[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 12 Dec 2011 15:23:18 +0400
From: Cyrill Gorcunov <gorcunov@...il.com>
To: Ingo Molnar <mingo@...e.hu>
Cc: Sasha Levin <levinsasha928@...il.com>,
Thomas Gleixner <tglx@...utronix.de>,
Pekka Enberg <penberg@...nel.org>,
LKML <linux-kernel@...r.kernel.org>
Subject: Re: [patch 0/3] kvm tool: Serial emulation overhaul
On Mon, Dec 12, 2011 at 12:12:22PM +0100, Ingo Molnar wrote:
>
> * Sasha Levin <levinsasha928@...il.com> wrote:
>
> > 50k cycles for every single byte is pretty much as good as it
> > will get with serial console. See slide #5 in Marcelo's KVM
> > Forum 2010 presentation[1] where he timed a heavyweight exit
> > to about 40k cycles.
>
> > [1]
> > http://www.linux-kvm.org/wiki/images/e/ea/2010-forum-mtosatti_walkthrough_entry_exit.pdf
>
> But what we do here is a PIO exit. That, according to Marcelo's
> measurements, is about 10K cycles, back to back. [*]
>
> So where does the extra overhead come from?
>
> We shouldn't care that there's virtio-console - the goal of
> tools/kvm it speed everything up as much as possible, so we
> should not jump to the next IO abstraction unless we know where
> every cycle was spent with simpler IO models ...
>
> Thanks,
>
> Ingo
>
> [*] Also, those 10K cycles include some significant Qemu
> overhead - a couple of thousand cycles - that should be much
> lower in the tools/kvm case.
>
FWIW, last time I was mcount'ing calltrace -- we spend a lot of time
due to timer signals, ie because of
#define TIMER_INTERVAL_NS 1000000 /* 1 msec */
so, do we really need it being that hight? This poll includes
enquiry if there some symbol a user typed in console. Maybe
we should reduce this rate?
Cyrill
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists