On Mon, Jan 11, 2021 at 4:25 AM Christian König
<ckoenig.leichtzumerken@xxxxxxxxx> wrote:
>
> Am 08.01.21 um 16:42 schrieb Nirmoy:
> >
> > On 1/8/21 4:25 PM, Christian König wrote:
> >> Am 18.12.20 um 15:40 schrieb Nirmoy Das:
> >>> Use amdgpu_sa_bo instead of amdgpu_bo.
> >>>
> >>> v2:
> >>> * do not initialize bo to get hint from compiler for -Wuninitialized
> >>> * pass NULL fence to amdgpu_sa_bo_free if fence is undefined.
> >>
> >> Found a major bug in this which is probably the reason why that never
> >> worked before.
> >>
> >> See below.
> >>
> >>>
> >>> Signed-off-by: Nirmoy Das <nirmoy.das@xxxxxxx>
> >>> ---
> >>> drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c | 56
> >>> +++++++------------------
> >>> drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 17 ++++----
> >>> drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 47 ++++++++++-----------
> >>> 3 files changed, 45 insertions(+), 75 deletions(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
> >>> index 8b989670ed66..13450a3df044 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
> >>> @@ -1057,7 +1057,7 @@ int amdgpu_uvd_ring_parse_cs(struct
> >>> amdgpu_cs_parser *parser, uint32_t ib_idx)
> >>> return 0;
> >>> }
> >>>
> >>> -static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct
> >>> amdgpu_bo *bo,
> >>> +static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct
> >>> amdgpu_sa_bo *bo,
> >>> bool direct, struct dma_fence **fence)
> >>> {
> >>> struct amdgpu_device *adev = ring->adev;
> >>> @@ -1071,19 +1071,6 @@ static int amdgpu_uvd_send_msg(struct
> >>> amdgpu_ring *ring, struct amdgpu_bo *bo,
> >>> unsigned offset_idx = 0;
> >>> unsigned offset[3] = { UVD_BASE_SI, 0, 0 };
> >>>
> >>> - amdgpu_bo_kunmap(bo);
> >>> - amdgpu_bo_unpin(bo);
> >>> -
> >>> - if (!ring->adev->uvd.address_64_bit) {
> >>> - struct ttm_operation_ctx ctx = { true, false };
> >>> -
> >>> - amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_VRAM);
> >>> - amdgpu_uvd_force_into_uvd_segment(bo);
> >>> - r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
> >>> - if (r)
> >>> - goto err;
> >>> - }
> >>> -
> >>> r = amdgpu_job_alloc_with_ib(adev, 64, direct ?
> >>> AMDGPU_IB_POOL_DIRECT :
> >>> AMDGPU_IB_POOL_DELAYED, &job);
> >>> if (r)
> >>> @@ -1101,7 +1088,7 @@ static int amdgpu_uvd_send_msg(struct
> >>> amdgpu_ring *ring, struct amdgpu_bo *bo,
> >>> data[3] = PACKET0(offset[offset_idx] + UVD_NO_OP, 0);
> >>>
> >>> ib = &job->ibs[0];
> >>> - addr = amdgpu_bo_gpu_offset(bo);
> >>> + addr = amdgpu_sa_bo_gpu_addr(bo);
> >>> ib->ptr[0] = data[0];
> >>> ib->ptr[1] = addr;
> >>> ib->ptr[2] = data[1];
> >>> @@ -1115,33 +1102,17 @@ static int amdgpu_uvd_send_msg(struct
> >>> amdgpu_ring *ring, struct amdgpu_bo *bo,
> >>> ib->length_dw = 16;
> >>>
> >>> if (direct) {
> >>> - r = dma_resv_wait_timeout_rcu(bo->tbo.base.resv,
> >>> - true, false,
> >>> - msecs_to_jiffies(10));
> >>> - if (r == 0)
> >>> - r = -ETIMEDOUT;
> >>> - if (r < 0)
> >>> - goto err_free;
> >>> -
> >>> r = amdgpu_job_submit_direct(job, ring, &f);
> >>> if (r)
> >>> goto err_free;
> >>> } else {
> >>> - r = amdgpu_sync_resv(adev, &job->sync, bo->tbo.base.resv,
> >>> - AMDGPU_SYNC_ALWAYS,
> >>> - AMDGPU_FENCE_OWNER_UNDEFINED);
> >>> - if (r)
> >>> - goto err_free;
> >>> -
> >>> r = amdgpu_job_submit(job, &adev->uvd.entity,
> >>> AMDGPU_FENCE_OWNER_UNDEFINED, &f);
> >>> if (r)
> >>> goto err_free;
> >>> }
> >>>
