Re: [PATCH 05/18] drm: Add helper to implement legacy dirtyfb

[Daniel Vetter <daniel@ffwll.ch>]

On Tue, Oct 02, 2018 at 04:30:30PM +0000, Deepak Singh Rawat wrote:
> > On Thu, Sep 27, 2018 at 05:30:07PM -0700, Deepak Rawat wrote:
> > > From: Rob Clark <robdclark@gmail.com>
> > >
> > > Add an atomic helper to implement dirtyfb support.  This is needed to
> > > support DSI command-mode panels with x11 userspace (ie. when we can't
> > > rely on pageflips to trigger a flush to the panel).
> > >
> > > v2: Modified the helper to use plane fb_damage_clips property and
> > > removed plane_state::dirty flag.
> > >
> > > v3:
> > > - Use uapi drm_mode_rect.
> > > - Support annotate flags.
> > >
> > > Cc: ville.syrjala@linux.intel.com
> > > Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> > > Cc: Pekka Paalanen <ppaalanen@gmail.com>
> > > Signed-off-by: Rob Clark <robdclark@gmail.com>
> > > Signed-off-by: Deepak Rawat <drawat@vmware.com>
> > > ---
> > >  drivers/gpu/drm/drm_damage_helper.c | 121
> > ++++++++++++++++++++++++++++
> > >  include/drm/drm_damage_helper.h     |   4 +
> > >  2 files changed, 125 insertions(+)
> > >
> > > diff --git a/drivers/gpu/drm/drm_damage_helper.c
> > b/drivers/gpu/drm/drm_damage_helper.c
> > > index bc734ddaf180..ecf26275543f 100644
> > > --- a/drivers/gpu/drm/drm_damage_helper.c
> > > +++ b/drivers/gpu/drm/drm_damage_helper.c
> > > @@ -26,6 +26,7 @@
> > >   *
> > >   * Authors:
> > >   * Deepak Rawat <drawat@vmware.com>
> > > + * Rob Clark <robdclark@gmail.com>
> > >   *
> > >
> > **********************************************************
> > ****************/
> > >
> > > @@ -70,6 +71,21 @@
> > >   * rectangles clipped to &drm_plane_state.src.
> > >   */
> > >
> > > +static void convert_clip_rect_to_rect(const struct drm_clip_rect *src,
> > > +				      struct drm_mode_rect *dest,
> > > +				      uint32_t num_clips, uint32_t src_inc)
> > > +{
> > > +	while (num_clips > 0) {
> > > +		dest->x1 = src->x1;
> > > +		dest->y1 = src->y1;
> > > +		dest->x2 = src->x2;
> > > +		dest->y2 = src->y2;
> > > +		src += src_inc;
> > > +		dest++;
> > > +		num_clips--;
> > > +	}
> > > +}
> > > +
> > >  /**
> > >   * drm_plane_enable_fb_damage_clips - Enables plane fb damage clips
> > property.
> > >   * @plane: Plane on which to enable damage clips property.
> > > @@ -123,6 +139,111 @@ void
> > drm_atomic_helper_check_plane_damage(struct drm_atomic_state *state,
> > >  }
> > >  EXPORT_SYMBOL(drm_atomic_helper_check_plane_damage);
> > >
> > > +/**
> > > + * drm_atomic_helper_dirtyfb - Helper for dirtyfb.
> > > + * @fb: DRM framebuffer.
> > > + * @file_priv: Drm file for the ioctl call.
> > > + * @flags: Dirty fb annotate flags.
> > > + * @color: Color for annotate fill.
> > > + * @clips: Dirty region.
> > > + * @num_clips: Count of clip in clips.
> > > + *
> > > + * A helper to implement &drm_framebuffer_funcs::dirty using damage
> > interface
> > > + * during plane update. Similar to ioctl, this helper do a full plane update
> > > + * when no clips are provided.
> > 
> > The above kerneldoc doesn't make sense to me ... we do have clips provided
> > here, it's just a different layout.
> 
> Thanks Daniel for the review. I meant if clips are NULL, which is possible.

Ah right, I've forgotten that the dirty ioctl allows num_clips == 0 iff
clips == NULL. Might be good to document that, for people like me :-)

