[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20070216194059.GF6425@infidigm.net>
Date: Fri, 16 Feb 2007 14:40:59 -0500
From: Jeff Muizelaar <jeff@...idigm.net>
To: Andi Kleen <andi@...stfloor.org>
Cc: linux-kernel@...r.kernel.org
Subject: Re: Using sched_clock for mmio-trace
On Fri, Feb 16, 2007 at 09:02:44PM +0100, Andi Kleen wrote:
> Jeff Muizelaar <jeff@...idigm.net> writes:
> >
> > The question is, what api should I be using? I need something that can
> > be called from inside interrupt handlers, and obviously the more
> > accurate and the lower the overhead the better.
>
> Use do_gettimeofday(). sched_clock() is not for general use
> and only for some very limited use cases and will give you
> unexpected results in several cases.
>
> There are a few cases where gtod is still a little slow, but these
> are being addressed. In many cases it is fast. In some hardware
> it stays slow, but there is not much that can be done about that
> because of the hardware design.
>
> It works fine from interrupt handlers and other strange contexts.
Ok, I'll probably use it then. How does the overhead of calling
do_gettimteofday() compare to doing an mmio read/write over PCI express?
i.e. is it going to be a performance problem if I call do_gettimeofday
for every mmio read/write?
Also, is there any good reason blk-trace doesn't use do_gettimeofday()?
-Jeff
-
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