitation |
| 7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 8 | * and/or sell copies of the Software, and to permit persons to whom the |
| 9 | * Software is furnished to do so, subject to the following conditions: |
| 10 | * |
| 11 | * The above copyright notice and this permission notice shall be included in |
| 12 | * all copies or substantial portions of the Software. |
| 13 | * |
| 14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 20 | * OTHER DEALINGS IN THE SOFTWARE. |
| 21 | * |
| 22 | */ |
| 23 | |
| 24 | #include <linux/firmware.h> |
| 25 | #include "amdgpu.h" |
| 26 | #include "amdgpu_vcn.h" |
| 27 | #include "amdgpu_pm.h" |
| 28 | #include "soc15.h" |
| 29 | #include "soc15d.h" |
| 30 | #include "soc15_hw_ip.h" |
| 31 | #include "vcn_v2_0.h" |
| 32 | |
| 33 | #include "vcn/vcn_5_0_0_offset.h" |
| 34 | #include "vcn/vcn_5_0_0_sh_mask.h" |
| 35 | #include "ivsrcid/vcn/irqsrcs_vcn_5_0.h" |
| 36 | #include "vcn_v5_0_0.h" |
| 37 | |
| 38 | #include <drm/drm_drv.h> |
| 39 | |
| 40 | static const struct amdgpu_hwip_reg_entry vcn_reg_list_5_0[] = { |
| 41 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_POWER_STATUS), |
| 42 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_STATUS), |
| 43 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_CONTEXT_ID), |
| 44 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_CONTEXT_ID2), |
| 45 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_DATA0), |
| 46 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_DATA1), |
| 47 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_CMD), |
| 48 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI), |
| 49 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO), |
| 50 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI2), |
| 51 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO2), |
| 52 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI3), |
| 53 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO3), |
| 54 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI4), |
| 55 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO4), |
| 56 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR), |
| 57 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR), |
| 58 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR2), |
| 59 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR2), |
| 60 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR3), |
| 61 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR3), |
| 62 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR4), |
| 63 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR4), |
| 64 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE), |
| 65 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE2), |
| 66 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE3), |
| 67 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE4), |
| 68 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_CTL), |
| 69 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_DATA), |
| 70 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_MASK), |
| 71 | SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_PAUSE) |
| 72 | }; |
| 73 | |
| 74 | static int amdgpu_ih_clientid_vcns[] = { |
| 75 | SOC15_IH_CLIENTID_VCN, |
| 76 | SOC15_IH_CLIENTID_VCN1 |
| 77 | }; |
| 78 | |
| 79 | static void vcn_v5_0_0_set_unified_ring_funcs(struct amdgpu_device *adev); |
| 80 | static void vcn_v5_0_0_set_irq_funcs(struct amdgpu_device *adev); |
| 81 | static int vcn_v5_0_0_set_pg_state(struct amdgpu_vcn_inst *vinst, |
| 82 | enum amd_powergating_state state); |
| 83 | static int vcn_v5_0_0_pause_dpg_mode(struct amdgpu_vcn_inst *vinst, |
| 84 | struct dpg_pause_state *new_state); |
| 85 | static void vcn_v5_0_0_unified_ring_set_wptr(struct amdgpu_ring *ring); |
| 86 | |
| 87 | /** |
| 88 | * vcn_v5_0_0_early_init - set function pointers and load microcode |
| 89 | * |
| 90 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 91 | * |
| 92 | * Set ring and irq function pointers |
| 93 | * Load microcode from filesystem |
| 94 | */ |
| 95 | static int vcn_v5_0_0_early_init(struct amdgpu_ip_block *ip_block) |
| 96 | { |
| 97 | struct amdgpu_device *adev = ip_block->adev; |
| 98 | int i, r; |
| 99 | |
| 100 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) |
| 101 | /* re-use enc ring as unified ring */ |
| 102 | adev->vcn.inst[i].num_enc_rings = 1; |
| 103 | |
| 104 | vcn_v5_0_0_set_unified_ring_funcs(adev); |
| 105 | vcn_v5_0_0_set_irq_funcs(adev); |
| 106 | |
| 107 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 108 | adev->vcn.inst[i].set_pg_state = vcn_v5_0_0_set_pg_state; |
| 109 | |
| 110 | r = amdgpu_vcn_early_init(adev, i); |
| 111 | if (r) |
| 112 | return r; |
| 113 | } |
| 114 | |
| 115 | return 0; |
| 116 | } |
| 117 | |
| 118 | /** |
| 119 | * vcn_v5_0_0_sw_init - sw init for VCN block |
| 120 | * |
| 121 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 122 | * |
| 123 | * Load firmware and sw initialization |
| 124 | */ |
| 125 | static int vcn_v5_0_0_sw_init(struct amdgpu_ip_block *ip_block) |
| 126 | { |
| 127 | struct amdgpu_ring *ring; |
| 128 | struct amdgpu_device *adev = ip_block->adev; |
| 129 | int i, r; |
| 130 | |
| 131 | for (i = 0; i < adev->vcn.num_vcn_inst; i++) { |
| 132 | struct amdgpu_vcn5_fw_shared *fw_shared; |
| 133 | |
| 134 | if (adev->vcn.harvest_config & (1 << i)) |
| 135 | continue; |
| 136 | |
| 137 | r = amdgpu_vcn_sw_init(adev, i); |
| 138 | if (r) |
| 139 | return r; |
| 140 | |
| 141 | amdgpu_vcn_setup_ucode(adev, i); |
| 142 | |
| 143 | r = amdgpu_vcn_resume(adev, i); |
| 144 | if (r) |
| 145 | return r; |
| 146 | |
| 147 | atomic_set(v: &adev->vcn.inst[i].sched_score, i: 0); |
| 148 | |
| 149 | /* VCN UNIFIED TRAP */ |
| 150 | r = amdgpu_irq_add_id(adev, client_id: amdgpu_ih_clientid_vcns[i], |
| 151 | VCN_5_0__SRCID__UVD_ENC_GENERAL_PURPOSE, source: &adev->vcn.inst[i].irq); |
| 152 | if (r) |
| 153 | return r; |
| 154 | |
| 155 | /* VCN POISON TRAP */ |
| 156 | r = amdgpu_irq_add_id(adev, client_id: amdgpu_ih_clientid_vcns[i], |
| 157 | VCN_5_0__SRCID_UVD_POISON, source: &adev->vcn.inst[i].irq); |
| 158 | if (r) |
| 159 | return r; |
| 160 | |
| 161 | ring = &adev->vcn.inst[i].ring_enc[0]; |
| 162 | ring->use_doorbell = true; |
| 163 | ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i; |
| 164 | |
| 165 | ring->vm_hub = AMDGPU_MMHUB0(0); |
| 166 | sprintf(buf: ring->name, fmt: "vcn_unified_%d" , i); |
| 167 | |
| 168 | r = amdgpu_ring_init(adev, ring, max_dw: 512, irq_src: &adev->vcn.inst[i].irq, irq_type: 0, |
| 169 | hw_prio: AMDGPU_RING_PRIO_0, sched_score: &adev->vcn.inst[i].sched_score); |
| 170 | if (r) |
| 171 | return r; |
| 172 | |
| 173 | fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr; |
| 174 | fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE); |
| 175 | fw_shared->sq.is_enabled = 1; |
| 176 | |
| 177 | if (amdgpu_vcnfw_log) |
| 178 | amdgpu_vcn_fwlog_init(vcn: &adev->vcn.inst[i]); |
| 179 | |
| 180 | if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) |
| 181 | adev->vcn.inst[i].pause_dpg_mode = vcn_v5_0_0_pause_dpg_mode; |
| 182 | } |
| 183 | |
| 184 | adev->vcn.supported_reset = |
| 185 | amdgpu_get_soft_full_reset_mask(ring: &adev->vcn.inst[0].ring_enc[0]); |
| 186 | if (!amdgpu_sriov_vf(adev)) |
| 187 | adev->vcn.supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; |
| 188 | |
| 189 | r = amdgpu_vcn_reg_dump_init(adev, reg: vcn_reg_list_5_0, ARRAY_SIZE(vcn_reg_list_5_0)); |
| 190 | if (r) |
| 191 | return r; |
| 192 | |
| 193 | r = amdgpu_vcn_sysfs_reset_mask_init(adev); |
| 194 | if (r) |
| 195 | return r; |
| 196 | |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | /** |
| 201 | * vcn_v5_0_0_sw_fini - sw fini for VCN block |
| 202 | * |
| 203 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 204 | * |
| 205 | * VCN suspend and free up sw allocation |
| 206 | */ |
| 207 | static int vcn_v5_0_0_sw_fini(struct amdgpu_ip_block *ip_block) |
| 208 | { |
| 209 | struct amdgpu_device *adev = ip_block->adev; |
| 210 | int i, r, idx; |
| 211 | |
| 212 | if (drm_dev_enter(dev: adev_to_drm(adev), idx: &idx)) { |
| 213 | for (i = 0; i < adev->vcn.num_vcn_inst; i++) { |
| 214 | struct amdgpu_vcn5_fw_shared *fw_shared; |
| 215 | |
| 216 | if (adev->vcn.harvest_config & (1 << i)) |
| 217 | continue; |
| 218 | |
| 219 | fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr; |
| 220 | fw_shared->present_flag_0 = 0; |
| 221 | fw_shared->sq.is_enabled = 0; |
| 222 | } |
| 223 | |
| 224 | drm_dev_exit(idx); |
| 225 | } |
| 226 | |
| 227 | for (i = 0; i < adev->vcn.num_vcn_inst; i++) { |
| 228 | r = amdgpu_vcn_suspend(adev, i); |
| 229 | if (r) |
| 230 | return r; |
| 231 | } |
| 232 | |
| 233 | amdgpu_vcn_sysfs_reset_mask_fini(adev); |
| 234 | |
| 235 | for (i = 0; i < adev->vcn.num_vcn_inst; i++) |
| 236 | amdgpu_vcn_sw_fini(adev, i); |
| 237 | |
| 238 | return 0; |
| 239 | } |
| 240 | |
| 241 | /** |
| 242 | * vcn_v5_0_0_hw_init - start and test VCN block |
| 243 | * |
| 244 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 245 | * |
| 246 | * Initialize the hardware, boot up the VCPU and do some testing |
| 247 | */ |
| 248 | static int vcn_v5_0_0_hw_init(struct amdgpu_ip_block *ip_block) |
| 249 | { |
| 250 | struct amdgpu_device *adev = ip_block->adev; |
| 251 | struct amdgpu_ring *ring; |
| 252 | int i, r; |
| 253 | |
| 254 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 255 | if (adev->vcn.harvest_config & (1 << i)) |
| 256 | continue; |
| 257 | |
| 258 | ring = &adev->vcn.inst[i].ring_enc[0]; |
| 259 | |
| 260 | adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell, |
| 261 | ((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i); |
| 262 | |
| 263 | r = amdgpu_ring_test_helper(ring); |
| 264 | if (r) |
| 265 | return r; |
| 266 | } |
| 267 | |
| 268 | return 0; |
| 269 | } |
| 270 | |
| 271 | /** |
| 272 | * vcn_v5_0_0_hw_fini - stop the hardware block |
| 273 | * |
| 274 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 275 | * |
| 276 | * Stop the VCN block, mark ring as not ready any more |
| 277 | */ |
| 278 | static int vcn_v5_0_0_hw_fini(struct amdgpu_ip_block *ip_block) |
| 279 | { |
| 280 | struct amdgpu_device *adev = ip_block->adev; |
| 281 | int i; |
| 282 | |
| 283 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 284 | struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[i]; |
| 285 | |
| 286 | if (adev->vcn.harvest_config & (1 << i)) |
| 287 | continue; |
| 288 | |
| 289 | cancel_delayed_work_sync(dwork: &vinst->idle_work); |
| 290 | |
| 291 | if (!amdgpu_sriov_vf(adev)) { |
| 292 | if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || |
| 293 | (vinst->cur_state != AMD_PG_STATE_GATE && |
| 294 | RREG32_SOC15(VCN, i, regUVD_STATUS))) { |
| 295 | vinst->set_pg_state(vinst, AMD_PG_STATE_GATE); |
| 296 | } |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | return 0; |
| 301 | } |
| 302 | |
| 303 | /** |
| 304 | * vcn_v5_0_0_suspend - suspend VCN block |
| 305 | * |
| 306 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 307 | * |
| 308 | * HW fini and suspend VCN block |
| 309 | */ |
| 310 | static int vcn_v5_0_0_suspend(struct amdgpu_ip_block *ip_block) |
| 311 | { |
| 312 | struct amdgpu_device *adev = ip_block->adev; |
| 313 | int r, i; |
| 314 | |
| 315 | r = vcn_v5_0_0_hw_fini(ip_block); |
| 316 | if (r) |
| 317 | return r; |
| 318 | |
| 319 | for (i = 0; i < adev->vcn.num_vcn_inst; i++) { |
| 320 | r = amdgpu_vcn_suspend(adev: ip_block->adev, i); |
| 321 | if (r) |
| 322 | return r; |
| 323 | } |
| 324 | |
| 325 | return r; |
| 326 | } |
| 327 | |
| 328 | /** |
| 329 | * vcn_v5_0_0_resume - resume VCN block |
| 330 | * |
| 331 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 332 | * |
| 333 | * Resume firmware and hw init VCN block |
| 334 | */ |
| 335 | static int vcn_v5_0_0_resume(struct amdgpu_ip_block *ip_block) |
| 336 | { |
| 337 | struct amdgpu_device *adev = ip_block->adev; |
| 338 | int r, i; |
| 339 | |
| 340 | for (i = 0; i < adev->vcn.num_vcn_inst; i++) { |
| 341 | r = amdgpu_vcn_resume(adev: ip_block->adev, i); |
| 342 | if (r) |
| 343 | return r; |
| 344 | } |
| 345 | |
| 346 | r = vcn_v5_0_0_hw_init(ip_block); |
| 347 | |
| 348 | return r; |
| 349 | } |
| 350 | |
| 351 | /** |
| 352 | * vcn_v5_0_0_mc_resume - memory controller programming |
| 353 | * |
| 354 | * @vinst: VCN instance |
| 355 | * |
| 356 | * Let the VCN memory controller know it's offsets |
| 357 | */ |
| 358 | static void vcn_v5_0_0_mc_resume(struct amdgpu_vcn_inst *vinst) |
| 359 | { |
| 360 | struct amdgpu_device *adev = vinst->adev; |
| 361 | int inst = vinst->inst; |
| 362 | uint32_t offset, size; |
| 363 | const struct common_firmware_header *hdr; |
| 364 | |
| 365 | hdr = (const struct common_firmware_header *)adev->vcn.inst[inst].fw->data; |
| 366 | size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8); |
| 367 | |
| 368 | /* cache window 0: fw */ |
| 369 | if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { |
| 370 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, |
| 371 | (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo)); |
| 372 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, |
| 373 | (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi)); |
| 374 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0); |
| 375 | offset = 0; |
| 376 | } else { |
| 377 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, |
| 378 | lower_32_bits(adev->vcn.inst[inst].gpu_addr)); |
| 379 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, |
| 380 | upper_32_bits(adev->vcn.inst[inst].gpu_addr)); |
| 381 | offset = size; |
| 382 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3); |
| 383 | } |
| 384 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size); |
| 385 | |
| 386 | /* cache window 1: stack */ |
| 387 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, |
| 388 | lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset)); |
| 389 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, |
| 390 | upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset)); |
| 391 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0); |
| 392 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE); |
| 393 | |
