[PATCHv1] drm/omap: get fbdev working for manually updated display

[Tomi Valkeinen]

On 01/05/2021 23:00, Sebastian Reichel wrote:
> Running a legacy application, which directly uses /dev/fb0
> currently results in display not being refreshed when the
> application mmaps the memory instead of calling write().
> 
> Deferred IO support will also schedule dirty_work with the
> damage collected from the mmap page writes and thus gets
> some more legacy applications working.
> 
> Delay frequency settings have been taken over from
> drm_fb_helper_generic_probe().
> 
> Reported-by: Carl Philipp Klemm <philipp at uvos.xyz>
> Tested-by: Carl Philipp Klemm <philipp at uvos.xyz>
> Signed-off-by: Sebastian Reichel <sebastian.reichel at collabora.com>
> ---
> Hi,
> 
> I still need to fix my Droid 4, so I only compile tested this
> myself.
> 
> -- Sebastian
> ---
>   drivers/gpu/drm/omapdrm/omap_fbdev.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/gpu/drm/omapdrm/omap_fbdev.c b/drivers/gpu/drm/omapdrm/omap_fbdev.c
> index 42eac6ad12bd..fc3897d2d7cc 100644
> --- a/drivers/gpu/drm/omapdrm/omap_fbdev.c
> +++ b/drivers/gpu/drm/omapdrm/omap_fbdev.c
> @@ -87,6 +87,11 @@ static const struct fb_ops omap_fb_ops = {
>   	.fb_imageblit = drm_fb_helper_sys_imageblit,
>   };
>   
> +static struct fb_deferred_io omap_fb_defio = {
> +	.delay		= HZ / 20,
> +	.deferred_io	= drm_fb_helper_deferred_io,
> +};
> +
>   static int omap_fbdev_create(struct drm_fb_helper *helper,
>   		struct drm_fb_helper_surface_size *sizes)
>   {
> @@ -176,6 +181,9 @@ static int omap_fbdev_create(struct drm_fb_helper *helper,
>   
>   	drm_fb_helper_fill_info(fbi, helper, sizes);
>   
> +	fbi->fbdefio = &omap_fb_defio;
> +	fb_deferred_io_init(fbi);
> +
>   	dev->mode_config.fb_base = dma_addr;
>   
>   	fbi->screen_buffer = omap_gem_vaddr(fbdev->bo);
> 
> base-commit: 152d32aa846835987966fd20ee1143b0e05036a0
> 

On DRA76 EVM, with parallel output, this causes cache-flush problem. 
When I run my simple fbtest, which just draws to the fb, I see lots of 
cache-line artifacts.

When I run kmstest (from kms++), I get a flood of list corruptions 
somewhere in omap_gem. Is it related to "Deferred I/O is not compatible 
with SHMEM"?

  Tomi