Date: Mon, 25 Feb 2002 18:13:17 +1100 (EST) From: Bruce Evans <bde@zeta.org.au> To: Matthew Dillon <dillon@apollo.backplane.com> Cc: Terry Lambert <tlambert2@mindspring.com>, Alfred Perlstein <alfred@FreeBSD.ORG>, Bosko Milekic <bmilekic@unixdaemons.com>, Seigo Tanimura <tanimura@r.dl.itc.u-tokyo.ac.jp>, <current@FreeBSD.ORG>, John Baldwin <jhb@FreeBSD.ORG> Subject: Re: Success! critical_enter()/critical_exit() revamp (was Re: malloc_bucket() idea (was Re: How to fix malloc.)) Message-ID: <20020225180234.C39518-100000@gamplex.bde.org> In-Reply-To: <200202250152.g1P1q1d27292@apollo.backplane.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 24 Feb 2002, Matthew Dillon wrote: > :cpu_switch() certainly needs to do this if it can be called with the > :interrupt enable flag[s] in different states. I need the sti's (actually > :enable_intr()'s because I don't want fast interrupts to be disabled > :during context switches. This works because enabling interrupts is sure > :to be safe, since we might be switching to a thread that will enable > :them. Some sort of lock is needed to prevent interrupts interfering > :with the switch. I think soft-masking them in critical_enter() is > :sufficient in your version too. > > I don't think we want to make sched_lock any more complex then it > already is, so at least for the foreseeable future we are not > going to be able to actually execute an interrupt handler until > the sched_lock is released in (typically) msleep(). I am rather Well, my kernel has been executing fast interrupt handlers while sched_lock is held for almost a year. It's actually less complicated with respect to sched_lock but more complicated with respect to fast interrupt handlers. > annoyed that two levels of procedure have to be called with the > sched_lock held (mi_switch() and cpu_switch()), leaving interrupts > disabled for a fairly long period of time, but I don't see any way > around it right now. The worst offenders for interrupt latency seemed to be calcru() and/or the sched_locking related to fork and/or exit. Latency was many thousand instructions (reasonable only on 100+ MIPS machines). sched_locking for calcru() is moostly bogus and should be easy to avoid, but not so for context switching. > Eventually (presumably) we will have per-cpu run queues. That combined > with interrupt stealing may resolve the problem for us. I am still > not convinced that making the various *pending* flags globals will > be more efficient, because it introduces significant cache mastership > issues. It might be easier to do this: OK. I don't care about this yet. Bruce To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020225180234.C39518-100000>