Re: memory leak in scsi_cmd_cache 2.6.15

[Jens Axboe <axboe@suse.de>]

On Fri, Jan 27 2006, Jamie Heilman wrote:
> Jens Axboe wrote:
> > On Sun, Jan 22 2006, Jamie Heilman wrote:
> > > Ariel wrote:
> > > > ata_piix seems like it's in common for all, but this is not a lot of 
> > > > systems, so it could just be a coincidence and the problem caused by 
> > > > something that's not chipset specific.
> > > 
> > > Hmm.  I just moved my sata_sil stuff out of the way and rebooted:
> > > 
> > > $ uptime; grep scsi_cmd_cache /proc/slabinfo
> > >  23:22:16 up 4 min,  1 user,  load average: 0.00, 0.03, 0.00
> > > scsi_cmd_cache      1200   1200    384   10    1 : tunables   54   27   8 : slabdata    120    120      0
> > > 
> > > My other workstation also runs 2.6.15.1 but uses sata_nv and doesn't
> > > exhibit the problem.
> > 
> > The SATA low level driver is very unlikely to play a role in this. But
> > you are both using md (raid1 to be specific) on top of scsi, I'd say
> > that's the best clue. I'd very much doubt the nvidia module as well.
> 
> True, my sata_nv box doesn't use md.  OTOH, my machines running raid1
> on an LSI (MPT Fusion driver) SCSI controller don't have this leak.

Those SCSI LLD don't support ordered flush barriers, that's why.

-- 
Jens Axboe