Re: Re: Re: Re: [PATCH -tip v6 06/22] [BUGFIX] x86: Prohibit probingon memcpy/memset

From: Jovi Zhangwei
Date: Tue Dec 24 2013 - 04:53:59 EST


On Tue, Dec 24, 2013 at 4:32 PM, Masami Hiramatsu
<masami.hiramatsu.pt@xxxxxxxxxxx> wrote:
> (2013/12/24 15:39), Jovi Zhangwei wrote:
>> On Mon, Dec 23, 2013 at 6:59 PM, Masami Hiramatsu
>> <masami.hiramatsu.pt@xxxxxxxxxxx> wrote:
>>> (2013/12/23 13:51), Jovi Zhangwei wrote:
>>>> On Fri, Dec 20, 2013 at 5:21 PM, Masami Hiramatsu
>>>> <masami.hiramatsu.pt@xxxxxxxxxxx> wrote:
>>>>> (2013/12/20 17:31), Jovi Zhangwei wrote:
>>>>>> On Fri, Dec 20, 2013 at 12:42 PM, Masami Hiramatsu
>>>>>> <masami.hiramatsu.pt@xxxxxxxxxxx> wrote:
>>>>>>> (2013/12/20 12:07), Jovi Zhangwei wrote:
>>>>>>>> On Fri, Dec 20, 2013 at 10:37 AM, Masami Hiramatsu
>>>>>>>> <masami.hiramatsu.pt@xxxxxxxxxxx> wrote:
>>>>>>>>> Hi Jovi,
>>>>>>>>>
>>>>>>>>> (2013/12/19 18:37), Jovi Zhangwei wrote:
>>>>>>>>>> Hi Masami,
>>>>>>>>>>
>>>>>>>>>> On Thu, Dec 19, 2013 at 5:04 PM, Masami Hiramatsu
>>>>>>>>>> <masami.hiramatsu.pt@xxxxxxxxxxx> wrote:
>>>>>>>>>>> memcpy/memset functions are fundamental functions and
>>>>>>>>>>> those are involved in kprobe's exception handling.
>>>>>>>>>>> Prohibit probing on them to avoid kernel crash.
>>>>>>>>>>>
>>>>>>>>>> Would you please let me know the LKML link of that bugfix, I cannot
>>>>>>>>>> find it in my LKML fold.
>>>>>>>>>
>>>>>>>>> Yeah, that was found in my testing environment.
>>>>>>>>>
>>>>>>>>>> No objection on this patch. :) just want to know more, It seems there
>>>>>>>>>> have no problem to probe memcpy in my box, maybe I didn't hit the
>>>>>>>>>> crash code path.
>>>>>>>>>
>>>>>>>>> Ah, I see. Originally the problem happened when I put a probe on
>>>>>>>>> __memcpy. And it looks the instances of memcpy and __memcpy are
>>>>>>>>> same on x86-64. Thus I decided to blacklist both. (memset/__memset too)
>>>>>>>>> Have you ever tried to probe __memcpy on your box?
>>>>>>>>>
>>>>>>>> Hmm, still no crash, __memcpy and __memset are both tested.
>>>>>>>>
>>>>>>>> I use below kprobe related config:
>>>>>>>>
>>>>>>>> CONFIG_KPROBES=y
>>>>>>>> CONFIG_JUMP_LABEL=y
>>>>>>>> CONFIG_OPTPROBES=y
>>>>>>>> CONFIG_KPROBES_ON_FTRACE=y
>>>>>>>
>>>>>>> Hmm, I've added some debugging options.
>>>>>>>
>>>>>>> CONFIG_SLUB_DEBUG=y
>>>>>>> CONFIG_X86_DEBUGCTLMSR=y
>>>>>>> CONFIG_PNP_DEBUG_MESSAGES=y
>>>>>>> CONFIG_DEBUG_INFO=y
>>>>>>> CONFIG_DEBUG_FS=y
>>>>>>> CONFIG_DEBUG_KERNEL=y
>>>>>>> CONFIG_DEBUG_STACK_USAGE=y
>>>>>>> CONFIG_DEBUG_MEMORY_INIT=y
>>>>>>> CONFIG_DEBUG_STACKOVERFLOW=y
>>>>>>> CONFIG_DEBUG_SPINLOCK=y
>>>>>>> CONFIG_DEBUG_MUTEXES=y
>>>>>>> CONFIG_DEBUG_LOCK_ALLOC=y
>>>>>>> CONFIG_DEBUG_LOCKDEP=y
>>>>>>> CONFIG_DEBUG_BUGVERBOSE=y
>>>>>>> CONFIG_DEBUG_RODATA=y
>>>>>>> CONFIG_DEBUG_BOOT_PARAMS=y
>>>>>>>
>>>>>>> I guess some of them might cause it.
>>>>>>>
>>>>>> I recompiled the kernel with those config enabled, unfortunately still no crash,
>>>>>> I tested on 3.13.0-rc4, a fedora kvm box.
>>>>>
>>>>> Hmm, it's very odd. I'm running 3.13.0-rc4 x86-64 on the fedora
>>>>> kvm box too. here is the full of my kconfig.
>>>>>
>>>> That configuration is still working for me, no crash for memcpy kprobe test.
>>>
>>> Would you do __memcpy test? or memcpy test? I only had a crash on the
>>> __memcpy(__memset).
>>>
>> Still no crash, use your kernel config.
>> memcpy and __memcpy have same address in /proc/kallsyms.
>>
>> Looks like a interesting problem.
>
> Agreed. In my case, those have same address, but only probing
> __memcpy caused a kernel crash. I'm not sure why, but it is
> safe to disable probing on it.
>
Shall we need dig further to address the root cause? IMO, the kprobe
should act same behavior when given same probe address, but it's look
so weird in your box. :)

Thanks,

Jovi
--
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/