[PATCH] drm/amdkfd: make sure ring buffer is flushed before update wptr

[Lazar, Lijo]

On 11/15/2024 3:28 PM, Victor Zhao wrote:
> In a consecutive packet submission, for example unmap and query status,
> when CP is reading wptr caused by unmap packet doorbell ring, if in some
> case CP operates slower (e.g. doorbell_mode=1) and wptr has been updated
> to next packet (query status), but the query status packet content has
> not been flushed to memory yet, it will cause CP fetched stalled data.
> 
> Adding mb to ensure ring buffer has been updated before updating wptr.
> Also adding a mb to ensure wptr updated before doorbell ring.
> 
> Signed-off-by: Victor Zhao <Victor.Zhao at amd.com>
> ---
>  drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
> index 4843dcb9a5f7..55d18aed257b 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
> @@ -306,12 +306,17 @@ int kq_submit_packet(struct kernel_queue *kq)
>  	if (amdgpu_amdkfd_is_fed(kq->dev->adev))
>  		return -EIO;
>  
> +	/* Make sure ring buffer is updated before wptr updated */
> +	mb();
> +

Maybe add a specific comment here to indicate this is especially needed
in DOORBELL_MODE=1 when CP fetches value from WPTR memory instead of
doorbell packet.

Reviewed-by: Lijo Lazar <lijo.lazar at amd.com>

Thanks,
Lijo
>  	if (kq->dev->kfd->device_info.doorbell_size == 8) {
>  		*kq->wptr64_kernel = kq->pending_wptr64;
> +		mb(); /* Make sure wptr updated before ring doorbell */
>  		write_kernel_doorbell64(kq->queue->properties.doorbell_ptr,
>  					kq->pending_wptr64);
>  	} else {
>  		*kq->wptr_kernel = kq->pending_wptr;
> +		mb(); /* Make sure wptr updated before ring doorbell */
>  		write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
>  					kq->pending_wptr);
>  	}