From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <49F0704D.5070408@domain.hid> Date: Thu, 23 Apr 2009 15:42:37 +0200 From: Gilles Chanteperdrix MIME-Version: 1.0 References: <1240479431.6990.30.camel@domain.hid> <1240482629.7599.45.camel@domain.hid> <49F05891.8040408@domain.hid> <1240493674.16510.36.camel@domain.hid> In-Reply-To: <1240493674.16510.36.camel@domain.hid> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Subject: Re: [Xenomai-help] adeos patch 2.6.28.9-x86-2.2-07 feedback List-Id: Help regarding installation and common use of Xenomai List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Vikesh Rambaran Cc: xenomai-help Vikesh Rambaran wrote: > Hi Gilles > > On Thu, 2009-04-23 at 14:01 +0200, Gilles Chanteperdrix wrote: >> Vikesh Rambaran wrote: >>> I am going to test with CONFIG_NO_HERTZ and ACPI_PROCESSOR disabled >>> as per Gilles suggestions on other emails. >> Yes, you should disable ACPI_PROCESSOR. I am surprised that it is not >> already done, because it is (or should be) written everywhere. >> > > Thanx for your reply. It is well documented on the mailing lists ..:) > > However, I have previously disabled ACPI completely (not just > ACPI_PROCESSOR) ONLY because it gave the lowest latencies across all the > platforms which i tested. Well that's the way it seemed for our > application. (unfortunately, i know this is not what's recommended on > the lists) > > To retest, I have rebuilt the kernel 2.6.28.9 with ACPI enabled, but > ACPI_PROCESSOR disabled. Our applications latency report goes up > to 250uS for the few minutes that it was run on the Core 2 PC. Ok. We definitely need a trace here. I would bet for SMI here, but it could also be an error in the I-pipe patch. So, please enable the I-pipe tracer, and run latency -f. You will find some documentation on the I-pipe tracer here: http://www.xenomai.org/index.php/I-pipe:Tracer > > On the Core Duo laptop, the above kernel and application give latencies > < 40uS. By the way, is it ok to use a kernel compiled for Core2 on a > Core Duo platform ? It depends, Core 2 duo is 64 bits whereas Core duo is 32 bits, so, obviously a 64 bits kernel will not run on a 32 bits machine. > > The latency test does not show any significant difference whether ACPI > is completely disabled or only ACPI_PROCESSOR is disabled. Disabling ACPI completely is known to fail on some platforms, and may cause SMI interrupts on others, which is why we advise people to enable it. But on some platform, it may not make any difference. -- Gilles.