From: "Michael S. Zick" Subject: Re: [parisc-linux] Good news for d380 but sad news for n4k with newest rwlock implementation ; -) Date: Wed, 13 Sep 2006 23:16:00 -0500 Message-ID: <200609132316.00952.mszick@morethan.org> References: <4508505F.8030407@scarlet.be> <200609131410.51839.mszick@morethan.org> <45086ED6.2070109@scarlet.be> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Cc: parisc-linux@lists.parisc-linux.org To: Joel Soete Return-Path: In-Reply-To: <45086ED6.2070109@scarlet.be> List-Id: parisc-linux developers list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: parisc-linux-bounces@lists.parisc-linux.org On Wed September 13 2006 15:49, Joel Soete wrote: > > Michael S. Zick wrote: > > On Wed September 13 2006 13:39, Joel Soete wrote: > >> Hello all, > >> > >> The good point is that k 2.6.18-rc5-pa1 (32bit smp) already run my kernel build loop for 10 days of test without any pb ;-) > >> That is progress - the D380 used to be the one that failed first - now that is around 5 times longer than ever before. > >> Otoh, the 2.6.18-rc7-pa0 (64bit smp) boot well on n4k but didn't survive 10 minutes (here there are an additional tar/untar > >> loop) to stress test. > >> This may also be good news - with the general problems taken care of, now we have something specific to the N4K to locate. And having it fail in 10 minutes is a lot nicer than failing in 10 days. > > > > Finally, I think we forced the kernel to give us some usable information! > > > > Besides the kernel version and kernel width (32-bit & 64-bit) your earlier > > posting of the /proc/cpuinfo messages shows the d380 is a 32-byte cache-line > for c110 yes but for d380 I would like to check (this a p8000 but runing 32bit kernel so?) > My bad, it was the c110 with the 32-byte cache-line, I lost track of what was what. > (specialy usefull to me to learn how to finish the jda stuff) > IIRC, Since spinlock_t is a compile-time size, Dave thought that all we could do was print a warning if the kernel was booted on a machine with a cache-line size different than what it was compiled for. To do that requires the cache-line size be put into that pdc data structure; which is part of the /proc/cpuinfo and dmesg changes. Since application programs may not have to be compile-time sized, having the cache-line size in /proc/cpuinfo could be useful for other purposes also. I seem to recall that Dave passed on those changes to Kyle, but I may be wrong in my recollection. > > and the n4k is a 64-byte cache-line machine. > > > Curious to read results on pa8800 system (if somebody could take try on such system?) > > > What patches where applied to this kernel? > > Willy's spinlock changes? > > The timer changes? > > The interrupt nesting changes? > > The /proc/cpuinfo changes? > > > on d180 only "Willy's spinlock changes" (as per pa1) > on n4k all. > Perhaps those changes should get another look for 64-bit problems before pulling them to locate the one that took the machine down so fast. > Joel Mike _______________________________________________ parisc-linux mailing list parisc-linux@lists.parisc-linux.org http://lists.parisc-linux.org/mailman/listinfo/parisc-linux