summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c')
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c163
1 files changed, 82 insertions, 81 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index b5011431d..cdb963cc2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -84,6 +84,8 @@ static void amdgpu_uvd_idle_work_handler(struct work_struct *work);
int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
{
+ struct amdgpu_ring *ring;
+ struct amd_sched_rq *rq;
unsigned long bo_size;
const char *fw_name;
const struct common_firmware_header *hdr;
@@ -187,6 +189,15 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
amdgpu_bo_unreserve(adev->uvd.vcpu_bo);
+ ring = &adev->uvd.ring;
+ rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL];
+ r = amd_sched_entity_init(&ring->sched, &adev->uvd.entity,
+ rq, amdgpu_sched_jobs);
+ if (r != 0) {
+ DRM_ERROR("Failed setting up UVD run queue.\n");
+ return r;
+ }
+
for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) {
atomic_set(&adev->uvd.handles[i], 0);
adev->uvd.filp[i] = NULL;
@@ -206,6 +217,8 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
if (adev->uvd.vcpu_bo == NULL)
return 0;
+ amd_sched_entity_fini(&adev->uvd.ring.sched, &adev->uvd.entity);
+
r = amdgpu_bo_reserve(adev->uvd.vcpu_bo, false);
if (!r) {
amdgpu_bo_kunmap(adev->uvd.vcpu_bo);
@@ -224,32 +237,30 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
int amdgpu_uvd_suspend(struct amdgpu_device *adev)
{
- struct amdgpu_ring *ring = &adev->uvd.ring;
- int i, r;
+ unsigned size;
+ void *ptr;
+ int i;
if (adev->uvd.vcpu_bo == NULL)
return 0;
- for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) {
- uint32_t handle = atomic_read(&adev->uvd.handles[i]);
- if (handle != 0) {
- struct fence *fence;
+ for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i)
+ if (atomic_read(&adev->uvd.handles[i]))
+ break;
- amdgpu_uvd_note_usage(adev);
+ if (i == AMDGPU_MAX_UVD_HANDLES)
+ return 0;
- r = amdgpu_uvd_get_destroy_msg(ring, handle, &fence);
- if (r) {
- DRM_ERROR("Error destroying UVD (%d)!\n", r);
- continue;
- }
+ cancel_delayed_work_sync(&adev->uvd.idle_work);
- fence_wait(fence, false);
- fence_put(fence);
+ size = amdgpu_bo_size(adev->uvd.vcpu_bo);
+ ptr = adev->uvd.cpu_addr;
- adev->uvd.filp[i] = NULL;
- atomic_set(&adev->uvd.handles[i], 0);
- }
- }
+ adev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
+ if (!adev->uvd.saved_bo)
+ return -ENOMEM;
+
+ memcpy(adev->uvd.saved_bo, ptr, size);
return 0;
}
@@ -258,25 +269,29 @@ int amdgpu_uvd_resume(struct amdgpu_device *adev)
{
unsigned size;
void *ptr;
- const struct common_firmware_header *hdr;
- unsigned offset;
if (adev->uvd.vcpu_bo == NULL)
return -EINVAL;
- hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
- offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
- memcpy(adev->uvd.cpu_addr, (adev->uvd.fw->data) + offset,
- (adev->uvd.fw->size) - offset);
-
- cancel_delayed_work_sync(&adev->uvd.idle_work);
-
size = amdgpu_bo_size(adev->uvd.vcpu_bo);
- size -= le32_to_cpu(hdr->ucode_size_bytes);
ptr = adev->uvd.cpu_addr;
- ptr += le32_to_cpu(hdr->ucode_size_bytes);
- memset(ptr, 0, size);
+ if (adev->uvd.saved_bo != NULL) {
+ memcpy(ptr, adev->uvd.saved_bo, size);
+ kfree(adev->uvd.saved_bo);
+ adev->uvd.saved_bo = NULL;
+ } else {
+ const struct common_firmware_header *hdr;
+ unsigned offset;
+
+ hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
+ offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
+ memcpy(adev->uvd.cpu_addr, (adev->uvd.fw->data) + offset,
+ (adev->uvd.fw->size) - offset);
+ size -= le32_to_cpu(hdr->ucode_size_bytes);
+ ptr += le32_to_cpu(hdr->ucode_size_bytes);
+ memset(ptr, 0, size);
+ }
return 0;
}
@@ -293,7 +308,8 @@ void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
amdgpu_uvd_note_usage(adev);
- r = amdgpu_uvd_get_destroy_msg(ring, handle, &fence);
+ r = amdgpu_uvd_get_destroy_msg(ring, handle,
+ false, &fence);
if (r) {
DRM_ERROR("Error destroying UVD (%d)!\n", r);
continue;
@@ -523,13 +539,6 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx,
return -EINVAL;
}
- r = reservation_object_wait_timeout_rcu(bo->tbo.resv, true, false,
- MAX_SCHEDULE_TIMEOUT);
- if (r < 0) {
- DRM_ERROR("Failed waiting for UVD message (%ld)!\n", r);
- return r;
- }
-
r = amdgpu_bo_kmap(bo, &ptr);
if (r) {
DRM_ERROR("Failed mapping the UVD message (%ld)!\n", r);
@@ -614,7 +623,6 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx)
{
struct amdgpu_bo_va_mapping *mapping;
