Re: [PATCH] soc: renesas: renesas-soc: release 'chipid' from ioremap()

[Dan Carpenter]

On Mon, Apr 03, 2023 at 09:12:55AM +0200, Geert Uytterhoeven wrote:
> Hi Dan,
> 
> On Mon, Apr 3, 2023 at 6:29 AM Dan Carpenter <error27@xxxxxxxxx> wrote:
> > On Fri, Mar 31, 2023 at 02:13:10PM +0200, Geert Uytterhoeven wrote:
> > > On Fri, Mar 31, 2023 at 12:14 PM Li Yang <lidaxian@xxxxxxxxxxx> wrote:
> > > > Smatch reports:
> > > >
> > > > drivers/soc/renesas/renesas-soc.c:536 renesas_soc_init() warn:
> > > > 'chipid' from ioremap() not released on lines: 475.
> > > >
> > > > If soc_dev_atrr allocation is failed, function renesas_soc_init()
> > > > will return without releasing 'chipid' from ioremap().
> > > >
> > > > Fix this by adding function iounmap().
> > > >
> > > > Fixes: cb5508e47e60 ("soc: renesas: Add support for reading product revision for RZ/G2L family")
> > > > Signed-off-by: Li Yang <lidaxian@xxxxxxxxxxx>
> > > > Reviewed-by: Dan Carpenter <error27@xxxxxxxxx>
> > >
> > > Thanks for your patch!
> > >
> > > > --- a/drivers/soc/renesas/renesas-soc.c
> > > > +++ b/drivers/soc/renesas/renesas-soc.c
> > > > @@ -471,8 +471,11 @@ static int __init renesas_soc_init(void)
> > > >         }
> > > >
> > > >         soc_dev_attr = kzalloc(sizeof(*soc_dev_attr), GFP_KERNEL);
> > > > -       if (!soc_dev_attr)
> > > > +       if (!soc_dev_attr) {
> > > > +               if (chipid)
> > > > +                       iounmap(chipid);
> > >
> > > We don't really care, as the system is dead at this point anyway.
> >
> > Why even have the check for NULL then?  The kzalloc() is small enough
> 
> Because else someone will submit a patch to add that check? ;-)
> 
> > to the point where it litterally cannot fail.
> 
> I still don't understand how it can be guaranteed that small allocations
> never fail... "while (1) kmalloc(16, GFP_KERNEL);"
> 

I read an lwn article on it and I think I once looked it up to try
figure out how small the definition of "small" was and it was
surprisingly large...  But I have no idea.  I think maybe small atomic
allocations can fail and GFP_KERNEL allocations sleep forever?  (These
guesses are worthless).

> Perhaps we need a different mechanism to annotate error handling code
> that cannot ever happen in a real product, so it can be thrown away by
> the compiler, while still pleasing the static checkers?  All these
> checks and error handling code do affect kernel size.  There are
> Linux products running on SoCs with 8 MiB of internal SRAM.

People sometimes call BUG_ON(!soc_dev_attr).  It's sort of rare these
days.  It would be easy to make a function which silences Smatch...

	__system_is_dead();

regards,
dan carpenter