From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <4854CF37.20001@domain.hid> Date: Sun, 15 Jun 2008 10:13:43 +0200 From: Philippe Gerum MIME-Version: 1.0 References: <4850DC9D.4010607@domain.hid> <2ff1a98a0806120208n48d15ed6m4e67b582dced1ba3@domain.hid> <48510808.9020304@domain.hid> <485174A7.7040505@domain.hid> <485227E2.3030603@domain.hid> <48523BE4.8070403@domain.hid> <48523E11.40702@domain.hid> In-Reply-To: <48523E11.40702@domain.hid> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Subject: Re: [Xenomai-help] High latencies on Core2Duo Reply-To: philippe.gerum@domain.hid List-Id: Help regarding installation and common use of Xenomai List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Sebastian Smolorz Cc: xenomai@xenomai.org Sebastian Smolorz wrote: > Philippe Gerum wrote: >> Sebastian Smolorz wrote: >>> Philippe Gerum wrote: >>>> Could you run another trace with that patch applied? TIA, >>> >>> Here it comes. >>> >> >> Thanks. Is the latency spot regularly happening at the same place >> after a few runs, >> or rather randomly? > > > I can for sure activate it with a kernel compilation with make -j2. > Without load, it is more or less random and more seldom. > Could you try booting with idle=poll just to make sure that we don't get trapped into long wakeup latency from some idling instruction? We explicitly avoid mwait unless forced to use it by a bootparam, but maybe something goes wrong with the default idle code as well. FYI, I have latency spots of the same magnitude on a Dell precision m65 laptop here too, but I also have latency figures < 5 us under extreme load on a four-way Opteron-based workstation (Fujitsu Siemens) with the very same pipeline+Xenomai combo, so there is a significant risk that a hw issue (possibly chipset) bites us here. TIA, -- Philippe.