Re: 2.5.40-ac4 kernel BUG at slab.c:1477!

[Andre Hedrick]

Only that if it is SG it must conform the ATA/ATAPI DMA on 64kb on 4b
boundaries.  Some what basic stuff iirc, but remember this is piped via
scsi and the eh-stratagy/joke leaves something to be desired.

One concern could be crossing SG lists with the device in PIO, that could
get hairy.

On Sun, 6 Oct 2002, Manfred Spraul wrote:

>  > This happens at random during boot when loading modules.
>  > About half of the time ide-scsi works fine.
>  > The system continues to boot after the BUG with /dev/hdc unaccessible.
> 
> from mm/slab.c:
> 
> 1475 if (xchg((unsigned long *)objp, RED_MAGIC1) != RED_MAGIC2)
> 1476     /* Either write before start, or a double free. */
> 1477     BUG();
> 
> You run an uniprocessor kernel, with slab debugging enabled, and the 
> red-zoning test notices a write before the beginning of the buffer 
> during scsi_probe_and_add_lun, with ide-scsi.
> 
> Andre: Do you know if ide-scsi makes any assumptions about memory 
> alignment of the input buffers? With slab debugging disabled, the 
> alignment is 32 or 64 bytes, with debugging enabled, it's just 4 byte 
> [actually sizeof(void*)] aligned.
> 
> Murray, could you apply the attached patch? It dumps the redzone value 
> during scsi_probe_and_add_lun. Hopefully this will help to find who 
> corrupts the buffers.
> 
> --
> 	Manfred
> 

Andre Hedrick
LAD Storage Consulting Group