Re: [PATCH] kprobes,x86: disable irq durinr optimized callback

From: Masami Hiramatsu
Date: Tue Apr 26 2011 - 20:51:52 EST


(2011/04/26 23:19), Jiri Olsa wrote:
> On Tue, Apr 26, 2011 at 09:46:25AM -0400, Steven Rostedt wrote:
>> On Tue, Apr 26, 2011 at 03:01:31PM +0200, Jiri Olsa wrote:
>>> hi,
>>>
>>> attached patch is disabling irqs during optimized callback,
>>> so we dont miss any in-irq kprobes as missed.
>>>
>>> Also I think there's small window where current_kprobe variable
>>> could be touched in non-safe way, but I was not able to hit
>>> any issue.
>>>
>>> I'm not sure wether this is a bug or if it was intentional to have
>>> irqs enabled during the pre_handler callback.
>>
>> That's not very convincing. Did you see if we actually did miss events.
>> If that's the case then it is a bug. The conversion to optimizing should
>> not cause events to be missed.
>
> yep, running following:
>
> # cd /debug/tracing/
> # echo "p mutex_unlock" >> kprobe_events
> # echo "p _raw_spin_lock" >> kprobe_events
> # echo "p smp_apic_timer_interrupt" >> ./kprobe_events
> # echo 1 > events/enable
>
> makes the optimized kprobes to be missed. They are not missed in
> same testcase for non-optimized kprobes. I should have mentioned
> that, sry ;)

Good catch! that's right! kprobes' int3 automatically disables
irq, but optimized path doesn't. And that causes unexpected
event loss.


>
>>
>>
>>>
>>> wbr,
>>> jirka
>>>
>>> ---
>>> Disabling irqs during optimized callback, so we dont miss
>>> any in-irq kprobes as missed.
>>>
>>> Interrupts are also disabled during non-optimized kprobes callbacks.
>>>
>>> Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
>>> ---
>>> arch/x86/kernel/kprobes.c | 3 +++
>>> 1 files changed, 3 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
>>> index c969fd9..917cb31 100644
>>> --- a/arch/x86/kernel/kprobes.c
>>> +++ b/arch/x86/kernel/kprobes.c
>>> @@ -1183,11 +1183,13 @@ static void __kprobes optimized_callback(struct optimized_kprobe *op,
>>> struct pt_regs *regs)
>>> {
>>> struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
>>> + unsigned long flags;
>>>
>>> /* This is possible if op is under delayed unoptimizing */
>>> if (kprobe_disabled(&op->kp))
>>> return;
>>>
>>> + local_irq_save(flags);
>>> preempt_disable();
>>
>> No reason to disable preemption if you disabled interrupts.

Right,

> ops, missed that.. attaching new patch


> ---
> Disabling irqs during optimized callback, so we dont miss
> any in-irq kprobes as missed.
>
> running following:
>
> # cd /debug/tracing/
> # echo "p mutex_unlock" >> kprobe_events
> # echo "p _raw_spin_lock" >> kprobe_events
> # echo "p smp_apic_timer_interrupt" >> ./kprobe_events
> # echo 1 > events/enable
>
> makes the optimized kprobes to be missed. None is missed
> if the kprobe optimatization is disabled.
>
> Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>

Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@xxxxxxxxxxx>


Ingo, could you pull this as a bugfix?

Thank you!


> ---
> arch/x86/kernel/kprobes.c | 5 +++--
> 1 files changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
> index c969fd9..f1a6244 100644
> --- a/arch/x86/kernel/kprobes.c
> +++ b/arch/x86/kernel/kprobes.c
> @@ -1183,12 +1183,13 @@ static void __kprobes optimized_callback(struct optimized_kprobe *op,
> struct pt_regs *regs)
> {
> struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
> + unsigned long flags;
>
> /* This is possible if op is under delayed unoptimizing */
> if (kprobe_disabled(&op->kp))
> return;
>
> - preempt_disable();
> + local_irq_save(flags);
> if (kprobe_running()) {
> kprobes_inc_nmissed_count(&op->kp);
> } else {
> @@ -1207,7 +1208,7 @@ static void __kprobes optimized_callback(struct optimized_kprobe *op,
> opt_pre_handler(&op->kp, regs);
> __this_cpu_write(current_kprobe, NULL);
> }
> - preempt_enable_no_resched();
> + local_irq_restore(flags);
> }
>
> static int __kprobes copy_optimized_instructions(u8 *dest, u8 *src)


--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@xxxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/