From mboxrd@z Thu Jan 1 00:00:00 1970 From: Philippe Gerum In-Reply-To: <1240482629.7599.45.camel@domain.hid> References: <1240479431.6990.30.camel@domain.hid> <1240482629.7599.45.camel@domain.hid> Content-Type: text/plain Date: Thu, 23 Apr 2009 13:48:08 +0200 Message-Id: <1240487288.6990.71.camel@domain.hid> Mime-Version: 1.0 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 On Thu, 2009-04-23 at 12:30 +0200, Vikesh Rambaran wrote: > Hi > > On Thu, 2009-04-23 at 11:37 +0200, Philippe Gerum wrote: > > On Thu, 2009-04-23 at 11:20 +0200, vikesh rambaran wrote: > > > Hi > > > > > > Thanx for the updated patch. Some feedback ... > > > > > > I tested it with Xenomai 2.4.7 under Ubuntu 8.04.2 on three different > > > platforms > > > (AMD Athlon Desktop, Intel Core 2 Desktop, Intel Core Duo Laptop) > > > > > > Typical system lockups when in the gnome desktop do NOT occur > > > anymore ! Great :) > > > > > > Initial observations when compared to Xenomai 2.4.5 with > > > adeos-ipipe-2.6.24-x86-2.0-07.patch. : > > > > > > 1. Latency test shows a slight increase of about 5 to 10 % on the > > > Intel platforms. > > > (Still need to do the comparison on the AMD platform). This translates > > > to less than 5 uS > > > so it's ok for our case. > > > > > > 2. On the Intel Core 2 Desktop, our application has a much more > > > significant max scheduling jitter > > > of about 125 uS (compared with < 60uS for Xenomia 2.4.5). This > > > 'increased jitter' does not > > > show up on the other two platforms. Still trying to trace this one. > > > > Time for an I-pipe tracer log, with latency -f. I guess this > > pathological jitter should trigger with standard tests as well; if it > > does not, well, there might be other regression spots in the APIs. > > > > .config would help as well. TIA, > > > Thanx Philippe > > To clarify, the latency test has not shown such significant jitter on > any of the platforms. Latency tests only show up to about 40uS > (normal). > > The high jitter shows up in 'our' application using 2.6.28.9-x86-2.2-07 > with Xenomai 2.4.7 on only one of the three platforms. > > [The same application does not show such high latencies using > Xenomai 2.4.5 with adeos-ipipe-2.6.24-x86-2.0-07.patch on any of the > three platforms] > Assuming that you have a mean to detect a latency spot situation from your application, then I would suggest to go straight to the point, and instrument your system as follow: 1) in your app, adapt the following pattern: if (latency_too_high) xntrace_user_freeze(whatever, 0); xntrace_user_freeze() is available to require the I-pipe tracer to take a snapshot from the recent kernel execution path (see include/nucleus/trace.h); with that information, you may find some hints about the source of the latency spot. 2) enable CONFIG_IPIPE_TRACE (including *_MCOUNT) 3) run the app, fetch the trace data from /proc/ipipe/trace/frozen after the latency spot showed up. We should find a "freeze" event tag close to the root of the issue. > Attached are the .config files of the last set of tests done. > Thanks. > I am going to test with CONFIG_NO_HERTZ and ACPI_PROCESSOR disabled > as per Gilles suggestions on other emails. > > > Keep well > Vicki > > > > > > -- Philippe.