[Intel-gfx] [PATCH v2 4/4] drm/i915: Handle dma fences in dirtyfb callback

[Ville Syrjälä]

On Thu, Jul 27, 2023 at 08:35:18AM +0300, Jouni Högander wrote:
> Take into account dma fences in dirtyfb callback. If there is no
> unsignaled dma fences perform flush immediately. If there are
> unsignaled dma fences perform invalidate and add callback which will
> queue flush when the fence gets signaled.
> 
> v2: Use dma_resv_get_singleton
> 
> Signed-off-by: Jouni Högander <jouni.hogander at intel.com>
> ---
>  drivers/gpu/drm/i915/display/intel_fb.c | 57 +++++++++++++++++++++++--
>  1 file changed, 54 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_fb.c b/drivers/gpu/drm/i915/display/intel_fb.c
> index 446bbf7986b6..56a21377680d 100644
> --- a/drivers/gpu/drm/i915/display/intel_fb.c
> +++ b/drivers/gpu/drm/i915/display/intel_fb.c
> @@ -7,6 +7,9 @@
>  #include <drm/drm_framebuffer.h>
>  #include <drm/drm_modeset_helper.h>
>  
> +#include <linux/dma-fence.h>
> +#include <linux/dma-resv.h>
> +
>  #include "i915_drv.h"
>  #include "intel_display.h"
>  #include "intel_display_types.h"
> @@ -1896,6 +1899,21 @@ static int intel_user_framebuffer_create_handle(struct drm_framebuffer *fb,
>  	return drm_gem_handle_create(file, &obj->base, handle);
>  }
>  
> +struct frontbuffer_fence_cb {
> +	struct dma_fence_cb base;
> +	struct intel_frontbuffer *front;
> +};
> +
> +static void intel_user_framebuffer_fence_wake(struct dma_fence *dma,
> +					      struct dma_fence_cb *data)
> +{
> +	struct frontbuffer_fence_cb *cb = container_of(data, typeof(*cb), base);
> +
> +	intel_frontbuffer_queue_flush(cb->front);
> +	kfree(cb);
> +	dma_fence_put(dma);
> +}
> +
>  static int intel_user_framebuffer_dirty(struct drm_framebuffer *fb,
>  					struct drm_file *file,
>  					unsigned int flags, unsigned int color,
> @@ -1903,11 +1921,44 @@ static int intel_user_framebuffer_dirty(struct drm_framebuffer *fb,
>  					unsigned int num_clips)
>  {
>  	struct drm_i915_gem_object *obj = intel_fb_obj(fb);
> +	struct intel_frontbuffer *front = to_intel_frontbuffer(fb);
> +	struct dma_fence *fence;
> +	struct frontbuffer_fence_cb *cb;
> +	int ret = 0;
> +
> +	if (dma_resv_test_signaled(obj->base.resv, dma_resv_usage_rw(false)))
> +		goto flush;
> +
> +	intel_frontbuffer_invalidate(front, ORIGIN_DIRTYFB);

Looks like this could be deferred until dma_fence_add_callback()
has informed us whether the fence has signalled in the meantime.
That way we only do the invalidate if we succesfully add the
callback. Could in theory drop the dma_resv_test_signaled() as well,
but maybe it's worth keeping that around to avoid the kmalloc()/etc.
whenever possible.

Apart from that and the few comments I gave to the other patch
it all looks pretty reasonable to me.

> +
> +	ret = dma_resv_get_singleton(obj->base.resv, dma_resv_usage_rw(false),
> +				     &fence);
> +	if (ret || !fence)
> +		goto flush;
> +
> +	cb = kmalloc(sizeof(*cb), GFP_KERNEL);
> +	if (!cb) {
> +		dma_fence_put(fence);
> +		ret = -ENOMEM;
> +		goto flush;
> +	}
>  
> -	i915_gem_object_flush_if_display(obj);
> -	intel_frontbuffer_flush(to_intel_frontbuffer(fb), ORIGIN_DIRTYFB);
> +	cb->front = front;
>  
> -	return 0;
> +	ret = dma_fence_add_callback(fence, &cb->base,
> +				     intel_user_framebuffer_fence_wake);
> +	if (ret) {
> +		intel_user_framebuffer_fence_wake(fence, &cb->base);
> +		if (ret == -ENOENT)
> +			ret = 0;
> +	}
> +
> +	return ret;
> +
> +flush:
> +	i915_gem_object_flush_if_display(obj);
> +	intel_frontbuffer_flush(front, ORIGIN_DIRTYFB);
> +	return ret;
>  }
>  
>  static const struct drm_framebuffer_funcs intel_fb_funcs = {
> -- 
> 2.34.1

-- 
Ville Syrjälä
Intel