Date: Fri, 11 Jun 2004 00:33:17 +0200 From: Ralf Baechle <ralf@linux-mips.org> To: juli mallett <jmallett@FreeBSD.org> Cc: freebsd-mips@FreeBSD.org Subject: Re: Some MIPS status goodies. Message-ID: <20040610223317.GA16814@linux-mips.org> In-Reply-To: <20040610221430.GA26579@FreeBSD.org> References: <Pine.LNX.4.33.0406101040240.2858-100000@servww6.ww.uni-erlangen.de> <20040610090712.GA71925@FreeBSD.org> <20040610160207.GB32315@linux-mips.org> <20040610221430.GA26579@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Jun 10, 2004 at 12:14:30PM -1000, juli mallett wrote: > > > R10K and IP28 support are probably a ways away, but I'm interested > > > in supporting these systems and more. Right now it's easiest to start > > > with the, much simpler, R4400 Indigo2 I have here, and get a good base > > > of this stuff, then move into supporting the more modern and complex > > > CPUs and machines - but be assured I am taking them into account! > > > > R10000 in non-coherent systems (Indigo 2 R10000, O2) is a rather hard to > > use processor due to unwanted behaviour of speculative execution that > > leads to memory corruption. The workarounds are fairly complex; efficient > > solutions involve compiler modifications. > > Yeah, I read about the compiler mods SGI had to make, with accessing > the stack as a barrier... It sounded like that was only for the > kernel though, yeah? Yes. The CPU can't speculate memory access for virtual addresses that aren't mapped, so for usermode the trick is to keep every page unmapped while doing DMA on it. Ralf
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040610223317.GA16814>