Re: [PATCH 2/3] of: irq: make callers of of_irq_parse_one() release the device node

From: Geert Uytterhoeven
Date: Sat Mar 04 2023 - 09:47:57 EST


Hi Jean-Jacques,

On Sat, Mar 4, 2023 at 11:34 AM Jean-Jacques Hiblot
<jjhiblot@xxxxxxxxxxxxxxx> wrote:
> On 02/03/2023 08:49, Geert Uytterhoeven wrote:
> > On Wed, Mar 1, 2023 at 7:53 PM Jean-Jacques Hiblot
> > <jjhiblot@xxxxxxxxxxxxxxx> wrote:
> >> of_irq_parse_one() does a get() on the device node returned in out_irq->np.
> >> Callers of of_irq_parse_one() must do a put() when they are done with it.
> >
> > What does "be done with it" really mean here?
> >
> >> Signed-off-by: Jean-Jacques Hiblot <jjhiblot@xxxxxxxxxxxxxxx>
> >
> >> --- a/arch/arm/mach-shmobile/regulator-quirk-rcar-gen2.c
> >> +++ b/arch/arm/mach-shmobile/regulator-quirk-rcar-gen2.c
> >> @@ -184,6 +184,7 @@ static int __init rcar_gen2_regulator_quirk(void)
> >> kfree(quirk);
> >> continue;
> >> }
> >> + of_node_put(argsa->np);
> >
> > The quirk object, which is a container of argsa, is still used below,
> > and stored in a linked list. I agree argsa->np is not dereferenced,
> > but the pointer itself is still compared to other pointers.
>
> I fail to see when the pointers are compared. It looks to me that only
> the args are compared. Am I missing something ?

You're right, in upstream, there is no such check.
In my local tree, I have converted the comparisons below to use a new
helper of_phandle_args_eq() (which does compare the np member, too),
but that change never went upstream, as the other user of that helper
was rejected.

> In any case, looking more closely at the code, I guess that indeed the
> of_node_put() shouldn't be added here because this code expects that the
> nodes never go away. That is probably a good assertion in case of PMICs

OK.

> > IIUIC, calling of_node_put() might cause the reference count to drop to
> > zero, and the underlying struct node object to be deallocated.
> > So when a future reference to the same DT node will be taken, a new
> > struct node object will be allocated, and the pointer comparison below
> > will fail?

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds