Date: Tue, 1 Sep 1998 09:35:12 -0700 (PDT) From: Matthew Jacob <mjacob@feral.com> To: Andrew Gallatin <gallatin@cs.duke.edu> Cc: freebsd-alpha@FreeBSD.ORG Subject: Re: Alpha Install - oops! Message-ID: <Pine.LNX.4.02.9809010933180.18190-100000@feral-gw> In-Reply-To: <13804.8316.576176.582569@grasshopper.cs.duke.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
I wouldn't claim that it does or doesn't happen- I just would suspect this as being what a piece of h/w would do. It's probably a lot of gates and probably a big performance penalty to avoid prefetch across certain boundaries. As much as possible, I prefer direct map as well. On Tue, 1 Sep 1998, Andrew Gallatin wrote: > > Matthew Jacob writes: > > > > > > I'd watch this for crossing page boundaries on reads if you're using any > > S/G map stuff. In fact, I'd have an extra mapping at the end of > > the S/G list that just remaps the first page so that any prefetch > > on a read won't get a fault but will just pick up known good data. > > Gee, I'd hope a GL would be smart enough not to prefectch past the end > of a S/G segment; I do know that the page-boundry DMA restriction is gone on GL's. > > However, I use the direct-map segment in my Myrinet drivers (on DU & > on *BSD), so I can't verify that it doesn't happen. > > Drew > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.4.02.9809010933180.18190-100000>