Date: Sat, 23 Apr 2005 13:46:25 +1000 From: Sam Lawrance <boris@brooknet.com.au> To: David Xu <davidxu@freebsd.org> Cc: freebsd-current@freebsd.org Subject: Re: kern/78474 for 5.4? (kernel stack swapout problem again) Message-ID: <1114227985.69709.6.camel@dirk.no.domain> In-Reply-To: <4269C309.30702@freebsd.org> References: <1114052573.1075.10.camel@dirk.no.domain> <20050422200102.G10333@carver.gumbysoft.com> <4269C309.30702@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 2005-04-23 at 11:37 +0800, David Xu wrote: > Doug White wrote: > > >On Thu, 21 Apr 2005, David Xu wrote: > > > > > > > >>Sam Lawrance wrote: > >> > >> > >> > >>>Will this problem: > >>> > >>>Swapped out procs not brought in immediately after child exits > >>>http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/78474 > >>> > >>>be dealt with for the release of 5.4? > >>> > >>>Perhaps I'm the only FreeBSD user with swapped out processes ;-) > >>> > >>>-Sam > >>> > >>> > >>> > >>> > >>> > >>> > >>I have noticed that current spinlock implementation no longer means that > >>CPU must be in critical region (critical_enter is called), even previous > >>hack TDP_WAKEPROC0 is no longer correct, :(, I think it is the time to > >>disable swapout. > >> > >> > > > >I'm sorry, I can't parse your double negative. On -CURRENT critical > >sections (entered with critical_enter()) no longer disable interrupts, but > >do inhibit preemption. But spinlocks still do critical_enter() (see > >spinlock_enter()). > > > > > > > I will commit the patch provided in the PR, it should work. I am just > worrying > TDP_WAKEPROC0 will not work if spinlock does not entering critical region. If it's any comfort, using that patch, I've not had any swap-in delay problems since i raised the PR in early March. Running with all the typical debugging options turned on, and no problems arose. -Sam
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1114227985.69709.6.camel>