| 394 | /* cache window 2: context */ |
| 395 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, |
| 396 | lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); |
| 397 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, |
| 398 | upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); |
| 399 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0); |
| 400 | WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE); |
| 401 | |
| 402 | /* non-cache window */ |
| 403 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW, |
| 404 | lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr)); |
| 405 | WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH, |
| 406 | upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr)); |
| 407 | WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0); |
| 408 | WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0, |
| 409 | AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn5_fw_shared))); |
| 410 | } |
| 411 | |
| 412 | /** |
| 413 | * vcn_v5_0_0_mc_resume_dpg_mode - memory controller programming for dpg mode |
| 414 | * |
| 415 | * @vinst: VCN instance |
| 416 | * @indirect: indirectly write sram |
| 417 | * |
| 418 | * Let the VCN memory controller know it's offsets with dpg mode |
| 419 | */ |
| 420 | static void vcn_v5_0_0_mc_resume_dpg_mode(struct amdgpu_vcn_inst *vinst, |
| 421 | bool indirect) |
| 422 | { |
| 423 | struct amdgpu_device *adev = vinst->adev; |
| 424 | int inst_idx = vinst->inst; |
| 425 | uint32_t offset, size; |
| 426 | const struct common_firmware_header *hdr; |
| 427 | |
| 428 | hdr = (const struct common_firmware_header *)adev->vcn.inst[inst_idx].fw->data; |
| 429 | size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8); |
| 430 | |
| 431 | /* cache window 0: fw */ |
| 432 | if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { |
| 433 | if (!indirect) { |
| 434 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 435 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), |
| 436 | (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect); |
| 437 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 438 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), |
| 439 | (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect); |
| 440 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 441 | VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); |
| 442 | } else { |
| 443 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 444 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect); |
| 445 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 446 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect); |
| 447 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 448 | VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect); |
| 449 | } |
| 450 | offset = 0; |
| 451 | } else { |
| 452 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 453 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), |
| 454 | lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect); |
| 455 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 456 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), |
| 457 | upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect); |
| 458 | offset = size; |
| 459 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 460 | VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), |
| 461 | AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect); |
| 462 | } |
| 463 | |
| 464 | if (!indirect) |
| 465 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 466 | VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect); |
| 467 | else |
| 468 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 469 | VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect); |
| 470 | |
| 471 | /* cache window 1: stack */ |
| 472 | if (!indirect) { |
| 473 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 474 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), |
| 475 | lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect); |
| 476 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 477 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), |
| 478 | upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect); |
| 479 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 480 | VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); |
| 481 | } else { |
| 482 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 483 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect); |
| 484 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 485 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect); |
| 486 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 487 | VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect); |
| 488 | } |
| 489 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 490 | VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect); |
| 491 | |
| 492 | /* cache window 2: context */ |
| 493 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 494 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), |
| 495 | lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); |
| 496 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 497 | VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), |
| 498 | upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect); |
| 499 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 500 | VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect); |
| 501 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 502 | VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect); |
| 503 | |
| 504 | /* non-cache window */ |
| 505 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 506 | VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), |
| 507 | lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect); |
| 508 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 509 | VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), |
| 510 | upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect); |
| 511 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 512 | VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect); |
| 513 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 514 | VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0), |
| 515 | AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn5_fw_shared)), 0, indirect); |
| 516 | |
| 517 | /* VCN global tiling registers */ |
| 518 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 519 | VCN, inst_idx, regUVD_GFX10_ADDR_CONFIG), |
| 520 | adev->gfx.config.gb_addr_config, 0, indirect); |
| 521 | |
| 522 | return; |
| 523 | } |
| 524 | |
| 525 | /** |
| 526 | * vcn_v5_0_0_disable_static_power_gating - disable VCN static power gating |
| 527 | * |
| 528 | * @vinst: VCN instance |
| 529 | * |
| 530 | * Disable static power gating for VCN block |
| 531 | */ |
| 532 | static void vcn_v5_0_0_disable_static_power_gating(struct amdgpu_vcn_inst *vinst) |
| 533 | { |
| 534 | struct amdgpu_device *adev = vinst->adev; |
| 535 | int inst = vinst->inst; |
| 536 | uint32_t data = 0; |
| 537 | |
| 538 | if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { |
| 539 | data = 1 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT; |
| 540 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 541 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0, |
| 542 | UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK); |
| 543 | |
| 544 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT; |
| 545 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 546 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 547 | 1 << UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS__SHIFT, |
| 548 | UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK); |
| 549 | |
| 550 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT; |
| 551 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 552 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 553 | 1 << UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS__SHIFT, |
| 554 | UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK); |
| 555 | |
| 556 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT; |
| 557 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 558 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 559 | 1 << UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS__SHIFT, |
| 560 | UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK); |
| 561 | } else { |
| 562 | data = 1 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT; |
| 563 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 564 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0, |
| 565 | UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK); |
| 566 | |
| 567 | data = 1 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT; |
| 568 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 569 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0, |
| 570 | UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK); |
| 571 | |
| 572 | data = 1 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT; |
| 573 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 574 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0, |
| 575 | UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK); |
| 576 | |
| 577 | data = 1 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT; |
| 578 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 579 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0, |
| 580 | UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK); |
| 581 | } |
| 582 | |
| 583 | data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS); |
| 584 | data &= ~0x103; |
| 585 | if (adev->pg_flags & AMD_PG_SUPPORT_VCN) |
| 586 | data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON | |
| 587 | UVD_POWER_STATUS__UVD_PG_EN_MASK; |
| 588 | |
| 589 | WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data); |
| 590 | return; |
| 591 | } |
| 592 | |
| 593 | /** |
| 594 | * vcn_v5_0_0_enable_static_power_gating - enable VCN static power gating |
| 595 | * |
| 596 | * @vinst: VCN instance |
| 597 | * |
| 598 | * Enable static power gating for VCN block |
| 599 | */ |
| 600 | static void vcn_v5_0_0_enable_static_power_gating(struct amdgpu_vcn_inst *vinst) |
| 601 | { |
| 602 | struct amdgpu_device *adev = vinst->adev; |
| 603 | int inst = vinst->inst; |
| 604 | uint32_t data; |
| 605 | |
| 606 | if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { |
| 607 | /* Before power off, this indicator has to be turned on */ |
| 608 | data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS); |
| 609 | data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK; |
| 610 | data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF; |
| 611 | WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data); |
| 612 | |
| 613 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT; |
| 614 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 615 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 616 | 1 << UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS__SHIFT, |
| 617 | UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK); |
| 618 | |
| 619 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT; |
| 620 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 621 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 622 | 1 << UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS__SHIFT, |
| 623 | UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK); |
| 624 | |
| 625 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT; |
| 626 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 627 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 628 | 1 << UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS__SHIFT, |
| 629 | UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK); |
| 630 | |
| 631 | data = 2 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT; |
| 632 | WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG, data); |
| 633 | SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, |
| 634 | 1 << UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS__SHIFT, |
| 635 | UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK); |
| 636 | } |
| 637 | return; |
| 638 | } |
| 639 | |
| 640 | /** |
| 641 | * vcn_v5_0_0_disable_clock_gating - disable VCN clock gating |
| 642 | * |
| 643 | * @vinst: VCN instance |
| 644 | * |
| 645 | * Disable clock gating for VCN block |
| 646 | */ |
| 647 | static void vcn_v5_0_0_disable_clock_gating(struct amdgpu_vcn_inst *vinst) |
| 648 | { |
| 649 | return; |
| 650 | } |
| 651 | |
| 652 | #if 0 |
| 653 | /** |
| 654 | * vcn_v5_0_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode |
| 655 | * |
| 656 | * @vinst: VCN instance |
| 657 | * @sram_sel: sram select |
| 658 | * @indirect: indirectly write sram |
| 659 | * |
| 660 | * Disable clock gating for VCN block with dpg mode |
| 661 | */ |
| 662 | static void vcn_v5_0_0_disable_clock_gating_dpg_mode(struct amdgpu_vcn_inst *vinst, |
| 663 | uint8_t sram_sel, |
| 664 | uint8_t indirect) |
| 665 | { |
| 666 | return; |
| 667 | } |
| 668 | #endif |
| 669 | |
| 670 | /** |
| 671 | * vcn_v5_0_0_enable_clock_gating - enable VCN clock gating |
| 672 | * |
| 673 | * @vinst: VCN instance |
| 674 | * |
| 675 | * Enable clock gating for VCN block |
| 676 | */ |
| 677 | static void vcn_v5_0_0_enable_clock_gating(struct amdgpu_vcn_inst *vinst) |
| 678 | { |
| 679 | return; |
| 680 | } |
| 681 | |
| 682 | /** |
| 683 | * vcn_v5_0_0_start_dpg_mode - VCN start with dpg mode |
| 684 | * |
| 685 | * @vinst: VCN instance |
| 686 | * @indirect: indirectly write sram |
| 687 | * |
| 688 | * Start VCN block with dpg mode |
| 689 | */ |
| 690 | static int vcn_v5_0_0_start_dpg_mode(struct amdgpu_vcn_inst *vinst, |
| 691 | bool indirect) |
| 692 | { |
| 693 | struct amdgpu_device *adev = vinst->adev; |
| 694 | int inst_idx = vinst->inst; |
| 695 | struct amdgpu_vcn5_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr; |
| 696 | struct amdgpu_ring *ring; |
| 697 | uint32_t tmp; |
| 698 | int ret; |
| 699 | |
| 700 | /* disable register anti-hang mechanism */ |
| 701 | WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1, |
| 702 | ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); |
