Re: 2.4.1-pre8 losing pages

[Peter Horton]

On Fri, Jan 26, 2001 at 07:48:05PM +0000, Russell King wrote:
> Peter Horton writes:
> > The corruption is dependent on having a swapped on swap partition. If I
> > "swapoff" the corruption goes away, but it comes back when I "swapon"
> > again. I feel this a kernel bug, but as I'm the only person out here who's
> > seeing it I'm at a loss ...
> 
> The reason I ask is that on an ARM box running plain 2.4.0 with swap
> enabled I get rather a lot of SEGVs.  Turn swap off, and I don't see
> any.
> 
> It sounds like it may be related.
> 

Okay, scratch that. It does still happen when there's no swap, but for
some reason it happens a lot less often. Looks like it's timing related,
it only fails when using 7200rpm drives, not older 5400rpm ones (even
though they too are using UDMA33). I've ruled out the filing system, the
IDE controller, the drives and the RAM, so that leaves the kernel or the
CPU - I'll try and beg/borrow/steal another CPU and try that. I can
compile kernels / run X whilst the test is running without a problem so it
looks like it's the bulk write that's the problem.

P.