> > > + *
> > > + * Returns: Zero on success, negative errno on failure.
> > > + */
> > > +int drm_atomic_helper_dirtyfb(struct drm_framebuffer *fb,
> > > +			      struct drm_file *file_priv, unsigned flags,
> > > +			      unsigned color, struct drm_clip_rect *clips,
> > > +			      unsigned num_clips)
> > > +{
> > > +	struct drm_modeset_acquire_ctx ctx;
> > > +	struct drm_property_blob *damage = NULL;
> > > +	struct drm_mode_rect *rects = NULL;
> > > +	struct drm_atomic_state *state;
> > > +	struct drm_plane *plane;
> > > +	int ret = 0;
> > > +
> > > +	/*
> > > +	 * When called from ioctl, we are interruptable, but not when called
> > > +	 * internally (ie. defio worker)
> > > +	 */
> > > +	drm_modeset_acquire_init(&ctx,
> > > +		file_priv ? DRM_MODESET_ACQUIRE_INTERRUPTIBLE : 0);
> > > +
> > > +	state = drm_atomic_state_alloc(fb->dev);
> > > +	if (!state) {
> > > +		ret = -ENOMEM;
> > > +		goto out;
> > > +	}
> > > +	state->acquire_ctx = &ctx;
> > > +
> > > +	if (clips) {
> > > +		uint32_t inc = 1;
> > > +
> > > +		if (flags & DRM_MODE_FB_DIRTY_ANNOTATE_COPY) {
> > > +			inc = 2;
> > > +			num_clips /= 2;
> > > +		}
> > > +
> > > +		rects = kcalloc(num_clips, sizeof(*rects), GFP_KERNEL);
> > > +		if (!rects) {
> > > +			ret = -ENOMEM;
> > > +			goto out;
> > > +		}
> > > +
> > > +		convert_clip_rect_to_rect(clips, rects, num_clips, inc);
> > > +		damage = drm_property_create_blob(fb->dev,
> > > +						  num_clips * sizeof(*rects),
> > > +						  rects);
> > > +		if (IS_ERR(damage)) {
> > > +			ret = PTR_ERR(damage);
> > > +			damage = NULL;
> > > +			goto out;
> > > +		}
> > > +	}
> > > +
> > > +retry:
> > > +	drm_for_each_plane(plane, fb->dev) {
> > > +		struct drm_plane_state *plane_state;
> > > +
> > > +		if (plane->state->fb != fb)
> > > +			continue;
> > > +
> > > +		plane_state = drm_atomic_get_plane_state(state, plane);
> > > +		if (IS_ERR(plane_state)) {
> > > +			ret = PTR_ERR(plane_state);
> > > +			goto out;
> > > +		}
> > > +
> > > +		drm_property_replace_blob(&plane_state-
> > >fb_damage_clips,
> > > +					  damage);
> > > +	}
> > > +
> > > +	ret = drm_atomic_commit(state);
> > 
> > This blocks, which matches what udl does (one of the very first
> > implementations of the fb->dirty callback). I got lost a bit in the vmwgfx
> > dirty implementation, so no idea whether that one blocked or not.
> > 
> > If the blocking here is a problem, then we need a separate worker
> > similarly to what the fbdev worker does (since using the atomic
> > nonblocking stuff has uapi side-effects).
> > 
> > Either way, probably something we need to document in the kernel-doc.
> > -Daniel
> 
> IIUC vmwgfx dirty implementation also block. vmwgfx dirty implementation
> will make sure that commands are flushed to virtual device and all the mutex
> lock are held until then.
> 
> With my testing on vmwgfx software only VM I see that non-blocking will cause
> rendering issues with so many dirty_fb ioctl()'s and kernel returns EBUSY(?).

Yup, drm_atomic_commit_nonblocking will not work here, wrong semantics. As
I mentioned, we'd need a separate struct_work and similar logic like in
the fbdev defio support in drm_fb_helper.c.

> Yes I agree I can update the kernel-doc to specify blocking and if someone
> need non-blocking version in future that can be easily done.

Sounds good, and much less work :-)
-Daniel

> 
> Thanks,
> Deepak
> > 
> > > +
> > > +out:
> > > +	if (ret == -EDEADLK) {
> > > +		drm_atomic_state_clear(state);
> > > +		ret = drm_modeset_backoff(&ctx);
> > > +		if (!ret)
> > > +			goto retry;
> > > +	}
> > > +
> > > +	drm_property_blob_put(damage);
> > > +	kfree(rects);
> > > +	drm_atomic_state_put(state);
> > > +
> > > +	drm_modeset_drop_locks(&ctx);
> > > +	drm_modeset_acquire_fini(&ctx);
> > > +
> > > +	return ret;
> > > +
> > > +}
> > > +EXPORT_SYMBOL(drm_atomic_helper_dirtyfb);
> > > +
> > >  /**
> > >   * drm_atomic_helper_damage_iter_init - Initialize the damage iterator.
> > >   * @iter: The iterator to initialize.
> > > diff --git a/include/drm/drm_damage_helper.h
> > b/include/drm/drm_damage_helper.h
> > > index af188799820f..e4fe7036cab2 100644
> > > --- a/include/drm/drm_damage_helper.h
> > > +++ b/include/drm/drm_damage_helper.h
> > > @@ -67,6 +67,10 @@ struct drm_atomic_helper_damage_iter {
> > >  void drm_plane_enable_fb_damage_clips(struct drm_plane *plane);
> > >  void drm_atomic_helper_check_plane_damage(struct drm_atomic_state
> > *state,
> > >  					  struct drm_plane_state
> > *plane_state);
> > > +int drm_atomic_helper_dirtyfb(struct drm_framebuffer *fb,
> > > +			      struct drm_file *file_priv, unsigned flags,
> > > +			      unsigned color, struct drm_clip_rect *clips,
> > > +			      unsigned num_clips);
> > >  void
> > >  drm_atomic_helper_damage_iter_init(struct
> > drm_atomic_helper_damage_iter *iter,
> > >  				   const struct drm_plane_state *old_state,
> > > --
> > > 2.17.1
> > >
> > 
> > --
> > Daniel Vetter
> > Software Engineer, Intel Corporation
> > https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fblog.ff
> > wll.ch&amp;data=02%7C01%7Cdrawat%40vmware.com%7C0835e0509d8849
> > 89c6c208d6277513ae%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C1%7C0%7
> > C636739781135270102&amp;sdata=6%2Fr%2FS%2B3k7nJDZKNd9tRnmofCGf
> > CG2TKqDy4tG6lojsA%3D&amp;reserved=0

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel