Is there any way to debug driver?

[Guan, Xiao-Feng]

Hello,

 

  I'm AMD driver developer and would like to ask a question. This issue
blocks us from making a better driver for WINE.

 

  Recently we have difficulty in debugging our driver with WINE. We used
to start wine-pthread in gdb to debug the applications. Or else we
cannot break into our driver. However, now we cannot build this
executable any more in wine-1.1.23 or git repository. I wonder if there
is any other method to get all symbols right in gdb.

 

  As we can see that, from version 1.1.23, all application fails to
start on AMD card. We are going to investigate why it happens. If it is
necessary, Would you please let us know a little more about the changes
of this version?

 

Thanks & regards.

 

Guan Xiaofeng

OpenGL Software Engineer, 

Graphic Product Group

Shanghai R&D Center, AMD Cop.

Xiao-Feng.Guan(a)amd.com <mailto:Xiao-Feng.Guan(a)amd.com> 

Tel [phone] ext 25746

[Henri Verbeet]

2009/6/17 Guan, Xiao-Feng <Xiao-Feng.Guan(a)amd.com>:
>   As we can see that, from version 1.1.23, all application fails to start on
> AMD card. We are going to investigate why it happens. If it is necessary,
> Would you please let us know a little more about the changes of this
> version?

There's a bug for that specific issue filed at
http://ati.cchtml.com/show_bug.cgi?id=1571. It contains an explanation
and test case. Current git versions of Wine avoid the bug by just not
attaching compressed and depth formats to an FBO, since they're not
supposed to be color-renderable anyway.

I'm afraid I can't help much with getting gdb working with Wine, I
don't use/like debuggers much. I think you're supposed to use winedbg
for Wine debugging though.

Henri

[Wang, Robin]

We also have tried using winedbg, but it cannot break into our driver either.

The only way we can break into our driver before is using "gdb wine-pthread", but now it is not available.

Do you have some suggestion on winedbg configuration to make it able to break into graphics drivers.

Thanks

-----Original Message-----
From: Henri Verbeet [mailto:hverbeet(a)gmail.com] 
Sent: Wednesday, June 17, 2009 9:27 PM
To: Guan, Xiao-Feng
Cc: wine-devel(a)winehq.org; Wang, Robin; Zhou, Jesse; Jin, Jian-Rong; Sun, Sunny; Boudier, Pierre
Subject: Re: Is there any way to debug driver?

2009/6/17 Guan, Xiao-Feng <Xiao-Feng.Guan(a)amd.com>:
>   As we can see that, from version 1.1.23, all application fails to start on
> AMD card. We are going to investigate why it happens. If it is necessary,
> Would you please let us know a little more about the changes of this
> version?

There's a bug for that specific issue filed at
http://ati.cchtml.com/show_bug.cgi?id=1571. It contains an explanation
and test case. Current git versions of Wine avoid the bug by just not
attaching compressed and depth formats to an FBO, since they're not
supposed to be color-renderable anyway.

I'm afraid I can't help much with getting gdb working with Wine, I
don't use/like debuggers much. I think you're supposed to use winedbg
for Wine debugging though.

Henri

[Roderick Colenbrander]

Have you tried: winedbg --gdb appname.exe? It will use gdb then and
still see win32 symbols I think.

Roderick


On Wed, Jun 17, 2009 at 5:49 PM, Wang, Robin<Robin.Wang(a)amd.com> wrote:
> We also have tried using winedbg, but it cannot break into our driver either.
>
> The only way we can break into our driver before is using "gdb wine-pthread", but now it is not available.
>
> Do you have some suggestion on winedbg configuration to make it able to break into graphics drivers.
>
> Thanks
>
> -----Original Message-----
> From: Henri Verbeet [mailto:hverbeet(a)gmail.com]
> Sent: Wednesday, June 17, 2009 9:27 PM
> To: Guan, Xiao-Feng
> Cc: wine-devel(a)winehq.org; Wang, Robin; Zhou, Jesse; Jin, Jian-Rong; Sun, Sunny; Boudier, Pierre
> Subject: Re: Is there any way to debug driver?
>
> 2009/6/17 Guan, Xiao-Feng <Xiao-Feng.Guan(a)amd.com>:
> >   As we can see that, from version 1.1.23, all application fails to start on
> > AMD card. We are going to investigate why it happens. If it is necessary,
> > Would you please let us know a little more about the changes of this
> > version?
>
> There's a bug for that specific issue filed at
> http://ati.cchtml.com/show_bug.cgi?id=1571. It contains an explanation
> and test case. Current git versions of Wine avoid the bug by just not
> attaching compressed and depth formats to an FBO, since they're not
> supposed to be color-renderable anyway.
>
> I'm afraid I can't help much with getting gdb working with Wine, I
> don't use/like debuggers much. I think you're supposed to use winedbg
> for Wine debugging though.
>
> Henri

