Skip to contentSkip to conversation

2.6.15-git breaks Xorg on em64t

DISCUSSION Dave Jones 5 participants 12 messages Jan 14, 2006 – Jan 16, 2006

lore
Share & tools
lore

Dave Jones

author
#0 lore raw

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)
- Reboot
- Total lockup
- NMI watchdog firing, and then lockup

I've tried backing out a handful of the x86-64 patches, and didn't get too far, as some of them are dependant on others, it quickly became a real mess to try to bisect where exactly it broke.

Any ideas for potential candidates to try & back out ?

Dave

#1 lore raw
Quoted 11 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)
- Reboot
- Total lockup
- NMI watchdog firing, and then lockup

Quoted 5 lines

I've tried backing out a handful of the x86-64 patches, and didn't get too far, as some of them are dependant on others, it quickly became a real mess to try to bisect where exactly it broke.

Any ideas for potential candidates to try & back out ?

Did you perhaps take a look at my report ? -git{6,7} were
bad for me, and the netconsole stack was, uhm, interesting.

http://www.ussg.iu.edu/hypermail/linux/kernel/0601.1/2130.html

-git8 made the problem disappear. Haven't tested more recent
snapshots.

Ciao,

--alessandro

"Somehow all you ever need is, never really quite enough, you know"

(Bruce Springsteen - "Reno")

#2 lore raw
In reply to Dave Jones
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Normally it should be impossible to cause machine checks from software on Intel systems.

Quoted 7 lines

- Reboot
- Total lockup
- NMI watchdog firing, and then lockup

I've tried backing out a handful of the x86-64 patches, and didn't get too far, as some of them are dependant on others, it quickly became a real mess to try to bisect where exactly it broke.\

Shouldn't be too bad - i did a binary search for something else and it worked pretty well.

Quoted 2 lines

Any ideas for potential candidates to try & back out ?

Does it work when you revert all x86-64 changes?

-Andi

author
#3 lore raw
Quoted 11 lines · Andi Kleen
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Normally it should be impossible to cause machine checks from software on Intel systems.

-git7+ is the only time I've ever seen one on this box.

Quoted 10 lines
Quoted 7 lines

- Reboot
- Total lockup
- NMI watchdog firing, and then lockup

I've tried backing out a handful of the x86-64 patches, and didn't get too far, as some of them are dependant on others, it quickly became a real mess to try to bisect where exactly it broke.\

Shouldn't be too bad - i did a binary search for something else and it worked pretty well.

The patches reverted, but I hit problems like modprobe segfaulting during boot, or reboots during APIC init.

Quoted 2 lines
Quoted 1 line

Any ideas for potential candidates to try & back out ?

Does it work when you revert all x86-64 changes?

-git6 which was the last one not to include the x86-64 changes boots and runs fine.

I'll try a latest -git with x86-64 backed out when I get a chance.

Dave

#4 lore raw
Quoted 10 lines · Dave Jones
Quoted 9 lines · Andi Kleen
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Quoted 4 lines
Quoted 2 lines

Normally it should be impossible to cause machine checks from software on Intel systems.

-git7+ is the only time I've ever seen one on this box.

What happens when you apply

ftp://ftp.firstfloor.org/pub/ak/x86_64/quilt/patches/page-table-setup

?

Anyways, since there doesn't seem to be much interest in pretesting of x86-64 patches anymore before merge such things happen occassionally.

-Andi

author
#5 lore raw
Quoted 11 lines · Andi Kleen
Quoted 10 lines · Dave Jones
Quoted 9 lines · Andi Kleen
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

Normally it should be impossible to cause machine checks from software on Intel systems.

-git7+ is the only time I've ever seen one on this box.

What happens when you apply

ftp://ftp.firstfloor.org/pub/ak/x86_64/quilt/patches/page-table-setup

What does this apply to ? Is it dependant on something else not merged yet ? I get rejects when I apply it to 2.6.15-git10

I don't have time to fix it up by hand right now (I was going to set a build going before I turned in for the night..)

I'll poke at it again tomorrow.

Another datapoint btw: I've another EM64T that works just fine. The one that fails is the only one that isn't using onboard VGA, this one has a PCIE Radeon. Given it happens when X is starting up, it could be that the X radeon driver does something special which is why others aren't seeing this.

Dave

#6 lore raw
Quoted 27 lines · Dave Jones
Quoted 11 lines · Andi Kleen
Quoted 10 lines · Dave Jones
Quoted 9 lines · Andi Kleen
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

Normally it should be impossible to cause machine checks from software on Intel systems.

-git7+ is the only time I've ever seen one on this box.

What happens when you apply

ftp://ftp.firstfloor.org/pub/ak/x86_64/quilt/patches/page-table-setup

What does this apply to ? Is it dependant on something else not merged yet ? I get rejects when I apply it to 2.6.15-git10

I don't have time to fix it up by hand right now (I was going to set a build going before I turned in for the night..)

Quoted 7 lines

I'll poke at it again tomorrow.

Another datapoint btw: I've another EM64T that works just fine. The one that fails is the only one that isn't using onboard VGA, this one has a PCIE Radeon. Given it happens when X is starting up, it could be that the X radeon driver does something special which is why others aren't seeing this.

On my i386 box with an ati radeon 9600 agp card of sorts trying to use the radeon driver (ever since I installed suse 10.0) causes an immediate hard lockup on startup of X. Sounds like your problem... Setting the driver to VESA or using the ati proprietory driver works fine (but only the ati driver actually is usable for me).

