2 * Copyright 2016 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
27 #include <linux/firmware.h>
28 #include <linux/module.h>
29 #include <linux/dmi.h>
30 #include <linux/pci.h>
31 #include <linux/debugfs.h>
32 #include <drm/drm_drv.h>
35 #include "amdgpu_pm.h"
36 #include "amdgpu_vcn.h"
40 #define FIRMWARE_RAVEN "amdgpu/raven_vcn.bin"
41 #define FIRMWARE_PICASSO "amdgpu/picasso_vcn.bin"
42 #define FIRMWARE_RAVEN2 "amdgpu/raven2_vcn.bin"
43 #define FIRMWARE_ARCTURUS "amdgpu/arcturus_vcn.bin"
44 #define FIRMWARE_RENOIR "amdgpu/renoir_vcn.bin"
45 #define FIRMWARE_GREEN_SARDINE "amdgpu/green_sardine_vcn.bin"
46 #define FIRMWARE_NAVI10 "amdgpu/navi10_vcn.bin"
47 #define FIRMWARE_NAVI14 "amdgpu/navi14_vcn.bin"
48 #define FIRMWARE_NAVI12 "amdgpu/navi12_vcn.bin"
49 #define FIRMWARE_SIENNA_CICHLID "amdgpu/sienna_cichlid_vcn.bin"
50 #define FIRMWARE_NAVY_FLOUNDER "amdgpu/navy_flounder_vcn.bin"
51 #define FIRMWARE_VANGOGH "amdgpu/vangogh_vcn.bin"
52 #define FIRMWARE_DIMGREY_CAVEFISH "amdgpu/dimgrey_cavefish_vcn.bin"
53 #define FIRMWARE_ALDEBARAN "amdgpu/aldebaran_vcn.bin"
54 #define FIRMWARE_BEIGE_GOBY "amdgpu/beige_goby_vcn.bin"
55 #define FIRMWARE_YELLOW_CARP "amdgpu/yellow_carp_vcn.bin"
56 #define FIRMWARE_VCN_3_1_2 "amdgpu/vcn_3_1_2.bin"
57 #define FIRMWARE_VCN4_0_0 "amdgpu/vcn_4_0_0.bin"
58 #define FIRMWARE_VCN4_0_2 "amdgpu/vcn_4_0_2.bin"
59 #define FIRMWARE_VCN4_0_3 "amdgpu/vcn_4_0_3.bin"
60 #define FIRMWARE_VCN4_0_4 "amdgpu/vcn_4_0_4.bin"
61 #define FIRMWARE_VCN4_0_5 "amdgpu/vcn_4_0_5.bin"
63 MODULE_FIRMWARE(FIRMWARE_RAVEN);
64 MODULE_FIRMWARE(FIRMWARE_PICASSO);
65 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
66 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
67 MODULE_FIRMWARE(FIRMWARE_RENOIR);
68 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
69 MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
70 MODULE_FIRMWARE(FIRMWARE_NAVI10);
71 MODULE_FIRMWARE(FIRMWARE_NAVI14);
72 MODULE_FIRMWARE(FIRMWARE_NAVI12);
73 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
74 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
75 MODULE_FIRMWARE(FIRMWARE_VANGOGH);
76 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
77 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
78 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
79 MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
80 MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
81 MODULE_FIRMWARE(FIRMWARE_VCN4_0_2);
82 MODULE_FIRMWARE(FIRMWARE_VCN4_0_3);
83 MODULE_FIRMWARE(FIRMWARE_VCN4_0_4);
84 MODULE_FIRMWARE(FIRMWARE_VCN4_0_5);
86 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
88 int amdgpu_vcn_early_init(struct amdgpu_device *adev)
90 char ucode_prefix[30];
94 amdgpu_ucode_ip_version_decode(adev, UVD_HWIP, ucode_prefix, sizeof(ucode_prefix));
95 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix);
96 r = amdgpu_ucode_request(adev, &adev->vcn.fw, fw_name);
98 amdgpu_ucode_release(&adev->vcn.fw);
103 int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
105 unsigned long bo_size;
106 const struct common_firmware_header *hdr;
107 unsigned char fw_check;
108 unsigned int fw_shared_size, log_offset;
111 INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
112 mutex_init(&adev->vcn.vcn_pg_lock);
113 mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
114 atomic_set(&adev->vcn.total_submission_cnt, 0);
115 for (i = 0; i < adev->vcn.num_vcn_inst; i++)
116 atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
118 if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
119 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
120 adev->vcn.indirect_sram = true;
123 * Some Steam Deck's BIOS versions are incompatible with the
124 * indirect SRAM mode, leading to amdgpu being unable to get
125 * properly probed (and even potentially crashing the kernel).
