Skip to contentSkip to conversation

Blast and data miscompare

DISCUSSION Leo Przybylski 4 participants 4 messages Oct 31, 2004 – Nov 11, 2004

lore
Share & tools
lore
#0 lore raw

Hello,

I have tried searching on this issue, but found nothing. I heard from a kernel developer at work that a memory error was discovered recently in the linux 2.6 kernel that causes data miscompare errors in the generic scsi driver when executing blast tests.

Does anyone know more about this???

Leo

#1 lore raw
Quoted 6 lines

I have tried searching on this issue, but found nothing. I heard from a kernel developer at work that a memory error was discovered recently in the linux 2.6 kernel that causes data miscompare errors in the generic scsi driver when executing blast tests.

Does anyone know more about this???

Not sure if it's the same problem. But we were seeing a miscompare on 2.4 due to a incorrect COW happening, followed by a hardware hash hole w/ PPC64.

To fix it we had to make sure that the PTE was cleared and the TLB's flushed before the new PTE was established.

Martin, was this fixed on 2.6?

Thanks,
Jake

#2 lore raw

--Jake Moilanen <[email protected]> wrote (on Tuesday, November 09, 2004 09:34:00 -0600):

Quoted 7 lines
Quoted 6 lines

I have tried searching on this issue, but found nothing. I heard from a kernel developer at work that a memory error was discovered recently in the linux 2.6 kernel that causes data miscompare errors in the generic scsi driver when executing blast tests.

Does anyone know more about this???

Quoted 8 lines

Not sure if it's the same problem. But we were seeing a miscompare on 2.4 due to a incorrect COW happening, followed by a hardware hash hole w/ PPC64.

To fix it we had to make sure that the PTE was cleared and the TLB's flushed before the new PTE was established.

Martin, was this fixed on 2.6?

Yup, was already fixed in 2.6, and is PPC64 only. Most of those errors tend to be caused by IO problems ...

M.

#3 lore raw
In reply to Jake Moilanen
Quoted 8 lines · Jake Moilanen

Not sure if it's the same problem. But we were seeing a miscompare on 2.4 due to a incorrect COW happening, followed by a hardware hash hole w/ PPC64.

To fix it we had to make sure that the PTE was cleared and the TLB's flushed before the new PTE was established.

Martin, was this fixed on 2.6?

It can't happen on 2.6; when BenH rewrote the PTE handling in 2.6 earlier this year, this was one of the things we made sure couldn't be a problem.

Paul.