Re: [PATCH v2 7/9] drm/simpledrm: Enable FB_DAMAGE_CLIPS property

["Noralf Trønnes" <noralf@tronnes.org>]

Den 09.11.2021 15.56, skrev Thomas Zimmermann:
> Hi,
> 
> thanks for looking through all this code.
> 
> Am 09.11.21 um 14:04 schrieb Noralf Trønnes:
>>
>>
>> Den 09.11.2021 13.38, skrev Thomas Zimmermann:
>>>
>>>
>>> Am 08.11.21 um 21:55 schrieb Noralf Trønnes:
>>>>
>>>>
>>>> Den 01.11.2021 15.15, skrev Thomas Zimmermann:
>>>>> Enable the FB_DAMAGE_CLIPS property to reduce display-update
>>>>> overhead. Also fixes a warning in the kernel log.
>>>>>
>>>>>     simple-framebuffer simple-framebuffer.0: [drm]
>>>>> drm_plane_enable_fb_damage_clips() not called
>>>>>
>>>>> Fix the computation of the blit rectangle. This wasn't an issue so
>>>>> far, as simpledrm always blitted the full framebuffer. The code now
>>>>> supports damage clipping and virtual screen sizes.
>>>>>
>>>>> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
>>>>> ---
>>>>>    drivers/gpu/drm/tiny/simpledrm.c | 30
>>>>> ++++++++++++++++++++++--------
>>>>>    1 file changed, 22 insertions(+), 8 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/tiny/simpledrm.c
>>>>> b/drivers/gpu/drm/tiny/simpledrm.c
>>>>> index 571f716ff427..e872121e9fb0 100644
>>>>> --- a/drivers/gpu/drm/tiny/simpledrm.c
>>>>> +++ b/drivers/gpu/drm/tiny/simpledrm.c
>>>>> @@ -642,7 +642,7 @@ simpledrm_simple_display_pipe_enable(struct
>>>>> drm_simple_display_pipe *pipe,
>>>>>        void *vmap = shadow_plane_state->data[0].vaddr; /* TODO: Use
>>>>> mapping abstraction */
>>>>>        struct drm_device *dev = &sdev->dev;
>>>>>        void __iomem *dst = sdev->screen_base;
>>>>> -    struct drm_rect clip;
>>>>> +    struct drm_rect src_clip, dst_clip;
>>>>>        int idx;
>>>>>          if (!fb)
>>>>> @@ -651,10 +651,14 @@ simpledrm_simple_display_pipe_enable(struct
>>>>> drm_simple_display_pipe *pipe,
>>>>>        if (!drm_dev_enter(dev, &idx))
>>>>>            return;
>>>>>    -    drm_rect_init(&clip, 0, 0, fb->width, fb->height);
>>>>> +    drm_rect_fp_to_int(&src_clip, &plane_state->src);
>>>>>    -    dst += drm_fb_clip_offset(sdev->pitch, sdev->format, &clip);
>>>>> -    drm_fb_blit_toio(dst, sdev->pitch, sdev->format->format, vmap,
>>>>> fb, &clip);
>>>>> +    dst_clip = plane_state->dst;
>>>>
>>>> I assume that src_clip and dst_clip are of the same size, since scaling
>>>> is not supported. What prevents dst_clip from being bigger than the
>>>> buffer that's being blitted into? Where is that bounds checking done?
>>>
>>> The value of dst_clip comes from plane_state->dst, which gets
>>> initialized in drm_atomic_helper_check_plane_state(). [1] The fields are
>>> taken from the crtc_{x,y,w,h} values by drm_plane_get_dest(). [2] For
>>> primary planes, the crtc_{x,y,w,h} values are initialized in
>>> __drm_atomic_helper_set_config() to the size of the display. [3] That
>>> size comes directly from the current video mode. [4] From all I can see
>>> this cannot overflow.
>>>
>>
>> Ok, that takes care of the upper bound.
>>
>> There's this in drm_atomic_helper_check_plane_state():
>>
>>     plane_state->visible = drm_rect_clip_scaled(src, dst, &clip);
>>
>>     if (!plane_state->visible)
>>         /*
>>          * Plane isn't visible; some drivers can handle this
>>          * so we just return success here.  Drivers that can't
>>          * (including those that use the primary plane helper's
>>          * update function) will return an error from their
>>          * update_plane handler.
>>          */
>>         return 0;
>>
>> drm_atomic_helper_damage_merged() checks ->visible and returns false if
>> it is not set so update() is good on the lower bound.
>>
>> Maybe it's necessary to check ->visible here in enable() before doing
>> the blit?
> 
> Is it even possible to create an invisible primary plane here? We cannot
> scale [1] and if the primary plane is smaller than the framebuffer, we
> hit the case at [2]. The only way I can see this is that the visibility
> test at [3] succeeds. That would require a framebuffer of size 0. That's
> impossible from the code in frambuffer_check(). [4] Unless there's some
> corner case, we're good here.