126 * Hence, check for these versions here - notice this is
127 * restricted to Vangogh (Deck's APU).
129 if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(3, 0, 2)) {
130 const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION);
132 if (bios_ver && (!strncmp("F7A0113", bios_ver, 7) ||
133 !strncmp("F7A0114", bios_ver, 7))) {
134 adev->vcn.indirect_sram = false;
136 "Steam Deck quirk: indirect SRAM disabled on BIOS %s\n", bios_ver);
140 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
141 adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
143 /* Bit 20-23, it is encode major and non-zero for new naming convention.
144 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
145 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
146 * is zero in old naming convention, this field is always zero so far.
147 * These four bits are used to tell which naming convention is present.
149 fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
151 unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
153 fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
154 enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
155 enc_major = fw_check;
156 dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
157 vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
158 DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
159 enc_major, enc_minor, dec_ver, vep, fw_rev);
161 unsigned int version_major, version_minor, family_id;
163 family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
164 version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
165 version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
166 DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
167 version_major, version_minor, family_id);
170 bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
171 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
172 bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
174 if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0)) {
175 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
176 log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
178 fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
179 log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
182 bo_size += fw_shared_size;
184 if (amdgpu_vcnfw_log)
185 bo_size += AMDGPU_VCNFW_LOG_SIZE;
187 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
188 if (adev->vcn.harvest_config & (1 << i))
191 r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
192 AMDGPU_GEM_DOMAIN_VRAM |
193 AMDGPU_GEM_DOMAIN_GTT,
194 &adev->vcn.inst[i].vcpu_bo,
195 &adev->vcn.inst[i].gpu_addr,
196 &adev->vcn.inst[i].cpu_addr);
198 dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
202 adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr +
203 bo_size - fw_shared_size;
204 adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr +
205 bo_size - fw_shared_size;
207 adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size;
209 if (amdgpu_vcnfw_log) {
210 adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
211 adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
212 adev->vcn.inst[i].fw_shared.log_offset = log_offset;
215 if (adev->vcn.indirect_sram) {
216 r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
217 AMDGPU_GEM_DOMAIN_VRAM |
218 AMDGPU_GEM_DOMAIN_GTT,
219 &adev->vcn.inst[i].dpg_sram_bo,
220 &adev->vcn.inst[i].dpg_sram_gpu_addr,
221 &adev->vcn.inst[i].dpg_sram_cpu_addr);
223 dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
232 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
236 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
237 if (adev->vcn.harvest_config & (1 << j))
240 amdgpu_bo_free_kernel(
241 &adev->vcn.inst[j].dpg_sram_bo,
242 &adev->vcn.inst[j].dpg_sram_gpu_addr,
243 (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
245 kvfree(adev->vcn.inst[j].saved_bo);
247 amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
248 &adev->vcn.inst[j].gpu_addr,
249 (void **)&adev->vcn.inst[j].cpu_addr);
251 amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
253 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
254 amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
257 amdgpu_ucode_release(&adev->vcn.fw);
258 mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
259 mutex_destroy(&adev->vcn.vcn_pg_lock);
264 /* from vcn4 and above, only unified queue is used */
265 static bool amdgpu_vcn_using_unified_queue(struct amdgpu_ring *ring)
267 struct amdgpu_device *adev = ring->adev;
270 if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0))
276 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
279 int vcn_config = adev->vcn.vcn_config[vcn_instance];
281 if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK))
283 else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK))
285 else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK))
291 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
297 bool in_ras_intr = amdgpu_ras_intr_triggered();
299 cancel_delayed_work_sync(&adev->vcn.idle_work);
301 /* err_event_athub will corrupt VCPU buffer, so we need to
302 * restore fw data and clear buffer in amdgpu_vcn_resume() */
306 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
