Bug 64776

Summary: [9.1.2]"GPU fault detected" whit "eclipse juno" crash system
Product: Mesa Reporter: mombelli.mauro
Component: Drivers/Gallium/radeonsiAssignee: Default DRI bug account <dri-devel>
Status: RESOLVED INVALID QA Contact:
Severity: normal    
Priority: medium    
Version: 9.1   
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments: dmesg with a nice error log
log of corg.. doesn't seems to catch something
radeonsi: Make sure disabled colour buffers are disabled in the CB state

Description mombelli.mauro 2013-05-20 00:01:30 UTC
Created attachment 79557 [details]
dmesg with a nice error log

hi,
after updating to mesa, ati-dri and mesa-libgl 9.1.2, everything work but when launching "eclipse juno" (even a fresh install) the monitor turn off, sometimes the system doesn't respond, sometimes the montor keep turning on and off, GUI in freezed but i can still use virtual consolle. No problem with steam, bzflag, flash, older version of eclipse or other java program. Also GPU extensive test have been done on windows system with no fault.

Work-around is falling back to 9.1.1

my board:

$ lspci | grep -i VGA
01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Pitcairn PRO [Radeon HD 7850]

here you will find attached dmesg and xorg log during one of the (rare) times when monitor was going on and off. Xorg seems to stop just before the system goes in this "loop state"
anyway dmesg seems to catch the problem
Comment 1 mombelli.mauro 2013-05-20 00:02:23 UTC
Created attachment 79558 [details]
log of corg.. doesn't seems to catch something
Comment 2 Alex Deucher 2013-05-20 13:18:36 UTC
What you are seeing is a GPU reset.  CB6 is writing to an invalid mapping at GPU page 0x00014EDB.  Can you bisect the mesa 9.1 branch between 9.1.1 and 9.1.2 to identify the commit that broke it?
Comment 3 mombelli.mauro 2013-05-20 13:29:10 UTC
I have no idea where to start.
I know how to compile code and debug program, but i have no clue on the
mesa's code structure, and how it works.
I don't know what CB6 is, i'm not sure about page meaning (is it similar to
"classic" RAM page?) etc..
can you tell me at least witch bunch of file/operation i have to debug, and
how?


2013/5/20 <bugzilla-daemon@freedesktop.org>

>   *Comment # 2 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c2> on bug
> 64776 <https://bugs.freedesktop.org/show_bug.cgi?id=64776> from Alex
> Deucher <agd5f@yahoo.com> *
>
> What you are seeing is a GPU reset.  CB6 is writing to an invalid mapping at
> GPU page 0x00014EDB.  Can you bisect the mesa 9.1 branch between 9.1.1 and
> 9.1.2 to identify the commit that broke it?
>
>  ------------------------------
> You are receiving this mail because:
>
>    - You reported the bug.
>
>
Comment 4 Alex Deucher 2013-05-20 13:44:35 UTC
(In reply to comment #3)
> I have no idea where to start.
> I know how to compile code and debug program, but i have no clue on the
> mesa's code structure, and how it works.
> I don't know what CB6 is, i'm not sure about page meaning (is it similar to
> "classic" RAM page?) etc..
> can you tell me at least witch bunch of file/operation i have to debug, and
> how?

CB6 is the 6th color buffer and the GPU has a VM page table just like the CPU.  That info is not really important for you, I mentioned it for other developers.  If you could bisect mesa using git, that would be great.  There are a lot of howtos for using git to bisect.  E.g.,
https://wiki.ubuntu.com/X/BisectingMesa

In your case, it would be something like:
git bisect start
git bisect bad mesa-9.1.2
git bisect good mesa-9.1.1
Comment 5 mombelli.mauro 2013-05-20 14:15:47 UTC
ok, never heard of this git's feature, really nice and useful!
I'll try to bisect when i'll have some spare time.


2013/5/20 <bugzilla-daemon@freedesktop.org>

>   *Comment # 4 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c4> on bug
> 64776 <https://bugs.freedesktop.org/show_bug.cgi?id=64776> from Alex
> Deucher <agd5f@yahoo.com> *
>
> (In reply to comment #3 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c3>)
> > I have no idea where to start.
> > I know how to compile code and debug program, but i have no clue on the
> > mesa's code structure, and how it works.
> > I don't know what CB6 is, i'm not sure about page meaning (is it similar to
> > "classic" RAM page?) etc..
> > can you tell me at least witch bunch of file/operation i have to debug, and
> > how?
>
>
> CB6 is the 6th color buffer and the GPU has a VM page table just like the CPU.
> That info is not really important for you, I mentioned it for other developers.
>  If you could bisect mesa using git, that would be great.  There are a lot of
> howtos for using git to bisect.  E.g.,https://wiki.ubuntu.com/X/BisectingMesa
>
> In your case, it would be something like:
> git bisect start
> git bisect bad mesa-9.1.2
> git bisect good mesa-9.1.1
>
>  ------------------------------
> You are receiving this mail because:
>
>    - You reported the bug.
>
>
Comment 6 Michel Dänzer 2013-05-22 11:27:32 UTC
Created attachment 79647 [details] [review]
radeonsi: Make sure disabled colour buffers are disabled in the CB state

Does this Mesa patch help?
Comment 7 mombelli.mauro 2013-05-23 00:01:40 UTC
uff tryed now but i have to recompile LLVM from git as my distribution has
only 3.2.. i'll try again next time with some spare time.
Thanks anyway


2013/5/22 <bugzilla-daemon@freedesktop.org>

>   *Comment # 6 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c6> on bug
> 64776 <https://bugs.freedesktop.org/show_bug.cgi?id=64776> from Michel
> Dänzer <michel@daenzer.net> *
>
> Created attachment 79647 [details] [review] <https://bugs.freedesktop.org/attachment.cgi?id=79647> [details] <https://bugs.freedesktop.org/attachment.cgi?id=79647&action=edit> [review] <https://bugs.freedesktop.org/page.cgi?id=splinter.html&bug=64776&attachment=79647>
> radeonsi: Make sure disabled colour buffers are disabled in the CB state
>
> Does this Mesa patch help?
>
>  ------------------------------
> You are receiving this mail because:
>
>    - You reported the bug.
>
>
Comment 8 mombelli.mauro 2013-05-29 20:07:12 UTC
i'm actually compiling versus tag 9-1.1, also tryed versus 9-1.3 (with a
git checkout TAG_NAME)
the autoconfig, then make and i get:
make[3]: *** Nessuna regola per generare l'obiettivo
«../../../src/mapi/entry.c», necessario per «entry.lo».  Stop.
(no rure to generate the «../../../src/mapi/entry.c», necessary for
«entry.lo».  Stop.

help?


2013/5/23 Mauro Mombelli <mombelli.mauro@gmail.com>

> uff tryed now but i have to recompile LLVM from git as my distribution has
> only 3.2.. i'll try again next time with some spare time.
> Thanks anyway
>
>
> 2013/5/22 <bugzilla-daemon@freedesktop.org>
>
>   *Comment # 6 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c6> on bug
>> 64776 <https://bugs.freedesktop.org/show_bug.cgi?id=64776> from Michel
>> Dänzer <michel@daenzer.net> *
>>
>> Created attachment 79647 [details] [review] <https://bugs.freedesktop.org/attachment.cgi?id=79647> [details] <https://bugs.freedesktop.org/attachment.cgi?id=79647&action=edit> [review] <https://bugs.freedesktop.org/page.cgi?id=splinter.html&bug=64776&attachment=79647>
>> radeonsi: Make sure disabled colour buffers are disabled in the CB state
>>
>> Does this Mesa patch help?
>>
>>  ------------------------------
>> You are receiving this mail because:
>>
>>    - You reported the bug.
>>
>>
>
Comment 9 Michel Dänzer 2013-05-30 07:16:27 UTC
(In reply to comment #8)
> (no rure to generate the «../../../src/mapi/entry.c», necessary for
> «entry.lo».  Stop.

That should only happen when switching between the master and 9.1 branches. In that case, the easiest solution is to make distclean and start from scratch.
Comment 10 mombelli.mauro 2013-05-30 21:24:22 UTC
Yeh, got it, I've tryed to compile from master, then from other tag
thinking git checkout woul clear everything. My bad, I'll try to compile
again when I can. Should I use newer tag than 1.1?
I think strage that compiling mAster was on error for me.. Normally aren't
the pull self-contined?
Il giorno 30/mag/2013 09:16, <bugzilla-daemon@freedesktop.org> ha scritto:

>   *Comment # 9 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c9> on bug
> 64776 <https://bugs.freedesktop.org/show_bug.cgi?id=64776> from Michel
> Dänzer <michel@daenzer.net> *
>
> (In reply to comment #8 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c8>)> (no rure to generate the «../../../src/mapi/entry.c», necessary for
> > «entry.lo».  Stop.
>
> That should only happen when switching between the master and 9.1 branches. In
> that case, the easiest solution is to make distclean and start from scratch.
>
>  ------------------------------
> You are receiving this mail because:
>
>    - You reported the bug.
>
>
Comment 11 Michel Dänzer 2013-05-31 09:14:08 UTC
(In reply to comment #10)
> Should I use newer tag than 1.1?

You wrote the problem happens with 9.1.2 but not with 9.1.1, so it's probably best to stay within those tags for now.


