Re: [PATCH v2 2/2] gpio: Add Cadence GPIO driver
From: Bartosz Golaszewski
Date: Tue Dec 18 2018 - 09:06:24 EST
wt., 18 gru 2018 o 14:54 Janek Kotas <jank@xxxxxxxxxxx> napisaÅ(a):
> > On 18 Dec 2018, at 13:50, Bartosz Golaszewski <bgolaszewski@xxxxxxxxxxxx> wrote:
> > pon., 17 gru 2018 o 23:22 Linus Walleij <linus.walleij@xxxxxxxxxx> napisaÅ(a):
> >> On Mon, Dec 17, 2018 at 4:51 PM Bartosz Golaszewski
> >> <bgolaszewski@xxxxxxxxxxxx> wrote:
> >>> The driver looks good but is there any particular reason not to use
> >>> regmap for register IO?
> >> I thought we only use regmap for MMIO when the register range is
> >> shared (as in a system controller) so that some registers are for this,
> >> some register or even bits in a register for some other driver, so they
> >> need the spinlock in the regmap to protect the register range.
> > This is what syscon is for. Regmap simply abstracts any register IO.
> > For instance: there's no locking in this driver. Are we sure it's not
> > needed? Regmap provides internal locking for you in the form of a
> > mutex or spinlock.
> > Also: it looks like the interrupts here are quite simple with a single
> > bit per interrupt in the status register and the same layout in the
> > mask register - it could probably profit from using the
> > regmap_irq_chip and not bother with reimplementing irq_chip callbacks.
> >> It is also nice for shadowing/caching of register contents I guess,
> >> wat does this driver get from regmap MMIO?
> > Code shrinkage IMO.
> > Note that I'm not blocking this from being merged - I just think that
> > using modern frameworks is always a good idea.
> I can reimplement the driver using regmap, but It seems in such case
> I wonât be able to use the Generic GPIO Infrastructure, would I?
> So I would need to provide functions for setting direction, etc.
> I think it would make the driver code bigger.
Indeed. If anything: gpio-mmio would need to be converted to using regmap.
So I guess nevermind my comment.