[PATCH 01/10] drm/amdgpu: send IVs to the KFD only after processing them v3

[Kuehling, Felix]

Patches 1-3 are Reviewed-by: Felix Kuehling <Felix.Kuehling at amd.com>

I applied all 10 patches and tested them with kfdtest on Fiji and
Vega10. It seems to not break anything obvious.

I think I found a problem in patch 9 and have a question about patch 8
regarding the context in which interrupt processing functions would run.
I sent separate emails for both.

Regards,
  Felix

On 2018-12-05 4:15 a.m., Christian König wrote:
> This allows us to filter out VM faults in the GMC code.
>
> v2: don't filter out all faults
> v3: fix copy&paste typo, send all IV to the KFD, don't change message level
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c | 38 +++++++++++--------------
>  1 file changed, 17 insertions(+), 21 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
> index 6b6524f04ce0..79b6f456f2c5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
> @@ -149,9 +149,6 @@ static void amdgpu_irq_callback(struct amdgpu_device *adev,
>  	if (!amdgpu_ih_prescreen_iv(adev))
>  		return;
>  
> -	/* Before dispatching irq to IP blocks, send it to amdkfd */
> -	amdgpu_amdkfd_interrupt(adev, (const void *) &ih->ring[ring_index]);
> -
>  	entry.iv_entry = (const uint32_t *)&ih->ring[ring_index];
>  	amdgpu_ih_decode_iv(adev, &entry);
>  
> @@ -371,39 +368,38 @@ void amdgpu_irq_dispatch(struct amdgpu_device *adev,
>  	unsigned client_id = entry->client_id;
>  	unsigned src_id = entry->src_id;
>  	struct amdgpu_irq_src *src;
> +	bool handled = false;
>  	int r;
>  
>  	trace_amdgpu_iv(entry);
>  
>  	if (client_id >= AMDGPU_IRQ_CLIENTID_MAX) {
>  		DRM_DEBUG("Invalid client_id in IV: %d\n", client_id);
> -		return;
> -	}
>  
> -	if (src_id >= AMDGPU_MAX_IRQ_SRC_ID) {
> +	} else	if (src_id >= AMDGPU_MAX_IRQ_SRC_ID) {
>  		DRM_DEBUG("Invalid src_id in IV: %d\n", src_id);
> -		return;
> -	}
>  
> -	if (adev->irq.virq[src_id]) {
> +	} else if (adev->irq.virq[src_id]) {
>  		generic_handle_irq(irq_find_mapping(adev->irq.domain, src_id));
> -	} else {
> -		if (!adev->irq.client[client_id].sources) {
> -			DRM_DEBUG("Unregistered interrupt client_id: %d src_id: %d\n",
> -				  client_id, src_id);
> -			return;
> -		}
>  
> -		src = adev->irq.client[client_id].sources[src_id];
> -		if (!src) {
> -			DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id);
> -			return;
> -		}
> +	} else if (!adev->irq.client[client_id].sources) {
> +		DRM_DEBUG("Unregistered interrupt client_id: %d src_id: %d\n",
> +			  client_id, src_id);
>  
> +	} else if ((src = adev->irq.client[client_id].sources[src_id])) {
>  		r = src->funcs->process(adev, src, entry);
> -		if (r)
> +		if (r < 0)
>  			DRM_ERROR("error processing interrupt (%d)\n", r);
> +		else if (r)
> +			handled = true;
> +
> +	} else {
> +		DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id);
>  	}
> +
> +	/* Send it to amdkfd as well if it isn't already handled */
> +	if (!handled)
> +		amdgpu_amdkfd_interrupt(adev, entry->iv_entry);
>  }
>  
>  /**