307 if (adev->vcn.harvest_config & (1 << i))
309 if (adev->vcn.inst[i].vcpu_bo == NULL)
312 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
313 ptr = adev->vcn.inst[i].cpu_addr;
315 adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
316 if (!adev->vcn.inst[i].saved_bo)
319 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
320 memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
327 int amdgpu_vcn_resume(struct amdgpu_device *adev)
333 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
334 if (adev->vcn.harvest_config & (1 << i))
336 if (adev->vcn.inst[i].vcpu_bo == NULL)
339 size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
340 ptr = adev->vcn.inst[i].cpu_addr;
342 if (adev->vcn.inst[i].saved_bo != NULL) {
343 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
344 memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
347 kvfree(adev->vcn.inst[i].saved_bo);
348 adev->vcn.inst[i].saved_bo = NULL;
350 const struct common_firmware_header *hdr;
353 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
354 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
355 offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
356 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
357 memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
358 le32_to_cpu(hdr->ucode_size_bytes));
361 size -= le32_to_cpu(hdr->ucode_size_bytes);
362 ptr += le32_to_cpu(hdr->ucode_size_bytes);
364 memset_io(ptr, 0, size);
370 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
372 struct amdgpu_device *adev =
373 container_of(work, struct amdgpu_device, vcn.idle_work.work);
374 unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
378 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
379 if (adev->vcn.harvest_config & (1 << j))
382 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
383 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
385 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
386 struct dpg_pause_state new_state;
389 unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
390 new_state.fw_based = VCN_DPG_STATE__PAUSE;
392 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
394 adev->vcn.pause_dpg_mode(adev, j, &new_state);
397 fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
401 if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
402 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
404 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
407 dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
409 schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
413 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
415 struct amdgpu_device *adev = ring->adev;
418 atomic_inc(&adev->vcn.total_submission_cnt);
420 if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
421 r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
424 dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
427 mutex_lock(&adev->vcn.vcn_pg_lock);
428 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
429 AMD_PG_STATE_UNGATE);
431 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
432 struct dpg_pause_state new_state;
434 if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
435 atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
436 new_state.fw_based = VCN_DPG_STATE__PAUSE;
438 unsigned int fences = 0;
441 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
442 fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
444 if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
445 new_state.fw_based = VCN_DPG_STATE__PAUSE;
447 new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
450 adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
452 mutex_unlock(&adev->vcn.vcn_pg_lock);
455 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
457 if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
458 ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
459 atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
461 atomic_dec(&ring->adev->vcn.total_submission_cnt);
463 schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
466 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
468 struct amdgpu_device *adev = ring->adev;
473 /* VCN in SRIOV does not support direct register read/write */
474 if (amdgpu_sriov_vf(adev))
477 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
478 r = amdgpu_ring_alloc(ring, 3);
481 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
482 amdgpu_ring_write(ring, 0xDEADBEEF);
483 amdgpu_ring_commit(ring);
484 for (i = 0; i < adev->usec_timeout; i++) {
485 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
486 if (tmp == 0xDEADBEEF)
491 if (i >= adev->usec_timeout)
497 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
499 struct amdgpu_device *adev = ring->adev;
504 if (amdgpu_sriov_vf(adev))
507 r = amdgpu_ring_alloc(ring, 16);
511 rptr = amdgpu_ring_get_rptr(ring);
513 amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
514 amdgpu_ring_commit(ring);
516 for (i = 0; i < adev->usec_timeout; i++) {
517 if (amdgpu_ring_get_rptr(ring) != rptr)
522 if (i >= adev->usec_timeout)
528 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
529 struct amdgpu_ib *ib_msg,
530 struct dma_fence **fence)
532 u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
