Re: Fw: Re: 2.6.9-rc2-mm4

[Jens Axboe <axboe@suse.de>]

On Mon, Oct 04 2004, Borislav Petkov wrote:
> On Monday 04 October 2004 19:36, Jens Axboe wrote:
> > On Mon, Oct 04 2004, Borislav Petkov wrote:
> > > Ok here we go,
> > >
> > > final results:
> > >
> > >  2.6.8-rc1: OK
> > >  2.6.8-rc2: OK
> > >  2.6.8-rc3: OK
> > >  2.6.8-rc3-bk1: OK
> > >  2.6.8-rc3-bk2: OK
> > >  2.6.8-rc3-bk3: OK
> > >  2.6.8-rc3-bk4: OK
> > >  2.6.8-rc4: BUG!
> > >
> > > So, assuming that everything went fine during testing, the bug got
> > > introduced in the transition between 2.6.8-rc3-bk4 and 2.6.8-rc4.
> >
> > That's some nice testing, thank you. Try backing out this hunk:
> >
> > diff -urp linux-2.6.8-rc3-bk4/drivers/block/scsi_ioctl.c
> > linux-2.6.8-rc4/drivers/block/scsi_ioctl.c ---
> > linux-2.6.8-rc3-bk4/drivers/block/scsi_ioctl.c 2004-08-03
> > 23:28:51.000000000 +0200 +++
> > linux-2.6.8-rc4/drivers/block/scsi_ioctl.c 2004-08-10 04:24:08.000000000
> > +0200 @@ -90,7 +90,7 @@ static int sg_set_reserved_size(request_
> >   if (size < 0)
> >    return -EINVAL;
> >   if (size > (q->max_sectors << 9))
> > -  return -EINVAL;
> > +  size = q->max_sectors << 9;
> >
> >   q->sg_reserved_size = size;
> >   return 0;
> >
> > It's the only thing that sticks out, and it could easily explain it if
> > your cd ripper starts issuing requests that are too big. Maybe even add
> > a printk() here, so it will look like this in the kernel you test:
> >
> >  if (size > (q->sectors << 9)) {
> >   printk("%u rejected\n", size);
> >   return -EINVAL;
> >  }
> >
> > to verify.
> 
> Yeah, that was it. Two lines in the log:
> 
> Oct 4 22:07:04 zmei kernel: 3145728 rejected
> Oct 4 22:07:04 zmei kernel: 3145728 rejected
> 
> Hmm, so this means that my dvd drive is sending too big requests. What do we 
> do: firmware upgrade?

It actually means we have a little discrepancy between what programs
expact of the api. What program are you using? They are supposed to read
back what value has been set with SG_GET_RESERVED_SIZE, I guess this one
does not.

What's a little extra strange is that this command apparently doesn't
have ->dxfer_len set to the correct size, or it should be caught in
sg_io() at issue time.

-- 
Jens Axboe