On Fri, Jul 04, 2008 at 09:47:03AM +1000, Benjamin Herrenschmidt wrote:
> On Thu, 2008-07-03 at 16:35 +0100, Tim Yamin wrote:
> > >> +static void
> > >> +mpc52xx_bmdma_start(struct ata_queued_cmd *qc)
> > >> +{
> > >> + struct ata_port *ap = qc->ap;
> > >> + struct mpc52xx_ata_priv *priv = ap->host->private_data;
> > >> +
> > >> + /* LocalBus lock */
> > >> + while (test_and_set_bit(0, &pata_mpc52xx_ata_dma_lock) != 0)
> > >> + ;
> > >
> > > Need to be able to bail on timeout.
> >
> > A deadlock can't occur within the PATA driver because you won't have
> > two DMA requests happening at once, so there is no point in adding a
> > timeout. And even if you do have a timeout, you'd have to drop the I/O
> > request somehow, so it's not really a good idea. If anything else
> > needs to touch the DMA lock, it should do so in a sensible fashion...
>
> But why a hand-coded lock with bitops ? Why not a real spinlock then ?
> The later is more efficient anyway.
Umm, yeah.... (don't know why I didn't clue into that). This definitely
should be a spinlock.
g.
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev
