[PATCH] drm/amdgpu:fix world switch hang

[Christian König]

Am 06.06.2017 um 11:27 schrieb Monk Liu:
> for SR-IOV, we must keep the pipeline-sync in the protection
> of COND_EXEC, otherwise the command consumed by CPG is not
> consistent when world switch triggerd, e.g.:
>
> world switch hit and the IB frame is skipped so the fence
> won't signal, thus CP will jump to the next DMAframe's pipeline-sync
> command, and it will make CP hang foever.
>
> after pipelin-sync moved into COND_EXEC the consistency can be
> guaranteed
>
> Change-Id: I3059535745e2e1215343299b17f27fe081fc01bc
> Signed-off-by: Monk Liu <Monk.Liu at amd.com>

Reviewed-by: Christian König <christian.koenig at amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 5 +++--
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 +++++--
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 2 +-
>   3 files changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
> index 025e03b..947bd34 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
> @@ -130,6 +130,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
>   
>   	unsigned i;
>   	int r = 0;
> +	bool need_pipe_sync = false;
>   
>   	if (num_ibs == 0)
>   		return -EINVAL;
> @@ -165,7 +166,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
>   	if (ring->funcs->emit_pipeline_sync && job &&
>   	    ((tmp = amdgpu_sync_get_fence(&job->sched_sync)) ||
>   	     amdgpu_vm_need_pipeline_sync(ring, job))) {
> -		amdgpu_ring_emit_pipeline_sync(ring);
> +		need_pipe_sync = true;
>   		fence_put(tmp);
>   	}
>   
> @@ -173,7 +174,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
>   		ring->funcs->insert_start(ring);
>   
>   	if (vm) {
> -		r = amdgpu_vm_flush(ring, job);
> +		r = amdgpu_vm_flush(ring, job, need_pipe_sync);
>   		if (r) {
>   			amdgpu_ring_undo(ring);
>   			return r;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index b065f8b..6d62e34 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -747,7 +747,7 @@ static bool amdgpu_vm_is_large_bar(struct amdgpu_device *adev)
>    *
>    * Emit a VM flush when it is necessary.
>    */
> -int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job)
> +int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_pipe_sync)
>   {
>   	struct amdgpu_device *adev = ring->adev;
>   	unsigned vmhub = ring->funcs->vmhub;
> @@ -770,12 +770,15 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job)
>   		vm_flush_needed = true;
>   	}
>   
> -	if (!vm_flush_needed && !gds_switch_needed)
> +	if (!vm_flush_needed && !gds_switch_needed && !need_pipe_sync)
>   		return 0;
>   
>   	if (ring->funcs->init_cond_exec)
>   		patch_offset = amdgpu_ring_init_cond_exec(ring);
>   
> +	if (need_pipe_sync)
> +		amdgpu_ring_emit_pipeline_sync(ring);
> +
>   	if (ring->funcs->emit_vm_flush && vm_flush_needed) {
>   		struct fence *fence;
>   
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> index 9aa00d9..cc0009f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> @@ -222,7 +222,7 @@ int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
>   int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
>   		      struct amdgpu_sync *sync, struct fence *fence,
>   		      struct amdgpu_job *job);
> -int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job);
> +int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job, bool need_pipe_sync);
>   void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vmhub,
>   			unsigned vmid);
>   void amdgpu_vm_reset_all_ids(struct amdgpu_device *adev);