Re: [SUGGESTION]: drop virtual merge accounting in I/O requests

[Mikulas Patocka <mpatocka@redhat.com>]

On Fri, 11 Jul 2008, Julian Calaby wrote:

> On Fri, Jul 11, 2008 at 07:56, Mikulas Patocka <mpatocka@redhat.com> wrote:
>> Hi
>>
>> I'm getting crashes with InitIO A100u2w controller on Sparc64 (I had to fix
>> the endianity issues in the driver, but that's unrelated).
>>
>> When I examined the crashes, it turned out that SCSI layer passed requests
>> with too many segments. The controller has at most 32 SG entries per
>> request. It sets shost->sg_tablesize to 32, but despite this, larger
>> requests were submitted to it --- this resulted in overwriting random memory
>> and crashes.
>
> [snip]
>
> Should this have been CC'd to linux-scsi too? This sounds like it
> could be a bug at their end.
>
> Thanks,
>
> -- 
> Julian Calaby

This is not a bug in scsi layer, those "nr_hw_segments" miscalculations 
are hapenning in block/blk-merge.c and they can show up with any block 
device driver that has small number of SG entries.

But I can CC it there, just for their information.

Mikulas