| 703 | |
| 704 | /* enable dynamic power gating mode */ |
| 705 | tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS); |
| 706 | tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; |
| 707 | tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; |
| 708 | WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp); |
| 709 | |
| 710 | if (indirect) |
| 711 | adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr; |
| 712 | |
| 713 | /* enable VCPU clock */ |
| 714 | tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); |
| 715 | tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK; |
| 716 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 717 | VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect); |
| 718 | |
| 719 | /* disable master interrupt */ |
| 720 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 721 | VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect); |
| 722 | |
| 723 | /* setup regUVD_LMI_CTRL */ |
| 724 | tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | |
| 725 | UVD_LMI_CTRL__REQ_MODE_MASK | |
| 726 | UVD_LMI_CTRL__CRC_RESET_MASK | |
| 727 | UVD_LMI_CTRL__MASK_MC_URGENT_MASK | |
| 728 | UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | |
| 729 | UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | |
| 730 | (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | |
| 731 | 0x00100000L); |
| 732 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 733 | VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect); |
| 734 | |
| 735 | vcn_v5_0_0_mc_resume_dpg_mode(vinst, indirect); |
| 736 | |
| 737 | tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); |
| 738 | tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; |
| 739 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 740 | VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect); |
| 741 | |
| 742 | /* enable LMI MC and UMC channels */ |
| 743 | tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT; |
| 744 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 745 | VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect); |
| 746 | |
| 747 | /* enable master interrupt */ |
| 748 | WREG32_SOC24_DPG_MODE(inst_idx, SOC24_DPG_MODE_OFFSET( |
| 749 | VCN, inst_idx, regUVD_MASTINT_EN), |
| 750 | UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect); |
| 751 | |
| 752 | if (indirect) { |
| 753 | ret = amdgpu_vcn_psp_update_sram(adev, inst_idx, ucode_id: 0); |
| 754 | if (ret) { |
| 755 | dev_err(adev->dev, "%s: vcn sram load failed %d\n" , __func__, ret); |
| 756 | return ret; |
| 757 | } |
| 758 | } |
| 759 | |
| 760 | ring = &adev->vcn.inst[inst_idx].ring_enc[0]; |
| 761 | |
| 762 | WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr); |
| 763 | WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); |
| 764 | WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4); |
| 765 | |
| 766 | tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE); |
| 767 | tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK); |
| 768 | WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp); |
| 769 | fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET; |
| 770 | WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0); |
| 771 | WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0); |
| 772 | |
| 773 | tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR); |
| 774 | WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp); |
| 775 | ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR); |
| 776 | |
| 777 | tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE); |
| 778 | tmp |= VCN_RB_ENABLE__RB1_EN_MASK; |
| 779 | WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp); |
| 780 | fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF); |
| 781 | |
| 782 | WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL, |
| 783 | ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT | |
| 784 | VCN_RB1_DB_CTRL__EN_MASK); |
| 785 | |
| 786 | /* Keeping one read-back to ensure all register writes are done, |
| 787 | * otherwise it may introduce race conditions. |
| 788 | */ |
| 789 | RREG32_SOC15(VCN, inst_idx, regUVD_STATUS); |
| 790 | |
| 791 | return 0; |
| 792 | } |
| 793 | |
| 794 | /** |
| 795 | * vcn_v5_0_0_start - VCN start |
| 796 | * |
| 797 | * @vinst: VCN instance |
| 798 | * |
| 799 | * Start VCN block |
| 800 | */ |
| 801 | static int vcn_v5_0_0_start(struct amdgpu_vcn_inst *vinst) |
| 802 | { |
| 803 | struct amdgpu_device *adev = vinst->adev; |
| 804 | int i = vinst->inst; |
| 805 | struct amdgpu_vcn5_fw_shared *fw_shared; |
| 806 | struct amdgpu_ring *ring; |
| 807 | uint32_t tmp; |
| 808 | int j, k, r; |
| 809 | |
| 810 | if (adev->vcn.harvest_config & (1 << i)) |
| 811 | return 0; |
| 812 | |
| 813 | if (adev->pm.dpm_enabled) |
| 814 | amdgpu_dpm_enable_vcn(adev, enable: true, inst: i); |
| 815 | |
| 816 | fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr; |
| 817 | |
| 818 | if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) |
| 819 | return vcn_v5_0_0_start_dpg_mode(vinst, indirect: adev->vcn.inst[i].indirect_sram); |
| 820 | |
| 821 | /* disable VCN power gating */ |
| 822 | vcn_v5_0_0_disable_static_power_gating(vinst); |
| 823 | |
| 824 | /* set VCN status busy */ |
| 825 | tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY; |
| 826 | WREG32_SOC15(VCN, i, regUVD_STATUS, tmp); |
| 827 | |
| 828 | /* enable VCPU clock */ |
| 829 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), |
| 830 | UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK); |
| 831 | |
| 832 | /* disable master interrupt */ |
| 833 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0, |
| 834 | ~UVD_MASTINT_EN__VCPU_EN_MASK); |
| 835 | |
| 836 | /* enable LMI MC and UMC channels */ |
| 837 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0, |
| 838 | ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); |
| 839 | |
| 840 | tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET); |
| 841 | tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK; |
| 842 | tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK; |
| 843 | WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp); |
| 844 | |
| 845 | /* setup regUVD_LMI_CTRL */ |
| 846 | tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL); |
| 847 | WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp | |
| 848 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | |
| 849 | UVD_LMI_CTRL__MASK_MC_URGENT_MASK | |
| 850 | UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | |
| 851 | UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK); |
| 852 | |
| 853 | vcn_v5_0_0_mc_resume(vinst); |
| 854 | |
| 855 | /* VCN global tiling registers */ |
| 856 | WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG, |
| 857 | adev->gfx.config.gb_addr_config); |
| 858 | |
| 859 | /* unblock VCPU register access */ |
| 860 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0, |
| 861 | ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK); |
| 862 | |
| 863 | /* release VCPU reset to boot */ |
| 864 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0, |
| 865 | ~UVD_VCPU_CNTL__BLK_RST_MASK); |
| 866 | |
| 867 | for (j = 0; j < 10; ++j) { |
| 868 | uint32_t status; |
| 869 | |
| 870 | for (k = 0; k < 100; ++k) { |
| 871 | status = RREG32_SOC15(VCN, i, regUVD_STATUS); |
| 872 | if (status & 2) |
| 873 | break; |
| 874 | mdelay(10); |
| 875 | if (amdgpu_emu_mode == 1) |
| 876 | msleep(msecs: 1); |
| 877 | } |
| 878 | |
| 879 | if (amdgpu_emu_mode == 1) { |
| 880 | r = -1; |
| 881 | if (status & 2) { |
| 882 | r = 0; |
| 883 | break; |
| 884 | } |
| 885 | } else { |
| 886 | r = 0; |
| 887 | if (status & 2) |
| 888 | break; |
| 889 | |
| 890 | dev_err(adev->dev, |
| 891 | "VCN[%d] is not responding, trying to reset the VCPU!!!\n" , i); |
| 892 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), |
| 893 | UVD_VCPU_CNTL__BLK_RST_MASK, |
| 894 | ~UVD_VCPU_CNTL__BLK_RST_MASK); |
| 895 | mdelay(10); |
| 896 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0, |
| 897 | ~UVD_VCPU_CNTL__BLK_RST_MASK); |
| 898 | |
| 899 | mdelay(10); |
| 900 | r = -1; |
| 901 | } |
| 902 | } |
| 903 | |
| 904 | if (r) { |
| 905 | dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n" , i); |
| 906 | return r; |
| 907 | } |
| 908 | |
| 909 | /* enable master interrupt */ |
| 910 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), |
| 911 | UVD_MASTINT_EN__VCPU_EN_MASK, |
| 912 | ~UVD_MASTINT_EN__VCPU_EN_MASK); |
| 913 | |
| 914 | /* clear the busy bit of VCN_STATUS */ |
| 915 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0, |
| 916 | ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT)); |
| 917 | |
| 918 | ring = &adev->vcn.inst[i].ring_enc[0]; |
| 919 | WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL, |
| 920 | ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT | |
| 921 | VCN_RB1_DB_CTRL__EN_MASK); |
| 922 | |
| 923 | WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr); |
| 924 | WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); |
| 925 | WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4); |
| 926 | |
| 927 | tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE); |
| 928 | tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK); |
| 929 | WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp); |
| 930 | fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET; |
| 931 | WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0); |
| 932 | WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0); |
| 933 | |
| 934 | tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR); |
| 935 | WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp); |
| 936 | ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR); |
| 937 | |
| 938 | tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE); |
| 939 | tmp |= VCN_RB_ENABLE__RB1_EN_MASK; |
| 940 | WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp); |
| 941 | fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF); |
| 942 | |
| 943 | /* Keeping one read-back to ensure all register writes are done, |
| 944 | * otherwise it may introduce race conditions. |
| 945 | */ |
| 946 | RREG32_SOC15(VCN, i, regUVD_STATUS); |
| 947 | |
| 948 | return 0; |
| 949 | } |
| 950 | |
| 951 | /** |
| 952 | * vcn_v5_0_0_stop_dpg_mode - VCN stop with dpg mode |
| 953 | * |
| 954 | * @vinst: VCN instance |
| 955 | * |
| 956 | * Stop VCN block with dpg mode |
| 957 | */ |
| 958 | static void vcn_v5_0_0_stop_dpg_mode(struct amdgpu_vcn_inst *vinst) |
| 959 | { |
| 960 | struct amdgpu_device *adev = vinst->adev; |
| 961 | int inst_idx = vinst->inst; |
| 962 | struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE}; |
| 963 | uint32_t tmp; |
| 964 | |
| 965 | vcn_v5_0_0_pause_dpg_mode(vinst, new_state: &state); |
| 966 | |
| 967 | /* Wait for power status to be 1 */ |
| 968 | SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1, |
| 969 | UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); |
| 970 | |
| 971 | /* wait for read ptr to be equal to write ptr */ |
| 972 | tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR); |
| 973 | SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF); |
| 974 | |
| 975 | /* disable dynamic power gating mode */ |
| 976 | WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0, |
| 977 | ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); |
| 978 | |
| 979 | /* Keeping one read-back to ensure all register writes are done, |
| 980 | * otherwise it may introduce race conditions. |
| 981 | */ |
| 982 | RREG32_SOC15(VCN, inst_idx, regUVD_STATUS); |
| 983 | |
| 984 | return; |
| 985 | } |
| 986 | |
| 987 | /** |
| 988 | * vcn_v5_0_0_stop - VCN stop |
| 989 | * |
| 990 | * @vinst: VCN instance |
| 991 | * |
| 992 | * Stop VCN block |
| 993 | */ |
| 994 | static int vcn_v5_0_0_stop(struct amdgpu_vcn_inst *vinst) |
| 995 | { |
| 996 | struct amdgpu_device *adev = vinst->adev; |
| 997 | int i = vinst->inst; |
| 998 | struct amdgpu_vcn5_fw_shared *fw_shared; |
| 999 | uint32_t tmp; |
| 1000 | int r = 0; |
| 1001 | |
| 1002 | if (adev->vcn.harvest_config & (1 << i)) |
| 1003 | return 0; |
| 1004 | |
| 1005 | fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr; |
| 1006 | fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF; |
| 1007 | |
| 1008 | if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { |
| 1009 | vcn_v5_0_0_stop_dpg_mode(vinst); |
| 1010 | r = 0; |
| 1011 | goto done; |
| 1012 | } |
| 1013 | |
| 1014 | /* wait for vcn idle */ |
| 1015 | r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7); |
| 1016 | if (r) |
| 1017 | goto done; |
| 1018 | |
| 1019 | tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK | |
| 1020 | UVD_LMI_STATUS__READ_CLEAN_MASK | |
| 1021 | UVD_LMI_STATUS__WRITE_CLEAN_MASK | |
| 1022 | UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK; |
| 1023 | r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp); |
| 1024 | if (r) |
| 1025 | goto done; |
| 1026 | |
| 1027 | /* disable LMI UMC channel */ |
| 1028 | tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2); |
| 1029 | tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK; |
| 1030 | WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp); |
| 1031 | tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK | |
| 1032 | UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK; |
| 1033 | r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp); |
| 1034 | if (r) |
| 1035 | goto done; |
| 1036 | |
| 1037 | /* block VCPU register access */ |
| 1038 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), |
| 1039 | UVD_RB_ARB_CTRL__VCPU_DIS_MASK, |
| 1040 | ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK); |
| 1041 | |
| 1042 | /* reset VCPU */ |
| 1043 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), |
| 1044 | UVD_VCPU_CNTL__BLK_RST_MASK, |
| 1045 | ~UVD_VCPU_CNTL__BLK_RST_MASK); |
| 1046 | |
| 1047 | /* disable VCPU clock */ |
| 1048 | WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0, |
| 1049 | ~(UVD_VCPU_CNTL__CLK_EN_MASK)); |
| 1050 | |
| 1051 | /* apply soft reset */ |
| 1052 | tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET); |
| 1053 | tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK; |
| 1054 | WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp); |
| 1055 | tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET); |
| 1056 | tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK; |
| 1057 | WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp); |
| 1058 | |
| 1059 | /* clear status */ |
| 1060 | WREG32_SOC15(VCN, i, regUVD_STATUS, 0); |
| 1061 | |
| 1062 | /* enable VCN power gating */ |
| 1063 | vcn_v5_0_0_enable_static_power_gating(vinst); |
| 1064 | |
| 1065 | /* Keeping one read-back to ensure all register writes are done, |
| 1066 | * otherwise it may introduce race conditions. |
| 1067 | */ |
| 1068 | RREG32_SOC15(VCN, i, regUVD_STATUS); |
| 1069 | |
| 1070 | done: |
| 1071 | if (adev->pm.dpm_enabled) |
| 1072 | amdgpu_dpm_enable_vcn(adev, enable: false, inst: i); |
| 1073 | |
| 1074 | return r; |
| 1075 | } |
| 1076 | |
| 1077 | /** |
| 1078 | * vcn_v5_0_0_pause_dpg_mode - VCN pause with dpg mode |
| 1079 | * |
| 1080 | * @vinst: VCN instance |
| 1081 | * @new_state: pause state |
| 1082 | * |
| 1083 | * Pause dpg mode for VCN block |
| 1084 | */ |
| 1085 | static int vcn_v5_0_0_pause_dpg_mode(struct amdgpu_vcn_inst *vinst, |
| 1086 | struct dpg_pause_state *new_state) |
| 1087 | { |
| 1088 | struct amdgpu_device *adev = vinst->adev; |
| 1089 | int inst_idx = vinst->inst; |
| 1090 | uint32_t reg_data = 0; |
| 1091 | int ret_code; |
| 1092 | |
| 1093 | /* pause/unpause if state is changed */ |
| 1094 | if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { |
| 1095 | DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d" , |
| 1096 | adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based); |
| 1097 | reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) & |
| 1098 | (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); |
| 1099 | |
| 1100 | if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { |
| 1101 | ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1, |
| 1102 | UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); |
| 1103 | |
| 1104 | if (!ret_code) { |
| 1105 | /* pause DPG */ |
| 1106 | reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; |
| 1107 | WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data); |
| 1108 | |
| 1109 | /* wait for ACK */ |
| 1110 | SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE, |
| 1111 | UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, |
| 1112 | UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); |
| 1113 | } |
| 1114 | } else { |
| 1115 | /* unpause dpg, no need to wait */ |
| 1116 | reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; |
| 1117 | WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data); |
| 1118 | } |
| 1119 | adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; |
| 1120 | } |
| 1121 | |
| 1122 | return 0; |
| 1123 | } |
| 1124 | |
| 1125 | /** |
| 1126 | * vcn_v5_0_0_unified_ring_get_rptr - get unified read pointer |
| 1127 | * |
| 1128 | * @ring: amdgpu_ring pointer |
| 1129 | * |
| 1130 | * Returns the current hardware unified read pointer |
| 1131 | */ |
| 1132 | static uint64_t vcn_v5_0_0_unified_ring_get_rptr(struct amdgpu_ring *ring) |
| 1133 | { |
| 1134 | struct amdgpu_device *adev = ring->adev; |
| 1135 | |
| 1136 | if (ring != &adev->vcn.inst[ring->me].ring_enc[0]) |
| 1137 | DRM_ERROR("wrong ring id is identified in %s" , __func__); |
| 1138 | |
| 1139 | return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR); |
| 1140 | } |
| 1141 | |
| 1142 | /** |
| 1143 | * vcn_v5_0_0_unified_ring_get_wptr - get unified write pointer |
| 1144 | * |
| 1145 | * @ring: amdgpu_ring pointer |
| 1146 | * |
| 1147 | * Returns the current hardware unified write pointer |
| 1148 | */ |
| 1149 | static uint64_t vcn_v5_0_0_unified_ring_get_wptr(struct amdgpu_ring *ring) |
| 1150 | { |
| 1151 | struct amdgpu_device *adev = ring->adev; |
| 1152 | |
| 1153 | if (ring != &adev->vcn.inst[ring->me].ring_enc[0]) |
| 1154 | DRM_ERROR("wrong ring id is identified in %s" , __func__); |
| 1155 | |
| 1156 | if (ring->use_doorbell) |
| 1157 | return *ring->wptr_cpu_addr; |
| 1158 | else |
| 1159 | return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR); |
| 1160 | } |
| 1161 | |
| 1162 | /** |
| 1163 | * vcn_v5_0_0_unified_ring_set_wptr - set enc write pointer |
| 1164 | * |
| 1165 | * @ring: amdgpu_ring pointer |
| 1166 | * |
| 1167 | * Commits the enc write pointer to the hardware |
| 1168 | */ |
| 1169 | static void vcn_v5_0_0_unified_ring_set_wptr(struct amdgpu_ring *ring) |
| 1170 | { |
| 1171 | struct amdgpu_device *adev = ring->adev; |
| 1172 | |
| 1173 | if (ring != &adev->vcn.inst[ring->me].ring_enc[0]) |
| 1174 | DRM_ERROR("wrong ring id is identified in %s" , __func__); |
| 1175 | |
| 1176 | if (ring->use_doorbell) { |
| 1177 | *ring->wptr_cpu_addr = lower_32_bits(ring->wptr); |
| 1178 | WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); |
| 1179 | } else { |
| 1180 | WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr)); |
| 1181 | } |
| 1182 | } |
| 1183 | |
| 1184 | static int vcn_v5_0_0_ring_reset(struct amdgpu_ring *ring, |
| 1185 | unsigned int vmid, |
| 1186 | struct amdgpu_fence *timedout_fence) |
| 1187 | { |
| 1188 | struct amdgpu_device *adev = ring->adev; |
| 1189 | struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[ring->me]; |
| 1190 | int r; |
| 1191 | |
| 1192 | amdgpu_ring_reset_helper_begin(ring, guilty_fence: timedout_fence); |
| 1193 | r = vcn_v5_0_0_stop(vinst); |
| 1194 | if (r) |
| 1195 | return r; |
| 1196 | r = vcn_v5_0_0_start(vinst); |
| 1197 | if (r) |
| 1198 | return r; |
| 1199 | return amdgpu_ring_reset_helper_end(ring, guilty_fence: timedout_fence); |
| 1200 | } |
| 1201 | |
| 1202 | static const struct amdgpu_ring_funcs vcn_v5_0_0_unified_ring_vm_funcs = { |
| 1203 | .type = AMDGPU_RING_TYPE_VCN_ENC, |
| 1204 | .align_mask = 0x3f, |
| 1205 | .nop = VCN_ENC_CMD_NO_OP, |
| 1206 | .get_rptr = vcn_v5_0_0_unified_ring_get_rptr, |
| 1207 | .get_wptr = vcn_v5_0_0_unified_ring_get_wptr, |
| 1208 | .set_wptr = vcn_v5_0_0_unified_ring_set_wptr, |
| 1209 | .emit_frame_size = |
| 1210 | SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + |
| 1211 | SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 + |
| 1212 | 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */ |
| 1213 | 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */ |
| 1214 | 1, /* vcn_v2_0_enc_ring_insert_end */ |
| 1215 | .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */ |
| 1216 | .emit_ib = vcn_v2_0_enc_ring_emit_ib, |
| 1217 | .emit_fence = vcn_v2_0_enc_ring_emit_fence, |
| 1218 | .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush, |
| 1219 | .test_ring = amdgpu_vcn_enc_ring_test_ring, |
| 1220 | .test_ib = amdgpu_vcn_unified_ring_test_ib, |
| 1221 | .insert_nop = amdgpu_ring_insert_nop, |
| 1222 | .insert_end = vcn_v2_0_enc_ring_insert_end, |
| 1223 | .pad_ib = amdgpu_ring_generic_pad_ib, |
| 1224 | .begin_use = amdgpu_vcn_ring_begin_use, |
| 1225 | .end_use = amdgpu_vcn_ring_end_use, |
| 1226 | .emit_wreg = vcn_v2_0_enc_ring_emit_wreg, |
| 1227 | .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait, |
