SDMA out-of-bounds write access of tiled surface

Marek Olšák maraeo at gmail.com
Thu Aug 25 13:01:25 UTC 2016


FYI, I've disabled SDMA texture copying for Carrizo in Mesa. The VM
faults should no longer occur.

Marek

On Fri, Aug 5, 2016 at 3:22 PM, Mads <mads at ab3.no> wrote:
> Just to update, R600_DEBUG=nodma is still needed when updating to
> xorg-server 1.18.4, plasma 5.7.3 and linux 4.7.0, and libdrm, mesa and llvm
> built from live sources from sometime this week.
>
> Should this info be in some bugzilla somewhere?
>
> - Mads
>
>
> On 2016-07-11 23:58, Nicolai Hähnle wrote:
>
>> The culprit SDMA buffer was actually submitted by the X server, I believe.
>>
>> Nicolai
>>
>>> On 11.07.2016 23:52, Mads wrote: Yes, also with 5.7 the corruption
>>> appears if I start /usr/bin/konsole or
>>> /usr/bin/dolphin (and I guess it will be apparent with more
>>> applications, if you want I could test some more). Although, I have no
>>> trouble using chromium or steam, playing all sorts of games. As you can
>>> see if you look through the thread, it might be reason to believe that
>>> plasmashell itself might be the culprit sending the data that corrupts
>>> the screen.
>>>
>>> - Mads
>>>
>>>> On 2016-07-11 23:42, Marek Olšák wrote:
>>>>
>>>> Can you see any corruption or is there just the message in dmesg? (It
>>>> might just be a page directory fetch and not a memory access)
>>>>
>>>> Marek


More information about the amd-gfx mailing list