[Stefan Dösinger]

Am Wednesday 17 June 2009 21:56:14 schrieb Roderick Colenbrander:
> Have you tried: winedbg --gdb appname.exe? It will use gdb then and
> still see win32 symbols I think.
> You can also start the app, put a Sleep(10000) somewhere into the early 
> wined3d init code, then attach gdb to the running process

[Ken Thomases]

On Jun 17, 2009, at 10:49 AM, Wang, Robin wrote:
> We also have tried using winedbg, but it cannot break into our  
> driver either.
>
> The only way we can break into our driver before is using "gdb wine- 
> pthread", but now it is not available.
>
> Do you have some suggestion on winedbg configuration to make it able  
> to break into graphics drivers.

As already mentioned by Ben Klein, just try "gdb wine" instead of "gdb  
wine-pthread".  The executable name has changed.  That's all.

Cheers,
Ken

[Guan, Xiao-Feng]

The gdb way always turns into some SIGTRAP signal from preloader and
finally terminates the loading process. It seems that the WINE binary
loader would not co-work with gdb.

We will try another way -- the WINE visual tests, to see if it is enough
for work.

Thanks very much.

Guan Xiaofeng
AMD Shanghai SW OpenGL Team
021-61601838-25746
Xiao-Feng.Guan(a)amd.com
 
 
-----Original Message-----
From: Ken Thomases [mailto:ken(a)codeweavers.com] 
Sent: Thursday, June 18, 2009 11:34 PM
To: Wang, Robin; wine-devel
Cc: Guan, Xiao-Feng; Boudier, Pierre; Zhou, Jesse; Jin, Jian-Rong; Sun,
Sunny
Subject: Re: Is there any way to debug driver?

On Jun 17, 2009, at 10:49 AM, Wang, Robin wrote:
> We also have tried using winedbg, but it cannot break into our  
> driver either.
>
> The only way we can break into our driver before is using "gdb wine- 
> pthread", but now it is not available.
>
> Do you have some suggestion on winedbg configuration to make it able  
> to break into graphics drivers.

As already mentioned by Ben Klein, just try "gdb wine" instead of "gdb  
wine-pthread".  The executable name has changed.  That's all.

Cheers,
Ken

[Roderick Colenbrander]

At what sort of points did you try to set breakpoints before? Note
that we are loading opengl dynamically which might make debugging a
little bit harder. What about plain winedbg? Today I tried 'winedbg
--gdb notepad' and indeed had some issues but plain winedbg worked
properly. Further 'gdb wine notepad' also worked properly except that
for some reason my back trace didn't include debug symbols.

It might help to explain what commands you are exactly using and what
sort of break points you attempt to set. Then we might be able to help
you better.

Roderick

On Mon, Jun 22, 2009 at 8:52 AM, Guan, Xiao-Feng<Xiao-Feng.Guan(a)amd.com> wrote:
> The gdb way always turns into some SIGTRAP signal from preloader and
> finally terminates the loading process. It seems that the WINE binary
> loader would not co-work with gdb.
>
> We will try another way -- the WINE visual tests, to see if it is enough
> for work.
>
> Thanks very much.
>
> Guan Xiaofeng
> AMD Shanghai SW OpenGL Team
> 021-61601838-25746
> Xiao-Feng.Guan(a)amd.com
>
> -----Original Message-----
> From: Ken Thomases [mailto:ken(a)codeweavers.com]
> Sent: Thursday, June 18, 2009 11:34 PM
> To: Wang, Robin; wine-devel
> Cc: Guan, Xiao-Feng; Boudier, Pierre; Zhou, Jesse; Jin, Jian-Rong; Sun,
> Sunny
> Subject: Re: Is there any way to debug driver?
>
> On Jun 17, 2009, at 10:49 AM, Wang, Robin wrote:
> > We also have tried using winedbg, but it cannot break into our
> > driver either.
> >
> > The only way we can break into our driver before is using "gdb wine-
> > pthread", but now it is not available.
> >
> > Do you have some suggestion on winedbg configuration to make it able
> > to break into graphics drivers.
>
> As already mentioned by Ben Klein, just try "gdb wine" instead of "gdb
> wine-pthread".  The executable name has changed.  That's all.
>
> Cheers,
> Ken

[Wang, Robin]

Hi Roderick,

Using winedbg, it is OK for us to break into libGL.so. But we cannot set break point in our dri drivers, which is loaded by libGL.so using dlopen().

Thanks,
Robin

-----Original Message-----
From: Roderick Colenbrander [mailto:thunderbird2k(a)gmail.com] 
Sent: Monday, June 22, 2009 8:15 PM
To: Guan, Xiao-Feng
Cc: Ken Thomases; Wang, Robin; wine-devel; Boudier, Pierre; Zhou, Jesse; Jin, Jian-Rong; Sun, Sunny
Subject: Re: Is there any way to debug driver?

At what sort of points did you try to set breakpoints before? Note
that we are loading opengl dynamically which might make debugging a
little bit harder. What about plain winedbg? Today I tried 'winedbg
--gdb notepad' and indeed had some issues but plain winedbg worked
properly. Further 'gdb wine notepad' also worked properly except that
for some reason my back trace didn't include debug symbols.

It might help to explain what commands you are exactly using and what
sort of break points you attempt to set. Then we might be able to help
you better.

Roderick

On Mon, Jun 22, 2009 at 8:52 AM, Guan, Xiao-Feng<Xiao-Feng.Guan(a)amd.com> wrote:
> The gdb way always turns into some SIGTRAP signal from preloader and
> finally terminates the loading process. It seems that the WINE binary
> loader would not co-work with gdb.
>
> We will try another way -- the WINE visual tests, to see if it is enough
> for work.
>
> Thanks very much.
>
> Guan Xiaofeng
> AMD Shanghai SW OpenGL Team
> 021-61601838-25746
> Xiao-Feng.Guan(a)amd.com
>
> -----Original Message-----
> From: Ken Thomases [mailto:ken(a)codeweavers.com]
> Sent: Thursday, June 18, 2009 11:34 PM
> To: Wang, Robin; wine-devel
> Cc: Guan, Xiao-Feng; Boudier, Pierre; Zhou, Jesse; Jin, Jian-Rong; Sun,
> Sunny
> Subject: Re: Is there any way to debug driver?
>
> On Jun 17, 2009, at 10:49 AM, Wang, Robin wrote:
> > We also have tried using winedbg, but it cannot break into our
> > driver either.
> >
> > The only way we can break into our driver before is using "gdb wine-
> > pthread", but now it is not available.
> >
> > Do you have some suggestion on winedbg configuration to make it able
> > to break into graphics drivers.
>
> As already mentioned by Ben Klein, just try "gdb wine" instead of "gdb
> wine-pthread".  The executable name has changed.  That's all.
>
> Cheers,
> Ken

[Eric Pouech]

Wang, Robin a écrit :
> Hi Roderick,
>
> Using winedbg, it is OK for us to break into libGL.so. But we cannot set break point in our dri drivers, which is loaded by libGL.so using dlopen().

please send me the output of running
WINEDEBUG=+dbghelp wine ....

and then at debugger prompt

Wine-dbg> break your_func_in_dri_driver
Wine-dbg> cont

TIA

-- 
Eric Pouech
"The problem with designing something completely foolproof is to underestimate the ingenuity of a complete idiot." (Douglas Adams)

[Roderick Colenbrander]

It was just mentioned on IRC that the environment variable
WINELOADERNOEXEC=1 needs to be set. After that you can just run 'gdb
wine' and continue the same way as using wine-preloader.

Roderick

On Mon, Jun 22, 2009 at 10:24 PM, Eric Pouech<eric.pouech(a)orange.fr> wrote:
> Wang, Robin a écrit :
> > Hi Roderick,
> >
> > Using winedbg, it is OK for us to break into libGL.so. But we cannot set
> > break point in our dri drivers, which is loaded by libGL.so using dlopen().
>
> please send me the output of running
> WINEDEBUG=+dbghelp wine ....
>
> and then at debugger prompt
>
> Wine-dbg> break your_func_in_dri_driver
> Wine-dbg> cont
>
> TIA
>
> --
> Eric Pouech
> "The problem with designing something completely foolproof is to
> underestimate the ingenuity of a complete idiot." (Douglas Adams)

[Guan, Xiao-Feng]

It works.
Thanks Roderick!

Guan Xiaofeng
AMD Shanghai SW OpenGL Team
021-61601838-25746
Xiao-Feng.Guan(a)amd.com
 
 

-----Original Message-----
From: Roderick Colenbrander [mailto:thunderbird2k(a)gmail.com] 
Sent: Wednesday, June 24, 2009 12:23 AM
To: Eric Pouech
Cc: Wang, Robin; Guan, Xiao-Feng; wine-devel; Zhou, Jesse; Sun, Sunny; Boudier, Pierre; Jin, Jian-Rong
Subject: Re: Is there any way to debug driver?

It was just mentioned on IRC that the environment variable
WINELOADERNOEXEC=1 needs to be set. After that you can just run 'gdb
wine' and continue the same way as using wine-preloader.

Roderick

On Mon, Jun 22, 2009 at 10:24 PM, Eric Pouech<eric.pouech(a)orange.fr> wrote:
> Wang, Robin a écrit :
> > Hi Roderick,
> >
> > Using winedbg, it is OK for us to break into libGL.so. But we cannot set
> > break point in our dri drivers, which is loaded by libGL.so using dlopen().
>
> please send me the output of running
> WINEDEBUG=+dbghelp wine ....
>
> and then at debugger prompt
>
> Wine-dbg> break your_func_in_dri_driver
> Wine-dbg> cont
>
> TIA
>
> --
> Eric Pouech
> "The problem with designing something completely foolproof is to
> underestimate the ingenuity of a complete idiot." (Douglas Adams)

[Ben Klein]

2009/6/17 Guan, Xiao-Feng <Xiao-Feng.Guan(a)amd.com>:
> Hello,
>
>   I’m AMD driver developer and would like to ask a question. This issue
> blocks us from making a better driver for WINE.
>
>   Recently we have difficulty in debugging our driver with WINE. We used to
> start wine-pthread in gdb to debug the applications. Or else we cannot break
> into our driver. However, now we cannot build this executable any more in
> wine-1.1.23 or git repository. I wonder if there is any other method to get
> all symbols right in gdb.

wine-pthread and wine-kthread got rolled together into a single "wine"
loader a few versions ago.

[Damjan Jovanovic]

On Wed, Jun 17, 2009 at 11:01 AM, Guan, Xiao-Feng<Xiao-Feng.Guan(a)amd.com> wrote:
> Hello,
>
>   As we can see that, from version 1.1.23, all application fails to start on
> AMD card. We are going to investigate why it happens. If it is necessary,
> Would you please let us know a little more about the changes of this
> version?

If it worked for you before and doesn't work in 1.1.23, it sounds like
you should do a regression test to find the patch that broke it:
http://wiki.winehq.org/RegressionTesting
> Thanks & regards.
>
> Guan Xiaofeng
>
> OpenGL Software Engineer,
>
> Graphic Product Group
>
> Shanghai R&D Center, AMD Cop.
>
> Xiao-Feng.Guan(a)amd.com
>
> Tel [phone] ext 25746

Regards and good luck
Damjan Jovanovic

[Roderick Colenbrander]

Well as Henri mentioned a bug report has been posted to the
(un)official amd bugzilla for it with details on the crash. It was
just related to the enabling of FBOs by default and some FBO related
tests.

Regarding debugging I would try to use winedbg and if it isn't
sufficient winedbg offers a gdb proxy method which could be useful as
well. Further it might also be useful to use the Wine ddraw/d3d8/d3d9
test suites. The tests consist of both conformance tests to windows
behavior and they also test tricky d3d functionality which popular
games require. Various tests fail at this point on the AMD drivers (on
windows some also don't always work) and before (on catalyst <=9.3 or
it was <=9.2) some tests even caused a system crash. It would be
useful if the tests could be run as they illustrate a lot of tricky
d3d9 situations.

Regards,
Roderick Colenbrander

On Wed, Jun 17, 2009 at 5:38 PM, Damjan Jovanovic<damjan.jov(a)gmail.com> wrote:
> On Wed, Jun 17, 2009 at 11:01 AM, Guan, Xiao-Feng<Xiao-Feng.Guan(a)amd.com> wrote:
> > Hello,
> >
> >   As we can see that, from version 1.1.23, all application fails to start on
> > AMD card. We are going to investigate why it happens. If it is necessary,
> > Would you please let us know a little more about the changes of this
> > version?
>
> If it worked for you before and doesn't work in 1.1.23, it sounds like
> you should do a regression test to find the patch that broke it:
> http://wiki.winehq.org/RegressionTesting
> > Thanks & regards.
> >
> > Guan Xiaofeng
> >
> > OpenGL Software Engineer,
> >
> > Graphic Product Group
> >
> > Shanghai R&D Center, AMD Cop.
> >
> > Xiao-Feng.Guan(a)amd.com
> >
> > Tel [phone] ext 25746
>
> Regards and good luck
> Damjan Jovanovic