Radeon HD 3870 + fglrx + 64 bit linux = crash

[zappbrannigan]

Hi!

As I mentioned before, i'm trying to run a Radeon HD 3870 with 64bit (amd64) Linux. If i use the fglrx driver and call xinit the system hang with a black screen. This is a hard crash and i can't see any output and no xorg.log will be written to disk. I've tested the fglrx 7.12 and 8.01 on Debian sid and Suse 10.3 with always the same results. The vesa driver works on these systems. Also the radeonhd driver works but corupts the system console. On 32 bit systems the fglrx driver works . The radeon bios is the first released version (compiled on 2007-10-17). The serial port on my worse asus p5k board doesn't work so i can't use a serial console to give more information.

Can anybody give me a hint?
Or has anybody a working 64bit/3870/fglrx configuration?

[Kano]

You can use netconsole too. Just be sure you use the interface with least shared interrupts.

[zappbrannigan]

You can use netconsole too. Just be sure you use the interface with least shared interrupts.
Thank you. I'll try this tomorrow.

[zappbrannigan]

I got this single line:
mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining
Is it possible to redirect the tty output with netconsole?

[zappbrannigan]

OK, I found a workaround for the system hang. I disabled the "Memory Remap Feature" in the bios. Who is to blame? The fglrx driver or ASUS? The Mainboard is a P5k with bios version 0805. The xorg.log file is too big to post but I saw this line:
(**) fglrx(0): ATI GART size: 256 MB I think this is too small since the HD3780 has 512 MB.
The workaround is not a solution because I lost 768 MB of my memory.

[seedofconspiracy]

Finally, someone with the same problem! I've been searching high and low as to why the fglrx 8.01 (and all other versions) crash my system. I'm running a Gigabyte X38-DS4 motherboard with an HD3870. I don't have a "Memory Remap Feature" in my BIOS, and from the looks of it, I don't think I'd want to use it either.

[bridgman]

How much system RAM do you guys have ?

[seedofconspiracy]

How much system RAM do you guys have ?

4GB of DDR2

[Melcar]

I have used HD3xxx cards before on 64bit distros (mainly Ubuntu and Suse) with both 7.12 and 8.01 drivers with no serious issues. This was with a X580 board, so maybe it's not the card/drivers but something with your particular MOBO?

[seedofconspiracy]

Well, actually, I had a similar result with 7.12 on an older AMD 939 / Opteron 165 / 2GB RAM system. I had an Nvidia 7950GT, which worked just fine. Who knows. At this point I've given up as I simply don't have the time to play with it. I'll just have to stick with a VM on top of Vista for now. I will however keep track of this thread to see if any solution comes about.

[zappbrannigan]

How much system RAM do you guys have ?
4GB of DDR2
Me too. I updated the damned ASUS mainboard BIOS with 0905 and run a test with Ubuntu 7.10 with the same (old) results. After work I'll buy a serial port card for more tests... Maybe a test with Vista 64 will follow.

[Kano]

http://www.mjmwired.net/kernel/Documentation/networking/netconsole.txt

[jens]

Wow, finally somebody with the same problem. I'm experiencing the same issue on a HD 3850, and I also found a workaround in disabling memory remapping. My hardware is:
Sapphire Radeon HD 3850 256mb, 4GB DDR2, Asus P5K-E/WiFi-AP, running 64bit ArchLinux.

On a side note, I also have Vista 64bit and XP 32bit installed on this system. In Vista, the driver at first seems to be working fine in 3d applications/games, but there is a crucial problem: severe graphical corruptions will appear usually after loading the next level or whatever makes something new load into the VRAM. I don't know at all if this could be related to the problem in Linux. But in XP the driver works flawlessly. I've just got the feeling that it could have something to do with the memory addressing in 32bit vs 64bit operating systems.

It would be interesting to see if more people experience this problem, maybe we can narrow down what hardware / os combinations are affected and hopefully draw a conclusion as to where the problem should be fixed.

[zappbrannigan]

Hello!
The new 8.02 driver fixes the hang of the first gl application but dosen't fix my 4GB problem. I run X -verbose 999
over a serial terminal and got this log:
http://paste.ubuntu-nl.org/56713/ (it's debian sid!). There is only one line difference between a run with/without memory remap (except timestamp): the last line. With enabled memory remap the system hang there. In line 499 the driver detects 256 mb video ram but the 3780 has 512mb.
This is not a mb error since the radeonhd driver works and a run with Vista x64 also works fine.

