Thread (15 messages) 15 messages, 2 authors, 2021-08-04

Re: [PATCH 00/11] Provide offset-adjusted framebuffer mappings

From: Thomas Zimmermann <tzimmermann@suse.de>
Date: 2021-08-04 18:22:24
Also in: dri-devel, linux-hyperv

Hi Sam

Am 03.08.21 um 18:21 schrieb Sam Ravnborg:
Hi Thomas,

On Tue, Aug 03, 2021 at 02:59:17PM +0200, Thomas Zimmermann wrote:
quoted
A framebuffer's offsets field might be non-zero to make the BO data
start at the specified offset within the BO's memory. Handle this
case in drm_gem_fb_vmap() and update all callers. So far, many drivers
ignore the offsets, which can lead to visual artifacts.

Patch 1 adds an optional argument to drm_gem_fb_vmap() to return the
offset-adjusted data address for use with shadow-buffered planes.

Patches 3 and 11 convert gud and vkms, which are the other callers of
drm_gem_fb_vmap(). For gud, it's just a cleanup. Vkms will handle the
framebuffer offsets correctly for its input and output framebuffers.

The other patches convert users of shadow-buffered planes to use the
data address. After conversion, each driver will use the correct data
for non-zero offsets.
quoted
   drm/ast: Use offset-adjusted shadow-plane mappings
   drm/gud: Get offset-adjusted mapping from drm_gem_fb_vmap()
   drm/hyperv: Use offset-adjusted shadow-plane mappings
   drm/mgag200: Use offset-adjusted shadow-plane mappings
   drm/cirrus: Use offset-adjusted shadow-plane mappings
   drm/gm12u320: Use offset-adjusted shadow-plane mappings
   drm/simpledrm: Use offset-adjusted shadow-plane mapping
   drm/udl: Use offset-adjusted shadow-plane mapping
   drm/vbox: Use offset-adjusted shadow-plane mappings
   drm/vkms: Use offset-adjusted shadow-plane mappings and output
Everything looked good while reading through the patches.
I cannot say if everything was properly converted but the patches looked
good.

So they are all:
Acked-by: Sam Ravnborg <redacted>
Thanks!
There was a few TODO comments visible aboput using the mapping api
properly. I assume this is coming in a later patch set..
There are indeed quite a few such comments. When we introduced the 
dma_buf_map type to solve the fbdev issue on sparc64, in many places I 
simply put the existing vaddr pointers into the map structure, and vice 
versa.

The code works as expected, but in the future we may want to use 
dma_buf_map for all framebuffer pointers. This would, for example, 
require format conversion helpers that operate on these structures. 
Adding that will require a number of changes throughout these helpers.

Best regards
Thomas
	Sam
-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help