Bug 65438

I can't reproduce this. Make sure your LLVM and Mesa snapshots are both up to date and built consistently.

If that doesn't help, can you isolate which LLVM change introduced the problem for you? The Git mirror at http://llvm.org/git/llvm.git might be useful for that.

well im sure everything is latest bit, i even did an emerge -C to the entire graphical stack [xserver mesa llvm clang libclc cairo libx11 xcb glamor gtk+] en re emerge with defaults cflags and the same things Qt4 works perfectly but gtk apps crash X[i guess is cairo related]. so im sure is not software related and since my 7770 can't reclock either i guess i got the short straw with this brand.

so can you provide me some help to do what you ask in gentoo? or howto debug the gpu properly? since gdb/valgrind/etc dont do anything against llvm[i guess is cuz is getting loaded by mesa and the output goes to the gpu]

im using FireBurn overlay

(In reply to comment #2)
> so can you provide me some help to do what you ask in gentoo?

I'm not familiar with Gentoo, but I'm sure there are many tutorials on Git bisect on the 'net, there might even be one specific to Gentoo.


> or howto debug the gpu properly? since gdb/valgrind/etc dont do anything
> against llvm[i guess is cuz is getting loaded by mesa and the output goes to
> the gpu]

It should work just like for any other X server crash, see http://wiki.x.org/wiki/Development/Documentation/ServerDebugging .

I have the same problem on a 7750 CAPE VERDE card.

When starting a xfce session X craches with the error message from above, starting kde works fine.

bisecting shows this bad commit:

8a72c73032707c7b3c042b716286a1ac33720991 is the first bad commit
commit 8a72c73032707c7b3c042b716286a1ac33720991
Author: Tom Stellard <thomas.stellard@amd.com>
Date:   Mon Jun 3 17:39:46 2013 +0000

    R600/SI: Fixup CopyToReg register class in PostprocessISelDAG()
    
    The CopyToReg nodes will sometimes try to copy a value from a VGPR to an
    SGPR.  This kind of copy is not possible, so we need to detect
    VGPR->SGPR copies and do something else.  The current strategy is to
    replace these copies with VGPR->VGPR copies and hope that all the users
    of CopyToReg can accept VGPRs as arguments.
    
    git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183132 91177308-0d34-0410-b5e6-96231b3b80d8

i confirm that commit 8a72c73032707c7b3c042b716286a1ac33720991
seems the culprit

btw hadack and michael many thanks for you help

btw hadack your 7750 can reclock? i mean profile mid/high or dynpm, in case it crashes the kernel like mine i made this bug report  https://bugzilla.kernel.org/show_bug.cgi?id=58621

Can you set the environment variable RADEON_DUMP_SHADERS=1 and re-post the output you get when starting xfce?

Created attachment 80567 [details]
xfce with RADEON_DUMP_SHADERS=1

heres the output with RADEON_DUMP_SHADERS

Rafael: I can reclock the card, but i have to go first to low and than i can set high, else it locks up. After boot it is in a strange state with clocks/voltage combination that no profile sets.

Created attachment 80630 [details] [review]
Possible Fix / Work Around

Can you try this patch?  It should fix the segfault, but it looks to me like the real bug might be in the CFG Structurizer.  There are a number of PHI nodes that take undef as an argument, and I'm not sure if that is correct.

Yes, with your patch xfce starts fine.
Do you need more info to find the real issue?

(In reply to comment #10)
> Yes, with your patch xfce starts fine.
> Do you need more info to find the real issue?

No, I think we have enough.

Tom committed his fix as revision 183937.

many thanks for the help very appreciated