> I think strage that compiling mAster was on error for me.. Normally aren't
> the pull self-contined?

Can't really say anything about that without at least seeing the error.
Comment 12 mombelli.mauro 2013-05-31 21:55:56 UTC
here the error tring to compile TAG 9.1.2 with patch (but error is in a
different file).
This is with autogen.sh without any option and with option ./autogen.sh
--prefix=/usr     --sysconfdir=/etc
--with-dri-driverdir=/usr/lib/xorg/modules/dri
--with-gallium-drivers=radeonsi
--with-dri-drivers=i915,i965,r200,radeon,nouveau,swrast
--enable-gallium-llvm     --enable-egl     --enable-gallium-egl
--with-egl-platforms=x11,drm,wayland     --enable-shared-glapi
--enable-gbm     --enable-glx-tls     --enable-dri     --enable-glx
--enable-osmesa     --enable-gles1     --enable-gles2
--enable-texture-float     --enable-xa     --enable-vdpau

gcc (GCC) 4.8.0 20130502 (prerelease)
with latest llvm from http://llvm.org/svn/llvm-project/llvm/trunk/


gallivm/lp_bld_debug.cpp: In member function 'virtual void
raw_debug_ostream::write_impl(const char*, size_t)':
gallivm/lp_bld_debug.cpp:106:40: warning: cast from type 'const char*' to
type 'char*' casts away qualifiers [-Wcast-qual]
       char *lastPtr = (char *)&Ptr[Size];
                                        ^
gallivm/lp_bld_debug.cpp: In function 'void lp_disassemble(const void*)':
gallivm/lp_bld_debug.cpp:206:64: error: no matching function for call to
'llvm::Target::createMCAsmInfo(std::string&) const'
    OwningPtr<const MCAsmInfo> AsmInfo(T->createMCAsmInfo(Triple));
                                                                ^
gallivm/lp_bld_debug.cpp:206:64: note: candidate is:
In file included from gallivm/lp_bld_debug.cpp:37:0:
/usr/include/llvm/Support/TargetRegistry.h:298:16: note: llvm::MCAsmInfo*
llvm::Target::createMCAsmInfo(const llvm::MCRegisterInfo&, llvm::StringRef)
const
     MCAsmInfo *createMCAsmInfo(const MCRegisterInfo &MRI,
                ^
/usr/include/llvm/Support/TargetRegistry.h:298:16: note:   candidate
expects 2 arguments, 1 provided




2013/5/31 <bugzilla-daemon@freedesktop.org>

>   *Comment # 11 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c11>on bug
> 64776 <https://bugs.freedesktop.org/show_bug.cgi?id=64776> from Michel
> Dänzer <michel@daenzer.net> *
>
> (In reply to comment #10 <https://bugs.freedesktop.org/show_bug.cgi?id=64776#c10>)
> > Should I use newer tag than 1.1?
>
>
> You wrote the problem happens with 9.1.2 but not with 9.1.1, so it's probably
> best to stay within those tags for now.
>
>
> > I think strage that compiling mAster was on error for me.. Normally aren't
> > the pull self-contined?
>
>
> Can't really say anything about that without at least seeing the error.
>
>  ------------------------------
> You are receiving this mail because:
>
>    - You reported the bug.
>
>
Comment 13 Laurent carlier 2013-05-31 22:03:16 UTC
(In reply to comment #12)
> here the error tring to compile TAG 9.1.2 with patch (but error is in a
> different file).
> This is with autogen.sh without any option and with option ./autogen.sh
> --prefix=/usr     --sysconfdir=/etc
> --with-dri-driverdir=/usr/lib/xorg/modules/dri
> --with-gallium-drivers=radeonsi
> --with-dri-drivers=i915,i965,r200,radeon,nouveau,swrast
> --enable-gallium-llvm     --enable-egl     --enable-gallium-egl
> --with-egl-platforms=x11,drm,wayland     --enable-shared-glapi
> --enable-gbm     --enable-glx-tls     --enable-dri     --enable-glx
> --enable-osmesa     --enable-gles1     --enable-gles2
> --enable-texture-float     --enable-xa     --enable-vdpau
> 
> gcc (GCC) 4.8.0 20130502 (prerelease)
> with latest llvm from http://llvm.org/svn/llvm-project/llvm/trunk/
> 

For mesa-9.1.x the proper llvm is here:
http://cgit.freedesktop.org/~tstellar/llvm/
Comment 14 Oded Gabbay 2016-05-24 14:05:01 UTC
Is this something that still happens/relevant ?
Comment 15 Timothy Arceri 2017-03-17 02:06:07 UTC
No updates in 4 years. Assuming this was fixed at some stage and closing bug.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.