533 struct amdgpu_device *adev = ring->adev;
534 struct dma_fence *f = NULL;
535 struct amdgpu_job *job;
536 struct amdgpu_ib *ib;
539 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
540 64, AMDGPU_IB_POOL_DIRECT,
546 ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
548 ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
549 ib->ptr[3] = addr >> 32;
550 ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
552 for (i = 6; i < 16; i += 2) {
553 ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
558 r = amdgpu_job_submit_direct(job, ring, &f);
562 amdgpu_ib_free(adev, ib_msg, f);
565 *fence = dma_fence_get(f);
571 amdgpu_job_free(job);
573 amdgpu_ib_free(adev, ib_msg, f);
577 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
578 struct amdgpu_ib *ib)
580 struct amdgpu_device *adev = ring->adev;
584 memset(ib, 0, sizeof(*ib));
585 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
586 AMDGPU_IB_POOL_DIRECT,
591 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
592 msg[0] = cpu_to_le32(0x00000028);
593 msg[1] = cpu_to_le32(0x00000038);
594 msg[2] = cpu_to_le32(0x00000001);
595 msg[3] = cpu_to_le32(0x00000000);
596 msg[4] = cpu_to_le32(handle);
597 msg[5] = cpu_to_le32(0x00000000);
598 msg[6] = cpu_to_le32(0x00000001);
599 msg[7] = cpu_to_le32(0x00000028);
600 msg[8] = cpu_to_le32(0x00000010);
601 msg[9] = cpu_to_le32(0x00000000);
602 msg[10] = cpu_to_le32(0x00000007);
603 msg[11] = cpu_to_le32(0x00000000);
604 msg[12] = cpu_to_le32(0x00000780);
605 msg[13] = cpu_to_le32(0x00000440);
606 for (i = 14; i < 1024; ++i)
607 msg[i] = cpu_to_le32(0x0);
612 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
613 struct amdgpu_ib *ib)
615 struct amdgpu_device *adev = ring->adev;
619 memset(ib, 0, sizeof(*ib));
620 r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
621 AMDGPU_IB_POOL_DIRECT,
626 msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
627 msg[0] = cpu_to_le32(0x00000028);
628 msg[1] = cpu_to_le32(0x00000018);
629 msg[2] = cpu_to_le32(0x00000000);
630 msg[3] = cpu_to_le32(0x00000002);
631 msg[4] = cpu_to_le32(handle);
632 msg[5] = cpu_to_le32(0x00000000);
633 for (i = 6; i < 1024; ++i)
634 msg[i] = cpu_to_le32(0x0);
639 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
641 struct dma_fence *fence = NULL;
645 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
649 r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
652 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
656 r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
660 r = dma_fence_wait_timeout(fence, false, timeout);
666 dma_fence_put(fence);
671 static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib,
672 uint32_t ib_pack_in_dw, bool enc)
674 uint32_t *ib_checksum;
676 ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */
677 ib->ptr[ib->length_dw++] = 0x30000002;
678 ib_checksum = &ib->ptr[ib->length_dw++];
679 ib->ptr[ib->length_dw++] = ib_pack_in_dw;
681 ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */
682 ib->ptr[ib->length_dw++] = 0x30000001;
683 ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3;
684 ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t);
689 static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
690 uint32_t ib_pack_in_dw)
693 uint32_t checksum = 0;
695 for (i = 0; i < ib_pack_in_dw; i++)
696 checksum += *(*ib_checksum + 2 + i);
698 **ib_checksum = checksum;
701 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
702 struct amdgpu_ib *ib_msg,
703 struct dma_fence **fence)
705 struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
706 unsigned int ib_size_dw = 64;
707 struct amdgpu_device *adev = ring->adev;
708 struct dma_fence *f = NULL;
709 struct amdgpu_job *job;
710 struct amdgpu_ib *ib;
711 uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
712 bool sq = amdgpu_vcn_using_unified_queue(ring);
713 uint32_t *ib_checksum;
714 uint32_t ib_pack_in_dw;
720 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
721 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
729 /* single queue headers */
731 ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t)
732 + 4 + 2; /* engine info + decoding ib in dw */
733 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false);
736 ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
737 ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
738 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
739 ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
740 memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
742 decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
743 decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
744 decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
746 for (i = ib->length_dw; i < ib_size_dw; ++i)
750 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw);
752 r = amdgpu_job_submit_direct(job, ring, &f);
756 amdgpu_ib_free(adev, ib_msg, f);
759 *fence = dma_fence_get(f);
765 amdgpu_job_free(job);
767 amdgpu_ib_free(adev, ib_msg, f);
771 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
773 struct dma_fence *fence = NULL;
777 r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
781 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
