Re: [PATCH 1/4] gpiolib: cdev: relocate debounce_period_us from struct gpio_desc

[Bartosz Golaszewski <brgl@xxxxxxxx>]

On Wed, Dec 13, 2023 at 4:59 PM Kent Gibson <warthog618@xxxxxxxxx> wrote:
>
> On Wed, Dec 13, 2023 at 04:40:12PM +0100, Bartosz Golaszewski wrote:
> > On Wed, Dec 13, 2023 at 3:27 PM Kent Gibson <warthog618@xxxxxxxxx> wrote:
> > >
> > > On Wed, Dec 13, 2023 at 03:54:53PM +0200, Andy Shevchenko wrote:
> > > > On Tue, Dec 12, 2023 at 01:42:50PM +0800, Kent Gibson wrote:
> > > > > Store the debounce period for a requested line locally, rather than in
> > > > > the debounce_period_us field in the gpiolib struct gpio_desc.
> > > > >
> > > > > Add a global tree of lines containing supplemental line information
> > > > > to make the debounce period available to be reported by the
> > > > > GPIO_V2_GET_LINEINFO_IOCTL and the line change notifier.
> > > >
> > > > ...
> > > >
> > > > >  struct line {
> > > > >     struct gpio_desc *desc;
> > > > > +   struct rb_node node;
> > > >
> > > > If you swap them, would it benefit in a code generation (bloat-o-meter)?
> > > >
> > >
> > > Didn't consider that placement within the scruct could impact code
> > > generation.
> > > Having the rb_nodes at the beginning of struct is preferable?
> > >
> >
> > I suppose it has something to do with 0 offset when using
> > container_of(). Not sure if that really matters though.
> >
>
> There are other fields that get the container_of() treatment, but node
> does look to be the one most used, so probably makes sense to put it
> first.
>
> > > > >  };
> > > >
> > > > ...
> > > >
> > > > > +struct supinfo {
> > > > > +   spinlock_t lock;
> > > > > +   struct rb_root tree;
> > > > > +};
> > > >
> > > > Same Q.
> > > >
> > >
> > > Same - I tend to put locks before the field(s) they cover.
> > > But if the node being first results in nicer code then happy to swap.
> > >
> > > > ...
> > > >
> > > > > +static struct supinfo supinfo;
> > > >
> > > > Why supinfo should be a struct to begin with? Seems to me as an unneeded
> > > > complication.
> > > >
> >
> > I think we should keep it as a struct but defined the following way:
> >
> > struct {
> >     spinlock_t lock;
> >     struct rb_root tree;
> > } supinfo;
>
> That is what I meant be merging the struct definition with the variable
> definition.  Or is there some other way to completely do away with the
> struct that I'm missing?

No, I also meant that.

>
> > >
> > > Yeah, that is a hangover from an earlier iteration where supinfo was
> > > contained in other object rather than being a global.
> > > Could merge the struct definition into the variable now.
> > >
> > > > ...
> > > >
> > > > > +                   pr_warn("%s: duplicate line inserted\n", __func__);
> > > >
> > > > I hope at bare minimum we have pr_fmt(), but even though this is poor message
> > > > that might require some information about exact duplication (GPIO chip label /
> > > > name, line number, etc). Generally speaking the __func__ in non-debug messages
> > > > _usually_ is a symptom of poorly written message.
> > > >
> > > > ...
> > >
> > > Yeah, I wasn't sure about the best way to log here.
> > >
> > > The details of chip or line etc don't add anything - seeing this error
> > > means there is a logic error in the code - we have inserted a line
> > > without erasing it.  Knowing which chip or line it happened to occur on
> > > wont help debug it.  It should never happen, but you can't just leave it
> > > unhandled, so I went with a basic log.
> > >
> >
> > We should yell loudly in that case - use one of the WARN() variants
> > that'll print a stack trace too and point you to the relevant line in
> > the code.
> >
>
> Ok, so any suggestion as to which WARN() variant would make the most sense?
>

Just a regular WARN(1, "msg ...");

> > > >
> > > > > +out_unlock:
> > > > > +   spin_unlock(&supinfo.lock);
> > > >
> > > > No use of cleanup.h?
> > > >
> > >
> > > Again, that is new to me, so no not yet.
> > >
> >
> > Yep, please use a guard, they're awesome. :)
> >
>
> Will do.
>
> Thanks,
> Kent.

Bart