On Wed, Sep 23, 2020 at 11:10:17AM +0200, Geert Uytterhoeven wrote:
> Hi Dan,
>
> On Wed, Sep 23, 2020 at 10:47 AM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> > This code needs to call iounmap() on the error paths.
>
> Thanks for your patch!
>
> > Fixes: 2ed29e15e4b2 ("ARM: shmobile: R-Mobile: Move pm-rmobile to drivers/soc/renesas/")
>
> This is not the commit that introduced the issue.
Duh... I don't know what I was thinking there.
>
> Fixes: 2173fc7cb681c38b ("ARM: shmobile: R-Mobile: Add DT support for
> PM domains")
>
> > Signed-off-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
>
> > --- a/drivers/soc/renesas/rmobile-sysc.c
> > +++ b/drivers/soc/renesas/rmobile-sysc.c
> > @@ -328,6 +328,7 @@ static int __init rmobile_init_pm_domains(void)
> > pmd = of_get_child_by_name(np, "pm-domains");
> > if (!pmd) {
> > pr_warn("%pOF lacks pm-domains node\n", np);
> > + iounmap(base);
>
> This one I can agree with, although that case is a bug in the DTS.
>
> > continue;
> > }
> >
> > @@ -341,6 +342,7 @@ static int __init rmobile_init_pm_domains(void)
> > of_node_put(pmd);
> > if (ret) {
> > of_node_put(np);
> > + iounmap(base);
>
> This one I cannot: in the (unlikely, only if OOM) case
> rmobile_add_pm_domains() returns an error, one or more PM subdomains may
> have been registered already. Hence if you call iounmap() here, the
> code will try to access unmapped registers later, leading to a crash.
>
It's actually impossible for this rmobile_add_pm_domains() to fail on
current kernels because small allocations never fail...
I'll send a v2. This is for a new static checker test that I added to
Smatch so I'm just sending a few of these out every day to collect
feedback for now. So thanks for reviewing this, it's very helpful.
regards,
dan carpenter
