On Mon, Jul 24, 2023 at 11:06:02AM +0000, Lu, Brent wrote:
> > > > +/* helper function to get the number of specific codec */
> >
> > ...and leak a lot of reference counts...
> >
> > > > +static int get_num_codecs(const char *hid) {
> > > > + struct acpi_device *adev = NULL;
> > > > + int dev_num = 0;
> > > > +
> > > > + do {
> > > > + adev = acpi_dev_get_next_match_dev(adev, hid, NULL, -1);
> > >
> > > Humm, I am a bit worried about reference counts.
> > >
> > > See
> > > https://elixir.bootlin.com/linux/latest/source/drivers/acpi/utils.c#L9
> > > 16, it's not clear to me where the get() is done.
> > >
> > > Adding Andy to make sure this is done right.
> >
> > Thank you for Cc'ing.
> >
> > Yes, the above code is problematic. One has to use the respective for_each macro
> > (defined nearby the used API).
> >
> > > > + if (adev)
> > > > + dev_num++;
> > > > + } while (adev != NULL);
> > > > +
> > > > + return dev_num;
> > > > +}
> Each invocation of acpi_dev_get_next_match_dev() calls acpi_dev_put() to release the
> adev from previous call. And the last call returns NULL. It seems to me the reference count
> should be fine. Is my understanding correct?
Ah, right. sorry for the confusion. That's why we have a macro
to not think about these details :-)
> I saw the macro for_each_acpi_dev_match and re-write the function as follow. Thanks for
> suggesting using the macro.
>
> /* helper function to get the number of specific codec */
> static int get_num_codecs(const char *hid) {
> struct acpi_device *adev;
> int dev_num = 0;
size_t here or at least unsigned int is more correct.
> for_each_acpi_dev_match(adev, hid, NULL, -1)
> dev_num++;
>
> return dev_num;
> }
Otherwise, yes, that's what I have in mind.
> Will test it in next few days.
--
With Best Regards,
Andy Shevchenko