784 r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
788 r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
792 r = dma_fence_wait_timeout(fence, false, timeout);
798 dma_fence_put(fence);
803 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
805 struct amdgpu_device *adev = ring->adev;
810 if (amdgpu_sriov_vf(adev))
813 r = amdgpu_ring_alloc(ring, 16);
817 rptr = amdgpu_ring_get_rptr(ring);
819 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
820 amdgpu_ring_commit(ring);
822 for (i = 0; i < adev->usec_timeout; i++) {
823 if (amdgpu_ring_get_rptr(ring) != rptr)
828 if (i >= adev->usec_timeout)
834 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
835 struct amdgpu_ib *ib_msg,
836 struct dma_fence **fence)
838 unsigned int ib_size_dw = 16;
839 struct amdgpu_job *job;
840 struct amdgpu_ib *ib;
841 struct dma_fence *f = NULL;
842 uint32_t *ib_checksum = NULL;
844 bool sq = amdgpu_vcn_using_unified_queue(ring);
850 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
851 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
857 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
862 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
864 ib->ptr[ib->length_dw++] = 0x00000018;
865 ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
866 ib->ptr[ib->length_dw++] = handle;
867 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
868 ib->ptr[ib->length_dw++] = addr;
869 ib->ptr[ib->length_dw++] = 0x0000000b;
871 ib->ptr[ib->length_dw++] = 0x00000014;
872 ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
873 ib->ptr[ib->length_dw++] = 0x0000001c;
874 ib->ptr[ib->length_dw++] = 0x00000000;
875 ib->ptr[ib->length_dw++] = 0x00000000;
877 ib->ptr[ib->length_dw++] = 0x00000008;
878 ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
880 for (i = ib->length_dw; i < ib_size_dw; ++i)
884 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
886 r = amdgpu_job_submit_direct(job, ring, &f);
891 *fence = dma_fence_get(f);
897 amdgpu_job_free(job);
901 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
902 struct amdgpu_ib *ib_msg,
903 struct dma_fence **fence)
905 unsigned int ib_size_dw = 16;
906 struct amdgpu_job *job;
907 struct amdgpu_ib *ib;
908 struct dma_fence *f = NULL;
909 uint32_t *ib_checksum = NULL;
911 bool sq = amdgpu_vcn_using_unified_queue(ring);
917 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
918 ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
924 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
929 ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
931 ib->ptr[ib->length_dw++] = 0x00000018;
932 ib->ptr[ib->length_dw++] = 0x00000001;
933 ib->ptr[ib->length_dw++] = handle;
934 ib->ptr[ib->length_dw++] = upper_32_bits(addr);
935 ib->ptr[ib->length_dw++] = addr;
936 ib->ptr[ib->length_dw++] = 0x0000000b;
938 ib->ptr[ib->length_dw++] = 0x00000014;
939 ib->ptr[ib->length_dw++] = 0x00000002;
940 ib->ptr[ib->length_dw++] = 0x0000001c;
941 ib->ptr[ib->length_dw++] = 0x00000000;
942 ib->ptr[ib->length_dw++] = 0x00000000;
944 ib->ptr[ib->length_dw++] = 0x00000008;
945 ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
947 for (i = ib->length_dw; i < ib_size_dw; ++i)
951 amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
953 r = amdgpu_job_submit_direct(job, ring, &f);
958 *fence = dma_fence_get(f);
964 amdgpu_job_free(job);
968 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
970 struct amdgpu_device *adev = ring->adev;
971 struct dma_fence *fence = NULL;
975 memset(&ib, 0, sizeof(ib));
976 r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
977 AMDGPU_IB_POOL_DIRECT,
982 r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
986 r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
990 r = dma_fence_wait_timeout(fence, false, timeout);
997 amdgpu_ib_free(adev, &ib, fence);
998 dma_fence_put(fence);
1003 int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1005 struct amdgpu_device *adev = ring->adev;
1008 if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(4, 0, 3)) {
1009 r = amdgpu_vcn_enc_ring_test_ib(ring, timeout);
1014 r = amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout);
1020 enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
1024 return AMDGPU_RING_PRIO_0;
1026 return AMDGPU_RING_PRIO_1;
1028 return AMDGPU_RING_PRIO_2;
1030 return AMDGPU_RING_PRIO_0;
1034 void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev)
1039 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1040 const struct common_firmware_header *hdr;
1042 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
1044 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1045 if (adev->vcn.harvest_config & (1 << i))
1047 /* currently only support 2 FW instances */
1049 dev_info(adev->dev, "More then 2 VCN FW instances!\n");
1052 idx = AMDGPU_UCODE_ID_VCN + i;
1053 adev->firmware.ucode[idx].ucode_id = idx;
1054 adev->firmware.ucode[idx].fw = adev->vcn.fw;
1055 adev->firmware.fw_size +=
1056 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
1058 if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
1059 IP_VERSION(4, 0, 3))
1062 dev_info(adev->dev, "Will use PSP to load VCN firmware\n");
1067 * debugfs for mapping vcn firmware log buffer.
