Yet again 2.6 test 9 and kt400 :(

[rageXLdude]

Hey everyone,

I guess there were numerous threads about the topic, but I've yet to get my r9000 to work in 2.6 test 9. I'm running Gentoo on kernel 2.6 test 9 (obviously), with via kt400 chipset.

I'm getting that dreaded ENODEV or whatever that is, cant remember verbatim, when initializing agpgart. Both kernel and driver internal agpgart were unable to allocate agp memory due to that bleeding error.

Recently I found module called via_agp. I was able to load it, and funny enough, I was getting 3d acceleration in wolfenstein, but glxinfo was still giving me SGI for some entries.

Does anyone have any idea how I might get my 3d to work, and what's with that via_agp and whether it must be loaded?

Thanks a bunch,

[ASCI Blue]

via_agp sounds like it's the agpgart module and yes it needs to be loaded before you load X. The fglrx module however does not, X will load it on it's own.

Since I don't know your skill level with linux when you use glxinfo look for the line OpenGL render string, it should say something to the effect of Radeon 9XXX and some other CPU related stuff. I'd have more accurate info but my Gentoo took a **** on me today so tomorrow maybe I'll have to restart compiling it.

[TheOneKEA]

OpenGL vendor string: Tungsten Graphics, Inc.
OpenGL renderer string: Mesa DRI Radeon 20020611 AGP 4x x86/MMX/SSE TCL
OpenGL version string: 1.2 Mesa 4.0.4

[rageXLdude]

Hmm... I've seen something about mm patches screwing up the agp, but I also tried development-sources, and still same result.
The problem I see is that when I just start X without any modules loaded, then it loads fglrx and agpgart, but no sign of via_agp. I find it very strange...

[TheOneKEA]

That's a bug in 2.6.0-test9-mm2 that I helped to fix. You need this patch:

diff -bBpur fix/drivers/char/drm/drm_agpsupport.h
linux-2.6.0-test9-mm2/drivers/char/drm/drm_agpsupport.h
--- fix/drivers/char/drm/drm_agpsupport.h 2003-11-06 00:49:53.299160210 +0100
+++ linux-2.6.0-test9-mm2/drivers/char/drm/drm_agpsupport.h 2003-11-06
00:27:43.000000000 +0100
@@ -37,8 +37,8 @@
#if __REALLY_HAVE_AGP


-#define DRM_AGP_GET symbol_get(agp_drm)
-#define DRM_AGP_PUT symbol_put(agp_drm)
+#define DRM_AGP_GET symbol_get(drm_agp)
+#define DRM_AGP_PUT symbol_put(drm_agp)

/**
* Pointer to the drm_agp_t structure made available by the agpgart module.
diff -bBpur fix/include/linux/agp_backend.h
linux-2.6.0-test9-mm2/include/linux/agp_backend.h
--- fix/include/linux/agp_backend.h 2003-11-06 00:50:03.247046605 +0100
+++ linux-2.6.0-test9-mm2/include/linux/agp_backend.h 2003-11-06
00:42:38.936727765 +0100
@@ -108,8 +108,7 @@ typedef struct {
int (*copy_info)(struct agp_kern_info *);
} drm_agp_t;

-/* Used by drm. */
-extern const drm_agp_t agp_drm;
+extern const drm_agp_t *drm_agp_p;

#endif /* __KERNEL__ */
#endif /* _AGP_BACKEND_H */

[wedgey__]

The problem I see is that when I just start X without any modules loaded, then it loads fglrx and agpgart, but no sign of via_agp. I find it very strange...

First, try loading all 3 modules before you start X. Occasionally it won't load all the modules, so it's better to load them yourself beforehand (just add them to /etc/modules.autoload.d/kernel-2.6). Also, remember that up until kernel 2.4, all you had to do was load the agpgart module to get AGP support. In 2.6 that's no longer the case, and I don't think X has been updated to take that into account yet. If via-agp won't load, check the output of dmesg, it should tell you what the problem is.

[rageXLdude]

wedge, I did try loading them all, and funny enough, acceleration was there, BUT after that, I couldnt unload them. It was saying some inept crap about modules being used by something, and it didnt say what. Another weird part is that glxinfo gave that SGI entry in the glx library part. Weird... I'll try that patch now.

[wedgey__]

You won't be able to unload them while X is running, and I think you'd also have to unload them in a particular order (fglrx -> via-agp -> agpgart). Have you run "opengl-update ati"?

[rageXLdude]

Heeey, I'm not THAT lame to try unloading video modules while X is running
J/k... Anyway, I have problems unloading them in console... fglrx unloads perfectly fine, but then agpgart says its being used by via_agp, and via_agp says that its just being used by some misterious force.

[rageXLdude]

Tried the patch, still ENOMEM error

This is internal agpgart log:

(II) fglrx(0): doing DRIScreenInit
drmOpenDevice: minor is 0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 7, (OK)
drmOpenDevice: minor is 0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 7, (OK)
drmOpenDevice: minor is 0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 7, (OK)
drmGetBusid returned ''
(II) fglrx(0): [drm] created "fglrx" driver at busid "PCI:1:0:0"
(II) fglrx(0): [drm] added 8192 byte SAREA at 0xd1a34000
(II) fglrx(0): [drm] mapped SAREA 0xd1a34000 to 0x40229000
(II) fglrx(0): [drm] framebuffer handle = 0xd8000000
(II) fglrx(0): [drm] added 1 reserved context for kernel
(II) fglrx(0): DRIScreenInit done
(II) fglrx(0): Kernel Module Version Information:
(II) fglrx(0): Name: fglrx
(II) fglrx(0): Version: 3.2.8
(II) fglrx(0): Date: Sep 21 2003
(II) fglrx(0): Desc: ATI Fire GL DRM kernel module
(II) fglrx(0): Kernel Module version matches driver.
(II) fglrx(0): Kernel Module Build Time Information:
(II) fglrx(0): Build-Kernel UTS_RELEASE: 2.6.0-test9-mm2
(II) fglrx(0): Build-Kernel MODVERSIONS: no
(II) fglrx(0): Build-Kernel __SMP__: no
(II) fglrx(0): Build-Kernel PAGE_SIZE: 0x1000
(II) fglrx(0): [drm] register handle = 0xe1000000
(EE) fglrx(0): [agp] unable to acquire AGP, error "xf86_ENOMEM"
(EE) fglrx(0): cannot init AGP
(II) fglrx(0): [drm] removed 1 reserved context for kernel
(II) fglrx(0): [drm] unmapping 8192 bytes of SAREA 0xd1a34000 at 0x40229000
(WW) fglrx(0): ***********************************************
(WW) fglrx(0): * DRI initialization failed! *
(WW) fglrx(0): * (maybe driver kernel module missing or bad) *
(WW) fglrx(0): * 2D acceleraton available (MMIO) *
(WW) fglrx(0): * no 3D acceleration available *
(WW) fglrx(0): ********************************************* *



This is external one:

(II) fglrx(0): doing DRIScreenInit
drmOpenDevice: minor is 0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is -1, (No such device)
drmOpenDevice: open result is -1, (No such device)
drmOpenDevice: Open failed
drmOpenDevice: minor is 0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is -1, (No such device)
drmOpenDevice: open result is -1, (No such device)
drmOpenDevice: Open failed
[drm] failed to load kernel module "agpgart"
drmOpenDevice: minor is 0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 7, (OK)
drmGetBusid returned ''

[wedgey__]

I wouldn't waste time trying the internal AGP support. If you get it working it'll almost certainly be with the external support. Do all three modules load without any errors in dmesg or elsewhere?

[rageXLdude]

Well, no errors when I load them all manually. In fact, only thing I have to do is load via_agp and then just startX, it goes on nicely. I'll go over log messages and dmesg again now.

Btw, do you know what might be causing portage to force me to install nvidia drivers any time I install anything app that requires direct rendering? I had that card for a while, but then I uninstalled it.. strange..

[rageXLdude]

Allright, here are the messages...

When I load via_agp and fglrx:

Linux agpgart interface v0.100 (c) Dave Jones
agpgart: Detected VIA KT400/KT400A/KT600 chipset
agpgart: Maximum main memory to use for agp memory: 203M
agpgart: AGP aperture is 128M @ 0xd0000000
fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies, Starnberg, GERMANY' taints kernel.
[fglrx] Maximum main memory to use for locked dma buffers: 198 MBytes.
[fglrx] module loaded - fglrx 3.2.8 [Sep 21 2003] on minor 0


