[PATCH v4 01/39] ARM: OMAP2+: gpmc: driver conversion
Mohammed, Afzal
afzal at ti.com
Tue May 8 02:18:37 EDT 2012
Hi Jon,
On Mon, May 07, 2012 at 21:32:58, Hunter, Jon wrote:
> >>> + /* no waitpin */
> >>> + case 0:
> >>> + break;
> >>> + default:
> >>> + dev_err(gpmc->dev, "multiple waitpins selected on CS:%u\n", cs);
> >>> + return -EINVAL;
> >>> + break;
> >>> + }
> >>
> >> Why not combined case 0 and default? Both are invalid configurations so
> >> just report invalid selection.
> >
> > Case 0 is not invalid, a case where waitpin is not used, default refers
> > to when a user selects multiple waitpins wrongly.
>
> Ok. Then for case 0, just return here. If the polarity is set, you could
> print an error here.
Different ways of doing things, this looks cleaner to me as already it is
checked, and time of execution in both cases would not differ much.
> >>> + if (gd->have_waitpin) {
> >>> + if (gd->waitpin != idx ||
> >>> + gd->waitpin_polarity != polarity) {
> >>> + dev_err(gpmc->dev, "error: conflict: waitpin %u with polarity %d on device %s.%d\n",
> >>> + gd->waitpin, gd->waitpin_polarity,
> >>> + gd->name, gd->id);
> >>> + return -EBUSY;
> >>> + }
> >>> + } else {
> >>
> >> Don't need the else as you are going to return in the above.
> >
> > Not always, only in case of error
>
> Ok, but seems that it can be simplified a little.
>
> What happens if a device uses more than one wait-pin? In other words a
> device with two chip-selects that uses two wait-pins?
Please re-read http://www.mail-archive.com/linux-omap@vger.kernel.org/msg67702.html
and your reply
Regards
Afzal
More information about the linux-mtd
mailing list