Date: Fri, 02 Jan 2009 12:39:43 -0600 From: Alan Cox <alc@cs.rice.edu> To: Mayur <mayur.shardul@gmail.com> Cc: Ed Schouten <ed@80386.nl>, Perforce Change Reviews <perforce@freebsd.org>, Mayur Shardul <mayur@freebsd.org> Subject: Re: PERFORCE change 155554 for review Message-ID: <495E5F6F.8030405@cs.rice.edu> In-Reply-To: <2ac427a0901011055x7e561f12w338efac38e6c5e71@mail.gmail.com> References: <200901011408.n01E8GYU036190@repoman.freebsd.org> <20090101141833.GF1176@hoeg.nl> <2ac427a0901011055x7e561f12w338efac38e6c5e71@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Mayur wrote: > Hi Ed, > This work is not targeted to reduce the contention but this seems an > interesting problem. Will look in to it. > Have a look at pmap_enter() and pmap_remove_pages() and more generally the use of the page queues mutex to synchronize access to the per-page pv list in the pmap. A simple and (likely) productive first step would be to introduce a new mutex in the pmap for synchronizing access to the per-page pv lists. The tricky part is that the pmap functions sometimes manipulate other per-page fields, like "dirty", that are also synchronized by the page queues lock. If you want to undertake this, I would be very supportive. Regards, Alan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?495E5F6F.8030405>