Mailing List Archive

Corey Richardson posted on Sun, 24 Apr 2011 21:03:06 -0400 as excerpted:

> (I was directed to this mailing list from someone in #gentoo. Redirect
> me if this is the wrong place. Thank you!)

FWIW, this list seems as good as any. The topic is reasonably desktop
related. The down side is that this is a relatively low-traffic list so
you might not get anyone that can be of much help. The general user list
is FAR higher traffic. You may or may not get better results there.

> I'm trying to use pyglet, a python library for multimedia (commonly used
> as a game library) using OpenGL. When running any sort of script
> utilizing pyglet, the following happens:
>
> "failed to create drawable" is printed to the shell twice The window
> pyglet opens is completely black, and hangs. To close the window I need
> to kill the python process.
>
> I've chatted with the Pyglet devs and determined that the error is in
> OpenGL and not pyglet. After googling around for this error, the only
> other instance I saw was someone using nouveau that needed some sort of
> driver upgrade.
>
> The output of glxinfo is attached, as well as the output from "eix
> media-libs/mesa", the implementation of OpenGL on my system.
>
> I tried reemerging mesa with -gallium in the USE, which had the same
> result as before except instead of a hang a Segmentation Fault was
> generated. When I emerged mesa back with gallium, it emitted an
> informational blurb that gallium only worked with i915 and another intel
> card, among the intels (with some other vendor's cards, but they didn't
> pertain to me). I am using an intel i915, and I believe my system is
> properly configured for that card. I would love to see the output of
> glxinfo from someone who has a working OpenGL for the i915.
>
> A simple test case (if you would like to test on your own system) is:
>
> python -c "import pyglet as p;w=p.window.Window();p.app.run();"
>
> This requires dev-python/pyglet to be emerged. It's a small library and
> is pure-python, so it should only take ~30 seconds.

Eh... not so here. The library itself might only take that, but portage
says I have USE flag conflicts, which will certainly require > 30s to
resolve.

As most desktop users I expect, I have USE=alsa set. That's a pyglet USE
flag that according to the portage error, triggers a dependency on alsa-
lib with USE=alisp. USE=alisp is certainly going to be rather less
commonly enabled, either in general or for alsa-libs itself, and it's not
enabled here, thus thus the conflict.

Of course, I might be able to set that, but I haven't checked, perhaps
that'll cascade into another dependency.

The point is, for most users, it's going to be a rather larger job than
the 30-second trivial emerge you suggested it might be. I might still do
it, tho; haven't decided yet.

> I am using AwesomeWM, if that is of consequence.

It shouldn't be in and of itself. Only if you are using a WM that uses
OpenGL for various effects (as does, say kwin, I'm a kde guy), and you
have those and thus the OpenGL requirment enabled, might it be
interesting, and I'm not familiar with AwesomeWM so don't know if that
applies or not.

> So, how can I further debug this error, and hopefully fix it? Request
> more information if you need it. Thank you for your time

> OpenGL vendor string: Mesa Project
> OpenGL renderer string: Software Rasterizer

This is highly suspect. I'm running a Radeon here, with the native
kernel/mesa/X freedomware drivers not the proprietary stuff, but I've seen
others point to this bit of glxinfo output to see if the system's properly
running hardware OpenGL or is relying on Mesa's software renderer only, as
well, so I believe it's a reliable indicator, regardless of hardware.

Here's what my Radeon system (with hardware OpenGL active) says, for
comparison:

OpenGL vendor string: Advanced Micro Devices, Inc.
OpenGL renderer string: Mesa DRI R600 (RV730 9495) 20090101 TCL DRI2

Those should specify the hardware maker (Intel in your case) and possibly
driver specific hardware and driver details. (Here, it's using the r600
sub-driver from the radeon driver, that subdriver supports the r6xx/r7xx
class chips, including the rv730 chip that's in my hardware, a Radeon
hd4650 based card. The Intel driver may report different details, but it
should definitely **NOT** say "Software Rasterizer", which indicates that
it's falling back to the mesa in-CPU software emulation code.)

*BUT*, that's with the "classic" driver (tho in kms mode). The gallium
driver for the r6xx/r7xx class radeons isn't as mature as for the earlier
r3xx-r5xx class radeons and intel hardware, and I've not experimented much
with it and know relatively little about it at this point. Gallium's on
my list to investigate further, but meanwhile, it's /possible/ (not likely
as it'd cause confusion, but possible) that the gallium glxinfo results
print software rasterizer here as well. When I get a chance to screw with
it some more I guess I'll know for sure, but meanwhile, I *SUSPECT* that
your posted strings indicate mesa fallback emulation rendering, NOT real
hardware OpenGL.

If that's the case as I suspect, I believe we have your problem. Many
OpenGL apps simply refuse to run with only the mesa software fallback, or,
in your case, appear to run, but simply black-background.

Back to the glxinfo output. Note that *other* bits of the output, like
the OpenGL version string just below the above, still correspond to the
mesa version. Again, here's mine, next mesa version but very similar to
yours, and can't be used to see whether hardware or software OpenGL is
enabled:

OpenGL version string: 2.1 Mesa 7.10.2

My netbook has Intel graphics, but I'm not as familiar with them as I am
the Radeons and am a bit fuzzy on which model it is (gma450, i915, i945, I
think I've seen all three numbers in different places but don't know how
they relate to each other), except that it's from the pre-poulsbo era.
I'll try to post its output as a followup, since that'll be at least
somewhat more directly comparable to what yours should be.

> [I] media-libs/mesa
> Installed versions: 7.10.1(11:58:19 PM 04/23/2011)(classic gallium
> nptl video_cards_intel -debug -gles -hardened -kernel_FreeBSD -llvm
> -motif -pic -selinux -video_cards_mach64 -video_cards_mga
> -video_cards_nouveau -video_cards_r128 -video_cards_radeon(tho I
don't believe likely due to the confusion it'd create because of the
previous use of those lines to check for it
> -video_cards_savage -video_cards_sis -video_cards_tdfx
> -video_cards_via -video_cards_vmware)


OK, looks like you have intel hardware configured here. And you have
classic and gallium USE flags enabled too. Good.

One thing I don't see you mention, is switching between the classic and
gallium OpenGL implementations. Again, I've not played with gallium much,
and it may be that this doesn't apply to intel at all, but at least for
radeons, eselect mesa is used to list and switch between the classic and
gallium drivers. You mention trying with gallium, but not specifically
switching to the classic drivers and trying that. Perhaps that's why you
got the segfault with mesa emerged -gallium, as if you hadn't eselect mesa
switched to the classic drivers, it could have tried to load libraries
that didn't exist in that case, thus triggering a segfault.

Maybe gallium's not working correctly on your hardware (at least with your
current software, kernel/xorg-server/mesa/ddx versions, ddx refers to the
x-driver, xf86-video-intel in your case) and you need to switch to the
classic implementation?

I'll do a followup with the netbook's glxinfo if I remember, but the above
should give you some pointers to investigate, meanwhile.

--
Duncan - List replies preferred. No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master." Richard Stallman

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 04/24/2011 11:34 PM, Duncan wrote:
> [snip]
> One thing I don't see you mention, is switching between the classic and
> gallium OpenGL implementations. Again, I've not played with gallium much,
> and it may be that this doesn't apply to intel at all, but at least for
> radeons, eselect mesa is used to list and switch between the classic and
> gallium drivers. You mention trying with gallium, but not specifically
> switching to the classic drivers and trying that. Perhaps that's why you
> got the segfault with mesa emerged -gallium, as if you hadn't eselect mesa
> switched to the classic drivers, it could have tried to load libraries
> that didn't exist in that case, thus triggering a segfault.

I think that was definitely a portion of the puzzle. I had never
selected gallium before, so it never wasn't using the classic driver.
Which makes the segfault even more mysterious. I selected gallium for
all the options and I now have a hang again (it went away when using
- -gallium, forgot to mention that). My updated glxinfo is attached. I
think it's odd that glxgears works but not pyglet. It may be like you
mentioned, not accepting the software fallback. Oh, and selecting
gallium didn't move OpenGL onto hardware, it's still running in software.

Could it possibly be a kernel option somewhere? KMS is on and I compiled
in Intel DRM support. Just flailing around wildly for causes.

- --
Corey Richardson
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJNtPiVAAoJEAFAbo/KNFvpg/gH/RtKgBOThCz9hAIdPUeSgz9r
QYHp3+FhlJZOp6qgY2u9IddXV5zVZgVL3Jh4YtQjmj1Fjq0m/cN5kQXhUhKwr6bh
aUswUla6Yg5Jy3XH6OjpLqVhkQ2cHkpJNpgae3gH/nki0DpFthEULoxLl24I1bHR
oBm6DkQM7iEayB5ZAd7tpQm0lVY5/gClUYoFFDpiUyKlAg4CXtv527Rormu+lyM3
NsFk6Efu873moIrLIgBDbCPhUujUXRLmPm93gQZTqt8YYMTxhebAtv0YLa9Y2Mf1
9QLKRZY0sW0cpk7JGz86p/cJirS/GojnfnMxcbWoYN2mdCqH/6+cbsg2l1CSSfE=
=mgUV
-----END PGP SIGNATURE-----

Duncan posted on Mon, 25 Apr 2011 03:34:38 +0000 as excerpted:

> I'll do a followup with the netbook's glxinfo if I remember, but the
> above should give you some pointers to investigate, meanwhile.

OK, on the netbook...

lspc reports Intel 945GME

The xorg log (Xorg.0.log) first reports "Driver for Intel Integrated
Graphics Chipsets: i810", THEN reports the Intel driver detecting chipset
945GME (as lspci listed), BUT THEN the Intel driver DRI2 driver is
reported as the i915.

Talking about the xorg log... check it too. If you're running hardware
OpenGL, it'll report something like this (FWIW, xorg-server 1.10.1):

LoadModule: "dri"
Loading /usr/lib/xorg/modules/extensions/libdri.so
Module dri: vendor="X.Org Foundation"
compiled for 1.10.1, module version = 1.0.0
ABI class: X.org Server Extension, version 5.0
Loading extension XFree86-DRI
LoadModule: "dri2"
Loading /usr/lib/xorg/modules/extensions/libdri2.so
Module dri2: vendor="X.Org Foundation"
compiled for 1.10.1, module version = 1.2.0
ABI class: X.Org Server Extension, version 5.0
Loading extension DRI2

Then later...

drmOpenDevice: node name is /dev/dri/card0

... followed by open result 8, (OK), another open, drmOpenMinor, busID,
etc.

Then later

intel(0): [DRI2] Setup complete
intel(0): [DRI2] DRI driver: i915

later...

intel(0): direct rendering: DRI2 Enabled

later...

GLX: Initialized DRI2 GL provider for screen 0


If instead it's saying DRI disabled, or doesn't mention it, you're
software-fallback rendered, for sure, as it's straight from the log.

FWIW, here's those lines as mentioned up-thread from the netbooks glxinfo:

OpenGL vendor string: Tungsten Graphics, Inc
OpenGL renderer string: Mesa DRI Intel(R) 945GME x86/MMX/SSE2

Kernel config:

Here I do a custom config, building everything I need directly into the
kernel, so I can disable module loading. It's easier that way as I don't
have to ensure that the modules subdirs stay in sync with the kernel.

For graphics, you'll want and/or I have:

DRM/DRI (CONFIG_DRM)

Intel 8xx/9xx/63x/64x/HD Graphics (CONFIG_DRM_I915)

I have modesetting on by default (CONFIG_DRM_I915_KMS)

Support for frame buffer devices (CONFIG_FB)

Inside that option:

Enable Video Mode Handling Helpers (CONFIG_FB_MODE_HELPERS)
(may not be necessary but can help with EDID detection and doesn't hurt)

Enable Tile Blitting Support (CONFIG_FB_TILEBLITTING)

**DO**NOT** enable the Intel framebuffer device options here as they
interfere with KMS.

Back out under graphics again, go into Console display driver support:

Framebuffer Console support (CONFIG_FRAMEBUFFER_CONSOLE)

You want that for KMS.

It's recommended to enable VGA text console (CONFIG_VGA_CONSOLE) as well,
for a fallback in case the kms drivers fail for some reason, or if you
wish to use UMS. (In the removable graphics card case, VGA text console
is also a backup in case you switch to a different card, but as Intel
graphics are normally integrated, it's not like that's a real option. But
it's still worthwhile in case there's a problem with the KMS framebuffer
or for trying UMS.)

Back out at the main graphics menu again, I have VGA Arbitration off as
it's only useful with multiple graphics cards, lowlevel video output
switch controls is on to support the key that controls that on my netbook,
and backlight and LCD device support is enabled, with both lowlevel LCD
and backlight controls enabled inside, but no specific drivers, as that's
what my hardware takes.

More interesting is the /dev/agpgart (AGP Support) option, with the Intel
chipset option checked inside it. My system's PCIE, so I'm not sure
that's actually necessary, but it's enabled and X including OpenGL works.
Maybe I'll experiment with it at some point, but that point isn't now.

And if you do still have AGP, and the BIOS has the fastwrite option, make
sure it's off. While it works on some systems, it's unsupported and the
xorg folks consider it unsupportable, because it's so crazy-buggy on so
many systems. Similarly, any fast-write options you might have in your
xorg.conf. Ensure they're off.

FWIW, on the netbook the only bit of xorg.conf config I have here is the
synaptics touchpad config file located in /etc/X11/xorg.conf.d/ . The
rest of the X config is entirely internal xorg defaults and auto-
detection. (The main system is a bit more complex as it has dual
monitors, and an xorg config that sets orientation, etc. But it too is
way smaller than it used to be.)

--
Duncan - List replies preferred. No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master." Richard Stallman

On Mon, 25 Apr 2011, Corey Richardson wrote:

> On 04/24/2011 11:34 PM, Duncan wrote:
>> [snip]
>> One thing I don't see you mention, is switching between the classic and
>> gallium OpenGL implementations. Again, I've not played with gallium much,
>> and it may be that this doesn't apply to intel at all, but at least for
>> radeons, eselect mesa is used to list and switch between the classic and
>> gallium drivers. You mention trying with gallium, but not specifically
>> switching to the classic drivers and trying that. Perhaps that's why you
>> got the segfault with mesa emerged -gallium, as if you hadn't eselect mesa
>> switched to the classic drivers, it could have tried to load libraries
>> that didn't exist in that case, thus triggering a segfault.
>
> I think that was definitely a portion of the puzzle. I had never
> selected gallium before, so it never wasn't using the classic driver.
> Which makes the segfault even more mysterious. I selected gallium for
> all the options and I now have a hang again (it went away when using
> - -gallium, forgot to mention that). My updated glxinfo is attached. I
> think it's odd that glxgears works but not pyglet. It may be like you
> mentioned, not accepting the software fallback. Oh, and selecting
> gallium didn't move OpenGL onto hardware, it's still running in software.
>
> Could it possibly be a kernel option somewhere? KMS is on and I compiled
> in Intel DRM support. Just flailing around wildly for causes.

You might want to prepare for the possibility that your card just
doesn't work with KMS, even different cards from the same brand.

At home, I have a Radeon FireGL V3400 that just refuses to work with any
kind of KMS, and I finally gave up. You can disable it either by
building a kernel that defaults to not having it on, or by passing a
kernel parameter from Grub ("nomodeset").

At work, I have a Radeon HD4350 which works great with KMS on console
VTY's and in X11, and hardware 3D works fine also. Both machines are
running Gentoo with the 1.9 Radeon drivers and the 2.6.38.2 kernel.

KMS is just going to have to be something that works when it works for
now.

--
+ Brent A. Busby +
+ Sr. UNIX Systems Admin + Vote for Cthulhu.
+ University of Chicago +
+ James Franck Institute + Why settle for the lesser evil?

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 04/25/2011 08:18 AM, Duncan wrote:
> Duncan posted on Mon, 25 Apr 2011 03:34:38 +0000 as excerpted:
>
>> I'll do a followup with the netbook's glxinfo if I remember, but the
>> above should give you some pointers to investigate, meanwhile.
>
> OK, on the netbook...
>
> lspc reports Intel 945GME
>
> The xorg log (Xorg.0.log) first reports "Driver for Intel Integrated
> Graphics Chipsets: i810", THEN reports the Intel driver detecting chipset
> 945GME (as lspci listed), BUT THEN the Intel driver DRI2 driver is
> reported as the i915.

Interesting, the DRI2 driver is reported as an i965 for me. I'll need to
investigate that further.

>
> Talking about the xorg log... check it too. If you're running hardware
> OpenGL, it'll report something like this (FWIW, xorg-server 1.10.1):
> [ snip ]
> intel(0): [DRI2] Setup complete
> intel(0): [DRI2] DRI driver: i915
>
> later...
>
> intel(0): direct rendering: DRI2 Enabled
>
> later...
>
> GLX: Initialized DRI2 GL provider for screen 0
>
>
> If instead it's saying DRI disabled, or doesn't mention it, you're
> software-fallback rendered, for sure, as it's straight from the log.
>

Yep, I have all the lines you do (with slightly different version
numbers), except the i965 as I reported earlier. I suspect that is the
cause of many of these problems, but I can't think why it'd be doing that.

> FWIW, here's those lines as mentioned up-thread from the netbooks glxinfo:
>
> OpenGL vendor string: Tungsten Graphics, Inc
> OpenGL renderer string: Mesa DRI Intel(R) 945GME x86/MMX/SSE2
>
> Kernel config:
>
> [ snip]

All of the kernel options check out, so it's not the kernel (although I
did need to go through and clean up a few things, thanks for giving me
an excuse ;-)

As for Gallium saying it's a i965 card, I think it was reading the last
option in eselect with it set. Not positive, but that's the only thing I
could come up with (I had both i915 and i965 set to gallium in eselect.
After changing that to classic, here's the new glxinfo I get:

OpenGL vendor string: Tungsten Graphics, Inc
OpenGL renderer string: Mesa DRI Mobile Intel® GM45 Express Chipset
OpenGL version string: 2.1 Mesa 7.10.1
OpenGL shading language version string: 1.20

Looks better, AND it solves the pyglet problem! Thank you Duncan.

Now that I think about gallium using i965, it doesn't seem as odd when
you consider that the DRI2 driver is i965. I find that very strange, any
ideas what that could be? Didn't really find anything online, perhaps my
google-foo is lacking.

- --
Corey Richardson
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJNtc3PAAoJEAFAbo/KNFvpvhUH/jv4n0hBWXRW4L61QZIe6Qnq
bh+u8z75a5KOoGAizeOdlmtN28Dg7SHJLMw7PAk4L6QkhiHFcMY7rPEbKbbTFj7K
+jCtWGkvt6oEWe6AeessrnjodSJGW7DdGFjT5yvU+ZU2OIfEFZyeuCJB4QznI0zv
aMI44OOdkJ2+BVeOjmq3joLlVFTIiROGk2IF65hCkZ4Y5AX3rm7zDUpq+voeJ8Y8
njNab62o0iGuvUJwm0xr9BCCxUATiNWoB6FbZSy12YlDs6CkJ0zrAkBwLeMlNQxZ
dYXSkAMEYAOi8IkuF6S2Pg/n3XA2W4tcYNiICTRkI88c79QuCpGOhYceLqcN+E0=
=Cu3b
-----END PGP SIGNATURE-----

Corey Richardson posted on Mon, 25 Apr 2011 15:38:55 -0400 as excerpted:

> As for Gallium saying it's a i965 card, I think it was reading the last
> option in eselect with it set. Not positive, but that's the only thing I
> could come up with (I had both i915 and i965 set to gallium in eselect.
> After changing that to classic, here's the new glxinfo I get:
>
> OpenGL vendor string: Tungsten Graphics, Inc OpenGL renderer string:
> Mesa DRI Mobile Intel® GM45 Express Chipset OpenGL version string: 2.1
> Mesa 7.10.1 OpenGL shading language version string: 1.20
>
> Looks better, AND it solves the pyglet problem! Thank you Duncan.
>
> Now that I think about gallium using i965, it doesn't seem as odd when
> you consider that the DRI2 driver is i965. I find that very strange, any
> ideas what that could be? Didn't really find anything online, perhaps my
> google-foo is lacking.

I expect it's simply not knowing about the eselect mesa bit, and having
something set wrong, originally. It would have been interesting to see
what the xorg log said before we started experimenting, as I suspect that
we fixed it in two stages, either setting something else wrong before we
found the first problem so fixing it didn't solve the entire problem, or
having two misconfigured settings in the first place and it took getting
them both correct to fix the problem.

FWIW, if you haven't already done so, you will probably want to enable
eselect's tab-completion option, among others. eselect bashcomp list will
give you a *BIG* list (199 items long, here) of bash-completion command
options you can activate if you use them on the command line. Each login
user can run eselect bashcomp and set their own prefs, so root doesn't
have to have the same set of active bash-completions as your normal user,
as someone (significant-other, whatever) else with a login on your
computer, etc.

Tab-completion for commands like eselect make them **MUCH** easier to use
effectively, just as tab-completion for filesystem paths makes them both
much easier to use effectively, AND much easier to use without error. If
you tab-complete an rm command, you're MUCH less likely to typo and rm the
wrong thing, and MUCH more likely to catch it if you do typo, due to the
completion giving you something unexpected, before you actually do the
removal!

With eselect tab completion enabled, it too becomes much easier to use
effectively and without error. =:^)

--
Duncan - List replies preferred. No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master." Richard Stallman