[blfs-dev] #6526: xf86-video-intel segfaults unless forcing uxa acceleration on current BLFS-svn.

Bruce Dubbs bruce.dubbs at gmail.com
Sun May 24 21:52:37 PDT 2015


Ken Moffat wrote:
> On Mon, May 25, 2015 at 04:44:22AM +0100, Ken Moffat wrote:
>
> only you didn't see it, because it was too big - must have been the
> attachment.  Oh well, you can git clone for yourselves if you want
> to :-)
>
> Here is the actualy mail without the attachment.
>
>> On Mon, May 25, 2015 at 04:14:04AM +0100, Ken Moffat wrote:
>>>
>>> Normally, I run ../gcc-5.1.0/contrib/test_summary but that is giving
>>> me no output at all. ???  This is from upgrading gcc on the existing
>>> system.
>>>
>>> Also, the build and tests completed in just over 10 minutes, which
>>> is a lot less than I was expecting (SandyBridge i3, -j4).  Colour me
>>> baffled and very much wishing I'd never touched this.
>>>
>> Well, I gave the git version of xf86-video-intel another try with
>> the patched compiler.  Needs autoreconf -fiv (don't do like I did on
>> my first encounter and run ./autogen.sh, that did not work).  As
>> Armin said, it builds ok, but I'm "once bitten, twice shy" so I
>> continued to enable uxa as well as the default sna.
>>
>> Tested it, without the conf file to force uxa : seems to be working
>> fine (with the released libdrm).
>>
>> I'm more confused than ever, and at a loss about which is the best
>> way forward.
>>
>> AFAICS, we seem to have the following options:

>> 2. Use the intel driver from git, and mention that gcc-5.1.0 needs
>> to be patched.

Tried that and it worked for both sna and uxa.  When I removed the 20-intel.conf 
file, I did see:

[ 17878.141] (II) modesetting: Driver for Modesetting Kernel Drivers: kms
[ 17878.141] (++) using VT number 1

[ 17878.141] (--) controlling tty is VT number 1, auto-enabling KeepTty
[ 17878.141] (II) intel(0): Using Kernel Mode Setting driver: i915, version 
1.6.0 20150130
[ 17878.141] (WW) Falling back to old probe method for modesetting

I'm not sure what that WW warning means, if anything.  When I left the 
20-intel.conf and just commented out the Option, it was not there.

>> We seem to be getting closer to using non-released versions of
>> everything :-(

Well, everything is a bit of an exaggeration.  Generally it's because upstream 
does not release "stable" packages in a timely manner (if at all).

>> Whatever we do, I think it would be a good idea to keep some info
>> about using uxa to work around sna problems.  But perhaps that
>> belongs in the wiki ?

I think it's OK in the book as it is until a stable release is made.

   -- Bruce




More information about the blfs-dev mailing list