Re: [PATCH] drm/vkms: guarantee vblank when capturing crc

[daniel]

On Sat, Aug 01, 2020 at 03:49:29PM -0300, Melissa Wen wrote:
> VKMS needs vblank interrupts enabled to capture CRC. When vblank is
> disabled, tests like kms_cursor_crc and kms_pipe_crc_basic getting stuck
> waiting for a capture that will not occur until vkms wakes up. This
> patch ensures that vblank remains enabled as long as the CRC capture is
> needed.
> 
> It clears the execution of the following kms_cursor_crc subtests:
> 1. pipe-A-cursor-[size,alpha-opaque, NxN-(on-screen, off-screen, sliding,
> random, fast-moving])] - successful when running individually.
> 2. pipe-A-cursor-dpms passes again
> 3. pipe-A-cursor-suspend also passes
> 
> The issue was initially tracked in the sequential execution of IGT
> kms_cursor_crc subtest: when running the test sequence or one of its
> subtests twice, the odd execs complete and the pairs get stuck in an
> endless wait. In the IGT code, calling a wait_for_vblank on preparing
> for CRC capture prevented the busy-wait. But the problem persisted in
> the pipe-A-cursor-dpms and -suspend subtests.
> 
> Checking the history, the pipe-A-cursor-dpms subtest was successful
> when, in vkms_atomic_commit_tail, instead of using the flip_done op, it
> used wait_for_vblanks. Another way to prevent blocking was
> wait_one_vblank when enabling crtc. However, in both cases,
> pipe-A-cursor-suspend persisted blocking in the 2nd start of CRC
> capture, which may indicate that something got stuck in the step of CRC
> setup. Indeed, wait_one_vblank in the crc setup was able to sync things
> and free all kms_cursor_crc subtests.
> 
> Besides, other alternatives to force enabling vblanks or prevent
> disabling them such as calling drm_crtc_put_vblank or modeset_enables
> before commit_planes + offdelay = 0, also unlock all subtests
> executions. These facts together converge on the lack of vblank to
> unblock the crc capture.
> 
> Finally, considering the vkms's dependence on vblank interruptions to
> perform tasks, this patch acquires a vblank ref when setup CRC source
> and releases ref when disabling crc capture, ensuring vblanks happen to
> compute CRC.
> 
> Cc: Rodrigo Siqueira <rodrigosiqueiramelo@xxxxxxxxx>
> Cc: Haneen Mohammed <hamohammed.sa@xxxxxxxxx>
> Co-developed-by: Sidong Yang <realwakka@xxxxxxxxx>
> Signed-off-by: Sidong Yang <realwakka@xxxxxxxxx>
> Co-developed-by: Daniel Vetter <daniel@xxxxxxxx>
> Signed-off-by: Daniel Vetter <daniel@xxxxxxxx>
> Signed-off-by: Melissa Wen <melissa.srw@xxxxxxxxx>

Excellent summary of the debug story.

> ---
>  drivers/gpu/drm/vkms/vkms_composer.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c
> index 4af2f19480f4..1161eaa383f1 100644
> --- a/drivers/gpu/drm/vkms/vkms_composer.c
> +++ b/drivers/gpu/drm/vkms/vkms_composer.c
> @@ -241,6 +241,14 @@ int vkms_set_crc_source(struct drm_crtc *crtc, const char *src_name)
>  
>  	ret = vkms_crc_parse_source(src_name, &enabled);
>  
> +	/* Ensure that vblank interruptions are enabled for crc capture */
> +	/* Enabling CRC: acquire vblank ref */

This comment just explains what the code does, that's not needed. The
first comment I think can be replaced if we create a little helper
function like

void vkms_set_composer(struct vkms_output, bool enable) {
	bool old_state;

	if (enabled)
		drm_crtc_vblank_get(crtc);

	spin_lock_irq(&out->lock);
	old_enable = out->composer_enabled;
	out->composer_enabled = enabled;
	spin_unlock_irq(&out->lock);

	if (old_enabled)
		drm_crtc_vblank_put(crtc);

}

This should also help as prep for the writeback work, where we have a 2nd
user that might need to enable the composer (maybe even need to switch to
refcounting the composer state then).

> +	if (enabled)
> +		drm_crtc_vblank_get(crtc);
> +	/* Disabling CRC: release vblank ref */
> +	if (!src_name)
> +		drm_crtc_vblank_put(crtc);

I'm not sure this correctly releases the vblank reference in all cases, I
think the suggestion in the helper function pseudo code should work
better. It does mean we temporarily elevate the vblank refcount if we go
enabled -> enabled, but that's not a problem since it's refcounted.

Cheers, Daniel

> +
>  	spin_lock_irq(&out->lock);
>  	out->composer_enabled = enabled;
>  	spin_unlock_irq(&out->lock);
> -- 
> 2.27.0
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch