Weird mtrr allocation error...

[SilverTab]

I tried searching the forums but didnt find a thread corresponding exactly to my problem (I did came accross the multiple mtrr allocation threads ...)

anyway, I have a Radeon 9559 and I have Ubuntu 5.10 installed with all the required driver ( I followed the howto from ubuntu forum...seemed to be working for most of the people)... I installed the driver, ran fglrxconfig, then rebooted, and I still get the Mesa driver when I use fglrxinfo....

Here is the result of dmesg | grep fglrx

[4294755.314000] fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies, Starnberg, GERMANY' taints kernel.
[4294755.319000] [fglrx] Maximum main memory to use for locked dma buffers: 930 MBytes.
[4294755.320000] [fglrx] module loaded - fglrx 8.16.20 [Aug 16 2005] on minor 0
[4294755.334000] [fglrx:firegl_addmap] *ERROR* mtrr allocation failed (-22)
[4294755.335000] [fglrx] Internal AGP support requested, but kernel AGP support active.
[4294755.335000] [fglrx] Have to use kernel AGP support to avoid conflicts.
[4294755.335000] [fglrx] Kernel AGP support doesn't provide agplock functionality.
[4294755.335000] [fglrx] AGP detected, AgpState   = 0x1f004a1b (hardware caps of chipset)
[4294755.335000] [fglrx:firegl_unlock] *ERROR* Process 8845 using kernel context 0

I though it was just a regular allocation problem so I tried to follow the howto posted on this forum, however, here is the weird part:
result of cat /proc/mtrr only shows:
reg00: base=0x00000000 ( 0MB), size=984064MB: write-back, count=1


which I assume is my ram....

nothing else is showing up...
anyone have an idea on what's going on?? why is nothing else showing up??

[SilverTab]

One more thing here's part of my /var/log/Xorg.0.log:

(EE) fglrx(0): [agp] unable to acquire AGP, error "xf86_EINVAL"
(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 0xf8bfb000 at 0xb7ab8000
(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 part seems to be the error??
(EE) fglrx(0): [agp] unable to acquire AGP, error "xf86_EINVAL"

[offby1]

I'm getting the same thing, but I have a bit more in my dmesg output (Note the highlights):

[4294739.156000] fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies, Starnberg, GERMANY' taints kernel.
[4294739.157000] [fglrx] Maximum main memory to use for locked dma buffers: 928 MBytes.
[4294739.157000] ACPI: PCI Interrupt 0000:01:00.0[A] -> GSI 16 (level, low) -> IRQ 16
[4294739.157000] [fglrx] module loaded - fglrx 8.18.6 [Oct 11 2005] on minor 0
 [4294739.180000] mtrr: type mismatch for e8000000,8000000 old: write-back new: write-combining
[4294739.180000] [fglrx:firegl_addmap] *ERROR* mtrr allocation failed (-22)
[4294739.181000] [fglrx] Maximum main memory to use for locked dma buffers: 928 MBytes.
[4294739.181000] [fglrx] Internal AGP support requested, but kernel AGP support active.
[4294739.181000] [fglrx] Have to use kernel AGP support to avoid conflicts.
[4294739.181000] [fglrx] AGP detected, AgpState   = 0x1f004a1b (hardware caps of chipset)
 [4294739.181000] mtrr: type mismatch for f8000000,4000000 old: write-back new: write-combining
[4294739.181000] [fglrx:firegl_unlock] *ERROR* Process 9418 using kernel context 0

[jec0s]

Did you make sure to completely remove the ubuntu drivers that might of been installed previously, including linux-restricted-modules..
Well.. right now.. I can't get X to run with new fglrx module anymore (funny cause it ran fine for a while) (EE) No devices detected.
I'm just gonna wait till its in ubuntu-backports now, cause its not building right with the installer..

[SilverTab]

Originally Posted by jec0s

Did you make sure to completely remove the ubuntu drivers that might of been installed previously, including linux-restricted-modules..
Well.. right now.. I can't get X to run with new fglrx module anymore (funny cause it ran fine for a while) (EE) No devices detected.
I'm just gonna wait till its in ubuntu-backports now, cause its not building right with the installer..

