Re: [Beignet] Beignet crashes on vanilla 3.17.1 with IVB hardware

Re: [Beignet] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Zhigang Gong]

Hi,

For IVB, I just checked the 3.18-rc1, it has the following patch:
commit c9224faa59c3071ecfa2d4b24592f4eb61e57069
Author: Brad Volkin <bradley.d.volkin@intel.com>
Date:   Tue Jun 17 14:10:34 2014 -0700

    drm/i915: Add some L3 registers to the parser whitelist

    Beignet needs these in order to program the L3 cache config for
    OpenCL workloads, particularly when using SLM.

    Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
    Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

So, beignet should work fine with 3.18 on IVB/BYT.

But for the HSW，I'm not quite sure when we could get a workable vanilla
kernel. Something I found at the intel-gfx mail list as below and it doesn't
sound good.

http://lists.freedesktop.org/archives/intel-gfx/2014-May/044694.html
http://lists.freedesktop.org/archives/intel-gfx/2014-May/045088.html

CC to intel-gfx mail list. Hope we can get an official anwser here.

Thanks,
Zhigang Gong.

On Thu, Oct 23, 2014 at 03:39:35PM +0300, Vasily Khoruzhick wrote:
> Hi,
> 
> As you maybe know, any application which uses beignet OpenCL
> implementation crashes on Ivy Bridge hardware when using vanilla
> 3.17.1 kernel.
> I guess it's due to batchbuffer security and patch to disable
> batchbuffer security is required, but guys, it fails since 3.16, and
> 3.17 was released quite a while ago.
> 
> Could you cooperate with i915 driver devs to make Beignet working on
> vanilla kernel without extra patches?
> 
> Thanks!
> 
> Regards,
> Vasily
> _______________________________________________
> Beignet mailing list
> Beignet@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/beignet
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Beignet] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Zhigang Gong]

I just checked again, and found both 3.17 and 3.17.1 should work
fine on IVB with beignet. I just tested beignet on IVB with kernel
3.17.1, all unit tests passed successfully. For IVB user, no need
to wait for 3.18.

Don't know which application you were testing on your IVB machine.
If it could be reproduced easily, please open a bug on fd.o.

Thanks,
Zhigang Gong.

On Fri, Oct 24, 2014 at 10:27:23AM +0800, Zhigang Gong wrote:
> Hi,
> 
> For IVB, I just checked the 3.18-rc1, it has the following patch:
> commit c9224faa59c3071ecfa2d4b24592f4eb61e57069
> Author: Brad Volkin <bradley.d.volkin@intel.com>
> Date:   Tue Jun 17 14:10:34 2014 -0700
> 
>     drm/i915: Add some L3 registers to the parser whitelist
> 
>     Beignet needs these in order to program the L3 cache config for
>     OpenCL workloads, particularly when using SLM.
> 
>     Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
>     Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> 
> So, beignet should work fine with 3.18 on IVB/BYT.
> 
> But for the HSW，I'm not quite sure when we could get a workable vanilla
> kernel. Something I found at the intel-gfx mail list as below and it doesn't
> sound good.
> 
> http://lists.freedesktop.org/archives/intel-gfx/2014-May/044694.html
> http://lists.freedesktop.org/archives/intel-gfx/2014-May/045088.html
> 
> CC to intel-gfx mail list. Hope we can get an official anwser here.
> 
> Thanks,
> Zhigang Gong.
> 
> On Thu, Oct 23, 2014 at 03:39:35PM +0300, Vasily Khoruzhick wrote:
> > Hi,
> > 
> > As you maybe know, any application which uses beignet OpenCL
> > implementation crashes on Ivy Bridge hardware when using vanilla
> > 3.17.1 kernel.
> > I guess it's due to batchbuffer security and patch to disable
> > batchbuffer security is required, but guys, it fails since 3.16, and
> > 3.17 was released quite a while ago.
> > 
> > Could you cooperate with i915 driver devs to make Beignet working on
> > vanilla kernel without extra patches?
> > 
> > Thanks!
> > 
> > Regards,
> > Vasily
> > _______________________________________________
> > Beignet mailing list
> > Beignet@lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/beignet
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Intel-gfx] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Vasily Khoruzhick]

Hi Zhigang,

Luxmark crashes with following backtrace:

Program received signal SIGSEGV, Segmentation fault.
0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
(gdb) bt
#0  0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
#1  0x0000000000482236 in slg::RenderEngine::Stop() ()
#2  0x000000000047be9b in slg::RenderSession::~RenderSession() ()
#3  0x0000000000468a77 in LuxMarkApp::Stop() ()
#4  0x0000000000468b46 in LuxMarkApp::InitRendering(LuxMarkAppMode,
char const*) ()
#5  0x000000000046edbe in MainWindow::event(QEvent*) ()
#6  0x00007ffff7099b9c in QApplicationPrivate::notify_helper(QObject*,
QEvent*) () from /usr/lib/libQtGui.so.4
#7  0x00007ffff70a05e8 in QApplication::notify(QObject*, QEvent*) ()
from /usr/lib/libQtGui.so.4
#8  0x00007ffff6b6af7d in QCoreApplication::notifyInternal(QObject*,
QEvent*) () from /usr/lib/libQtCore.so.4
#9  0x00007ffff6b6e341 in
QCoreApplicationPrivate::sendPostedEvents(QObject*, int, QThreadData*)
() from /usr/lib/libQtCore.so.4
#10 0x00007ffff6b99e63 in ?? () from /usr/lib/libQtCore.so.4
#11 0x00007ffff4d05a1d in g_main_context_dispatch () from
/usr/lib/libglib-2.0.so.0
#12 0x00007ffff4d05d08 in ?? () from /usr/lib/libglib-2.0.so.0
#13 0x00007ffff4d05dbc in g_main_context_iteration () from
/usr/lib/libglib-2.0.so.0
#14 0x00007ffff6b99fad in
QEventDispatcherGlib::processEvents(QFlags<QEventLoop::ProcessEventsFlag>)
() from /usr/lib/libQtCore.so.4
#15 0x00007ffff713d9c6 in ?? () from /usr/lib/libQtGui.so.4
#16 0x00007ffff6b69ad1 in
QEventLoop::processEvents(QFlags<QEventLoop::ProcessEventsFlag>) ()
from /usr/lib/libQtCore.so.4
#17 0x00007ffff6b69e35 in
QEventLoop::exec(QFlags<QEventLoop::ProcessEventsFlag>) () from
/usr/lib/libQtCore.so.4
#18 0x00007ffff6b6f3c7 in QCoreApplication::exec() () from
/usr/lib/libQtCore.so.4
#19 0x000000000045afc8 in main ()

And convert from imagemagick crashes in libgbe.so:

Program received signal SIGSEGV, Segmentation fault.
0x00007fffeb0df18c in gbe::buildRegInfo(gbe::ir::BasicBlock&,
gbe::vector<gbe::RegInfoForMov>&) () from /usr/lib/beignet//libgbe.so
(gdb) bt
#0  0x00007fffeb0df18c in gbe::buildRegInfo(gbe::ir::BasicBlock&,
gbe::vector<gbe::RegInfoForMov>&) () from /usr/lib/beignet//libgbe.so
#1  0x00007fffeb0e048d in
gbe::GenWriter::removeLOADIs(gbe::ir::Liveness const&,
gbe::ir::Function&) () from /usr/lib/beignet//libgbe.so
#2  0x00007fffeb0fb823 in
gbe::GenWriter::emitFunction(llvm::Function&) () from
/usr/lib/beignet//libgbe.so
#3  0x00007fffeb100b63 in
gbe::GenWriter::runOnFunction(llvm::Function&) () from
/usr/lib/beignet//libgbe.so
#4  0x00007fffec0129e1 in
llvm::FPPassManager::runOnFunction(llvm::Function&) () from
/usr/lib/beignet//libgbe.so
#5  0x00007fffebe11083 in (anonymous
namespace)::CGPassManager::runOnModule(llvm::Module&) () from
/usr/lib/beignet//libgbe.so
#6  0x00007fffec015538 in
llvm::legacy::PassManagerImpl::run(llvm::Module&) () from
/usr/lib/beignet//libgbe.so
#7  0x00007fffeb11fe0e in gbe::llvmToGen(gbe::ir::Unit&, char const*,
void const*, int, bool) () from /usr/lib/beignet//libgbe.so
#8  0x00007fffeb0c4fc1 in gbe::Program::buildFromLLVMFile(char const*,
void const*, std::string&, int) () from /usr/lib/beignet//libgbe.so
#9  0x00007fffeb17d0ba in gbe::genProgramNewFromLLVM(unsigned int,
char const*, void const*, void const*, unsigned long, char*, unsigned
long*, int) () from /usr/lib/beignet//libgbe.so
#10 0x00007fffeb0c8e09 in gbe::programNewFromSource(unsigned int, char
const*, unsigned long, char const*, char*, unsigned long*) () from
/usr/lib/beignet//libgbe.so
#11 0x00007fffef0eac96 in cl_program_build () from /usr/lib/beignet/libcl.so
#12 0x00007fffef0e39de in clBuildProgram () from /usr/lib/beignet/libcl.so
#13 0x00007fffef32c4cb in clBuildProgram () from /usr/lib/libOpenCL.so
#14 0x00007ffff7a461ec in ?? () from /usr/lib/libMagickCore-6.Q16HDRI.so.2
#15 0x00007ffff7a4689f in InitOpenCLEnvInternal () from
/usr/lib/libMagickCore-6.Q16HDRI.so.2
#16 0x00007ffff7a46b83 in ?? () from /usr/lib/libMagickCore-6.Q16HDRI.so.2
#17 0x00007ffff7a4797a in ?? () from /usr/lib/libMagickCore-6.Q16HDRI.so.2
#18 0x00007ffff7a4808c in InitOpenCLEnv () from
/usr/lib/libMagickCore-6.Q16HDRI.so.2
#19 0x00007ffff7941558 in ?? () from /usr/lib/libMagickCore-6.Q16HDRI.so.2
#20 0x00007ffff7948d3e in AccelerateResizeImage () from
/usr/lib/libMagickCore-6.Q16HDRI.so.2
#21 0x00007ffff7a95da0 in ResizeImage () from
/usr/lib/libMagickCore-6.Q16HDRI.so.2
#22 0x00007ffff768d310 in MogrifyImage () from
/usr/lib/libMagickWand-6.Q16HDRI.so.2
#23 0x00007ffff7691bc9 in MogrifyImages () from
/usr/lib/libMagickWand-6.Q16HDRI.so.2
#24 0x00007ffff761d03b in ConvertImageCommand () from
/usr/lib/libMagickWand-6.Q16HDRI.so.2
#25 0x00007ffff7686ff7 in MagickCommandGenesis () from
/usr/lib/libMagickWand-6.Q16HDRI.so.2
#26 0x00000000004008a7 in ?? ()
#27 0x00007ffff703a040 in __libc_start_main () from /usr/lib/libc.so.6
#28 0x00000000004008fb in ?? ()

Reproducibility is 100%

Regards
Vasily

On Fri, Oct 24, 2014 at 10:05 AM, Zhigang Gong
<zhigang.gong@linux.intel.com> wrote:
> I just checked again, and found both 3.17 and 3.17.1 should work
> fine on IVB with beignet. I just tested beignet on IVB with kernel
> 3.17.1, all unit tests passed successfully. For IVB user, no need
> to wait for 3.18.
>
> Don't know which application you were testing on your IVB machine.
> If it could be reproduced easily, please open a bug on fd.o.
>
> Thanks,
> Zhigang Gong.
>
> On Fri, Oct 24, 2014 at 10:27:23AM +0800, Zhigang Gong wrote:
>> Hi,
>>
>> For IVB, I just checked the 3.18-rc1, it has the following patch:
>> commit c9224faa59c3071ecfa2d4b24592f4eb61e57069
>> Author: Brad Volkin <bradley.d.volkin@intel.com>
>> Date:   Tue Jun 17 14:10:34 2014 -0700
>>
>>     drm/i915: Add some L3 registers to the parser whitelist
>>
>>     Beignet needs these in order to program the L3 cache config for
>>     OpenCL workloads, particularly when using SLM.
>>
>>     Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
>>     Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>
>> So, beignet should work fine with 3.18 on IVB/BYT.
>>
>> But for the HSW，I'm not quite sure when we could get a workable vanilla
>> kernel. Something I found at the intel-gfx mail list as below and it doesn't
>> sound good.
>>
>> http://lists.freedesktop.org/archives/intel-gfx/2014-May/044694.html
>> http://lists.freedesktop.org/archives/intel-gfx/2014-May/045088.html
>>
>> CC to intel-gfx mail list. Hope we can get an official anwser here.
>>
>> Thanks,
>> Zhigang Gong.
>>
>> On Thu, Oct 23, 2014 at 03:39:35PM +0300, Vasily Khoruzhick wrote:
>> > Hi,
>> >
>> > As you maybe know, any application which uses beignet OpenCL
>> > implementation crashes on Ivy Bridge hardware when using vanilla
>> > 3.17.1 kernel.
>> > I guess it's due to batchbuffer security and patch to disable
>> > batchbuffer security is required, but guys, it fails since 3.16, and
>> > 3.17 was released quite a while ago.
>> >
>> > Could you cooperate with i915 driver devs to make Beignet working on
>> > vanilla kernel without extra patches?
>> >
>> > Thanks!
>> >
>> > Regards,
>> > Vasily
>> > _______________________________________________
>> > Beignet mailing list
>> > Beignet@lists.freedesktop.org
>> > http://lists.freedesktop.org/mailman/listinfo/beignet
>> _______________________________________________
>> Intel-gfx mailing list
>> Intel-gfx@lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/beignet

