Re: [PATCH] barrier patch set

[Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>]

On Saturday 20 of March 2004 02:48, Johannes Stezenbach wrote:
> Bartlomiej Zolnierkiewicz wrote:
> > On Saturday 20 of March 2004 00:01, Matthias Andree wrote:
> > > BTW, speaking of identify-device, hdparm -i (which uses
> > > HDIO_GET_IDENTITY) always returns "WriteCache=enabled" while hdparm -I
> > > that uses HDIO_DRIVE_CMD with WIN_PIDENTIFY reports the "correct" state
> > > that I've previously set with -W0. This is an i386 machine w/
> > > 2.6.5-rc1.
> > >
> > > Is HDIO_GET_IDENTITY working correctly?
> >
> > There were reports that on some drives you can't disable write cache
> > and even (?) that some drives lie (WC still enabled but marked as
> > disabled).

Doh, I misunderstood the question.

Correct answer is: everything is fine, RTFM (man hdparm). ;-)

> hdparm -i and -I ultimately both interpret WIN_IDENTIFY result, and both
> test bit 0x0020 of word 85. So it's unclear to me why they report a
> different write cache setting. I added a hexdump to dump_identity()
> in hdparm.c, and found that bit 0x0020 of word 85 is always set.

or WIN_PIDENTIFY to be strict but

-i returns _cached_ (read when the device was probed) identify data
(uses HDIO_GET_IDENTIFY ioctl)
-I reads _current_ data directly from the device
(uses HDIO_DRIVE_CMD ioctl)

> BTW, 'cat /proc/ide/hda/identify' or 'hdparm -Istdin
> </dev/ide/hda/identify' reports the same value as hdparm -I, and that is
> consistent with
> the value I set with hdparm -W x.
>
>
> So, is HDIO_GET_IDENTITY broken?

No.

Regards,
Bartlomiej