Yup I tried that, no luck....

One interesting thing though, I found someone on ubuntu forum who had the same card as I have (9550)

and he gets the exact same errors...

lspci |grep ATI
0000:01:00.0 VGA compatible controller: ATI Technologies Inc: Unknown device 4153
0000:01:00.1 Display controller: ATI Technologies Inc: Unknown device 4173

[hubick]

When the driver attempts to allocate it's own write-combining MTRR region(s) for your video RAM, it is conflicting with the existing write-back region.

I wrote some information about how MTRR's work, and how I fixed mine, here:
http://www.rage3d.com/board/showthread.php?t=33821469

In short, you need to remove the existing MTRR(s), find out the address(s) of your video RAM the driver is trying to create MTRR(s) for, and recreate your MTRR table with a write-combining region(s) in that spot.

[SilverTab]

Originally Posted by hubick

When the driver attempts to allocate it's own write-combining MTRR region(s) for your video RAM, it is conflicting with the existing write-back region.

I wrote some information about how MTRR's work, and how I fixed mine, here:
http://www.rage3d.com/board/showthread.php?t=33821469

In short, you need to remove the existing MTRR(s), find out the address(s) of your video RAM the driver is trying to create MTRR(s) for, and recreate your MTRR table with a write-combining region(s) in that spot.

ok thanks for the head out!...i'll read that

although I was told that these messages:
0000:01:00.0 VGA compatible controller: ATI Technologies Inc: Unknown device 4153
0000:01:00.1 Display controller: ATI Technologies Inc: Unknown device 4173

mean that my card isnt recognized by the kernel, so the kernel could be the prob....

im just not so sure what to try at this point

[e88z4]

Originally Posted by SilverTab

ok thanks for the head out!...i'll read that

although I was told that these messages:
0000:01:00.0 VGA compatible controller: ATI Technologies Inc: Unknown device 4153
0000:01:00.1 Display controller: ATI Technologies Inc: Unknown device 4173

mean that my card isnt recognized by the kernel, so the kernel could be the prob....

im just not so sure what to try at this point

I have exactly the same problem as you since Kernel 2.6.12. The previous kernel 2.6 was fine. The way I fix it is by assigning MTRR manually and reload fglrx through a script just before the X start.

[SilverTab]

Originally Posted by e88z4

I have exactly the same problem as you since Kernel 2.6.12. The previous kernel 2.6 was fine. The way I fix it is by assigning MTRR manually and reload fglrx through a script just before the X start.

anyway I can reach you?? (ICQ/AIM/MSN/Jabber/email...)

I'm definitly a noob when it comes to hardware in linux! ..


A little help would be greatly appreciated since you're the first person I find who had the exact same problem that I have, and managed to fix it...

[Septor]

Do you have both of your AGP modules loaded SilverTab? There should be an "agpgart" module as well as an "xxxx-agp" module (where xxxx=intel, nvidia or whatever your chipset is).

[SilverTab]

Originally Posted by Septor

Do you have both of your AGP modules loaded SilverTab? There should be an "agpgart" module as well as an "xxxx-agp" module (where xxxx=intel, nvidia or whatever your chipset is).

mmmmm how do I verify this?? (as I stated before, i'm not so good when it comes to hardware config etc. in linux...

[hubick]

Originally Posted by SilverTab

mmmmm how do I verify this?? (as I stated before, i'm not so good when it comes to hardware config etc. in linux...

lsmod

[SilverTab]

Originally Posted by Septor

Do you have both of your AGP modules loaded SilverTab? There should be an "agpgart" module as well as an "xxxx-agp" module (where xxxx=intel, nvidia or whatever your chipset is).

I found this line from lsmod:
agpgart 32328 2 fglrx,intel_agp

[mooninite]

Eh... intel_agp = fglrx suckage

CD into your /lib/modules/kernel-version/kernel/drivers/char/agp (i think) and rename the intel_agp.ko file... reboot and see if that helps.

[SilverTab]

Originally Posted by mooninite

Eh... intel_agp = fglrx suckage

CD into your /lib/modules/kernel-version/kernel/drivers/char/agp (i think) and rename the intel_agp.ko file... reboot and see if that helps.

should I rename it to something specific?? or just any name but intel_agp.ko.... (I located the file, its in:
/lib/modules/2.6.12-9-386/kernel/drivers/char/agp/intel-agp.ko)

[mooninite]

Originally Posted by SilverTab

should I rename it to something specific?? or just any name but intel_agp.ko.... (I located the file, its in:
/lib/modules/2.6.12-9-386/kernel/drivers/char/agp/intel-agp.ko)

It doesn't matter what name... yes, anything besides intel_agp.ko will work. Just name it "intel_agp.ko.old" for an example. You will notice that it is no longer modprobed and dmesg will show that fglrx reports that it is using its agpgart. (even though the kernel agpgart loads)

[armandchinkwiejoe]

Hi,

Same problem here. /proc/mtrr only has one entry. Strange thing is it is different then the one from SuSE 9.1.
And in SuSE 9.1 with echo "reg01 ......" |>/proc/mtrr you couls add lines but SuSE 10.0 does not complain but also does not add that line to /proc/mtrr.
In SuSE 9.1 video=vesafb:nomtrr worked great. In 10.0 no results.

Also using the internal AGP module (i have a Intel) does not help. Same error 'cannot initiate AGP' or something like that.

Any new ideas?

Regards,

Armand

[complex]

I faced with the same problem 2 days ago. Finally I managed to setup a working configuration.

HARDWARE:

Gigabyte GA-8IPE1000 Pro2 mainboard with latest BIOS update
Intel Pentium 4 3.0GHz HT processor
Gigabyte ATI Radeon 9550

SOFTWARE:

Debian 3.1 - Sarge, stable (partially unstable to install OpenOffice.org 2.0)XFree86 4.3.0.1
ATI Proprietary Display Driver 8.18.6 (built using the debian-way)
Custom 2.6.13.4 kernel.
The fglrx kernel module has been built from source package.

RELEVANT KERNEL CONFIGURATION OPTIONS:

CONFIG_MTRR=y
CONFIG_AGP=y # All AGP support modules are turned off! Only AGP support is ON!!!
CONFIG_DRM is not set

SYMPTOM:

dmesg and XFree86.0.log reports MTRR errors and disables direct rendering (accelerated OpenGL). 2D acceleration works fine.

TEST:

$ cat /proc/mtrr
reg00: base=0x00000000 ( 0MB), size=984064MB: write-back, count=1

The exact size may differ slightly, but always a large value, abour 1TB!
(984064MB is about 984GByte ~ 1TB)

This value is surely wrong, since a tipical machine has about 128Mb..2Gb memory.

CAUSE:

It seems that some BIOSes sets wrong MTRR values. The mainboard above is a real-world example. The wrong MTRR write-back region is larger than the 4Gbyte address space. This conflicts with ANY new MTRRs set by X or other driver/application.

SOLUTION:

We must clear the wrong and conficting MTRR entry on every boot with an rcS.d init script.

Create a text file named fix_mtrr in /etc/init.d and copy this text into it:

#!/bin/sh
# Fix wrong MTRR setting
echo "disable=0" >| /proc/mtrr
echo "base=0x0 size=0x40000000 type=write-back" >| /proc/mtrr

Replace 0x40000000 with the size of your main memory (not video memory)! Help:

0x08000000 = 128Mb
0x10000000 = 256Mb
0x40000000 = 1Gb
0x60000000 = 1.5Gb
0x80000000 = 2Gb

Add execute privileges:

chmod +x /etc/init.d/fix_mtrr

Create a symlink: /etc/rcS.d/S02fix_mtrr
The symlink must point to: ../init.d/fix_mtrr

Reboot your system.

It the kernel options above are set correctly and your ATI driver is configured properly, then 2D and 3D acceleration sould work fine.

TEST:

Your can run:

$ cat /proc/mtrr

Your should get something like:

reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1
reg01: base=0xd0000000 (3328MB), size= 128MB: write-combining, count=1
reg02: base=0xf0000000 (3840MB), size= 128MB: write-combining, count=1

Your can run:

$ fglrxinfo

You sould get something like:

display: :0.0 screen: 0
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON 9550 Generic
OpenGL version string: 1.3.5395 (X4.3.0-8.18.6)

The above solution should work in ANY situation when the BIOS sets wrong MTRRs.

