Re: [PATCH] clocksource: acpi_pm: fix return value of __setup handler

From: Dan Carpenter
Date: Wed Mar 09 2022 - 03:56:59 EST


On Tue, Mar 08, 2022 at 08:03:01PM -0800, Randy Dunlap wrote:
> __setup() handlers should return 1 to obsolete_checksetup() in
> init/main.c to indicate that the boot option has been handled.
> A return of 0 causes the boot option/value to be listed as an Unknown
> kernel parameter and added to init's (limited) environment strings.
>
> The __setup() handler interface isn't meant to handle negative return
> values -- they are non-zero, so they mean "handled" (like a return
> value of 1 does), but that's just a quirk. So return 1 from
> parse_pmtmr(). Also print a warning message if kstrtouint() returns
> an error.

Everyone probably predicted that this API was/is going to cause bugs?

Smatch treats every (struct obs_kernel_param)->setup_func function as
the same so it's impossible to write a static checker for this because
in do_early_param() it expects zero on success and obsolete_checksetup()
has expects 1 on success and also on failure.

init/main.c
732 static int __init do_early_param(char *param, char *val,
733 const char *unused __always_unused,
734 void *arg __always_unused)
735 {
736 const struct obs_kernel_param *p;
737
738 for (p = __setup_start; p < __setup_end; p++) {
739 if ((p->early && parameq(param, p->str)) ||
740 (strcmp(param, "console") == 0 &&
741 strcmp(p->str, "earlycon") == 0)
742 ) {
743 if (p->setup_func(val) != 0)
744 pr_warn("Malformed early option '%s'\n", param);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

745 }
746 }
747 /* We accept everything at this stage. */
748 return 0;
749 }

regards,
dan carpenter