[PATCH] ARM: Fix incorrect use of smp_processor_id() by syzbot report

[Dmitry Vyukov]

On Fri, Mar 12, 2021 at 5:13 AM <qiang.zhang at windriver.com> wrote:
>
> From: Zqiang <qiang.zhang at windriver.com>
>
> BUG: using smp_processor_id() in preemptible [00000000] code:
> syz-executor.0/15841
> caller is debug_smp_processor_id+0x20/0x24
> lib/smp_processor_id.c:64
>
> The smp_processor_id() is used in a code segment when
> preemption has been disabled, otherwise, when preemption
> is enabled this pointer is usually no longer useful
> since it may no longer point to per cpu data of the
> current processor.
>
> Reported-by: syzbot <syzbot+a7ee43e564223f195c84 at syzkaller.appspotmail.com>
> Fixes: f5fe12b1eaee ("ARM: spectre-v2: harden user aborts in kernel space")
> Signed-off-by: Zqiang <qiang.zhang at windriver.com>
> ---
>  arch/arm/include/asm/system_misc.h | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/arch/arm/include/asm/system_misc.h b/arch/arm/include/asm/system_misc.h
> index 66f6a3ae68d2..61916dc7d361 100644
> --- a/arch/arm/include/asm/system_misc.h
> +++ b/arch/arm/include/asm/system_misc.h
> @@ -21,8 +21,10 @@ typedef void (*harden_branch_predictor_fn_t)(void);
>  DECLARE_PER_CPU(harden_branch_predictor_fn_t, harden_branch_predictor_fn);
>  static inline void harden_branch_predictor(void)
>  {
> +       preempt_disable();
>         harden_branch_predictor_fn_t fn = per_cpu(harden_branch_predictor_fn,
>                                                   smp_processor_id());
> +       preempt_enable();
>         if (fn)
>                 fn();
>  }

Hi Qiang,

If the CPU can change here, what if it changes right after preempt_enable()?
Disabling preemption just around reading the callback looks like a
no-op. Shouldn't we disable preemption at least around reading and
calling the callback?

On the second look, the fn seems to be const after init, so maybe we
need to use raw_smp_processor_id() instead with an explanatory
comment?