From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S262862AbUCXF7c (ORCPT ); Wed, 24 Mar 2004 00:59:32 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263010AbUCXF7c (ORCPT ); Wed, 24 Mar 2004 00:59:32 -0500 Received: from ppp-217-133-42-200.cust-adsl.tiscali.it ([217.133.42.200]:2690 "EHLO dualathlon.random") by vger.kernel.org with ESMTP id S262862AbUCXF72 (ORCPT ); Wed, 24 Mar 2004 00:59:28 -0500 Date: Wed, 24 Mar 2004 07:00:20 +0100 From: Andrea Arcangeli To: linux-kernel@vger.kernel.org Subject: 2.6.5-rc2-aa2 Message-ID: <20040324060020.GA2065@dualathlon.random> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4.1i X-GPG-Key: 1024D/68B9CB43 13D9 8355 295F 4823 7C49 C012 DFA1 686E 68B9 CB43 X-PGP-Key: 1024R/CB4660B9 CC A0 71 81 F4 A0 63 AC C0 4B 81 1D 8C 15 C8 E5 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org This update covers the paging of nonlinear vmas, and the ppc architecture (the former is well tested, the latter not but at least it now compiles on ppc and ppc64). The next target is the merging of the prio_tree, but that will be a separated patch. After that this whole thing should be mergeable into mainline. Of course I can consider applying a filter to change 'anon_vma_t' to 'struct anon_vma' (even if I personally prefer the typedefs). BTW, I put the set_page_dirty back into the page_map_lock because we've to run that under the page_map_lock anyways if the pte was dirty during swapping/paging. I mean, as far as we run it there, then we can run it in the other s390-only places too. URL: http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.6/2.6.5-rc2-aa2.gz http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.6/2.6.5-rc2-aa2/ Files 2.6.5-rc2-aa1/anon_vma.gz and 2.6.5-rc2-aa2/anon_vma.gz differ Merged Hugh's s390 fixes for physical referenced/dirty bit handling. Implemented nonlinear paging support. Tested with two hundred threads with 1.7G of shm mapped each trashing in a loop. Verified the nonlinear pages are swapped out and swapped in correctly. No races during high load so far. Nice swap behaviour is not expectable for nonlinear mappings, nonlinear is optimal for either small pageable mappings or big mlocked mappings. Fixed ppc/ppc64/arm compilation by merging Hugh's anobjrmap-6 (I verified it compiles with a cross compiler but it's untested on real hardware). Avoid oopsing in fork too, if somebody forget VM_RESERVED (only warn). Renamd bitflag and fixed include guards per Christoph's suggestion. Files 2.6.5-rc2-aa1/extraversion and 2.6.5-rc2-aa2/extraversion differ Rediffed. Files 2.6.5-rc2-aa1/irq-safe-pagecache-lock.patch.gz and 2.6.5-rc2-aa2/irq-safe-pagecache-lock.patch.gz differ Files 2.6.5-rc2-aa1/stop-using-dirty-pages.patch.gz and 2.6.5-rc2-aa2/stop-using-dirty-pages.patch.gz differ Only in 2.6.5-rc2-aa1: stop-using-locked-pages-fix-2.patch.gz Only in 2.6.5-rc2-aa1: stop-using-locked-pages-fix.patch.gz Files 2.6.5-rc2-aa1/stop-using-locked-pages.patch.gz and 2.6.5-rc2-aa2/stop-using-locked-pages.patch.gz differ Only in 2.6.5-rc2-aa1: tag-writeback-pages-fix.patch.gz Only in 2.6.5-rc2-aa1: tag-writeback-pages-missing-filesystems.patch.gz Files 2.6.5-rc2-aa1/tag-writeback-pages.patch.gz and 2.6.5-rc2-aa2/tag-writeback-pages.patch.gz differ Files 2.6.5-rc2-aa1/unslabify-pgds-and-pmds.patch.gz and 2.6.5-rc2-aa2/unslabify-pgds-and-pmds.patch.gz differ Merged new versions from 2.6.5-rc2-mm1.