1069 #if defined(CONFIG_DEBUG_FS)
1070 static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf,
1071 size_t size, loff_t *pos)
1073 struct amdgpu_vcn_inst *vcn;
1075 volatile struct amdgpu_vcn_fwlog *plog;
1076 unsigned int read_pos, write_pos, available, i, read_bytes = 0;
1077 unsigned int read_num[2] = {0};
1079 vcn = file_inode(f)->i_private;
1083 if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log)
1086 log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1088 plog = (volatile struct amdgpu_vcn_fwlog *)log_buf;
1089 read_pos = plog->rptr;
1090 write_pos = plog->wptr;
1092 if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE)
1095 if (!size || (read_pos == write_pos))
1098 if (write_pos > read_pos) {
1099 available = write_pos - read_pos;
1100 read_num[0] = min_t(size_t, size, available);
1102 read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos;
1103 available = read_num[0] + write_pos - plog->header_size;
1104 if (size > available)
1105 read_num[1] = write_pos - plog->header_size;
1106 else if (size > read_num[0])
1107 read_num[1] = size - read_num[0];
1112 for (i = 0; i < 2; i++) {
1114 if (read_pos == AMDGPU_VCNFW_LOG_SIZE)
1115 read_pos = plog->header_size;
1116 if (read_num[i] == copy_to_user((buf + read_bytes),
1117 (log_buf + read_pos), read_num[i]))
1120 read_bytes += read_num[i];
1121 read_pos += read_num[i];
1125 plog->rptr = read_pos;
1130 static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = {
1131 .owner = THIS_MODULE,
1132 .read = amdgpu_debugfs_vcn_fwlog_read,
1133 .llseek = default_llseek
1137 void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i,
1138 struct amdgpu_vcn_inst *vcn)
1140 #if defined(CONFIG_DEBUG_FS)
1141 struct drm_minor *minor = adev_to_drm(adev)->primary;
1142 struct dentry *root = minor->debugfs_root;
1145 sprintf(name, "amdgpu_vcn_%d_fwlog", i);
1146 debugfs_create_file_size(name, S_IFREG | 0444, root, vcn,
1147 &amdgpu_debugfs_vcnfwlog_fops,
1148 AMDGPU_VCNFW_LOG_SIZE);
1152 void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn)
1154 #if defined(CONFIG_DEBUG_FS)
1155 volatile uint32_t *flag = vcn->fw_shared.cpu_addr;
1156 void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1157 uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size;
1158 volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr;
1159 volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr
1160 + vcn->fw_shared.log_offset;
1161 *flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG);
1162 fw_log->is_enabled = 1;
1163 fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF);
1164 fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32);
1165 fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE);
1167 log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog);
1168 log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE;
1169 log_buf->rptr = log_buf->header_size;
1170 log_buf->wptr = log_buf->header_size;
1171 log_buf->wrapped = 0;
1175 int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev,
1176 struct amdgpu_irq_src *source,
1177 struct amdgpu_iv_entry *entry)
1179 struct ras_common_if *ras_if = adev->vcn.ras_if;
1180 struct ras_dispatch_if ih_data = {
1187 if (!amdgpu_sriov_vf(adev)) {
1188 ih_data.head = *ras_if;
1189 amdgpu_ras_interrupt_dispatch(adev, &ih_data);
1191 if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
1192 adev->virt.ops->ras_poison_handler(adev);
1195 "No ras_poison_handler interface in SRIOV for VCN!\n");
1201 int amdgpu_vcn_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
1205 r = amdgpu_ras_block_late_init(adev, ras_block);
1209 if (amdgpu_ras_is_supported(adev, ras_block->block)) {
1210 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1211 if (adev->vcn.harvest_config & (1 << i) ||
1212 !adev->vcn.inst[i].ras_poison_irq.funcs)
1215 r = amdgpu_irq_get(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
1223 amdgpu_ras_block_late_fini(adev, ras_block);
1227 int amdgpu_vcn_ras_sw_init(struct amdgpu_device *adev)
1230 struct amdgpu_vcn_ras *ras;
1235 ras = adev->vcn.ras;
1236 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
1238 dev_err(adev->dev, "Failed to register vcn ras block!\n");
1242 strcpy(ras->ras_block.ras_comm.name, "vcn");
1243 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
1244 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
1245 adev->vcn.ras_if = &ras->ras_block.ras_comm;
1247 if (!ras->ras_block.ras_late_init)
1248 ras->ras_block.ras_late_init = amdgpu_vcn_ras_late_init;
1253 int amdgpu_vcn_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
1254 enum AMDGPU_UCODE_ID ucode_id)
1256 struct amdgpu_firmware_info ucode = {
1257 .ucode_id = (ucode_id ? ucode_id :
1258 (inst_idx ? AMDGPU_UCODE_ID_VCN1_RAM :
1259 AMDGPU_UCODE_ID_VCN0_RAM)),
1260 .mc_addr = adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1261 .ucode_size = ((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1262 (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr),
1265 return psp_execute_ip_fw_load(&adev->psp, &ucode);