Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit d71518b5 authored by Christian König's avatar Christian König Committed by Alex Deucher
Browse files

drm/amdgpu: cleanup in kernel job submission 



Add a job_alloc_with_ib helper and proper job submission.

Signed-off-by: default avatarChristian König <christian.koenig@amd.com>
Reviewed-by: default avatarAlex Deucher <alexander.deucer@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent a0332b56

drivers/gpu/drm/amd/amdgpu/amdgpu.h

+4 −9
Original line number Diff line number Diff line
@@ -797,14 +797,11 @@ extern struct amd_sched_backend_ops amdgpu_sched_ops; 


int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,

		     struct amdgpu_job **job);

int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,

			     struct amdgpu_job **job);

void amdgpu_job_free(struct amdgpu_job *job);

int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,

					 struct amdgpu_ring *ring,

					 struct amdgpu_ib *ibs,

					 unsigned num_ibs,

					 int (*free_job)(struct amdgpu_job *),

					 void *owner,

					 struct fence **fence);

int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,

		      void *owner, struct fence **f);



struct amdgpu_ring {

	struct amdgpu_device		*adev;
@@ -987,7 +984,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, 
		       uint64_t addr);

void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,

		      struct amdgpu_bo_va *bo_va);

int amdgpu_vm_free_job(struct amdgpu_job *job);



/*

 * context related structures
@@ -1244,7 +1240,6 @@ struct amdgpu_job { 
	uint32_t		num_ibs;

	void			*owner;

	struct amdgpu_user_fence uf;

	int (*free_job)(struct amdgpu_job *job);

};

#define to_amdgpu_job(sched_job)		\

		container_of((sched_job), struct amdgpu_job, base)

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

+1 −9
Original line number Diff line number Diff line
@@ -752,12 +752,6 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, 
	return 0;

}



static int amdgpu_cs_free_job(struct amdgpu_job *job)

{

	amdgpu_job_free(job);

	return 0;

}



static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,

			    union drm_amdgpu_cs *cs)

{
@@ -771,12 +765,10 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, 
	job->base.sched = &ring->sched;

	job->base.s_entity = &p->ctx->rings[ring->idx].entity;

	job->owner = p->filp;

	job->free_job = amdgpu_cs_free_job;



	fence = amd_sched_fence_create(job->base.s_entity, p->filp);

	if (!fence) {

		amdgpu_cs_free_job(job);

		kfree(job);

		amdgpu_job_free(job);

		return -ENOMEM;

	}

drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c

+38 −38
Original line number Diff line number Diff line
@@ -45,11 +45,26 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs, 
	(*job)->adev = adev;

	(*job)->ibs = (void *)&(*job)[1];

	(*job)->num_ibs = num_ibs;

	(*job)->free_job = NULL;



	return 0;

}



int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,

			     struct amdgpu_job **job)

{

	int r;



	r = amdgpu_job_alloc(adev, 1, job);

	if (r)

		return r;



	r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);

	if (r)

		kfree(*job);



	return r;

}



void amdgpu_job_free(struct amdgpu_job *job)

{

	unsigned i;
@@ -58,7 +73,27 @@ void amdgpu_job_free(struct amdgpu_job *job) 
		amdgpu_ib_free(job->adev, &job->ibs[i]);



	amdgpu_bo_unref(&job->uf.bo);

	/* TODO: Free the job structure here as well */

	kfree(job);

}



int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,

		      void *owner, struct fence **f)

{

	struct amdgpu_device *adev = job->adev;



	job->ring = ring;

	job->base.sched = &ring->sched;

	job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;

	job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);

	if (!job->base.s_fence)

		return -ENOMEM;



	*f = fence_get(&job->base.s_fence->base);



	job->owner = owner;

	amd_sched_entity_push_job(&job->base);



	return 0;

}



static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
@@ -106,10 +141,7 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job) 
	}



err:

	if (job->free_job)

		job->free_job(job);



	kfree(job);

	amdgpu_job_free(job);

	return fence;

}
@@ -117,35 +149,3 @@ struct amd_sched_backend_ops amdgpu_sched_ops = { 
	.dependency = amdgpu_sched_dependency,

	.run_job = amdgpu_sched_run_job,

};
 


int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,

					 struct amdgpu_ring *ring,

					 struct amdgpu_ib *ibs,

					 unsigned num_ibs,

					 int (*free_job)(struct amdgpu_job *),

					 void *owner,

					 struct fence **f)

{

	struct amdgpu_job *job =

		kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);

	if (!job)

		return -ENOMEM;

	job->base.sched = &ring->sched;

	job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;

	job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);

	if (!job->base.s_fence) {

		kfree(job);

		return -ENOMEM;

	}

	*f = fence_get(&job->base.s_fence->base);



	job->adev = adev;

	job->ring = ring;

	job->ibs = ibs;

	job->num_ibs = num_ibs;

	job->owner = owner;

	job->free_job = free_job;

	amd_sched_entity_push_job(&job->base);



	return 0;

}

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c

+12 −22
Original line number Diff line number Diff line
@@ -1012,9 +1012,10 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, 
		       struct fence **fence)

{

	struct amdgpu_device *adev = ring->adev;

	struct amdgpu_job *job;



	uint32_t max_bytes;

	unsigned num_loops, num_dw;

	struct amdgpu_ib *ib;

	unsigned i;

	int r;
@@ -1026,20 +1027,12 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, 
	while (num_dw & 0x7)

		num_dw++;



	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);

	if (!ib)

		return -ENOMEM;



	r = amdgpu_ib_get(adev, NULL, num_dw * 4, ib);

	if (r) {

		kfree(ib);

	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, &job);

	if (r)

		return r;

	}



	ib->length_dw = 0;



	if (resv) {

		r = amdgpu_sync_resv(adev, &ib->sync, resv,

		r = amdgpu_sync_resv(adev, &job->ibs[0].sync, resv,

				     AMDGPU_FENCE_OWNER_UNDEFINED);

		if (r) {

			DRM_ERROR("sync failed (%d).\n", r);
@@ -1050,27 +1043,24 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, 
	for (i = 0; i < num_loops; i++) {

		uint32_t cur_size_in_bytes = min(byte_count, max_bytes);



		amdgpu_emit_copy_buffer(adev, ib, src_offset, dst_offset,

					cur_size_in_bytes);

		amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_offset,

					dst_offset, cur_size_in_bytes);



		src_offset += cur_size_in_bytes;

		dst_offset += cur_size_in_bytes;

		byte_count -= cur_size_in_bytes;

	}



	amdgpu_ring_pad_ib(ring, ib);

	WARN_ON(ib->length_dw > num_dw);

	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,

						 &amdgpu_vm_free_job,

						 AMDGPU_FENCE_OWNER_UNDEFINED,

						 fence);

	amdgpu_ring_pad_ib(ring, &job->ibs[0]);

	WARN_ON(job->ibs[0].length_dw > num_dw);

	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_UNDEFINED, fence);

	if (r)

		goto error_free;



	return 0;



error_free:

	amdgpu_ib_free(adev, ib);

	kfree(ib);

	amdgpu_job_free(job);

	return r;

}

drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c

+12 −25
Original line number Diff line number Diff line
@@ -823,14 +823,6 @@ 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)
@@ -838,7 +830,8 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, 
	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;
@@ -862,15 +855,12 @@ 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(adev, 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;
@@ -882,12 +872,9 @@ 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);

	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_UNDEFINED, &f);

	if (r)

		goto err2;

		goto err_free;



	ttm_eu_fence_buffer_objects(&ticket, &head, f);
@@ -897,10 +884,10 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, 
	fence_put(f);



	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;