So that looks fine so far... now, here's when I startx

Debug: sleeping function called from invalid context at mm/slab.c:1868
in_atomic():1, irqs_disabled():0
Call Trace:
[<c011ea9b>] __might_sleep+0xab/0xd0
[<c014466c>] kmem_cache_alloc+0x6c/0x70
[<d1ab0e38>] __ke_alloc_wait_queue_head_struct+0x18/0x30 [fglrx]
[<d1ab3dd1>] firegl_setup_dev+0x91/0x1b0 [fglrx]
[<d1ab3e4c>] firegl_setup_dev+0x10c/0x1b0 [fglrx]
[<d1ab3364>] firegl_open+0x174/0x230 [fglrx]
[<d1ab32a2>] firegl_open+0xb2/0x230 [fglrx]
[<d1ab0ba6>] firegl_stub_open+0x106/0x1c0 [fglrx]
[<d1ab0aa0>] firegl_stub_open+0x0/0x1c0 [fglrx]
[<c0164388>] chrdev_open+0xe8/0x210
[<c01bdfb0>] devfs_open+0x0/0xf0
[<c01be098>] devfs_open+0xe8/0xf0
[<c01594a1>] dentry_open+0x151/0x270
[<c0159348>] filp_open+0x68/0x70
[<c015983b>] sys_open+0x5b/0x90
[<c03a429f>] syscall_call+0x7/0xb

mtrr: 0xd8000000,0x4000000 overlaps existing 0xd8000000,0x1000000
[fglrx:firegl_addmap] *ERROR* mtrr allocation failed (-22)
[fglrx:firegl_unlock] *ERROR* Process 3495 using kernel context 0
mtrr: 0xd8000000,0x4000000 overlaps existing 0xd8000000,0x1000000


any ideas?

EDIT: I do have mtrr support in kernel

[wedgey__]

Well, no errors when I load them all manually. In fact, only thing I have to do is load via_agp and then just startX, it goes on nicely. I'll go over log messages and dmesg again now.

If it works like that, just add agpgart, via-agp and fglrx to your /etc/modules.autoload.d/kernel-2.6 and you won't need to worry about it.

Btw, do you know what might be causing portage to force me to install nvidia drivers any time I install anything app that requires direct rendering? I had that card for a while, but then I uninstalled it.. strange..

I'm not 100% sure, but try looking in /var/cache/edb/virtuals for a line that starts with "virtual/opengl", and remove any nvidia-* entries that appear after it.

Debug: sleeping function called from invalid context at mm/slab.c:1868
in_atomic():1, irqs_disabled():0
Call Trace:
[<c011ea9b>] __might_sleep+0xab/0xd0
[<c014466c>] kmem_cache_alloc+0x6c/0x70
[<d1ab0e38>] __ke_alloc_wait_queue_head_struct+0x18/0x30 [fglrx]
[<d1ab3dd1>] firegl_setup_dev+0x91/0x1b0 [fglrx]
[<d1ab3e4c>] firegl_setup_dev+0x10c/0x1b0 [fglrx]
[<d1ab3364>] firegl_open+0x174/0x230 [fglrx]
[<d1ab32a2>] firegl_open+0xb2/0x230 [fglrx]
[<d1ab0ba6>] firegl_stub_open+0x106/0x1c0 [fglrx]
[<d1ab0aa0>] firegl_stub_open+0x0/0x1c0 [fglrx]
[<c0164388>] chrdev_open+0xe8/0x210
[<c01bdfb0>] devfs_open+0x0/0xf0
[<c01be098>] devfs_open+0xe8/0xf0
[<c01594a1>] dentry_open+0x151/0x270
[<c0159348>] filp_open+0x68/0x70
[<c015983b>] sys_open+0x5b/0x90
[<c03a429f>] syscall_call+0x7/0xb

Those are actually harmless. I think it's just a debug message someone forgot to remove.

mtrr: 0xd8000000,0x4000000 overlaps existing 0xd8000000,0x1000000
[fglrx:firegl_addmap] *ERROR* mtrr allocation failed (-22)
[fglrx:firegl_unlock] *ERROR* Process 3495 using kernel context 0
mtrr: 0xd8000000,0x4000000 overlaps existing 0xd8000000,0x1000000

I get those as well, but they don't seem to affect the working of the driver, so I'm ignoring them

[rageXLdude]

Lucky bastard :P

What confuses me the most is that when I load via_agp, it allocates memory properly, but the driver obviously for some reason cannot take advantage of that, tries to allocate memory again, and obviously fails

Btw, thx for nvidia tip, it worked