Re: Re: Regarding systemtap support for AArch64

[Masami Hiramatsu <masami dot hiramatsu dot pt at hitachi dot com>]

(2013/10/07 18:50), Sandeepa Prabhu wrote:
> On 5 October 2013 08:54, Masami Hiramatsu
> <masami.hiramatsu.pt@hitachi.com> wrote:
>> (2013/10/04 12:24), Sandeepa Prabhu wrote:
>>>>>>  - Is it really need to use spinlock to protect break_hook?
>>>>> Any cpu can remove breakpoint hooks right, and traversal happen in
>>>>> debug exception context so mutex are not safe (can sleep/schedule out)
>>>>> in debug exception.
>>>>
>>>> I don't think we need to remove the breakpoint hooks after starting
>>>> up the kernel. If we use the spinlock there, we'll pay a big cost
>>>> because of the lock contention.
>>> Not in kprobes. But kgdb can remove breakpoint handler and use same
>>> API. or atleast while providing an api we should not assume race
>>> cannot happen right?
>>
>> In that case, we'd better add a wrapper handler for kgdb so that
>> the list isn't updated even if the kgdb removes its handler.
>>
>>> And there wont be much lock contention, i'ts only if the debug
>>> framework (like kgdb) is wrapping-up, not is normal use-case.
>>
>> Hmm, it seems that the spinlock is locked while handling a breakpoint.
>> This will cause a bad performance issue when we put many kprobes
>> on SMP system.
> arm maintainers prefer a reader/writer spin-locks, so there wont be
> lock contention in debug path, each instance of kprobe hook trap (on
> any CPU) would be a reader, not blocking.

OK for the first step, and it eventually should be fixed to lockless.
(depends on the performance improvement)

>> [...]
>>>>>>  - probes-*.c is not good name for the simulator. those should have
>>>>>>    better name.
>>>>> May be decode-arm64.c? Originally I had decode-* but the logic is
>>>>> limited to kprobes and uprobes only so renamed that way.  Other cases
>>>>> like jump_labels, use different decoding, and may not share same code.
>>>>
>>>> The decoding table will be different from other usecase, but I think
>>>> simulator code can be shared (and must be). It may just get the address,
>>>> instruction, and registers, not the kprobe.
>>> When we wrote at Linaro, our plan was to share simulation calls
>>> between kprobes and uprobes, and planned to use 'struct kprobes' for
>>> both the frameworks, this is how it was done on "arch/arm/kernel/"
>>> effectively.
>>
>> Uh, I should review arm32 again...
>>
>>> If more frameworks can use it (as it seems) I change it to accept
>>> opcode, pc value and saved pt_regs and avoid kprobe struct altogether.
>>> Also, we are starting on uprobes at Linaro, so it won't be too long
>>> before we start thinking about that too ;-)
>>
>> Since kprobes data structure includes many information which is not related
>> to the simulation, I'd like to keep it away from that.
> Yup, can simulate without that. I will avoid kprobe struct from
> simulators and keep it cleaner :-)

Good ;)

[...]
>>> Question:
>>> I am working on v2 patchset based on comments, for next week to post,
>>> do you have basic aarch64 setup (fast-model/hardware), ARM v8-ARM etc?
>>> I mean, how about sharing some efforts with me(Linaro) going further?
>>
>> Yeah, I have a foundation-model simulator, I just need to set it up.
>>
>>> Most work shall go through LAKML so you may have to subscribe to that
>>> ;),  but do you mind working on Linaro hosted public git ? we can lay
>>> out a plan then. After kprobes, we have much work on uprobes in queue
>>> (both 32-bit and 64-bit user-space) and your insights help us, since
>>> you are one of the maintainers of both subsystems.
>>
>> Oh, OK. I'll subscribe it, and, yeah, linaro public git should be
>> better place to work with :)
> Thanks!!  I will reach out to leads in Linaro regarding read/write
> access and stuff (guess it's read-only right now)
> You can have a look at following Linaro cards (agile projecting tracking):
>    https://cards.linaro.org/browse/KWG-13
>    https://cards.linaro.org/browse/CARD-564
> If you find them interesting, you may subscribe to these by creating a
> login, and clicking on  more -> 'watch this issue'.

Looks nice :) I'll do that.

Thank you!


-- 
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com