Any new hints?

[uraslacker]

Any new hints?

Though you'll have to build it from source, I've had luck getting the radeonhd driver to (sort of) work with 4GB of RAM, an HD2400Pro, and an x86_64 version of Xubuntu.

You can follow some of the links I've given on this thread (http://ubuntuforums.org/showthread.php?t=633860) on Ubuntu Forums.

To give a quick summary, I...

Installed the radeonhd driver from GIT sources.
Passed the iommu=noaperture option to the kernel on boot.
Set the following options in the xorg.conf "Device" section:

Option "MonitorLayout" "TMDS"
Option "NoRandR"



There are other suggestions mentioned on Xorg's radeonhd Wiki, and in the bugs that I'm tracking on my final post at the Ubuntu Forums post that I refer to above.

NOTE: This is not "stable", though it is functional. I still see occasional glitches that require a reboot to fix. Most often this is caused by either allowing DPMS to shut off the LCD or just shutting if off manually. The radeonhd driver does not handle this well....yet. Still, this is better than the hard crash that the fglrx driver causes:)

Hope this helps! Cheers!

[zappbrannigan]

OK, I found the problem. The 64bit driver can't handle my mtrr table.

$ cat /proc/mtrr
reg00: base=0xd0000000 (3328MB), size= 256MB: uncachable, count=1
reg01: base=0xe0000000 (3584MB), size= 512MB: uncachable, count=1
reg02: base=0x00000000 ( 0MB), size=4096MB: write-back, count=1
reg03: base=0x100000000 (4096MB), size= 512MB: write-back, count=1
reg04: base=0x120000000 (4608MB), size= 256MB: write-back, count=1

The fglrx driver (8-01) works with this table:

reg00: base=0x00000000 ( 0MB), size=2048MB: write-back, count=1
reg01: base=0x80000000 (2048MB), size=1024MB: write-back, count=1
reg02: base=0xc0000000 (3072MB), size= 256MB: write-back, count=1
reg03: base=0x100000000 (4096MB), size= 512MB: write-back, count=1
reg04: base=0x120000000 (4608MB), size= 256MB: write-back, count=1


But after an update to fglrx 8-02 I got this (fgl_glxgears):
http://img240.imageshack.us/img240/3193/fglrx802qf5.jpg (http://imageshack.us)
:confused:

@uraslacker: thanks, but the (debian) radeonhd driver works since weeks.

[zappbrannigan]

But after an update to fglrx 8-02 I got this (fgl_glxgears):
http://img240.imageshack.us/img240/3193/fglrx802qf5.jpg (http://imageshack.us)
:confused:

OK, this is fixed under 8-03 !

[maligor]

The Bug is very much a Asus bios bug, it's present in P5K WS 1104 BIOS aswell. Remapping the mtrrs to what you mention worked here aswell.

It's also a fglrx bug since it clearly doesn't detect broken mtrr's, it also makes fglrx look rather bad since it happens on X11 startup.

[uncle_fungus]

Unfortunately it doesn't just affect the HD series cards either. I had to use this fix to get both an X1800XT and an X1900XT to work in my P5K-E with any more than 2GB of RAM.

[alec]

https://bugs.launchpad.net/ubuntu/+source/linux-restricted-modules-2.6.24/+bug/224404

And same problem here
Gigabyte EX38 mobo, 4GB DDR2 and 3870 radeon. Works fine except with fglrx in 64bit linux.
Who is to blame? Kernel? AMD? Intel?

I hoped I could switch to linux 100%

[kingtaurus]

Just quick point, this isn't just an AMD/ATI problem because according to this bug report: https://bugs.launchpad.net/linux/+bug/210780 this also will cause problems with intel cards (not an outright crash).

https://bugs.launchpad.net/ubuntu/+source/linux-restricted-modules-2.6.24/+bug/224404

And same problem here
Gigabyte EX38 mobo, 4GB DDR2 and 3870 radeon. Works fine except with fglrx in 64bit linux.
Who is to blame? Kernel? AMD? Intel?

I hoped I could switch to linux 100%

[alec]

Just quick point, this isn't just an AMD/ATI problem because according to this bug report: https://bugs.launchpad.net/linux/+bug/210780 this also will cause problems with intel cards (not an outright crash).

Those bug reports lead me to think its either a 64-bit kernel issue, or a driver for recent intel chipsets (p35, x38, x48, maybe integrated versions too)
I hope somebody is doing something

[laurie]

My system has a Radeon HD3870 + fglrx + 64 bit on a gigabyte ga-ma790fx-ds5 & 4GB ram woks well with multiple boot fedora9/suse10.3/ubuntu8.4/windows-xp suse10.3 is default. I'm new to the linux scene maybe someone can advise what tests or info I can pass on to help.

[alec]

My system has a Radeon HD3870 + fglrx + 64 bit on a gigabyte ga-ma790fx-ds5 & 4GB ram woks well with multiple boot fedora9/suse10.3/ubuntu8.4/windows-xp suse10.3 is default. I'm new to the linux scene maybe someone can advise what tests or info I can pass on to help.

Yes, your difference is AMD chipset. So far all reports have been on intel new chipsets

[Workaphobia]

Thank you SO MUCH for the information on this 4 GB problem. I was driving myself crazy trying to find a configuration that would work on my new system, but vesa, radeonhd, and fglrx all gave me varying degrees of graphics corruption when switching between X and console. More importantly, fglrx would crash my system unless I put Option NoAccel in my xorg.conf, for exactly the reasons described in this thread. Disabling the memory remapping feature in my BIOS solved the problem, and does indeed appear that Asus's bios is at fault.

Specs:
Intel Q6600
Asus P5K Deluxe
Mushkin 4GB RAM @ 1066Mhz
Asus EAH3650 (ATI 3650)
Gentoo Linux, amd64 architecture
Driver version 8.493
Kernel 22.6.24-r8
Xorg X11 7.2

When I had mtrr enabled in the kernel, I did in fact get a message at boot indicating that the bios's information was incorrect and was therefore ignored. The Option NoMTRR line I had in my xorg.conf apparently wasn't enough, or was ignored by all but the vesa driver.

The MTRR table in /proc before I disabled remapping was identical to the one provided earlier by zappbrannigan.

As an afterthought, I'd like to take this opportunity to embarass Asus by reminding everyone how awful their webpage is. The bios update download page is literally an unprofessional blank containing no information, as if someone forgot to administer their CMS.

I was on the verge of ditching this card. Thank you again.

[alec]

Did I understand correctly? If I fix mtrr tables there will be no performance loss or usable memory size loss?

I already went on to search a good x86 distro. But I like to have the most out of my rig, so x64 is always preferred.

[Redeeman]

I have a gigabyte X48-DQ6 board, and my mtrr is as follows:

reg00: base=0x00000000 ( 0MB), size=4096MB: write-back, count=1
reg01: base=0xe0000000 (3584MB), size= 512MB: uncachable, count=1
reg02: base=0xd0000000 (3328MB), size= 256MB: uncachable, count=1
reg03: base=0x100000000 (4096MB), size=1024MB: write-back, count=1
reg04: base=0x130000000 (4864MB), size= 256MB: uncachable, count=1
reg05: base=0xcff00000 (3327MB), size= 1MB: uncachable, count=1


if i use the "nv" driver, i get this kernel message when starting X: "mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining"

I actually do have a slight problem. It appears my graphics card (a quadro fx 3400m) Doesent properly work, as in, untill the nvidia drivers are loaded, all output is completely garbled (this goes during POST and grub and stuff aswell). I figured this is just the card being weird, since it works with other nvidia cards..

I am planing to throw in a radeon hd card at some very near point..

Should i expect these problems?

[kingtaurus]

I would think so. I don't know if this bug affects the radeonhd (open source driver), but if you are planning on using the fglrx driver.

I have a gigabyte X48-DQ6 board, and my mtrr is as follows:


if i use the "nv" driver, i get this kernel message when starting X: "mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining"

I actually do have a slight problem. It appears my graphics card (a quadro fx 3400m) Doesent properly work, as in, untill the nvidia drivers are loaded, all output is completely garbled (this goes during POST and grub and stuff aswell). I figured this is just the card being weird, since it works with other nvidia cards..

I am planing to throw in a radeon hd card at some very near point..

Should i expect these problems?