Re: [PATCH v3 1/5] gem/vram: pin to vram in vmap

[Thomas Zimmermann]

Hi

Am 27.06.19 um 14:23 schrieb Gerd Hoffmann:
> drm clients like the generic framebuffer emulation keep a permanent
> vmap active, which in turn has a permanent pin.  This pin needs to
> be in vram, otherwise we can't display the framebuffer.
> 
> Signed-off-by: Gerd Hoffmann <kraxel@xxxxxxxxxx>
> ---
>  drivers/gpu/drm/drm_gem_vram_helper.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c
> index 4de782ca26b2..c724876c6f2a 100644
> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
> @@ -592,7 +592,7 @@ void *drm_gem_vram_driver_gem_prime_vmap(struct drm_gem_object *gem)
>  	int ret;
>  	void *base;
>  
> -	ret = drm_gem_vram_pin(gbo, 0);
> +	ret = drm_gem_vram_pin(gbo, DRM_GEM_VRAM_PL_FLAG_VRAM);

I have a patch set that converts ast and mgag200 to generic framebuffer
emulation with a shadow FB. The actual BO is mapped by the fbdev code on
demand to update its content. Permanently mapping the fb console's BO
would consume too much display memory. This requires the pin function's
placement flag to be 0, so the BO is mapped in system memory by default.
The proposed patch breaks this.

Some ideas for solving this:

 1) Introduce a default_placement field in struct drm_gem_vram_helper
where this flag can be configured. I'd favor this option.

 2) Introduce a separate callback function for pinning to vram. The
driver would have to set the correct function pointers.

 3) Pin the fb console buffer manually from within the bochs driver.

Best regards
Thomas

>  	if (ret)
>  		return NULL;
>  	base = drm_gem_vram_kmap(gbo, true, NULL);
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Linux GmbH, Maxfeldstrasse 5, 90409 Nuernberg, Germany
GF: Felix ImendÃrffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG NÃrnberg)

Attachment: signature.asc
Description: OpenPGP digital signature