You could try that and if the others work for you like for me then it is likely the problem is indeed 100% in the radeon driver somewhere...

Best regards,

Anton

Signature · 4 lines
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @)
Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK
Linux NTFS maintainer / IRC: #ntfs on irc.freenode.net
WWW: http://linux-ntfs.sf.net/ & http://www-stu.christs.cam.ac.uk/~aia21/
#7 lore raw
In reply to Dave Jones
Quoted 7 lines

Another datapoint btw: I've another EM64T that works just fine. The one that fails is the only one that isn't using onboard VGA, this one has a PCIE Radeon. Given it happens when X is starting up, it could be that the X radeon driver does something special which is why others aren't seeing this.

It might be due to the DRM update that went through, but I can't think what might have caused it, if you backout the DRM merge does it help any?

did the previous kernel have DRM support for that card?

Dave.

author
#8 lore raw
Quoted 11 lines · Dave Airlie
Quoted 7 lines

Another datapoint btw: I've another EM64T that works just fine. The one that fails is the only one that isn't using onboard VGA, this one has a PCIE Radeon. Given it happens when X is starting up, it could be that the X radeon driver does something special which is why others aren't seeing this.

It might be due to the DRM update that went through, but I can't think what might have caused it, if you backout the DRM merge does it help any?

As it turns out, -git11 with all the DRM bits backed out gives me a working X again.

Quoted 1 line

did the previous kernel have DRM support for that card?

No. This is 1002:5b60 / 1002:5b70 based card.

I had previously missed the 5b60 part in lspci output, so thinking there was no 5b70 addition, I hadn't considered this as a suspect.
Mea Culpa. Looks like Andi is off the hook :-)

Any ideas for any debugging I can add ?

Dave

#9 lore raw
Quoted 16 lines
Quoted 10 lines
Quoted 6 lines

Another datapoint btw: I've another EM64T that works just fine. The one that fails is the only one that isn't using onboard VGA, this one has a PCIE Radeon. Given it happens when X is starting up, it could be that the X radeon driver does something special which is why others aren't seeing this.

It might be due to the DRM update that went through, but I can't think what might have caused it, if you backout the DRM merge does it help any?

As it turns out, -git11 with all the DRM bits backed out gives me a working X again.

Quoted 1 line

did the previous kernel have DRM support for that card?

Quoted 7 lines

No. This is 1002:5b60 / 1002:5b70 based card.

I had previously missed the 5b60 part in lspci output, so thinking there was no 5b70 addition, I hadn't considered this as a suspect.
Mea Culpa. Looks like Andi is off the hook :-)

Any ideas for any debugging I can add ?

Disable dri in your xorg.conf first, (remove the Load "dri"), if that works which it most likely will, then send me an Xorg.0.log, and a drm debug trace (modprobe drm debug=1),

I'm going to be looking at a 64-bit machine with PCIE radeon in the next day or two, I'll be getting one myself post LCA more than likely...

I think more than likely we are hitting a problem where the DRM sets up the radeon RAM controller and the X server sets it up differently... benh has fixes for this but they need to go into the X server driver....

Dave.

Quoted 4 lines

Dave

#10 lore raw
In reply to Dave Jones
Quoted 15 lines · Dave Jones
Quoted 14 lines · Andi Kleen
Quoted 13 lines · Dave Jones
Quoted 11 lines · Andi Kleen
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Normally it should be impossible to cause machine checks from software on Intel systems.

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

-git7+ is the only time I've ever seen one on this box.

What happens when you apply

ftp://ftp.firstfloor.org/pub/ak/x86_64/quilt/patches/page-table-setup

What does this apply to ? Is it dependant on something else not merged yet ? I get rejects when I apply it to 2.6.15-git10

To the other patches in the quilt queue (that is the x86-64 pending tree that everybody is supposed to test, but nobody does)

I see there was a one liner reject with the memory hot add patch in there. I reordered it now to come first. It was for -git9.

-Andi

author
#11 lore raw
Quoted 22 lines · Andi Kleen
Quoted 21 lines · Dave Jones
Quoted 19 lines · Andi Kleen
Quoted 10 lines · Dave Jones
Quoted 9 lines · Andi Kleen
Quoted 7 lines · Dave Jones

Andi,
Sometime in the last week something was introduced to Linus'
tree which makes my dual EM64T go nuts when X tries to start. By "go nuts", I mean it does various random things, seen so far..
- Machine check. (I'm convinced this isn't a hardware problem despite the new addition telling me otherwise :)

Quoted 4 lines
Quoted 2 lines

Normally it should be impossible to cause machine checks from software on Intel systems.

-git7+ is the only time I've ever seen one on this box.

What happens when you apply

ftp://ftp.firstfloor.org/pub/ak/x86_64/quilt/patches/page-table-setup

Quoted 8 lines
Quoted 2 lines

What does this apply to ? Is it dependant on something else not merged yet ? I get rejects when I apply it to 2.6.15-git10

To the other patches in the quilt queue (that is the x86-64 pending tree that everybody is supposed to test, but nobody does)

I see there was a one liner reject with the memory hot add patch in there. I reordered it now to come first. It was for -git9.

It didn't solve my problem. It hung at X startup again, and then
the NMI watchdog triggered. I'll try reverting DaveA's DRM bits, but I'm not optimistic, as a) this card isn't supported yet, and b) whilst it's compiled in, it was disabled by X at startup due to me using dual-head.

Dave