Greetings, Complex

[armandchinkwiejoe]

Yeah,
I just figured out the same thing last night.
echo 'disable=0' |>/proc/mtrr
echo 'base=0x00000000 size=0x400000000 type=write-back' |> /proc/mtrr
This is main memory
echo 'base=0xd0000000 size=0x8000000 type=write-combining' |> /proc/mtrr
This is the vga card.

Works great!
Armand

[aqua]

Hi guys
I wonder if you could help me a little more with this mtrr problem!?
I created the file /etc/init.d/fix_mtrr like this:

#!/bin/sh
 # Fix wrong MTRR setting
 echo "disable=0" >| /proc/mtrr
 echo "base=0x00000000  size=0x20000000 type=write-back" >| /proc/mtrr
 echo "base=0x00000000  size=0x08000000 type=wtite-combining" >| /proc/mtrr

And the link: /etc/init.d/rcS.d/S02fix_mtrr.
I have 512Mb mem and 9800Pro (Which has Xt core R360) 128Mb.
Nothing changes and the mtrr file still has the same one line:
reg00: base=0x00000000 ( 0MB), size=983552MB: write-back, count=1
I've SuSE 10, don't know if that's the problem. Can you tell me if I'm doing it right or not? I've tried with menu.1st: nomtrr , mtrr, wram 128 and all other changes I've found i other threads with no luck. And I don't understand if it's bios reporting wrong, then how come memtest or win2k or win xp and even win98 gets the right amount of memory!? Another question:" Does enabling or disabling bios and video cache in bios have anything to do with this problem"? Thanks

[e88z4]

Originally Posted by aqua

Hi guys
I wonder if you could help me a little more with this mtrr problem!?
I created the file /etc/init.d/fix_mtrr like this:

#!/bin/sh
 # Fix wrong MTRR setting
 echo "disable=0" >| /proc/mtrr
 echo "base=0x00000000  size=0x20000000 type=write-back" >| /proc/mtrr
 echo "base=0x00000000  size=0x08000000 type=wtite-combining" >| /proc/mtrr

And the link: /etc/init.d/rcS.d/S02fix_mtrr.
I have 512Mb mem and 9800Pro (Which has Xt core R360) 128Mb.
Nothing changes and the mtrr file still has the same one line:
reg00: base=0x00000000 ( 0MB), size=983552MB: write-back, count=1
I've SuSE 10, don't know if that's the problem. Can you tell me if I'm doing it right or not? I've tried with menu.1st: nomtrr , mtrr, wram 128 and all other changes I've found i other threads with no luck. And I don't understand if it's bios reporting wrong, then how come memtest or win2k or win xp and even win98 gets the right amount of memory!? Another question:" Does enabling or disabling bios and video cache in bios have anything to do with this problem"? Thanks

You have to insert the script just before the X start. The way I did it is editing the my GDM initalization script. I also reload fglrx module after MTRR is reallocated.
At least this method works for me.

[armandchinkwiejoe]

[quote=aqua]Hi guys
I wonder if you could help me a little more with this mtrr problem!?
I created the file /etc/init.d/fix_mtrr like this:

#!/bin/sh
# Fix wrong MTRR setting
echo "disable=0" >| /proc/mtrr
echo "base=0x00000000 size=0x20000000 type=write-back" >| /proc/mtrr
echo "base=0x00000000 size=0x08000000 type=wtite-combining" >| /proc/mtrr

And the link: /etc/init.d/rcS.d/S02fix_mtrr.
I have 512Mb mem and 9800Pro (Which has Xt core R360) 128Mb.
[End QUATE]

I use echo "disable=0" |> /proc/mtrr. | comes before >.
And also the base address of the vga card can not be the same as the memory. Check out lspci -v to find the base for the vga.

The file fix_mtrr is put in /etc/init.d and edit boot.local (in init.d) the execute that file. That is the way SuSE recommends it.

If you dare:
init 3
log is as root
cat /proc/mtrr
echo "disable=0" |>/proc/mtrr
cat /proc/mtrr (this will be extremely slow!)
echo "base=0x00000000 size=0x20000000 type=write-back" |>/proc/mtrr
cat /proc/mtrr

Then you can see if the command is correct.

Armand

[aqua]

Thanks guys. I just installed Mandriva 2005, rpm'd ATI's rpm, reboot, changed "radeon" to "fglrx", reboot, and glxgears=12000+! Couldn't be easier. At last a working Linux!

[GameQber]

(in response to complex's solution)
UPDATE:
After following what you said, the command "cat /proc/mtrr" now gives me:

reg01: base=0x40000000 (1024MB), size=983552MB: write-back, count=1

when it originally gave me:
reg00: base=0x00000000 ( 0MB), size=984064MB: write-back, count=1
reg01: base=0x40000000 (1024MB), size=983552MB: write-back, count=1


At startup, I saw the following:

Starting fix_mtrr: [FAILED]


Not sure what's going on here. I still have no 3D hardware acceleration (obviously).
Also, I'm fairly new to Linux, just so you know.

[alzen]

Originally Posted by complex

SOLUTION:

We must clear the wrong and conficting MTRR entry on every boot with an rcS.d init script.

Create a text file named fix_mtrr in /etc/init.d and copy this text into it:

#!/bin/sh
# Fix wrong MTRR setting
echo "disable=0" >| /proc/mtrr
echo "base=0x0 size=0x40000000 type=write-back" >| /proc/mtrr

Replace 0x40000000 with the size of your main memory (not video memory)! Help:

0x08000000 = 128Mb
0x10000000 = 256Mb
0x40000000 = 1Gb
0x60000000 = 1.5Gb
0x80000000 = 2Gb

Add execute privileges:

chmod +x /etc/init.d/fix_mtrr

Create a symlink: /etc/rcS.d/S02fix_mtrr
The symlink must point to: ../init.d/fix_mtrr

Reboot your system.

It the kernel options above are set correctly and your ATI driver is configured properly, then 2D and 3D acceleration sould work fine.

man i just want to thank u so much ; maybe it will sound a little funny but not for me, maybe u saw my posts on this my ATI card was just doing 2D and no OpenGL; that situation continued till today and i still do not know if it hangs or not(i guess it will hang sooner or later, it is ATI after all ; u know, i thought many times that drivers are working good "that time") but at least i can try this drivers.

i had it a little easier on Arch Linux, i mean i did not have to do all this linking stuff; but the most important part is that the drivers are working. thanks again man.

[valentine3]

Originally Posted by alzen

i had it a little easier on Arch Linux, i mean i did not have to do all this linking stuff; but the most important part is that the drivers are working. thanks again man.

I also use Arch and i was wondering how you got the script to run on boot. Thanks

[alzen]

Originally Posted by valentine3

I also use Arch and i was wondering how you got the script to run on boot. Thanks

very simple. u must create that script, place it in /etc/rc.d/, do chmod +x to make it executable and simply place it's name into daemons line in ur /etc/rc.conf, mine looks like:

DAEMONS=(syslog-ng hotplug !pcmcia network netfs crond firestarter httpd fix_mtrr)

btw. i am just after first lockup, ATI is ATI, do not buy that s#!% , i have some trace what may be causing lockups but guess i will post it in the other topic; just my noobish trace.

[valentine3]

That works great. I finally have 3d after a month of failure. Thanks

[etherealpanda]

This script kept failing for me on Ubuntu 5.10
/etc/init.d/rcS.d/S02fix_mtrr
It said something along the lines of not having write access.

I was not thinking to clearly so It took me a few hours to try changing it to S03fix_mtrr.
This corrected the error. Just incase anyone else ran into this problem

[flabdablet]

Your fix worked nicely for me too (Abit 1S7-E2 mobo, Radeon 9550, Ubuntu 5.10 Breezy).

One thing I don't understand, though: what is the >| construct you're using to write stuff into /proc/mtrr? I hadn't seen it before, assumed it was a typo for a plain old > and got error messages about read-only file systems until I fixed my script to use >| just like you said.

I can't find >| in the bash manual. How exactly is it different from >, and where is it documented?