Re: scsi disk size reporting in dmesg

[Randy.Dunlap]

On Thu, 20 Oct 2005, Dale Blount wrote:

> On Wed, 2005-10-19 at 22:09 -0700, Randy.Dunlap wrote:
> > On Mon, 17 Oct 2005 15:24:04 -0400 Dale Blount wrote:
> >
> > > Hello,
> > >
> > > I just added 2 external 1TB+ scsi devices to my i686 linux server
> > > running 2.6.13.4 connected to external LSI MPT card.  fdisk and df both
> > > show the sizes correctly (see below), but I'm worried that dmesg reports
> > > them incorrectly.
> > >
> > > SCSI device sda: 2460934144 512-byte hdwr sectors (160487 MB)
> > > SCSI device sdb: 3790438400 512-byte hdwr sectors (841193 MB)
> > >
> > > I don't think it's as simple as a variable overflow because both
> > > sdkp->capacity and mb look to be cast as unsigned long longs.  I know a
> > > workaround is to present less data per LUN, but I'd like to use it as
> > > it's setup currently if possible.  Is this just printing incorrectly or
> > > will I run into trouble when the device gets more full?
> >
> > The casts to (unsigned long long) just fix the printk() args to match
> > the format strings (and eliminate warnings).
> >
> > Looks to me like sdkp->capacity is correct.  The <mb> value looks
> > way off.  Since it's just printed here for user info, I don't see
> > how it can be a problem later on.
> >
>
> That's what I was hoping, but I didn't know for sure.  I figured I'd
> better ask to make sure my data wouldn't get truncated when the disk got
> fuller.  Between my first post and now, I've tested it by filling the
> drive with data and testing the md5sums for each file and it seems to
> work.  Other than the odd MB size reported, it seems to work just fine.
>
> > I don't see the error just yet.  Are there any other SCSI device-
> > related messages near these?  And just to confirm, but you must
> > have CONFIG_LBD (Large Block Device) enabled, right?
> >
>
> No, there are no other related messages near these other than the
> standard vendor/versions.  I did not enable CONFIG_LBD since the help
> says "bigger than 2TB", and I partitioned the storage system to present
> disks smaller than that.

I guessed at CONFIG_LBD=y last night.  Today I did the arithmetic
with CONFIG_LBD=n... and found the overflow.

In drivers/scsi/sd.c, approx. line #1260:

		sector_t sz = sdkp->capacity * (hard_sector/256);

sz is 32 bits.
capacity is 32 bits with a value of 2,460,934,144 == 0x92ae_e000.
Then we multiply that by 2 (512 / 256)... and it overflows.
Should be hex 1_255d_c000, but we drop the '1'.
The rest of the calculation (with this new value) does result
in the 160487 MB, matching your log message.

I think that you can fix it (patch it) by changing that line
to make 'sz' be type 'u64' for now.

		u64 sz = sdkp->capacity * (hard_sector/256);

But those divides by 1250 and 1950 are still voodoo to me.

-- 
~Randy
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/