Date: Mon, 21 Apr 2003 15:22:04 -0400 From: Jake Burkholder <jake@locore.ca> To: Marcel Moolenaar <marcel@xcllnt.net> Cc: freebsd-hackers@freebsd.org Subject: Re: Is pmap_kextract() allowed to fault? Message-ID: <20030421192204.GH50754@locore.ca> In-Reply-To: <20030421183322.GA557@dhcp01.pn.xcllnt.net> References: <20030421055332.GA4680@dhcp01.pn.xcllnt.net> <20030421132449.GA50754@locore.ca> <20030421183322.GA557@dhcp01.pn.xcllnt.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Apparently, On Mon, Apr 21, 2003 at 11:33:22AM -0700, Marcel Moolenaar said words to the effect of; > On Mon, Apr 21, 2003 at 09:24:49AM -0400, Jake Burkholder wrote: > > Apparently, On Sun, Apr 20, 2003 at 10:53:32PM -0700, > > Marcel Moolenaar said words to the effect of; > > > > > Gang, > > > > > > On ia64 pmap_kextract() uses the tpa instruction which given a > > > virtual address returns the physical address based on the > > > translation registers and cache (ie TLB). This can fault when > > > there's currently no mapping for the virtual address. > > > > > > Since all other architectures have a non-faulting implementation > > > (AFAICT), I'm a bit worried that we might get into trouble on > > > ia64. I couldn't find anything about pmap_kextract(), so maybe > > > anybody can enlighten me: > > > > > > 1. Is pmap_kextract() allowed to fault? > > > > It depends what kind of fault. Will tpa fail if it causes a tlb fault > > and the page is not in the vhpt (or whatever the fault handler searches), > > or will it end up calling vm_fault and actually trying to fault in the > > page? > > It will end up calling vm_fault() if so required. Yes, this is very bad. Especially for things like /dev/kmem, where you want to validate an address passed from userland. Jake > > > > 2. Is pmap_kextract() used often enough that using the cpu's TLB > > > is a possible performance speedup even if there are costly faults > > > that can sometimes happen? > > > > I doubt it. > > Ok, thanks. I think I'll use a non-trapping implementation then. > There's just too much circumstantiality... > > -- > 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?20030421192204.GH50754>