Ok, thanks.

Noralf.

> 
> However, I think that drm_atomic_check_plane_state() could later be
> improved to handle the 'visible' field in a more sophisticated way.
> 
> Best regards
> Thomas
> 
> [1]
> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_simple_kms_helper.c#L227
> 
> [2]
> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_atomic_helper.c#L854
> 
> [3]
> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_atomic_helper.c#L844
> 
> [4]
> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_framebuffer.c#L185
> 
> 
>>
>> Noralf.
>>
>>> Best regards
>>> Thomas
>>>
>>> [1]
>>> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_atomic_helper.c#L807
>>>
>>>
>>> [2]
>>> https://elixir.bootlin.com/linux/latest/source/include/drm/drm_plane.h#L257
>>>
>>> [3]
>>> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_atomic.c#L1590
>>>
>>>
>>> [4]
>>> https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/drm_modes.c#L788
>>>
>>>
>>>
>>>>
>>>> Noralf.
>>>>
>>>>> +    if (!drm_rect_intersect(&dst_clip, &src_clip))
>>>>> +        return;
>>>>> +
>>>>> +    dst += drm_fb_clip_offset(sdev->pitch, sdev->format, &dst_clip);
>>>>> +    drm_fb_blit_toio(dst, sdev->pitch, sdev->format->format, vmap,
>>>>> fb, &src_clip);
>>>>>          drm_dev_exit(idx);
>>>>>    }
>>>>> @@ -686,20 +690,28 @@ simpledrm_simple_display_pipe_update(struct
>>>>> drm_simple_display_pipe *pipe,
>>>>>        struct drm_framebuffer *fb = plane_state->fb;
>>>>>        struct drm_device *dev = &sdev->dev;
>>>>>        void __iomem *dst = sdev->screen_base;
>>>>> -    struct drm_rect clip;
>>>>> +    struct drm_rect damage_clip, src_clip, dst_clip;
>>>>>        int idx;
>>>>>          if (!fb)
>>>>>            return;
>>>>>    -    if (!drm_atomic_helper_damage_merged(old_plane_state,
>>>>> plane_state, &clip))
>>>>> +    if (!drm_atomic_helper_damage_merged(old_plane_state,
>>>>> plane_state, &damage_clip))
>>>>> +        return;
>>>>> +
>>>>> +    drm_rect_fp_to_int(&src_clip, &plane_state->src);
>>>>> +    if (!drm_rect_intersect(&src_clip, &damage_clip))
>>>>> +        return;
>>>>> +
>>>>> +    dst_clip = plane_state->dst;
>>>>> +    if (!drm_rect_intersect(&dst_clip, &src_clip))
>>>>>            return;
>>>>>          if (!drm_dev_enter(dev, &idx))
>>>>>            return;
>>>>>    -    dst += drm_fb_clip_offset(sdev->pitch, sdev->format, &clip);
>>>>> -    drm_fb_blit_toio(dst, sdev->pitch, sdev->format->format, vmap,
>>>>> fb, &clip);
>>>>> +    dst += drm_fb_clip_offset(sdev->pitch, sdev->format, &dst_clip);
>>>>> +    drm_fb_blit_toio(dst, sdev->pitch, sdev->format->format, vmap,
>>>>> fb, &src_clip);
>>>>>          drm_dev_exit(idx);
>>>>>    }
>>>>> @@ -794,6 +806,8 @@ static int simpledrm_device_init_modeset(struct
>>>>> simpledrm_device *sdev)
>>>>>        if (ret)
>>>>>            return ret;
>>>>>    +    drm_plane_enable_fb_damage_clips(&pipe->plane);
>>>>> +
>>>>>        drm_mode_config_reset(dev);
>>>>>          return 0;
>>>>>
>>>
>