Date: Mon, 21 Apr 2003 13:42:21 -0700 From: Marcel Moolenaar <marcel@xcllnt.net> To: Perforce Change Reviews <perforce@freebsd.org> Subject: Re: PERFORCE change 29365 for review Message-ID: <20030421204221.GA830@athlon.pn.xcllnt.net> In-Reply-To: <200304212028.h3LKSg2v082838@repoman.freebsd.org> References: <200304212028.h3LKSg2v082838@repoman.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Apr 21, 2003 at 01:28:42PM -0700, Marcel Moolenaar wrote: > http://perforce.freebsd.org/chv.cgi?CH=29365 > > Change 29365 by marcel@marcel_nfs on 2003/04/21 13:27:48 > > Use a non-trapping implementation for pmap_kextract. > This has been deliberately made non-inlined for now > to help profiling, debugging and whatsnot. Be extra > careful (probably over-correctingly so :-) with the > address passed to us, because it may be an address > given to use from user space. > > Note that an inlined or partially inlined version > should be used eventually. It appears that UMA is > a regular user of pmap_kextract and rudimentary > testing showed that it's mostly for direct mapped > KVAs. Doing the test for direct mapped KVAs inlined > will probably be worth it. > > Valuable insights by: jake JFYI: I noticed a reduction in the number of Nested TLB faults, so it appears that not depending on the cpu's TLB for pmap_kextract has a side-effect of not polluting the translation cache. I also have a nagging suspicion that it may have been the cause of not being able to run a MP kernel on the HP boxes in the FreeBSD cluster, even though SMP works (roughly) on the BigSur. On the HP boxes the BSP and AP get entangled in a life-lock where the AP holds sched_lock, while waiting for Giant that the BSP holds. The BSP is likely in hardclock and the AP is typically in UMA (I may have got the details slightly wrong). Since the AP has mostly a clean translation cache it's probably faulting a lot and may possibly end up in vm_fault. Again, this is just a nagging suspicion... -- Marcel Moolenaar USPA: A-39004 marcel@xcllnt.net
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030421204221.GA830>