> >>> - amdgpu_bo_fence(bo, f, false);
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> + amdgpu_sa_bo_free(adev, &bo, f);
> >>>
> >>> if (fence)
> >>> *fence = dma_fence_get(f);
> >>> @@ -1153,8 +1124,7 @@ static int amdgpu_uvd_send_msg(struct
> >>> amdgpu_ring *ring, struct amdgpu_bo *bo,
> >>> amdgpu_job_free(job);
> >>>
> >>> err:
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> + amdgpu_sa_bo_free(adev, &bo, NULL);
> >>> return r;
> >>> }
> >>>
> >>> @@ -1165,16 +1135,17 @@ int amdgpu_uvd_get_create_msg(struct
> >>> amdgpu_ring *ring, uint32_t handle,
> >>> struct dma_fence **fence)
> >>> {
> >>> struct amdgpu_device *adev = ring->adev;
> >>> - struct amdgpu_bo *bo = NULL;
> >>> + struct amdgpu_sa_bo *bo;
> >>> uint32_t *msg;
> >>> int r, i;
> >>>
> >>> - r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
> >>> - AMDGPU_GEM_DOMAIN_VRAM,
> >>> - &bo, NULL, (void **)&msg);
> >>> + r = amdgpu_sa_bo_new(&adev->ib_pools[AMDGPU_IB_POOL_DIRECT],
> >>> + &bo, 1024, PAGE_SIZE);
> >>
> >> This 1024 here is wrong. The size is in bytes and not double words.
> >> So this needs to be 4096.
> >>
> >> This only worked because amdgpu_bo_create_reserved always allocated a
> >> full page instead of 1024 bytes.
> >>
> >> Same problem on all other occasions.
> >
> >
> > I wonder why it worked on newer cards.
>
> Maybe just coincident, but there must be more problems. Even with the
> size fixed I get a hang of the IB tests during boot.
>
Maybe related to the recent change to move some of the UVD buffers
from vram to gtt?
Alex
> Christian.
>
> >
> >
> >>
> >> Additional to that the direct pool should only be used if we submit
> >> directly, e.g. during a GPU reset.
> >>
> >> Otherwise we need to use the normal pool.
> >
> >
> > I will resend with above changes.
> >
> >
> > Thanks,
> >
> > Nirmoy
> >
> >
> >>
> >> Regards,
> >> Christian.
> >>
> >>
> >>> +
> >>> if (r)
> >>> return r;
> >>>
> >>> + msg = amdgpu_sa_bo_cpu_addr(bo);
> >>> /* stitch together an UVD create msg */
> >>> msg[0] = cpu_to_le32(0x00000de4);
> >>> msg[1] = cpu_to_le32(0x00000000);
> >>> @@ -1197,16 +1168,17 @@ int amdgpu_uvd_get_destroy_msg(struct
> >>> amdgpu_ring *ring, uint32_t handle,
> >>> bool direct, struct dma_fence **fence)
> >>> {
> >>> struct amdgpu_device *adev = ring->adev;
> >>> - struct amdgpu_bo *bo = NULL;
> >>> + struct amdgpu_sa_bo *bo;
> >>> uint32_t *msg;
> >>> int r, i;
> >>>
> >>> - r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
> >>> - AMDGPU_GEM_DOMAIN_VRAM,
> >>> - &bo, NULL, (void **)&msg);
> >>> + r = amdgpu_sa_bo_new(&adev->ib_pools[AMDGPU_IB_POOL_DIRECT],
> >>> + &bo, 1024, PAGE_SIZE);
> >>> +
> >>> if (r)
> >>> return r;
> >>>
> >>> + msg = amdgpu_sa_bo_cpu_addr(bo);
> >>> /* stitch together an UVD destroy msg */
> >>> msg[0] = cpu_to_le32(0x00000de4);
> >>> msg[1] = cpu_to_le32(0x00000002);
> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
> >>> index 0d5284b936e4..bce29d6975d3 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
> >>> @@ -81,7 +81,7 @@ MODULE_FIRMWARE(FIRMWARE_VEGA20);
> >>>
> >>> static void amdgpu_vce_idle_work_handler(struct work_struct *work);
> >>> static int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring,
> >>> uint32_t handle,
> >>> - struct amdgpu_bo *bo,
> >>> + struct amdgpu_sa_bo *bo,
> >>> struct dma_fence **fence);
> >>> static int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring,
> >>> uint32_t handle,
> >>> bool direct, struct dma_fence **fence);
> >>> @@ -437,7 +437,7 @@ void amdgpu_vce_free_handles(struct
> >>> amdgpu_device *adev, struct drm_file *filp)
> >>> * Open up a stream for HW test
> >>> */
> >>> static int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring,
> >>> uint32_t handle,
> >>> - struct amdgpu_bo *bo,
> >>> + struct amdgpu_sa_bo *bo,
> >>> struct dma_fence **fence)
> >>> {
> >>> const unsigned ib_size_dw = 1024;
> >>> @@ -454,7 +454,7 @@ static int amdgpu_vce_get_create_msg(struct
> >>> amdgpu_ring *ring, uint32_t handle,
> >>>
> >>> ib = &job->ibs[0];
> >>>
> >>> - addr = amdgpu_bo_gpu_offset(bo);
> >>> + addr = amdgpu_sa_bo_gpu_addr(bo);
> >>>
> >>> /* stitch together an VCE create msg */
> >>> ib->length_dw = 0;
> >>> @@ -1130,16 +1130,16 @@ int amdgpu_vce_ring_test_ring(struct
> >>> amdgpu_ring *ring)
> >>> int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout)
> >>> {
> >>> struct dma_fence *fence = NULL;
> >>> - struct amdgpu_bo *bo = NULL;
> >>> + struct amdgpu_sa_bo *bo = NULL;
> >>> + struct amdgpu_device *adev = ring->adev;
> >>> long r;
> >>>
> >>> /* skip vce ring1/2 ib test for now, since it's not reliable */
> >>> if (ring != &ring->adev->vce.ring[0])
> >>> return 0;
> >>>
> >>> - r = amdgpu_bo_create_reserved(ring->adev, 512, PAGE_SIZE,
> >>> - AMDGPU_GEM_DOMAIN_VRAM,
> >>> - &bo, NULL, NULL);
> >>> + r = amdgpu_sa_bo_new(&adev->ib_pools[AMDGPU_IB_POOL_DIRECT],
> >>> + &bo, 512, PAGE_SIZE);
> >>> if (r)
> >>> return r;
> >>>
> >>> @@ -1158,8 +1158,7 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring
> >>> *ring, long timeout)
> >>> r = 0;
> >>>
> >>> error:
> >>> + amdgpu_sa_bo_free(adev, &bo, fence);
> >>> dma_fence_put(fence);
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> return r;
> >>> }
> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> >>> index 4a77c7424dfc..1e46b2f895ba 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> >>> @@ -488,7 +488,7 @@ int amdgpu_vcn_dec_sw_ring_test_ring(struct
> >>> amdgpu_ring *ring)
> >>> }
> >>>
> >>> static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
> >>> - struct amdgpu_bo *bo,
> >>> + struct amdgpu_sa_bo *bo,
> >>> struct dma_fence **fence)
> >>> {
> >>> struct amdgpu_device *adev = ring->adev;
> >>> @@ -504,7 +504,8 @@ static int amdgpu_vcn_dec_send_msg(struct
> >>> amdgpu_ring *ring,
> >>> goto err;
> >>>
> >>> ib = &job->ibs[0];
> >>> - addr = amdgpu_bo_gpu_offset(bo);
> >>> + addr = amdgpu_sa_bo_gpu_addr(bo);
> >>> +
> >>> ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
> >>> ib->ptr[1] = addr;
> >>> ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
> >>> @@ -521,9 +522,7 @@ static int amdgpu_vcn_dec_send_msg(struct
> >>> amdgpu_ring *ring,
> >>> if (r)
> >>> goto err_free;
> >>>
> >>> - amdgpu_bo_fence(bo, f, false);
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> + amdgpu_sa_bo_free(adev, &bo, f);
> >>>
> >>> if (fence)
> >>> *fence = dma_fence_get(f);
> >>> @@ -535,25 +534,27 @@ static int amdgpu_vcn_dec_send_msg(struct
> >>> amdgpu_ring *ring,
> >>> amdgpu_job_free(job);
> >>>
> >>> err:
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> + amdgpu_sa_bo_free(adev, &bo, NULL);
> >>> return r;
> >>> }
> >>>
> >>> static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring,
> >>> uint32_t handle,
> >>> - struct amdgpu_bo **bo)
> >>> + struct amdgpu_sa_bo **bo)
> >>> {
> >>> struct amdgpu_device *adev = ring->adev;
> >>> uint32_t *msg;
> >>> int r, i;
> >>>
> >>> *bo = NULL;
> >>> - r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
> >>> - AMDGPU_GEM_DOMAIN_VRAM,
> >>> - bo, NULL, (void **)&msg);
> >>> +
> >>> + r = amdgpu_sa_bo_new(&adev->ib_pools[AMDGPU_IB_POOL_DIRECT],
> >>> + bo, 1024, PAGE_SIZE);
> >>> +
> >>> if (r)
> >>> return r;
> >>>
> >>> + msg = amdgpu_sa_bo_cpu_addr(*bo);
> >>> +
> >>> msg[0] = cpu_to_le32(0x00000028);
> >>> msg[1] = cpu_to_le32(0x00000038);
> >>> msg[2] = cpu_to_le32(0x00000001);
> >>> @@ -575,18 +576,19 @@ static int
> >>> amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t hand
> >>> }
> >>>
> >>> static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring
> >>> *ring, uint32_t handle,
> >>> - struct amdgpu_bo **bo)
> >>> + struct amdgpu_sa_bo **bo)
> >>> {
> >>> struct amdgpu_device *adev = ring->adev;
> >>> uint32_t *msg;
> >>> int r, i;
> >>>
> >>> *bo = NULL;
> >>> - r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
> >>> - AMDGPU_GEM_DOMAIN_VRAM,
> >>> - bo, NULL, (void **)&msg);
> >>> + r = amdgpu_sa_bo_new(&adev->ib_pools[AMDGPU_IB_POOL_DIRECT],
> >>> + bo, 1024, PAGE_SIZE);
> >>> +
> >>> if (r)
> >>> return r;
> >>> + msg = amdgpu_sa_bo_cpu_addr(*bo);
> >>>
> >>> msg[0] = cpu_to_le32(0x00000028);
> >>> msg[1] = cpu_to_le32(0x00000018);
> >>> @@ -603,7 +605,7 @@ static int amdgpu_vcn_dec_get_destroy_msg(struct
> >>> amdgpu_ring *ring, uint32_t han
> >>> int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long
> >>> timeout)
> >>> {
> >>> struct dma_fence *fence = NULL;
> >>> - struct amdgpu_bo *bo;
> >>> + struct amdgpu_sa_bo *bo;
> >>> long r;
> >>>
> >>> r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
> >>> @@ -633,7 +635,7 @@ int amdgpu_vcn_dec_ring_test_ib(struct
> >>> amdgpu_ring *ring, long timeout)
> >>> }
> >>>
> >>> static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
> >>> - struct amdgpu_bo *bo,
> >>> + struct amdgpu_sa_bo *bo,
> >>> struct dma_fence **fence)
> >>> {
> >>> struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
> >>> @@ -651,7 +653,7 @@ static int amdgpu_vcn_dec_sw_send_msg(struct
> >>> amdgpu_ring *ring,
> >>> goto err;
> >>>
> >>> ib = &job->ibs[0];
> >>> - addr = amdgpu_bo_gpu_offset(bo);
> >>> + addr = amdgpu_sa_bo_gpu_addr(bo);
> >>> ib->length_dw = 0;
> >>>
> >>> ib->ptr[ib->length_dw++] = sizeof(struct
> >>> amdgpu_vcn_decode_buffer) + 8;
> >>> @@ -671,9 +673,7 @@ static int amdgpu_vcn_dec_sw_send_msg(struct
> >>> amdgpu_ring *ring,
> >>> if (r)
> >>> goto err_free;
> >>>
> >>> - amdgpu_bo_fence(bo, f, false);
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> + amdgpu_sa_bo_free(adev, &bo, f);
> >>>
> >>> if (fence)
> >>> *fence = dma_fence_get(f);
> >>> @@ -685,15 +685,14 @@ static int amdgpu_vcn_dec_sw_send_msg(struct
> >>> amdgpu_ring *ring,
> >>> amdgpu_job_free(job);
> >>>
> >>> err:
> >>> - amdgpu_bo_unreserve(bo);
> >>> - amdgpu_bo_unref(&bo);
> >>> + amdgpu_sa_bo_free(adev, &bo, NULL);
> >>> return r;
> >>> }
> >>>
> >>> int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long
> >>> timeout)
> >>> {
> >>> struct dma_fence *fence = NULL;
> >>> - struct amdgpu_bo *bo;
> >>> + struct amdgpu_sa_bo *bo;
> >>> long r;
> >>>
> >>> r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
> >>> --
> >>> 2.29.2
> >>>
> >>
> > _______________________________________________
> > amd-gfx mailing list
> > amd-gfx@xxxxxxxxxxxxxxxxxxxxx
> > https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
