Date: Fri, 5 Feb 1999 21:15:57 +1100 From: Bruce Evans <bde@zeta.org.au> To: adam@veda.is, mike@smith.net.au Cc: adam@FreeBSD.ORG, cvs-all@FreeBSD.ORG, cvs-committers@FreeBSD.ORG, nate@mt.sri.com, phk@critter.freebsd.dk Subject: Re: cvs commit: src/sys/i386/conf GENERIC Message-ID: <199902051015.VAA01285@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>> > This should be backed out, the correct fix for these systems is to >> > >> > sysctl -w kern.timecounter.method=1 This seems backwards. How can a fix for too-small h/w timecounter wraparound time affect systems using the TSC for a timecounter? The wraparound time for a 500MHz TSC is a measly 1169 years. I don't understand how the fix helps for i8254 timecounters either -- clkintr() in effect reads the i8254, and if clkintr() doesn't get called often enough (every 2/hz - delta) seconds, then there isn't much chance of extra i8254 reads due to method=1. Perhaps there are just a few critical i8254 reads after apm wakes things up (internal state of i8254_get_timecount() is hosed at wakeup time but it may get fixed as a side effect of method=1). >> Fair enough, but it should be documented somewhere obvious (i.e. as an >> agreeable comment in GENERIC) to keep POLA. I certainly didn't expect the >> removal of apm to mess up the timecounter. >The code in question can actually detect the need for this sysctl >adjustment. I again request that it make it as required. How can it do that? I can't detect any need for this sysctl :-). Bruce 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?199902051015.VAA01285>