From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from rune.pobox.com (rune.pobox.com [208.210.124.79]) by ozlabs.org (Postfix) with ESMTP id 6C8DD67CEB for ; Sun, 8 Oct 2006 09:37:02 +1000 (EST) Date: Sat, 7 Oct 2006 18:36:52 -0500 From: Nathan Lynch To: Benjamin Herrenschmidt Subject: Re: vsdo_datapage.h WTF? Message-ID: <20061007233652.GB26378@localdomain> References: <1160148208.4795.30.camel@pmac.infradead.org> <1160170374.22232.135.camel@localhost.localdomain> <20061007145811.GA4043@krispykreme> <1160258075.7122.7.camel@localhost.localdomain> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1160258075.7122.7.camel@localhost.localdomain> Cc: linuxppc-dev@ozlabs.org, David Woodhouse , Anton Blanchard , paulus@samba.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Benjamin Herrenschmidt wrote: > On Sun, 2006-10-08 at 00:58 +1000, Anton Blanchard wrote: > > Hi, > > > > > No. There -used- to be this systemcfg structure that was exposed to > > > userspace on ppc64. I made it deprecated. However, for the sake of > > > backward compatbility, I kept vdso datapage with the exact same layout > > > for the first part of it (before the syscall maps) and haven't removed > > > the mmap call for it, but it will be gone soon. > > > > We need a way to export cache size/associativity information (sysfs, > > /proc/device-tree parsing?) and a plan to fix the applications using it > > first. > > The application using it can still use its own copy of the old header > tho. And that's why I haven't removed the /proc file and will probably > not do in the near future. > > Regarding cache infos, I noticed the lack of anything useful in the > device-tree on Power5 LPAR or did I miss something ? Which means that on > recent machines, the kernel doesn't even know. On shared-processor partitions device nodes for caches are not provided at all. In dedicated mode the caches can be correlated to their respective processors by looking at the "l2-cache" properties iirc.