| 1228 | .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, |
| 1229 | .reset = vcn_v5_0_0_ring_reset, |
| 1230 | }; |
| 1231 | |
| 1232 | /** |
| 1233 | * vcn_v5_0_0_set_unified_ring_funcs - set unified ring functions |
| 1234 | * |
| 1235 | * @adev: amdgpu_device pointer |
| 1236 | * |
| 1237 | * Set unified ring functions |
| 1238 | */ |
| 1239 | static void vcn_v5_0_0_set_unified_ring_funcs(struct amdgpu_device *adev) |
| 1240 | { |
| 1241 | int i; |
| 1242 | |
| 1243 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 1244 | if (adev->vcn.harvest_config & (1 << i)) |
| 1245 | continue; |
| 1246 | |
| 1247 | adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v5_0_0_unified_ring_vm_funcs; |
| 1248 | adev->vcn.inst[i].ring_enc[0].me = i; |
| 1249 | } |
| 1250 | } |
| 1251 | |
| 1252 | /** |
| 1253 | * vcn_v5_0_0_is_idle - check VCN block is idle |
| 1254 | * |
| 1255 | * @ip_block: Pointer to the amdgpu_ip_block structure |
| 1256 | * |
| 1257 | * Check whether VCN block is idle |
| 1258 | */ |
| 1259 | static bool vcn_v5_0_0_is_idle(struct amdgpu_ip_block *ip_block) |
| 1260 | { |
| 1261 | struct amdgpu_device *adev = ip_block->adev; |
| 1262 | int i, ret = 1; |
| 1263 | |
| 1264 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 1265 | if (adev->vcn.harvest_config & (1 << i)) |
| 1266 | continue; |
| 1267 | |
| 1268 | ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE); |
| 1269 | } |
| 1270 | |
| 1271 | return ret; |
| 1272 | } |
| 1273 | |
| 1274 | /** |
| 1275 | * vcn_v5_0_0_wait_for_idle - wait for VCN block idle |
| 1276 | * |
| 1277 | * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. |
| 1278 | * |
| 1279 | * Wait for VCN block idle |
| 1280 | */ |
| 1281 | static int vcn_v5_0_0_wait_for_idle(struct amdgpu_ip_block *ip_block) |
| 1282 | { |
| 1283 | struct amdgpu_device *adev = ip_block->adev; |
| 1284 | int i, ret = 0; |
| 1285 | |
| 1286 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 1287 | if (adev->vcn.harvest_config & (1 << i)) |
| 1288 | continue; |
| 1289 | |
| 1290 | ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, |
| 1291 | UVD_STATUS__IDLE); |
| 1292 | if (ret) |
| 1293 | return ret; |
| 1294 | } |
| 1295 | |
| 1296 | return ret; |
| 1297 | } |
| 1298 | |
| 1299 | /** |
| 1300 | * vcn_v5_0_0_set_clockgating_state - set VCN block clockgating state |
| 1301 | * |
| 1302 | * @ip_block: amdgpu_ip_block pointer |
| 1303 | * @state: clock gating state |
| 1304 | * |
| 1305 | * Set VCN block clockgating state |
| 1306 | */ |
| 1307 | static int vcn_v5_0_0_set_clockgating_state(struct amdgpu_ip_block *ip_block, |
| 1308 | enum amd_clockgating_state state) |
| 1309 | { |
| 1310 | struct amdgpu_device *adev = ip_block->adev; |
| 1311 | bool enable = state == AMD_CG_STATE_GATE; |
| 1312 | int i; |
| 1313 | |
| 1314 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 1315 | struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[i]; |
| 1316 | |
| 1317 | if (adev->vcn.harvest_config & (1 << i)) |
| 1318 | continue; |
| 1319 | |
| 1320 | if (enable) { |
| 1321 | if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE) |
| 1322 | return -EBUSY; |
| 1323 | vcn_v5_0_0_enable_clock_gating(vinst); |
| 1324 | } else { |
| 1325 | vcn_v5_0_0_disable_clock_gating(vinst); |
| 1326 | } |
| 1327 | } |
| 1328 | |
| 1329 | return 0; |
| 1330 | } |
| 1331 | |
| 1332 | static int vcn_v5_0_0_set_pg_state(struct amdgpu_vcn_inst *vinst, |
| 1333 | enum amd_powergating_state state) |
| 1334 | { |
| 1335 | int ret = 0; |
| 1336 | |
| 1337 | if (state == vinst->cur_state) |
| 1338 | return 0; |
| 1339 | |
| 1340 | if (state == AMD_PG_STATE_GATE) |
| 1341 | ret = vcn_v5_0_0_stop(vinst); |
| 1342 | else |
| 1343 | ret = vcn_v5_0_0_start(vinst); |
| 1344 | |
| 1345 | if (!ret) |
| 1346 | vinst->cur_state = state; |
| 1347 | |
| 1348 | return ret; |
| 1349 | } |
| 1350 | |
| 1351 | /** |
| 1352 | * vcn_v5_0_0_process_interrupt - process VCN block interrupt |
| 1353 | * |
| 1354 | * @adev: amdgpu_device pointer |
| 1355 | * @source: interrupt sources |
| 1356 | * @entry: interrupt entry from clients and sources |
| 1357 | * |
| 1358 | * Process VCN block interrupt |
| 1359 | */ |
| 1360 | static int vcn_v5_0_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source, |
| 1361 | struct amdgpu_iv_entry *entry) |
| 1362 | { |
| 1363 | uint32_t ip_instance; |
| 1364 | |
| 1365 | switch (entry->client_id) { |
| 1366 | case SOC15_IH_CLIENTID_VCN: |
| 1367 | ip_instance = 0; |
| 1368 | break; |
| 1369 | case SOC15_IH_CLIENTID_VCN1: |
| 1370 | ip_instance = 1; |
| 1371 | break; |
| 1372 | default: |
| 1373 | DRM_ERROR("Unhandled client id: %d\n" , entry->client_id); |
| 1374 | return 0; |
| 1375 | } |
| 1376 | |
| 1377 | DRM_DEBUG("IH: VCN TRAP\n" ); |
| 1378 | |
| 1379 | switch (entry->src_id) { |
| 1380 | case VCN_5_0__SRCID__UVD_ENC_GENERAL_PURPOSE: |
| 1381 | amdgpu_fence_process(ring: &adev->vcn.inst[ip_instance].ring_enc[0]); |
| 1382 | break; |
| 1383 | case VCN_5_0__SRCID_UVD_POISON: |
| 1384 | amdgpu_vcn_process_poison_irq(adev, source, entry); |
| 1385 | break; |
| 1386 | default: |
| 1387 | DRM_ERROR("Unhandled interrupt: %d %d\n" , |
| 1388 | entry->src_id, entry->src_data[0]); |
| 1389 | break; |
| 1390 | } |
| 1391 | |
| 1392 | return 0; |
| 1393 | } |
| 1394 | |
| 1395 | static const struct amdgpu_irq_src_funcs vcn_v5_0_0_irq_funcs = { |
| 1396 | .process = vcn_v5_0_0_process_interrupt, |
| 1397 | }; |
| 1398 | |
| 1399 | /** |
| 1400 | * vcn_v5_0_0_set_irq_funcs - set VCN block interrupt irq functions |
| 1401 | * |
| 1402 | * @adev: amdgpu_device pointer |
| 1403 | * |
| 1404 | * Set VCN block interrupt irq functions |
| 1405 | */ |
| 1406 | static void vcn_v5_0_0_set_irq_funcs(struct amdgpu_device *adev) |
| 1407 | { |
| 1408 | int i; |
| 1409 | |
| 1410 | for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { |
| 1411 | if (adev->vcn.harvest_config & (1 << i)) |
| 1412 | continue; |
| 1413 | |
| 1414 | adev->vcn.inst[i].irq.num_types = adev->vcn.inst[i].num_enc_rings + 1; |
| 1415 | adev->vcn.inst[i].irq.funcs = &vcn_v5_0_0_irq_funcs; |
| 1416 | } |
| 1417 | } |
| 1418 | |
| 1419 | static const struct amd_ip_funcs vcn_v5_0_0_ip_funcs = { |
| 1420 | .name = "vcn_v5_0_0" , |
| 1421 | .early_init = vcn_v5_0_0_early_init, |
| 1422 | .sw_init = vcn_v5_0_0_sw_init, |
| 1423 | .sw_fini = vcn_v5_0_0_sw_fini, |
| 1424 | .hw_init = vcn_v5_0_0_hw_init, |
| 1425 | .hw_fini = vcn_v5_0_0_hw_fini, |
| 1426 | .suspend = vcn_v5_0_0_suspend, |
| 1427 | .resume = vcn_v5_0_0_resume, |
| 1428 | .is_idle = vcn_v5_0_0_is_idle, |
| 1429 | .wait_for_idle = vcn_v5_0_0_wait_for_idle, |
| 1430 | .set_clockgating_state = vcn_v5_0_0_set_clockgating_state, |
| 1431 | .set_powergating_state = vcn_set_powergating_state, |
| 1432 | .dump_ip_state = amdgpu_vcn_dump_ip_state, |
| 1433 | .print_ip_state = amdgpu_vcn_print_ip_state, |
| 1434 | }; |
| 1435 | |
| 1436 | const struct amdgpu_ip_block_version vcn_v5_0_0_ip_block = { |
| 1437 | .type = AMD_IP_BLOCK_TYPE_VCN, |
| 1438 | .major = 5, |
| 1439 | .minor = 0, |
| 1440 | .rev = 0, |
| 1441 | .funcs = &vcn_v5_0_0_ip_funcs, |
| 1442 | }; |
| 1443 | |