From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from hancock.sc.steeleye.com (nat9.steeleye.com [65.114.3.137]) by dsl2.external.hp.com (Postfix) with ESMTP id 6D9D84841 for ; Sat, 23 Aug 2003 23:18:12 -0600 (MDT) Received: from mulgrave-w.il.steeleye.com (il-ppp.sc.steeleye.com [172.17.6.240]) by hancock.sc.steeleye.com (8.11.6/linuxconf) with ESMTP id h7O5HuI30521; Sun, 24 Aug 2003 01:17:57 -0400 Subject: Re: [parisc-linux] Re: Problems with kernel mmap (failing tst-mmap-eofsync in glibc on parisc) From: James Bottomley To: "David S. Miller" Cc: hugh@veritas.com, willy@debian.org, Linux Kernel , PARISC list , drepper@redhat.com In-Reply-To: <20030823172251.4e656f9a.davem@redhat.com> References: <20030822110144.5f7b83c5.davem@redhat.com> <20030822113106.0503a665.davem@redhat.com> <1061578568.2053.313.camel@mulgrave> <20030822121955.619a14eb.davem@redhat.com> <1061591255.1784.636.camel@mulgrave> <20030822154100.06314c8e.davem@redhat.com> <1061600974.2090.809.camel@mulgrave> <20030823144330.5ddab065.davem@redhat.com> <1061677283.1992.471.camel@mulgrave> <20030823155312.63f996f6.davem@redhat.com> <1061680279.1785.534.camel@mulgrave> <20030823172251.4e656f9a.davem@redhat.com> Content-Type: text/plain Date: 24 Aug 2003 00:17:55 -0500 Message-Id: <1061702282.1992.1153.camel@mulgrave> Mime-Version: 1.0 Sender: parisc-linux-admin@lists.parisc-linux.org Errors-To: parisc-linux-admin@lists.parisc-linux.org List-Help: List-Post: List-Subscribe: , List-Id: parisc-linux developers list List-Unsubscribe: , List-Archive: On Sat, 2003-08-23 at 19:22, David S. Miller wrote: > You're not answering my question :( > > I know that when the case _DOES_ happen, your optimization is > worthwhile, my question is not about this. My question is about > how often does this case happen. Well the case we're arguing about depends on whether glibc uses mmaped file descriptors for libio. If it doesn't (which is how my machine is configured) then you're right, we never get to loop the i_mmap list because it's usually empty when flush_dcache_page is called. The only things I can really come up with are contrived cases with glibc mmaped file objects...then it is a big win if we open lots of mappings---which I guess would mainly happen with multi-threaded applications. Do you have any applications you'd like me to run comparisons on? James