From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rick Jones Subject: Re: Soft Lockups on 2.6.28-rc8 under netperf bulk receive workload Date: Wed, 17 Dec 2008 11:44:09 -0800 Message-ID: <49495689.9090205@hp.com> References: <4948566E.1050309@hp.com> <494934C5.5000003@chelsio.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Cc: Linux Network Development list , linux-ia64@vger.kernel.org To: Divy Le Ray Return-path: In-Reply-To: <494934C5.5000003@chelsio.com> Sender: linux-ia64-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Divy Le Ray wrote: > Can you please reconfigure your kernel with the following kernel hacking > options enabled, and run your tests again? > > Kernel hacking > Kernel debugging > Detect soft lockups > RT mutex debugging > Spinlock and rw-lock debugging: basic checks > Mutex debugging: basic checks Got the above > Lock debugging: detect incorrect freeing of live locks > Lock debugging: prove locking correctness I don't see those two above in a make menuconfig > Spinlock debugging: sleep-inside-spinlock checking > Compile the kernel with debug info Got those above. > Compile the kernel with frame pointers That doesn't appear in make menuconfig in kernel hacking either. I built a kernel with those I could find enabled, and I think it invoked the spirit of Heisenberg - no more soft lockups appearing on the console. I think it has also _significantly_ dropped the throughput. I've since learned how to tell caliper one doesn't care about the per-process stuff and just lump it all together. The cycles and cgprof profiles for the debug run can be found under the same URL as before: ftp://ftp.netperf.org/lockup with what I hope are obvious names. rick jones