[PATCH v4] drm/amd/amdgpu:Fix compute ring unable to detect hang.

Christian König ckoenig.leichtzumerken at gmail.com
Thu Sep 19 08:14:24 UTC 2019


Am 19.09.19 um 10:00 schrieb Jesse Zhang:
> When compute fence did not signal, compute ring cannot detect hardware
> hang because its timeout value is set to be infinite by default.
>
> In SR-IOV and passthrough mode, if user does not declare custome timeout
> value for compute ring, then use gfx ring timeout value as default. So
> that when there is a ture hardware hang, compute ring can detect it.
>
> Change-Id: I794ec0868c6c0aad407749457260ecfee0617c10
> Signed-off-by: Jesse Zhang <zhexi.zhang at amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c |  5 +----
>   drivers/gpu/drm/amd/amdgpu/soc15.c        | 10 ++++++++++
>   2 files changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> index cbcaa7c..963b6d1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> @@ -468,10 +468,7 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
>   			 * For sriov case, always use the timeout
>   			 * as gfx ring
>   			 */

Please also remove the comment since that is now stale.

Apart from that looks good to me,
Christian.

> -			if (!amdgpu_sriov_vf(ring->adev))
> -				timeout = adev->compute_timeout;
> -			else
> -				timeout = adev->gfx_timeout;
> +			timeout = adev->compute_timeout;
>   			break;
>   		case AMDGPU_RING_TYPE_SDMA:
>   			timeout = adev->sdma_timeout;
> diff --git a/drivers/gpu/drm/amd/amdgpu/soc15.c b/drivers/gpu/drm/amd/amdgpu/soc15.c
> index 7c7e9f5..6cd5548 100644
> --- a/drivers/gpu/drm/amd/amdgpu/soc15.c
> +++ b/drivers/gpu/drm/amd/amdgpu/soc15.c
> @@ -687,6 +687,16 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev)
>   	adev->rev_id = soc15_get_rev_id(adev);
>   	adev->nbio.funcs->detect_hw_virt(adev);
>   
> +	/*
> +	 * If running under SR-IOV or passthrough mode and user did not set
> +	 * custom value for compute ring timeout, set timeout to be the same
> +	 * as gfx ring timeout to avoid compute ring cannot detect an true
> +	 * hang.
> +	 */
> +	if ((amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev)) &&
> +		(adev->compute_timeout == MAX_SCHEDULE_TIMEOUT))
> +		adev->compute_timeout = adev->gfx_timeout;
> +
>   	if (amdgpu_sriov_vf(adev))
>   		adev->virt.ops = &xgpu_ai_virt_ops;
>   



More information about the amd-gfx mailing list