Re: 2.3.42: Strange memory corruption

Re: 2.3.42: Strange memory corruption

[Rik van Riel]

On Wed, 2 Feb 2000, Patrick Mau wrote:

> I have a really strange memory corruption problem with
> 2.3.42. My system configuration is as follows:

> After a few md5sums I get different checksums.
> Now I check the file with bzip2 and get a CRC error.
> 
> Now I make a copy (plain 'cp') of that file.
> 
> Then I repeat the above with the copy and get different
> checksums and also CRC errors from bzip2. (I expected that).
> 
> --> Here comes the strange part <---
> 
> Now I reboot into 2.3.30. Same config.
> And BOTH files, even the 'corrupt' copy, are
> correct.  I can run the md5sum test AND can uncompress
> BOTH of them. I can always reproduce that.

This looks a bit like there might be a race with the
pagetable mapping or read()ing of the file. It would
explain the three `suspicious' segfaults I've seen in
the last few days...

regards,

Rik
--
The Internet is not a network of computers. It is a network
of people. That is its real strength.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: 2.3.42: Strange memory corruption

[Manfred Spraul]

Rik van Riel wrote:
> 
> This looks a bit like there might be a race with the
> pagetable mapping or read()ing of the file. It would
> explain the three `suspicious' segfaults I've seen in
> the last few days...
> 

The TLB flush code contains various races if IPI's are sent between
switch_mm() and switch_to(). I and Ingo have written patches that fix
these problems, and we are waiting for Linus' reply.

I've posted my patch and a longer description to linux-kernel last
Friday "[PATCH] new tlb flush code".

--
	Manfred
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/