struct amdgpu_bo *bo;
- struct amdgpu_ib *ib;
uint32_t cmd, lo, hi;
uint64_t start, end;
uint64_t addr;
@@ -636,9 +644,10 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx)
addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE;
start += addr;
- ib = &ctx->parser->ibs[ctx->ib_idx];
- ib->ptr[ctx->data0] = start & 0xFFFFFFFF;
- ib->ptr[ctx->data1] = start >> 32;
+ amdgpu_set_ib_value(ctx->parser, ctx->ib_idx, ctx->data0,
+ lower_32_bits(start));
+ amdgpu_set_ib_value(ctx->parser, ctx->ib_idx, ctx->data1,
+ upper_32_bits(start));
cmd = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->idx) >> 1;
if (cmd < 0x4) {
@@ -700,7 +709,7 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx)
static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx,
int (*cb)(struct amdgpu_uvd_cs_ctx *ctx))
{
- struct amdgpu_ib *ib = &ctx->parser->ibs[ctx->ib_idx];
+ struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx];
int i, r;
ctx->idx++;
@@ -746,7 +755,7 @@ static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx,
static int amdgpu_uvd_cs_packets(struct amdgpu_uvd_cs_ctx *ctx,
int (*cb)(struct amdgpu_uvd_cs_ctx *ctx))
{
- struct amdgpu_ib *ib = &ctx->parser->ibs[ctx->ib_idx];
+ struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx];
int r;
for (ctx->idx = 0 ; ctx->idx < ib->length_dw; ) {
@@ -788,7 +797,7 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
[0x00000003] = 2048,
[0x00000004] = 0xFFFFFFFF,
};
- struct amdgpu_ib *ib = &parser->ibs[ib_idx];
+ struct amdgpu_ib *ib = &parser->job->ibs[ib_idx];
int r;
if (ib->length_dw % 16) {
@@ -821,22 +830,14 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
return 0;
}
-static int amdgpu_uvd_free_job(
- struct amdgpu_job *job)
-{
- amdgpu_ib_free(job->adev, job->ibs);
- kfree(job->ibs);
- return 0;
-}
-
-static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
- struct amdgpu_bo *bo,
- struct fence **fence)
+static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
+ bool direct, struct fence **fence)
{
struct ttm_validate_buffer tv;
struct ww_acquire_ctx ticket;
struct list_head head;
- struct amdgpu_ib *ib = NULL;
+ struct amdgpu_job *job;
+ struct amdgpu_ib *ib;
struct fence *f = NULL;
struct amdgpu_device *adev = ring->adev;
uint64_t addr;
@@ -860,15 +861,12 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
if (r)
goto err;
- ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
- if (!ib) {
- r = -ENOMEM;
- goto err;
- }
- r = amdgpu_ib_get(ring, NULL, 64, ib);
+
+ r = amdgpu_job_alloc_with_ib(adev, 64, &job);
if (r)
- goto err1;
+ goto err;
+ ib = &job->ibs[0];
addr = amdgpu_bo_gpu_offset(bo);
ib->ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
ib->ptr[1] = addr;
@@ -880,12 +878,19 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
ib->ptr[i] = PACKET2(0);
ib->length_dw = 16;
- r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
- &amdgpu_uvd_free_job,
- AMDGPU_FENCE_OWNER_UNDEFINED,
- &f);
- if (r)
- goto err2;
+ if (direct) {
+ r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f);
+ job->fence = f;
+ if (r)
+ goto err_free;
+
+ amdgpu_job_free(job);
+ } else {
+ r = amdgpu_job_submit(job, ring, &adev->uvd.entity,
+ AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+ if (r)
+ goto err_free;
+ }
ttm_eu_fence_buffer_objects(&ticket, &head, f);
@@ -893,16 +898,12 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
*fence = fence_get(f);
amdgpu_bo_unref(&bo);
fence_put(f);
- if (amdgpu_enable_scheduler)
- return 0;
- amdgpu_ib_free(ring->adev, ib);
- kfree(ib);
return 0;
-err2:
- amdgpu_ib_free(ring->adev, ib);
-err1:
- kfree(ib);
+
+err_free:
+ amdgpu_job_free(job);
+
err:
ttm_eu_backoff_reservation(&ticket, &head);
return r;
@@ -957,11 +958,11 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
amdgpu_bo_kunmap(bo);
amdgpu_bo_unreserve(bo);
- return amdgpu_uvd_send_msg(ring, bo, fence);
+ return amdgpu_uvd_send_msg(ring, bo, true, fence);
}
int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
- struct fence **fence)
+ bool direct, struct fence **fence)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_bo *bo;
@@ -999,7 +1000,7 @@ int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
amdgpu_bo_kunmap(bo);
amdgpu_bo_unreserve(bo);
- return amdgpu_uvd_send_msg(ring, bo, fence);
+ return amdgpu_uvd_send_msg(ring, bo, direct, fence);
}
static void amdgpu_uvd_idle_work_handler(struct work_struct *work)
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-01 10:52:58 +0000