Re: [Qemu-devel] QEMU Profiling

[Wei-Ren Chen]

On Tue, Feb 26, 2013 at 05:16:48PM +0400, Евард Вадим wrote:
> Hi list,
> Our team is now surveying various binary translation and emulation tools, and
> as part of this work we have to measure time spent in different parts of the
> process.
> More precisely, I'd like to make this work on x86_64-linux-user qemu without
> any accelerations.
> If article [1] is still relevant, I see the following places to spend time:
> 1. Locate end of translation block
> 2. Disassemble it to IR
> 3. Assemble x86_64 machine code
> 4. Execute it
> 5. Process syscalls
> 6. .. What else?

  The flow above is right, I would suggest you reading the source code
to get more familar with QEMU.
 
> What is the best method to measure these times you'd suggest?
> ./configure mentions 4 things I found relevant:
> 1. --enable-debug-tcg (didn't look at it yet)
> 2. --enable-trace-backend=B
> 3. --enable-profiler (some things like qemu_time seem relevant, didn't look
> closely yet)
> 4. --enable-gprof (I encountered problems when building and/or running qemu
> with gprof)

  You can get profiling about QEMU itself by using usual measurement, I
guess. The problem is how you get the time spent is the code cache. I
guess you have to spot where QEMU jump into the code cache, and
when/where it leave the code cache. Then insert your profiling code
there.

HTH,
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