Re: [Intel-gfx] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Zhigang Gong]

Hi,

Luxmark (both 2.0/2.1) works fine on my IVB machine. The back trace
you provided below doesn't indicate it's a beignet related problem.
It hadn't enter beignet domain and just crashed in luxmark internal.

On Fri, Oct 24, 2014 at 12:04:29PM +0300, Vasily Khoruzhick wrote:
> Hi Zhigang,
> 
> Luxmark crashes with following backtrace:
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
> (gdb) bt
> #0  0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
> #1  0x0000000000482236 in slg::RenderEngine::Stop() ()
> #2  0x000000000047be9b in slg::RenderSession::~RenderSession() ()
> #3  0x0000000000468a77 in LuxMarkApp::Stop() ()
> #4  0x0000000000468b46 in LuxMarkApp::InitRendering(LuxMarkAppMode,
> char const*) ()

After a quick analysis, I confirm that the second case is indeed a beignet
bug. Beignet lacks of some llvm intrinsics support such as
llvm.uadd.with.overflow.i32().

Will fix it next week.

Thanks,
Zhigang Gong.

> 
> And convert from imagemagick crashes in libgbe.so:
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x00007fffeb0df18c in gbe::buildRegInfo(gbe::ir::BasicBlock&,
> gbe::vector<gbe::RegInfoForMov>&) () from /usr/lib/beignet//libgbe.so
> (gdb) bt
> #0  0x00007fffeb0df18c in gbe::buildRegInfo(gbe::ir::BasicBlock&,
> gbe::vector<gbe::RegInfoForMov>&) () from /usr/lib/beignet//libgbe.so
> #1  0x00007fffeb0e048d in
> gbe::GenWriter::removeLOADIs(gbe::ir::Liveness const&,
> gbe::ir::Function&) () from /usr/lib/beignet//libgbe.so
> #2  0x00007fffeb0fb823 in
> gbe::GenWriter::emitFunction(llvm::Function&) () from
> /usr/lib/beignet//libgbe.so
> #3  0x00007fffeb100b63 in

Re: [Intel-gfx] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Vasily Khoruzhick]

Hi Zhigang,

On Fri, Oct 24, 2014 at 12:13 PM, Zhigang Gong
<zhigang.gong@linux.intel.com> wrote:
> Hi,
>
> Luxmark (both 2.0/2.1) works fine on my IVB machine. The back trace
> you provided below doesn't indicate it's a beignet related problem.
> It hadn't enter beignet domain and just crashed in luxmark internal.

I'm testing with Luxmark-1.3.1. Luxmark 2.0 works fine.

> On Fri, Oct 24, 2014 at 12:04:29PM +0300, Vasily Khoruzhick wrote:
>> Hi Zhigang,
>>
>> Luxmark crashes with following backtrace:
>>
>> Program received signal SIGSEGV, Segmentation fault.
>> 0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
>> (gdb) bt
>> #0  0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
>> #1  0x0000000000482236 in slg::RenderEngine::Stop() ()
>> #2  0x000000000047be9b in slg::RenderSession::~RenderSession() ()
>> #3  0x0000000000468a77 in LuxMarkApp::Stop() ()
>> #4  0x0000000000468b46 in LuxMarkApp::InitRendering(LuxMarkAppMode,
>> char const*) ()
>
> After a quick analysis, I confirm that the second case is indeed a beignet
> bug. Beignet lacks of some llvm intrinsics support such as
> llvm.uadd.with.overflow.i32().
>
> Will fix it next week.

Ok, thank you!

> Thanks,
> Zhigang Gong.

Regards
Vasily

Re: [Intel-gfx] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Zhigang Gong]

Vasily,

Could you try ImageMagick(convert) again with latest git master
beignet(git-e46764f). It should work now.

Thanks,
Zhigang Gong.

On Fri, Oct 24, 2014 at 04:36:49PM +0300, Vasily Khoruzhick wrote:
> Hi Zhigang,
> 
> On Fri, Oct 24, 2014 at 12:13 PM, Zhigang Gong
> <zhigang.gong@linux.intel.com> wrote:
> > Hi,
> >
> > Luxmark (both 2.0/2.1) works fine on my IVB machine. The back trace
> > you provided below doesn't indicate it's a beignet related problem.
> > It hadn't enter beignet domain and just crashed in luxmark internal.
> 
> I'm testing with Luxmark-1.3.1. Luxmark 2.0 works fine.
> 
> > On Fri, Oct 24, 2014 at 12:04:29PM +0300, Vasily Khoruzhick wrote:
> >> Hi Zhigang,
> >>
> >> Luxmark crashes with following backtrace:
> >>
> >> Program received signal SIGSEGV, Segmentation fault.
> >> 0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
> >> (gdb) bt
> >> #0  0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
> >> #1  0x0000000000482236 in slg::RenderEngine::Stop() ()
> >> #2  0x000000000047be9b in slg::RenderSession::~RenderSession() ()
> >> #3  0x0000000000468a77 in LuxMarkApp::Stop() ()
> >> #4  0x0000000000468b46 in LuxMarkApp::InitRendering(LuxMarkAppMode,
> >> char const*) ()
> >
> > After a quick analysis, I confirm that the second case is indeed a beignet
> > bug. Beignet lacks of some llvm intrinsics support such as
> > llvm.uadd.with.overflow.i32().
> >
> > Will fix it next week.
> 
> Ok, thank you!
> 
> > Thanks,
> > Zhigang Gong.
> 
> Regards
> Vasily
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/beignet

Re: [Intel-gfx] Beignet crashes on vanilla 3.17.1 with IVB hardware

[Vasily Khoruzhick]

Hi Zhigang,

On Tue, Oct 28, 2014 at 7:31 AM, Zhigang Gong
<zhigang.gong@linux.intel.com> wrote:
> Vasily,
>
> Could you try ImageMagick(convert) again with latest git master
> beignet(git-e46764f). It should work now.

It works fine now. Thank you!

Regards,
Vasily

>
> Thanks,
> Zhigang Gong.
>
> On Fri, Oct 24, 2014 at 04:36:49PM +0300, Vasily Khoruzhick wrote:
>> Hi Zhigang,
>>
>> On Fri, Oct 24, 2014 at 12:13 PM, Zhigang Gong
>> <zhigang.gong@linux.intel.com> wrote:
>> > Hi,
>> >
>> > Luxmark (both 2.0/2.1) works fine on my IVB machine. The back trace
>> > you provided below doesn't indicate it's a beignet related problem.
>> > It hadn't enter beignet domain and just crashed in luxmark internal.
>>
>> I'm testing with Luxmark-1.3.1. Luxmark 2.0 works fine.
>>
>> > On Fri, Oct 24, 2014 at 12:04:29PM +0300, Vasily Khoruzhick wrote:
>> >> Hi Zhigang,
>> >>
>> >> Luxmark crashes with following backtrace:
>> >>
>> >> Program received signal SIGSEGV, Segmentation fault.
>> >> 0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
>> >> (gdb) bt
>> >> #0  0x00000000004cd8b0 in slg::PathOCLRenderEngine::StopLockLess() ()
>> >> #1  0x0000000000482236 in slg::RenderEngine::Stop() ()
>> >> #2  0x000000000047be9b in slg::RenderSession::~RenderSession() ()
>> >> #3  0x0000000000468a77 in LuxMarkApp::Stop() ()
>> >> #4  0x0000000000468b46 in LuxMarkApp::InitRendering(LuxMarkAppMode,
>> >> char const*) ()
>> >
>> > After a quick analysis, I confirm that the second case is indeed a beignet
>> > bug. Beignet lacks of some llvm intrinsics support such as
>> > llvm.uadd.with.overflow.i32().
>> >
>> > Will fix it next week.
>>
>> Ok, thank you!
>>
>> > Thanks,
>> > Zhigang Gong.
>>
>> Regards
>> Vasily
>> _______________________________________________
>> Intel-gfx mailing list
>> Intel-gfx@lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/beignet