Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[Andi Kleen]

On Sat, Jul 17, 2004 at 06:26:03PM +0200, R. J. Wysocki wrote:
> I got this on a dual Opteron system on 2.6.8-rc1 with the latest x86-64 
> patchset from Andi:

Does it happen with x86_64-2.6.8-1 too ? 

-Andi

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[R. J. Wysocki]

[-- Attachment #1: Type: text/plain, Size: 1236 bytes --]

On Saturday 17 of July 2004 20:12, Andi Kleen wrote:
> On Sat, Jul 17, 2004 at 06:26:03PM +0200, R. J. Wysocki wrote:
> > I got this on a dual Opteron system on 2.6.8-rc1 with the latest x86-64
> > patchset from Andi:
>
> Does it happen with x86_64-2.6.8-1 too ?

It did not happen when I was running that kernel, but I had only run it for a 
couple of times.  It generally does not happen with this one either.  It's 
happened only once and I reported it immediately.  But ...

... I saw something very similar on 2.6.7-rc1 and I'm now looking at the log 
(attached - it's partially corrupted, because /var was on /dev/sdb5 that 
failed too).  The hardware configuration was similar to the current one, 
AFAIR.

Well, it looks like the whole SCSI bus sometimes goes south for some reason 
and it may very well be a hardware problem that manifests itself in such a 
(strange?) way.  Or not.  Anyway, it certainly had not happened on this 
hardware _before_ 2.6.7-rc1.

I have no idea what to do to make it happen again (suggestions welcome).

rjw

-- 
Rafael J. Wysocki
----------------------------
For a successful technology, reality must take precedence over public 
relations, for nature cannot be fooled.
					-- Richard P. Feynman

[-- Attachment #2: 2.6.7-rc1.log.gz --]
[-- Type: application/x-gzip, Size: 3833 bytes --]

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[R. J. Wysocki]

Andi,

On Saturday 17 of July 2004 21:09, R. J. Wysocki wrote:
> On Saturday 17 of July 2004 20:12, Andi Kleen wrote:
> > On Sat, Jul 17, 2004 at 06:26:03PM +0200, R. J. Wysocki wrote:
> > > I got this on a dual Opteron system on 2.6.8-rc1 with the latest x86-64
> > > patchset from Andi:
> >
> > Does it happen with x86_64-2.6.8-1 too ?
>
> It did not happen when I was running that kernel, but I had only run it for
> a couple of times.  It generally does not happen with this one either. 
> It's happened only once and I reported it immediately.  But ...
>
[snip]

I had this problem again this morning.  I was unpacking the kernel tarball to 
/dev/sda8 and it went south (the tarball had been partially unpacked before 
the partition was remounted r-o).  Then, I got back to 2.6.7 and ran fsck - 
now it found some errors (obviously) and fixed them.  Next (on 2.6.7), I 
unpacked the kernel to /dev/sda8 (again) and compiled the 2.6.8-rc2.  I ran 
it, unpacked the kernel to /dev/sda8 (again) and compiled it - everything 
worked.  Then, I applied your patch on top of the newly created 2.6.8-rc2 
tree and compiled the kernel.  After installing and running it I tried to 
unpack the kernel to /dev/sda8 (again) and it went south, so I got back to 
the "plain" 2.6.8-rc2, ran fsck and fixed the partition, unpacked the kernel 
to /dev/sda8 - and it all worked.

So, it seems, there's something in your patch that causes this misbehavior.

rjw

-- 
Rafael J. Wysocki
----------------------------
For a successful technology, reality must take precedence over public 
relations, for nature cannot be fooled.
					-- Richard P. Feynman

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[R. J. Wysocki]

On Sunday 18 of July 2004 14:48, R. J. Wysocki wrote:
> Andi,
>
> On Saturday 17 of July 2004 21:09, R. J. Wysocki wrote:
> > On Saturday 17 of July 2004 20:12, Andi Kleen wrote:
> > > On Sat, Jul 17, 2004 at 06:26:03PM +0200, R. J. Wysocki wrote:
> > > > I got this on a dual Opteron system on 2.6.8-rc1 with the latest
> > > > x86-64 patchset from Andi:
> > >
> > > Does it happen with x86_64-2.6.8-1 too ?
> >
> > It did not happen when I was running that kernel, but I had only run it
> > for a couple of times.  It generally does not happen with this one
> > either. It's happened only once and I reported it immediately.  But ...
>
> [snip]
>
> I had this problem again this morning.  I was unpacking the kernel tarball
> to /dev/sda8 and it went south (the tarball had been partially unpacked
> before the partition was remounted r-o).  Then, I got back to 2.6.7 and ran
> fsck - now it found some errors (obviously) and fixed them.  Next (on
> 2.6.7), I unpacked the kernel to /dev/sda8 (again) and compiled the
> 2.6.8-rc2.  I ran it, unpacked the kernel to /dev/sda8 (again) and compiled
> it - everything worked.  Then, I applied your patch on top of the newly
> created 2.6.8-rc2 tree and compiled the kernel.  After installing and
> running it I tried to unpack the kernel to /dev/sda8 (again) and it went
> south, so I got back to the "plain" 2.6.8-rc2, ran fsck and fixed the
> partition, unpacked the kernel to /dev/sda8 - and it all worked.
>
> So, it seems, there's something in your patch that causes this misbehavior.

To clarify: in the above "your patch" means "x86_64-2.6.8rc1-1".  I should 
have called it by name.  Sorry for the confusion,

rjw

-- 
Rafael J. Wysocki
----------------------------
For a successful technology, reality must take precedence over public 
relations, for nature cannot be fooled.
					-- Richard P. Feynman

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[Andi Kleen]

> To clarify: in the above "your patch" means "x86_64-2.6.8rc1-1".  I should 
> have called it by name.  Sorry for the confusion,

Hmm. Do you have CONFIG_IOMMU_DEBUG on or iommu=force set? 

-Andi

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[R. J. Wysocki]

On Tuesday 20 of July 2004 14:04, Andi Kleen wrote:
> > To clarify: in the above "your patch" means "x86_64-2.6.8rc1-1".  I
> > should have called it by name.  Sorry for the confusion,
>
> Hmm. Do you have CONFIG_IOMMU_DEBUG on or iommu=force set?

No.  Unless iommu=force is by default (I haven't turned it on specifically).

rjw

-- 
Rafael J. Wysocki
[tel. (+48) 605 053 693]
----------------------------
For a successful technology, reality must take precedence over public 
relations, for nature cannot be fooled.
					-- Richard P. Feynman

Re: 2.6.8-rc1: Possible SCSI-related problem on dual Opteron w/ NUMA

[Andi Kleen]

> I had this problem again this morning.  I was unpacking the kernel tarball to 
> /dev/sda8 and it went south (the tarball had been partially unpacked before 
> the partition was remounted r-o).  Then, I got back to 2.6.7 and ran fsck - 
> now it found some errors (obviously) and fixed them.  Next (on 2.6.7), I 
> unpacked the kernel to /dev/sda8 (again) and compiled the 2.6.8-rc2.  I ran 
> it, unpacked the kernel to /dev/sda8 (again) and compiled it - everything 
> worked.  Then, I applied your patch on top of the newly created 2.6.8-rc2 
> tree and compiled the kernel.  After installing and running it I tried to 
> unpack the kernel to /dev/sda8 (again) and it went south, so I got back to 
> the "plain" 2.6.8-rc2, ran fsck and fixed the partition, unpacked the kernel 
> to /dev/sda8 - and it all worked.
> 
> So, it seems, there's something in your patch that causes this misbehavior.

In which patch eactly? x86_64-2.6.8rc1-1 or x86_64-2.6.8rc1-2 ? 

If it started with -2 can you check if -1 has the problem too?

-Andi