[Qemu-devel] Summary of MTTCG Call 3/8/2015

[Qemu-devel] Summary of MTTCG Call 3/8/2015

[Alex Bennée]

Hi,

Seeing as the last summary was useful I thought I'd send out some fresh
notes:

Alex discussed his barrier tests which where posted to the list last
week. There is some behaviour that needs explaining w.r.t to barrier
only approaches which should fail in real hardware but don't. Alvise
prompted he notices the Load Acquire/Store Release semantics were new
for ARMv8 and Alex replied we'd need to look at the barrier stuff later
but they likely need testing on non-x86 backends where we'll need to be
careful to express full barrier semantics.

Fred talked about progress on the main patch set which is re-based on
the recent -rc0 release. He hopes to post the latest set of patches this
week.

Alvise also expects to post updated patch series for the
atomic/exclusive code this week. He's maintaining two trees at the
moment, one based direct off mainline QEMU and then a MTTCG aware
version. There was some discussion about how to proceed with the other
backends. The general consensus was as long as the patches didn't break
the other builds we could submit to the mainline list and ask for help
from other TCG experts. The changes are expected to be fairly
mechanical.

There was some general discussion about the current status of MTTCG
and the consensus was we are generally happy with the functionality.
There where a number of comments about making the MTTCG build friendly
to being disabled on different platforms (v4 was broken on
non-x86/non-arm). We shall see what the community thinks once the next
set of patches goes out.

Mark asked for advanced copies of the presentation Alex is giving at
KVMForum and he replied he'd hope to get wider circulation this week
after an internal review. He would also contact Fred offline to discuss
the demo options for the talk.

Finally it was agreed with KVM Forum only 2 weeks away to have the next
conference on Monday.


-- 
Alex Bennée

Re: [Qemu-devel] Summary of MTTCG Call 3/8/2015

[Claudio Fontana]

Thank you Alex!

Ciao

Claudio

-----Original Message-----
From: Alex Bennée [mailto:alex.bennee@linaro.org] 
Sent: Mittwoch, 5. August 2015 14:25
To: mttcg@listserver.greensocs.com; Alvise Rigo; pbonzini@redhat.com; Frederic Konrad; Jan Kiszka
Cc: Claudio Fontana; Mark Burton; qemu-devel@nongnu.org
Subject: Summary of MTTCG Call 3/8/2015

Hi,

Seeing as the last summary was useful I thought I'd send out some fresh
notes:

Alex discussed his barrier tests which where posted to the list last
week. There is some behaviour that needs explaining w.r.t to barrier
only approaches which should fail in real hardware but don't. Alvise
prompted he notices the Load Acquire/Store Release semantics were new
for ARMv8 and Alex replied we'd need to look at the barrier stuff later
but they likely need testing on non-x86 backends where we'll need to be
careful to express full barrier semantics.

Fred talked about progress on the main patch set which is re-based on
the recent -rc0 release. He hopes to post the latest set of patches this
week.

Alvise also expects to post updated patch series for the
atomic/exclusive code this week. He's maintaining two trees at the
moment, one based direct off mainline QEMU and then a MTTCG aware
version. There was some discussion about how to proceed with the other
backends. The general consensus was as long as the patches didn't break
the other builds we could submit to the mainline list and ask for help
from other TCG experts. The changes are expected to be fairly
mechanical.

There was some general discussion about the current status of MTTCG
and the consensus was we are generally happy with the functionality.
There where a number of comments about making the MTTCG build friendly
to being disabled on different platforms (v4 was broken on
non-x86/non-arm). We shall see what the community thinks once the next
set of patches goes out.

Mark asked for advanced copies of the presentation Alex is giving at
KVMForum and he replied he'd hope to get wider circulation this week
after an internal review. He would also contact Fred offline to discuss
the demo options for the talk.

Finally it was agreed with KVM Forum only 2 weeks away to have the next
conference on Monday.


-- 
Alex Bennée