Re: [PATCH] tpm: update to use wait_event calls

[Nish Aravamudan]

On 4/11/06, Kylene Jo Hall <kjhall@xxxxxxxxxx> wrote:
> On Mon, 2006-04-10 at 15:03 -0700, Andrew Morton wrote:
> >
> > > +           interruptible_sleep_on_timeout(&chip->vendor.int_queue,
> > > +                                          HZ *
> > > +                                          chip->vendor.timeout_a /
> > > +                                          1000);
> > >
> > > ...
> > >
> > > +           interruptible_sleep_on_timeout(queue, HZ * timeout / 1000);
> >
> > Please don't use the sleep_on functions.  They are racy unless (iirc) both
> > the waker and wakee are holding lock_kernel().  If the race hits, we miss a
> > wakeup.
> >
> > These should be converted to the not-racy wait_event_interruptible().
>
> Changed in this patch.
>
> Use wait_event_interruptible_timeout in place of
> interruptible_sleep_on_timeout due to its racy nature.
>
> Signed-off-by: Kylie Hall <kjhall@xxxxxxxxxx>
> ---
>  drivers/char/tpm/tpm_tis.c |   15 +++++++++------
>  1 files changed, 9 insertions(+), 6 deletions(-)
>
> --- linux-2.6.17-rc1/drivers/char/tpm/tpm_tis.c 2006-04-11 12:18:35.573996500 -0500
> +++ linux-2.6.16-44/drivers/char/tpm/tpm_tis.c  2006-04-11 14:00:04.341229250 -0500
> @@ -95,10 +95,10 @@ static int request_locality(struct tpm_c
>                  chip->vendor.iobase + TPM_ACCESS(l));
>
>         if (chip->vendor.irq) {
> -               interruptible_sleep_on_timeout(&chip->vendor.int_queue,
> -                                              HZ *
> -                                              chip->vendor.timeout_a /
> -                                              1000);
> +               wait_event_interruptible_timeout(chip->vendor.int_queue,
> +                                                (check_locality(chip, l) >= 0),
> +                                                HZ * chip->vendor.timeout_a /
> +                                                1000);
>                 if (check_locality(chip, l) >= 0)
>                         return l;

Rather than check the condition you slept on right away, couldn't you
just store the return value of wait_event_interruptible_timeout()? If
it's positive, the condition should be true, if it's negative, then
you got a signal, if it's 0, then you timed out. Same would go for the
other change.

Thanks,
Nish
-
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/