[Qemu-devel] oprofile on qemu

[Qemu-devel] oprofile on qemu

[Xin Tong]

Is there a way to profile QEMU's code cache ?

Thanks


Xin

Re: [Qemu-devel] oprofile on qemu

[陳韋任]

On Tue, Nov 22, 2011 at 08:23:32AM -0500, Xin Tong wrote:
> Is there a way to profile QEMU's code cache ?

  What do you try to profile? The execution time in the code
cache?

P.S. The subject is a little misleading. Oprofile is a tool.
     Do you specify you want use Oprofile only?

Regards,
chenwj

-- 
Wei-Ren Chen (陳韋任)
Computer Systems Lab, Institute of Information Science,
Academia Sinica, Taiwan (R.O.C.)
Tel:886-2-2788-3799 #1667
Homepage: http://people.cs.nctu.edu.tw/~chenwj

Re: [Qemu-devel] oprofile on qemu

[Xin Tong]

I oprofiled QEMU with some workloads, i.e. SPECjbb on Ubuntu Linux, i
see QEMU spent about 60 - 70% of the time in the code cache ( code
cache size is 256MB ). but I want to know which TB takes the most
amount of time.

Thanks

Xin


On Wed, Nov 23, 2011 at 3:10 AM, 陳韋任 <chenwj@iis.sinica.edu.tw> wrote:
> On Tue, Nov 22, 2011 at 08:23:32AM -0500, Xin Tong wrote:
>> Is there a way to profile QEMU's code cache ?
>
>  What do you try to profile? The execution time in the code
> cache?
>
> P.S. The subject is a little misleading. Oprofile is a tool.
>     Do you specify you want use Oprofile only?
>
> Regards,
> chenwj
>
> --
> Wei-Ren Chen (陳韋任)
> Computer Systems Lab, Institute of Information Science,
> Academia Sinica, Taiwan (R.O.C.)
> Tel:886-2-2788-3799 #1667
> Homepage: http://people.cs.nctu.edu.tw/~chenwj
>

Re: [Qemu-devel] oprofile on qemu

[Max Filippov]

> I oprofiled QEMU with some workloads, i.e. SPECjbb on Ubuntu Linux, i
> see QEMU spent about 60 - 70% of the time in the code cache ( code
> cache size is 256MB ). but I want to know which TB takes the most
> amount of time.

I doubt that it's possible to profile TB code by external tools without QEMU assistance.
Probably it's easier to account individual TB timing from within QEMU, like it is now done for translation,
see places marked with #ifdef CONFIG_PROFILER.

With TB chaining disabled, TB execution starts with tcg_qemu_tb_exec, there's one call to it in cpu_exec().
TB may exit normally, or via longjmp, so there are two places in cpu_exec() where it must be caught.

Virtual/physical mapping must also be taken care of, but it depends on your application.

Thanks.
-- Max