Re: [PATCH 4/4] gpiolib: cdev: replace locking wrappers for gpio_device with guards

[Kent Gibson <warthog618@xxxxxxxxx>]

On Wed, Dec 20, 2023 at 02:47:45PM +0100, Bartosz Golaszewski wrote:
> On Wed, Dec 20, 2023 at 2:28 PM Kent Gibson <warthog618@xxxxxxxxx> wrote:
> >
> > On Wed, Dec 20, 2023 at 02:19:37PM +0100, Bartosz Golaszewski wrote:
> > > On Wed, Dec 20, 2023 at 1:53 PM Kent Gibson <warthog618@xxxxxxxxx> wrote:
> > > >
> > > > On Wed, Dec 20, 2023 at 01:30:57PM +0100, Bartosz Golaszewski wrote:
> > > > > On Wed, Dec 20, 2023 at 1:23 PM Kent Gibson <warthog618@xxxxxxxxx> wrote:
> > > > > >
> > > > > > >
> > > > > >
> > > > > > It would be read and write guards for the gpio_device.
> > > > > > cdev would only be using the read flavour.
> > > > > > And possibly named something other than read/write as the purpose is to
> > > > > > prevent (read) or allow (write) object removal.
> > > > > >
> > > > > > I though that would be clearer than having to reference gpiolib.h to see
> > > > > > what gdev->sem covers, and allow you to change the locking
> > > > > > mechanism later and not have to update cdev.
> > > > > >
> > > > >
> > > > > I still prefer open-coded guards here for clarity. I hope that with
> > > > > SRCU in gpiolib.c, we'll get rid of locking in cdev entirely anyway.
> > > > >
> > > >
> > > > Ok, it is your object so I should use it the way you want it used.
> > > >
> > > > Btw, before I go pushing out a v2, do you have an answer on whether
> > > > gpio_ioctl() requires a guard, as mentioned in the cover letter?
> > > > Is the fact there is an active ioctl on the chardev sufficient in
> > > > itself to keep the gpio_device alive?
> > > >
> > >
> > > AFAICT: no. I think it's a bug (good catch!).
> >
> > The wrappers made that harder to pick up.
> > It kind of stood out as the exception after changing the other ioctls
> > over to guards - where was the guard for that one?
> >
>
> Yeah, it makes sense. This is precisely why guards are so much better
> than hand-coding locks.
>
> > > Can you extend your
> > > series with a backportable bugfix that would come first?
> > >
> >
> > Sure.  That would still use the guard(rwsem_read)?
> > I mean you don't to go adding a wrapper for the fix, just to
> > subsequently remove it, right?
> >
>
> In master - sure. But we definitely do want to backport that to stable
> branches and for that we need to use the old wrapper.
>

Ok, so cleanup.h is too recent for backporting.
Adding and then removing a wrapper it is then.

Cheers,
Kent.