[OpenIndiana-discuss] NVIDIA again

[Alan Coopersmith]

On 08/17/16 11:31 AM, Alexander Pyhalov wrote:
> Apostolos Syropoulos via openindiana-discuss писал 17.08.2016 21:03:
>> Hello,
>>
>> I reported my problem to the NVIDIA people who responed:
>>
>> https://devtalk.nvidia.com/default/topic/955405/solaris/
>>
>> It seems that something is wrong with the Xorg binary.
>>
>
> Hello.
> They could just managed to build it against newer Xorg. However, I don't
> understand, how in this case it works at all.
> As you can see, the mentioned symbol ( enableIndirectGLX ) is defined in Xorg
> 1.17 headers
> (https://cgit.freedesktop.org/xorg/xserver/tree/include/opaque.h?h=server-1.17-branch),
>
> but not in 1.14 headers
> (https://cgit.freedesktop.org/xorg/xserver/tree/include/opaque.h?h=server-1.14-branch),
> which we ship.
> So, our Xorg couldn't know about it.
>
> If you run elfdump on nvidia libraries, do you see it there?

Note that on Solarish OS'es, the backtrace will sometimes warn you about the
last linker lookup failure, even when it's not relevant - it could be the
driver looked for that symbol, didn't find it because your Xorg is an older
release, and went on about it's business only to crash later.

(I think, but have not had time to confirm that this is because the code I
  stuck in osinit.c years ago to catch actual errors here fails to check in
  the signal handler what signal was received so prints it for every signal:
https://cgit.freedesktop.org/xorg/xserver/commit?id=98f4179156391752e6688339487458ad7828abf4
  If correct, the fix would be to wrap the first chunk of that in
  OsSigHandler inside an "if (signo == SIGQUIT)" block.)

Still, that backtrace looks like some invalid function pointer got put
into the list of callbacks for the wakeup routine - either directly or
by something else overwriting a crucial bit of memory with 0x32.

-- 
	-Alan Coopersmith-              alan.coopersmith at oracle.com
	 Oracle Solaris Engineering - http://blogs.oracle.com/alanc