From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Date: Mon, 19 Feb 2007 14:15:20 -0600 To: Benjamin Herrenschmidt Subject: Re: [PATCH] powerpc: Introduce address space "slices" Message-ID: <20070219201520.GA6776@lixom.net> References: <1171867418.18571.3.camel@localhost.localdomain> <20070219153315.GA5036@lixom.net> <1171913963.18571.33.camel@localhost.localdomain> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1171913963.18571.33.camel@localhost.localdomain> From: olof@lixom.net (Olof Johansson) Cc: linuxppc-dev list , "cbe-oss-dev@ozlabs.org" List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Tue, Feb 20, 2007 at 06:39:23AM +1100, Benjamin Herrenschmidt wrote: > On Mon, 2007-02-19 at 09:33 -0600, Olof Johansson wrote: > > On Mon, Feb 19, 2007 at 05:43:38PM +1100, Benjamin Herrenschmidt wrote: > > > powerpc: Introduce address space "slices" > > > > > > This patch provide some infrastructure that will allow proper creation > > > of special VMAs with different page sizes on powerpc. > > > > > > The basic issue is to be able to do what hugetlbfs does but with > > > different page sizes for some other special filesystems, more > > > specifically, my need is: > > > > > > - hugetlbfs should still work of course :-) > > > > > > - SPE local store mappings using 64K pages on a 4K base page size > > > kernel on Cell > > > > Why? What is the reason they can't use 4K pages? > > Reduce TLB/ERAT trashing. SPE MMUs are fairly small, and in setups where > an SPE maps all the others, we take a performance hit due to trashing > with 4K pages. > > However, going to full 64K base page size has other > drawbacks, especially in setups with little main memory. Right. I was mostly wondering what the underlying reason for the requirement was. Thanks, -Olof