Date: Wed, 25 Apr 2001 18:33:17 -0700 (PDT) From: Matthew Jacob <mjacob@feral.com> To: John Baldwin <jhb@FreeBSD.org> Cc: cvs-all@FreeBSD.org, cvs-committers@FreeBSD.org Subject: RE: cvs commit: src/sys/alpha/alpha mp_machdep.c Message-ID: <Pine.BSF.4.21.0104251832590.6780-100000@beppo.feral.com> In-Reply-To: <XFMail.010425181100.jhb@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
My opinion is that I'd rather have the smp patch in than out. FWIW, which ain't much. On Wed, 25 Apr 2001, John Baldwin wrote: > > On 25-Apr-01 John Baldwin wrote: > > > > On 25-Apr-01 John Baldwin wrote: > >> > >> On 25-Apr-01 John Baldwin wrote: > >>> jhb 2001/04/25 10:24:57 PDT > >>> > >>> Modified files: > >>> sys/alpha/alpha mp_machdep.c > >>> Log: > >>> - Make the dumping of console messages from the secondary CPU's to the > >>> kernel console be #ifdef DIAGNOSTIC. > >>> - Don't set ktr_mask in release_aps(). > >> > >> Unfortunately, top of the tree still panic's (kind of cute when two cpu's > >> panic > >> at the exact same time and clobber each other's panic messages.). I'm > >> working > >> on smp.patch next which doesn't panic but changes many things. I'm still > >> trying to figure out why each CPU is doling out 115% of p_cpticks for each > >> schedcpu() update. > > > > Well, it turns out that on the dual 300 21164, the overhead of witness and > > invariants (probably mostly witness) delays schedcpu() enough that p_cpticks > > can end up as high as 146 for a second even though stathz is 128. Turning > > off > > witness made this go away. :( Anyways, I'll be posting my smp.patch which > > is > > known to work on both UP and SMP alpha and x86. In fact, for the alpha we > > can > > even run an SMP kernel on UP machines, so for 5.0 we should be able to enable > > SMP in GENERIC if that is desired. I'll post a URL and description of the > > patch to -smp shortly. > > Well, apparently I was looking at the wrong xterm or smoking crack, because I'm > still getting somewhat bogus p_pctcpu's even without witness: > > 10 root -16 0 0K 0K RUN 1 4:20 115.04% 115.04% idle: cpu > 11 root -16 0 0K 0K CPU0 0 4:21 114.84% 114.84% idle: cpu > 437 jhb 20 0 2056K 1912K pause 0 0:00 2.23% 0.88% tcsh > 436 root 4 0 4568K 3112K select 0 0:00 0.62% 0.24% sshd > 13 root -48 -167 0K 0K *Giant 0 0:02 0.00% 0.00% swi6: tty:s > 407 jhb 4 0 3040K 2088K select 1 0:01 0.00% 0.00% top > > Notably, the idle processes are rather bogus. The problem is that we are > ending up with p_cpticks up around 145 or 146 when schedcpu() is run, even > though stathz is 128 on the alpha. Not sure why this is broken, but having > somewhat off p_pctcpu but an otherwise stable system is much better than > panic'ing at boot, so I'd like to commit smp.patch even with this known bug. > > -- > > John Baldwin <jhb@FreeBSD.org> -- http://www.FreeBSD.org/~jhb/ > PGP Key: http://www.baldwin.cx/~john/pgpkey.asc > "Power Users Use the Power to Serve!" - http://www.FreeBSD.org/ > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0104251832590.6780-100000>