2018-07-12 15:53:53

by Nicholas Mc Guire

Subject: [PATCH] ARM: STi: at least warn if of_iomap fails

The call to of_iomap() is unchecked but scu_enable(), which the returned
address is passed on to, assumes a valid mapping. If the mapping is
invalid this could probably lead to undefined system state so at least
a warning should be issued.

Signed-off-by: Nicholas Mc Guire <[email protected]>
Fixes: commit 65ebcc115889 ("ARM: sti: Add STiH415 SOC support")
---
Problem was found by an experimental coccinelle script

Patch was compile tested with: multi_v7_defconfig (implies
CONFIG_ARCH_STI=y, CONFIG_SMP=y)

Patch is against 4.18-rc3 (localversion-next is next-20180712)

arch/arm/mach-sti/platsmp.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/arch/arm/mach-sti/platsmp.c b/arch/arm/mach-sti/platsmp.c
index 231f19e..89ae76f 100644
--- a/arch/arm/mach-sti/platsmp.c
+++ b/arch/arm/mach-sti/platsmp.c
@@ -109,6 +109,7 @@ static void __init sti_smp_prepare_cpus(unsigned int max_cpus)

if (np) {
scu_base = of_iomap(np, 0);
+ WARN_ON(!scu_base);
scu_enable(scu_base);
of_node_put(np);
}
--
2.1.4

2018-07-16 07:56:26

by Patrice CHOTARD

Subject: Re: [PATCH] ARM: STi: at least warn if of_iomap fails

Hi Nicholas

On 07/12/2018 05:48 PM, Nicholas Mc Guire wrote:
> The call to of_iomap() is unchecked but scu_enable(), which the returned
> address is passed on to, assumes a valid mapping. If the mapping is
> invalid this could probably lead to undefined system state so at least
> a warning should be issued.
>
> Signed-off-by: Nicholas Mc Guire <[email protected]>
> Fixes: commit 65ebcc115889 ("ARM: sti: Add STiH415 SOC support")
> ---
> Problem was found by an experimental coccinelle script
>
> Patch was compile tested with: multi_v7_defconfig (implies
> CONFIG_ARCH_STI=y, CONFIG_SMP=y)
>
> Patch is against 4.18-rc3 (localversion-next is next-20180712)
>
> arch/arm/mach-sti/platsmp.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/arch/arm/mach-sti/platsmp.c b/arch/arm/mach-sti/platsmp.c
> index 231f19e..89ae76f 100644
> --- a/arch/arm/mach-sti/platsmp.c
> +++ b/arch/arm/mach-sti/platsmp.c
> @@ -109,6 +109,7 @@ static void __init sti_smp_prepare_cpus(unsigned int max_cpus)
>
> if (np) {
> scu_base = of_iomap(np, 0);
> + WARN_ON(!scu_base);
> scu_enable(scu_base);
> of_node_put(np);
> }
>

I should prefer to exit with an error to avoid unpredictable behavior,
something like this is better :

scu_base = of_iomap(np, 0);
of_node_put(np);
if (!scu_base) {
pr_err("No SCU remap\n");
return;
}

scu_enable(scu_base);


Thanks

2018-07-16 08:49:30

by Nicholas Mc Guire

Subject: Re: [PATCH] ARM: STi: at least warn if of_iomap fails

On Mon, Jul 16, 2018 at 07:55:08AM +0000, Patrice CHOTARD wrote:
> Hi Nicholas
>
> On 07/12/2018 05:48 PM, Nicholas Mc Guire wrote:
> > The call to of_iomap() is unchecked but scu_enable(), which the returned
> > address is passed on to, assumes a valid mapping. If the mapping is
> > invalid this could probably lead to undefined system state so at least
> > a warning should be issued.
> >
> > Signed-off-by: Nicholas Mc Guire <[email protected]>
> > Fixes: commit 65ebcc115889 ("ARM: sti: Add STiH415 SOC support")
> > ---
> > Problem was found by an experimental coccinelle script
> >
> > Patch was compile tested with: multi_v7_defconfig (implies
> > CONFIG_ARCH_STI=y, CONFIG_SMP=y)
> >
> > Patch is against 4.18-rc3 (localversion-next is next-20180712)
> >
> > arch/arm/mach-sti/platsmp.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/arch/arm/mach-sti/platsmp.c b/arch/arm/mach-sti/platsmp.c
> > index 231f19e..89ae76f 100644
> > --- a/arch/arm/mach-sti/platsmp.c
> > +++ b/arch/arm/mach-sti/platsmp.c
> > @@ -109,6 +109,7 @@ static void __init sti_smp_prepare_cpus(unsigned int max_cpus)
> >
> > if (np) {
> > scu_base = of_iomap(np, 0);
> > + WARN_ON(!scu_base);
> > scu_enable(scu_base);
> > of_node_put(np);
> > }
> >
>
> I should prefer to exit with an error to avoid unpredictable behavior,
> something like this is better :
>
> scu_base = of_iomap(np, 0);
> of_node_put(np);
> if (!scu_base) {
> pr_err("No SCU remap\n");
> return;
> }
>
> scu_enable(scu_base);
>
even though it would be possible to locate that then from that message
would it not be prefereable to provide the infos upfront like:

pr_err("SCU remap failed at %s:%s():%d\n",
__FILE__, __func__, __LINE__);

if that makes sense I?ll resend - or is that considered too verbouse ?

thx!
hofrat