[gentoo-user] Glxgears crashes X

[gentoo-user] Glxgears crashes X

[Philip Webb]

Another day of probing finally got DRI working on my new machine:
the basic problem was that Mesa needed recompiling with 'i810' enabled.

Naturally (wry smile), this led to another show-stopper:
X starts with no errors in the log, but 'glxgears' & 'glxinfo' crash X.
There is a brief "back-trace", but it shows nothing immediately useful.
There are many reports via Google & Gentoo Forum of similar experiences,
some of them as recent as Aug-Sep 2007, all involving Nvidia or Radeon,
but none revealing a cause, let alone a solution.

Might the problem involve a conflict between kernel/Xorg modules ?
Is there some required combination of entries in  xorg.conf ? --
'man xorg.conf' is not the most helpful of documents,
it not being clear what many of the options etc really do.

The processor is an Intel G33 using drivers 'i810' & 'i915' ;
kernel 2.6.22-r8 ; 64-bit system .

Has anyone else encountered this or have any suggestions ?

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071027 Philip Webb wrote:
> X starts with no errors in the log, but 'glxgears' & 'glxinfo' crash X.
> The processor is an Intel G33 using drivers 'i810' & 'i915' ;
> kernel 2.6.22-r8 ; 64-bit system .

After sleeping on the problem & reading Forum discussion re 'x11-drm',
I've tried a few more things, none of which makes any difference:
permutating some options under 'Device' in  xorg.conf ,
checking all relevant items are enabled in the kernel
& finally unmerging 'x11-drm', which seems rather flakey
& whose removal had not the slightest effect (wry smile).

However, I did notice  2  lines on the screen after the crashes:

  Unrecognised deviceID 29c2
  backtrace ...
  ... /usr/lib64/dri/i915_dri.so ...  (provided by pkg 'mesa')

This suggests there is some very basic problem re recognising acceleration
on the G33 , caused by a defect in the driver(s) being used.

Can anyone confirm that this is the correct diagnosis ?
It looks like material for a bug report, but Gentoo or LKML ?

A lot of people have been reporting difficulties getting DRI to work here,
tho' only a few seem to have encountered my problem of a crash out of X.

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Benno Schulenberg]

Philip Webb wrote:
>   Unrecognised deviceID 29c2
>   backtrace ...
>   ... /usr/lib64/dri/i915_dri.so ...  (provided by pkg 'mesa')

What versions of Mesa and xf86-video-i810 are you running?  And what 
is in the Section "Device" of your /etc/X11/xorg.conf?

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071029 Benno Schulenberg wrote:
> Philip Webb wrote:
>>   Unrecognised deviceID 29c2
>>   backtrace ...
>>   ... /usr/lib64/dri/i915_dri.so ...  (provided by pkg 'mesa')
> What versions of Mesa and xf86-video-i810 are you running?

  'mesa-progs-6.5.2' (latest available).
  'xf86-video-i810-2.1.0' ( 2.1.1 is in testing (~)).

> And what is in the Section "Device" of your /etc/X11/xorg.conf?

  Section "Device"
      ### Available Driver options are:-
      ### Values: <i>: integer, <f>: float, <bool>: "True"/"False",
      ### <string>: "String", <freq>: "<f> Hz/kHz/MHz"
      ### [arg]: arg optional
      Option     "Accel" "On"           # [<bool>]
      #Option     "SWcursor"           	# [<bool>]
      #Option     "ColorKey"           	# <i>
      #Option     "CacheLines"         	# <i>
      #Option     "Dac6Bit"            	# [<bool>]
      #Option      "DRI" "true"         # [<bool>]
      #Option     "NoDDC"              	# [<bool>]
      #Option     "ShowCache"          	# [<bool>]
      #Option     "XvMCSurfaces"       	# <i>
      #Option     "PageFlip"           	# [<bool>]
 	Identifier  "Card0"
 	Driver      "intel"
 	VendorName  "Intel Corporation"
 	BoardName   "Integrated Graphics Controller"
 	BusID       "PCI:0:2:0"
  EndSection

This is the result of 'X -config' (with a small format change).
I tried permutations of Options suggested on Forum posts without success
(I haven't asked on the Forum myself): eg adding 'Option "DRI" "true"',
which seems merely to duplicate "Accel" above.

I am a bit confused by the requirement for  2  drivers  i810  i915 .
The same dir has  i810_dri.so  i915tex_dri.so  i965_dri.so  (all from Mesa).
Is it using the wrong driver ?  Is there a way to test the others ?

Thanks again for your response.

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Benno Schulenberg]

Philip Webb wrote:
> 071029 Benno Schulenberg wrote:
> > What versions of Mesa and xf86-video-i810 are you running?
>
>   'mesa-progs-6.5.2' (latest available).

Get the one from testing, 7.0.1.  But what you need is 
media-libs/mesa; mesa-progs is just glxinfo/glxgears.

>   'xf86-video-i810-2.1.0' ( 2.1.1 is in testing (~)).

Get the newest one, as there's much development in the intel driver.

>   Section "Device"
> [...]
>  	Driver      "intel"

Looks good.

> I tried permutations of Options suggested on Forum posts without
> success (I haven't asked on the Forum myself): eg adding 'Option
> "DRI" "true"', which seems merely to duplicate "Accel" above.

The man page of i810 says only NoAccel and DRI exist, and both 
default to "use it", so your settings don't change anything.

> I am a bit confused by the requirement for  2  drivers  i810 
> i915 . The same dir has  i810_dri.so  i915tex_dri.so  i965_dri.so
>  (all from Mesa). Is it using the wrong driver ?

No, it should auto-detect which of those drivers it needs.  It's 
just that your lib is too old for the newer G33,

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071029 Benno Schulenberg wrote:
> you need media-libs/mesa; mesa-progs is just glxinfo/glxgears.
>>   'xf86-video-i810-2.1.0' ( 2.1.1 is in testing (~)).
> Get the newest one, as there's much development in the intel driver.

I was using 'mesa' + 'mesa-progs' (read/write error by me).

I updated to the latest versions  7.0.1  7.0.1  2.1.1  & rebooted
& there is no change: X crashes as soon as 'glxinfo' starts
with the same error message "unrecognised deviceID 29c2"
while using  /usr/lib64/dri/i915_dri.so .

I note your comment re continuing development
& suspect it hasn't caught up with the hardware yet (smile).
Is there any way of testing the other  3  versions in the dir,
which I listed in the previous msg (one is 'i915tex') ?

I'ld like to get it working, if possible,
but software acceleration with Core 2 Duo & Intel G33 is faster
than hardware acceleration with Athlon 2500+ & Nvidia MX400 (current box):
former  1491 - 1530  vs  latter  1212 fps .
An upto-date Nvidia EN8500GT card costs  CAD 104 + tax ,
too much merely for Xscreensavers, but affordable if needed otherwise.

Your & others' further comments & suggestions are very welcome.

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Benno Schulenberg]

Philip Webb wrote:
> I updated to the latest versions  7.0.1  7.0.1  2.1.1  & rebooted
> & there is no change: X crashes as soon as 'glxinfo' starts
> with the same error message "unrecognised deviceID 29c2"
> while using  /usr/lib64/dri/i915_dri.so .

What kernel version are you running?  Try updating to the most 
recent one.  If then you still get "unrecognised deviceID 29c2", 
build the kernel modules from source (just the 1.7 DRM part):

  http://dri.freedesktop.org/wiki/Building

> Is there any way of testing the other  3  versions in the dir,
> which I listed in the previous msg (one is 'i915tex') ?

There's no need.  It knows which one to choose.  :)

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071030 Benno Schulenberg wrote:
> What kernel version are you running?  Try updating to the most recent.
> If then you still get "unrecognised deviceID 29c2", 
> build the kernel modules from source (just the 1.7 DRM part):
>   http://dri.freedesktop.org/wiki/Building

The kernel is 2.6.22-r8 : I will try 2.6.23 tomorrow.
I suspect the driver(s) haven't caught up with the hardware yet.

I checked the wiki paragraph & the subdir it refers to
& the modules are already built & 'lsmod' shows them installed:
is there any sense in rebuilding them ?

>> Is there any way of testing the other  3  versions in the dir,
>> which I listed in the previous msg (one is 'i915tex') ?
> There's no need.  It knows which one to choose.  :)

That's good to know (smile).

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Benno Schulenberg]

Philip Webb wrote:
> The kernel is 2.6.22-r8 : I will try 2.6.23 tomorrow.

No need, sorry, I was pointing you in the wrong direction.  Kernel 
2.6.22 already knows about 29c2 for i915.

> I checked the wiki paragraph & the subdir it refers to
> & the modules are already built & 'lsmod' shows them installed:
> is there any sense in rebuilding them ?

No.  But do build the Mesa modules from git, the 1.2.2 and 1.4 part:

  http://dri.freedesktop.org/wiki/Building

These three lines were added to i915/intel_context.h since 7.0.1:

#define PCI_CHIP_G33_G                  0x29C2
#define PCI_CHIP_Q35_G                  0x29B2
#define PCI_CHIP_Q33_G                  0x29D2

So, time to install git and do a big pull.  :)

(Don't build libdrm, the x11-libs/libdrm-2.3.0 you have is fine.)

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[James Ausmus]

[-- Attachment #1: Type: text/plain, Size: 1271 bytes --]

On 10/30/07, Benno Schulenberg <benno.schulenberg@gmail.com> wrote:
> Philip Webb wrote:
> > The kernel is 2.6.22-r8 : I will try 2.6.23 tomorrow.
>
> No need, sorry, I was pointing you in the wrong direction.  Kernel
> 2.6.22 already knows about 29c2 for i915.
>
> > I checked the wiki paragraph & the subdir it refers to
> > & the modules are already built & 'lsmod' shows them installed:
> > is there any sense in rebuilding them ?
>
> No.  But do build the Mesa modules from git, the 1.2.2 and 1.4 part:
>
>   http://dri.freedesktop.org/wiki/Building
>
> These three lines were added to i915/intel_context.h since 7.0.1:
>
> #define PCI_CHIP_G33_G                  0x29C2
> #define PCI_CHIP_Q35_G                  0x29B2
> #define PCI_CHIP_Q33_G                  0x29D2
>
> So, time to install git and do a big pull.  :)

You don't have to go to all that, the media-libs/mesa-7.0.1 works fine
*IF* you apply the attached patch (either hack the ebuild, or CTRL-Z
immediately after emerge gets done unpacking the source). I had the
same problem you did, did the Googling, found the patch, and
everything works fine for me now.

HTH

-James


>
> (Don't build libdrm, the x11-libs/libdrm-2.3.0 you have is fine.)
>
> Benno
> --
> gentoo-user@gentoo.org mailing list
>
>

[-- Attachment #2: g33.patch --]
[-- Type: application/octet-stream, Size: 3889 bytes --]

diff --git a/src/mesa/drivers/dri/i915/i915_texstate.c b/src/mesa/drivers/dri/i915/i915_texstate.c
index 9f0c949..a19d4b6 100644
--- a/src/mesa/drivers/dri/i915/i915_texstate.c
+++ b/src/mesa/drivers/dri/i915/i915_texstate.c
@@ -491,12 +491,19 @@ static void i915SetTexImages( i915ContextPtr i915,
       abort();
    }
 
-
-   if (i915->intel.intelScreen->deviceID == PCI_CHIP_I945_G ||
-       i915->intel.intelScreen->deviceID == PCI_CHIP_I945_GM)
-      i945LayoutTextureImages( i915, tObj );	 
-   else
-      i915LayoutTextureImages( i915, tObj );
+   switch (i915->intel.intelScreen->deviceID) {
+   case PCI_CHIP_I945_G:
+   case PCI_CHIP_I945_GM:
+   case PCI_CHIP_I945_GME:
+   case PCI_CHIP_G33_G:
+   case PCI_CHIP_Q33_G:
+   case PCI_CHIP_Q35_G:
+       i945LayoutTextureImages( i915, tObj );
+       break;
+   default:
+       i915LayoutTextureImages( i915, tObj );
+       break;
+   }
 
    t->Setup[I915_TEXREG_MS3] = 
       (((tObj->Image[0][t->intel.base.firstLevel]->Height - 1) << MS3_HEIGHT_SHIFT) |
diff --git a/src/mesa/drivers/dri/i915/intel_context.c b/src/mesa/drivers/dri/i915/intel_context.c
index e747fc6..11c23f2 100644
--- a/src/mesa/drivers/dri/i915/intel_context.c
+++ b/src/mesa/drivers/dri/i915/intel_context.c
@@ -123,6 +123,14 @@ const GLubyte *intelGetString( GLcontext *ctx, GLenum name )
 	 chipset = "Intel(R) 945G"; break;
       case PCI_CHIP_I945_GM:
 	 chipset = "Intel(R) 945GM"; break;
+      case PCI_CHIP_I945_GME:
+	 chipset = "Intel(R) 945GME"; break;
+      case PCI_CHIP_G33_G:
+	 chipset = "Intel(R) G33"; break;
+      case PCI_CHIP_Q35_G:
+	 chipset = "Intel(R) Q35"; break;
+      case PCI_CHIP_Q33_G:
+	 chipset = "Intel(R) Q33"; break;
       default:
 	 chipset = "Unknown Intel Chipset"; break;
       }
diff --git a/src/mesa/drivers/dri/i915/intel_context.h b/src/mesa/drivers/dri/i915/intel_context.h
index c48b074..3b50107 100644
--- a/src/mesa/drivers/dri/i915/intel_context.h
+++ b/src/mesa/drivers/dri/i915/intel_context.h
@@ -454,6 +454,10 @@ extern int INTEL_DEBUG;
 #define PCI_CHIP_I915_GM		0x2592
 #define PCI_CHIP_I945_G			0x2772
 #define PCI_CHIP_I945_GM		0x27A2
+#define PCI_CHIP_I945_GME		0x27AE
+#define PCI_CHIP_G33_G			0x29C2
+#define PCI_CHIP_Q35_G			0x29B2
+#define PCI_CHIP_Q33_G			0x29D2
 
 
 /* ================================================================
diff --git a/src/mesa/drivers/dri/i915/intel_screen.c b/src/mesa/drivers/dri/i915/intel_screen.c
index 67e176a..ca8610b 100644
--- a/src/mesa/drivers/dri/i915/intel_screen.c
+++ b/src/mesa/drivers/dri/i915/intel_screen.c
@@ -514,6 +514,10 @@ static GLboolean intelCreateContext( const __GLcontextModes *mesaVis,
    case PCI_CHIP_I915_GM:
    case PCI_CHIP_I945_G:
    case PCI_CHIP_I945_GM:
+   case PCI_CHIP_I945_GME:
+   case PCI_CHIP_G33_G:
+   case PCI_CHIP_Q35_G:
+   case PCI_CHIP_Q33_G:
       return i915CreateContext( mesaVis, driContextPriv, 
 			       sharedContextPrivate );
  
diff --git a/src/mesa/drivers/dri/i915/intel_tex.c b/src/mesa/drivers/dri/i915/intel_tex.c
index 98ddc79..5bd2806 100644
--- a/src/mesa/drivers/dri/i915/intel_tex.c
+++ b/src/mesa/drivers/dri/i915/intel_tex.c
@@ -677,7 +677,11 @@ static void intelUploadTexImage( intelContextPtr intel,
    /* Time for another vtbl entry:
     */
    else if (intel->intelScreen->deviceID == PCI_CHIP_I945_G ||
-            intel->intelScreen->deviceID == PCI_CHIP_I945_GM) {
+            intel->intelScreen->deviceID == PCI_CHIP_I945_GM ||
+            intel->intelScreen->deviceID == PCI_CHIP_I945_GME ||
+            intel->intelScreen->deviceID == PCI_CHIP_G33_G ||
+            intel->intelScreen->deviceID == PCI_CHIP_Q33_G ||
+            intel->intelScreen->deviceID == PCI_CHIP_Q35_G) {
       GLuint row_len = image->Width * image->TexFormat->TexelBytes;
       GLubyte *dst = (GLubyte *)(t->BufAddr + offset);
       GLubyte *src = (GLubyte *)image->Data;

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071030 James Ausmus wrote:
> On 10/30/07, Benno Schulenberg <benno.schulenberg@gmail.com> wrote:
>> build the Mesa modules from git, the 1.2.2 and 1.4 part:
>>   http://dri.freedesktop.org/wiki/Building
>> These three lines were added to i915/intel_context.h since 7.0.1:
>>   #define PCI_CHIP_G33_G                  0x29C2
>>   #define PCI_CHIP_Q35_G                  0x29B2
>>   #define PCI_CHIP_Q33_G                  0x29D2
> You don't have to go to all that, the media-libs/mesa-7.0.1 works fine
> *IF* you apply the attached patch (either hack the ebuild
> or CTRL-Z immediately after emerge gets done unpacking the source).
> I had the same problem you did, did the Googling, found the patch
> and everything works fine for me now.

I did try unpacking the MesaLibs distfile, patching & repacking,
but of course Portage objected that the file size was incorrect !
I've looked at the Mesa ebuild, but have no idea how to hack it
nor am I confident of applying the patch during a suspended emerge.
I could try the 2nd, but could you fill out a few more details of the 1st,
which I will note for future occasions ?
This certainly looks like the solution to the DRI problem.

Meanwhile, what I believed to be a small irritant is proving a show-stopper.
Leaving aside DRI, the display from the new machine spills off the screen:
it's way off to the left (eg Gkrellm is invisible on the KDE desktop)
& loses about half the KDE panel when placed at top/bottom/right-side;
mouse-X starts left of centre & the fonts & icons are badly oversized.
KDE Control Centre tells me the display is 1680x1050 @ 60 Hz (as it sb).
Elsewhere, it does appear the screen's physical size ( 430 x 270 mm )
is being recognised, tho' perhaps it's not being applied correctly.

I believed the problem was due to faulty thro'put by my KVM switch,
but plugging the monitor directly into the new machine make no difference.
I don't see any obvious solution in the monitor manual (Samsung 206BW).
The current machine using an Nvidia card & direct monitor plug-in
produces none of these defects, suggesting the cause is the G33 chip.
I'm not keen to buy another Nvidia card, if I can avoid it ...

Any advice here too wb very welcome.

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Dale]

Philip Webb wrote:
> I did try unpacking the MesaLibs distfile, patching & repacking,
> but of course Portage objected that the file size was incorrect !
> <<snip >>
>
> Any advice here too wb very welcome.
>
>   

You may be able to use the --digest feature so it doesn't check the
tarball before using it.  I think that may work.

Hope that helps.

Dale

:-)  :-)  :-) 
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071031 Philip Webb wrote:
> Meanwhile, what I believed to be a small irritant is proving a show-stopper.
> Leaving aside DRI, the display from the new machine spills off the screen:

See previous msg for full horror story, but I did check the DPI :
both machines show 99x98 in answer to 'xdpyinfo'.
There have been a lot of confused & frustrated users on the Forum recently
& a surprising lack of resolution of problems in this area:
perhaps I've fallen over the cutting into the bleeding edge with my G33.

One suggestion (repeated twice) was to try  xf86-video-i810-1.7.4 ,
which is said to lack a bug introduced in newer versions.  Mb tomorrow ...

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] new box DRI problem : more

[Philip Webb]

071031 Philip Webb wrote:
> 071030 James Ausmus wrote:
>> the media-libs/mesa-7.0.1 works fine
>> *IF* you apply the attached patch (either hack the ebuild
>> or CTRL-Z immediately after emerge gets done unpacking the source).
>> I had the same problem you did, did the Googling, found the patch
>> and everything works fine for me now.

I managed to hit  ^z  at the right moment & the emerge succeeded.
Afterwards, 'glxgears' ran safely revealing all of "1964 fps"
(it's  1491  with software acceleration on the new machine
&  1212  with h/ware acceleration on the current machine).
So thanks for the patch, but the big showstopper remains,
for which I will send another msg to the list.

-- 
========================,,============================================
SUPPORT     ___________//___,  Philip Webb : purslow@chass.utoronto.ca
ELECTRIC   /] [] [] [] [] []|  Centre for Urban & Community Studies
TRANSIT    `-O----------O---'  University of Toronto
-- 
gentoo-user@gentoo.org mailing list