From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on archive.lwn.net X-Spam-Level: X-Spam-Status: No, score=-6.0 required=5.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, MAILING_LIST_MULTI,RCVD_IN_DNSWL_HI autolearn=unavailable autolearn_force=no version=3.4.2 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by archive.lwn.net (Postfix) with ESMTP id 2D29F7D099 for ; Mon, 7 Jan 2019 19:25:30 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1728138AbfAGTZR (ORCPT ); Mon, 7 Jan 2019 14:25:17 -0500 Received: from mga12.intel.com ([192.55.52.136]:63398 "EHLO mga12.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1727397AbfAGTZR (ORCPT ); Mon, 7 Jan 2019 14:25:17 -0500 X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga001.jf.intel.com ([10.7.209.18]) by fmsmga106.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 07 Jan 2019 11:25:16 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.56,451,1539673200"; d="scan'208";a="124944401" Received: from ray.jf.intel.com (HELO [10.7.201.126]) ([10.7.201.126]) by orsmga001.jf.intel.com with ESMTP; 07 Jan 2019 11:25:16 -0800 Subject: Re: [PATCH] drop_caches: Allow unmapping pages To: Matthew Wilcox , Vincent Whitchurch Cc: akpm@linux-foundation.org, viro@zeniv.linux.org.uk, linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, linux-kernel@vger.kernel.org, mcgrof@kernel.org, keescook@chromium.org, corbet@lwn.net, linux-doc@vger.kernel.org, Vincent Whitchurch References: <20190107130239.3417-1-vincent.whitchurch@axis.com> <20190107141545.GX6310@bombadil.infradead.org> From: Dave Hansen Openpgp: preference=signencrypt Autocrypt: addr=dave.hansen@intel.com; keydata= mQINBE6HMP0BEADIMA3XYkQfF3dwHlj58Yjsc4E5y5G67cfbt8dvaUq2fx1lR0K9h1bOI6fC oAiUXvGAOxPDsB/P6UEOISPpLl5IuYsSwAeZGkdQ5g6m1xq7AlDJQZddhr/1DC/nMVa/2BoY 2UnKuZuSBu7lgOE193+7Uks3416N2hTkyKUSNkduyoZ9F5twiBhxPJwPtn/wnch6n5RsoXsb ygOEDxLEsSk/7eyFycjE+btUtAWZtx+HseyaGfqkZK0Z9bT1lsaHecmB203xShwCPT49Blxz VOab8668QpaEOdLGhtvrVYVK7x4skyT3nGWcgDCl5/Vp3TWA4K+IofwvXzX2ON/Mj7aQwf5W iC+3nWC7q0uxKwwsddJ0Nu+dpA/UORQWa1NiAftEoSpk5+nUUi0WE+5DRm0H+TXKBWMGNCFn c6+EKg5zQaa8KqymHcOrSXNPmzJuXvDQ8uj2J8XuzCZfK4uy1+YdIr0yyEMI7mdh4KX50LO1 pmowEqDh7dLShTOif/7UtQYrzYq9cPnjU2ZW4qd5Qz2joSGTG9eCXLz5PRe5SqHxv6ljk8mb ApNuY7bOXO/A7T2j5RwXIlcmssqIjBcxsRRoIbpCwWWGjkYjzYCjgsNFL6rt4OL11OUF37wL QcTl7fbCGv53KfKPdYD5hcbguLKi/aCccJK18ZwNjFhqr4MliQARAQABtEVEYXZpZCBDaHJp c3RvcGhlciBIYW5zZW4gKEludGVsIFdvcmsgQWRkcmVzcykgPGRhdmUuaGFuc2VuQGludGVs LmNvbT6JAjgEEwECACIFAlQ+9J0CGwMGCwkIBwMCBhUIAgkKCwQWAgMBAh4BAheAAAoJEGg1 lTBwyZKwLZUP/0dnbhDc229u2u6WtK1s1cSd9WsflGXGagkR6liJ4um3XCfYWDHvIdkHYC1t MNcVHFBwmQkawxsYvgO8kXT3SaFZe4ISfB4K4CL2qp4JO+nJdlFUbZI7cz/Td9z8nHjMcWYF IQuTsWOLs/LBMTs+ANumibtw6UkiGVD3dfHJAOPNApjVr+M0P/lVmTeP8w0uVcd2syiaU5jB aht9CYATn+ytFGWZnBEEQFnqcibIaOrmoBLu2b3fKJEd8Jp7NHDSIdrvrMjYynmc6sZKUqH2 I1qOevaa8jUg7wlLJAWGfIqnu85kkqrVOkbNbk4TPub7VOqA6qG5GCNEIv6ZY7HLYd/vAkVY E8Plzq/NwLAuOWxvGrOl7OPuwVeR4hBDfcrNb990MFPpjGgACzAZyjdmYoMu8j3/MAEW4P0z F5+EYJAOZ+z212y1pchNNauehORXgjrNKsZwxwKpPY9qb84E3O9KYpwfATsqOoQ6tTgr+1BR CCwP712H+E9U5HJ0iibN/CDZFVPL1bRerHziuwuQuvE0qWg0+0SChFe9oq0KAwEkVs6ZDMB2 P16MieEEQ6StQRlvy2YBv80L1TMl3T90Bo1UUn6ARXEpcbFE0/aORH/jEXcRteb+vuik5UGY 5TsyLYdPur3TXm7XDBdmmyQVJjnJKYK9AQxj95KlXLVO38lcuQINBFRjzmoBEACyAxbvUEhd GDGNg0JhDdezyTdN8C9BFsdxyTLnSH31NRiyp1QtuxvcqGZjb2trDVuCbIzRrgMZLVgo3upr MIOx1CXEgmn23Zhh0EpdVHM8IKx9Z7V0r+rrpRWFE8/wQZngKYVi49PGoZj50ZEifEJ5qn/H Nsp2+Y+bTUjDdgWMATg9DiFMyv8fvoqgNsNyrrZTnSgoLzdxr89FGHZCoSoAK8gfgFHuO54B lI8QOfPDG9WDPJ66HCodjTlBEr/Cwq6GruxS5i2Y33YVqxvFvDa1tUtl+iJ2SWKS9kCai2DR 3BwVONJEYSDQaven/EHMlY1q8Vln3lGPsS11vSUK3QcNJjmrgYxH5KsVsf6PNRj9mp8Z1kIG qjRx08+nnyStWC0gZH6NrYyS9rpqH3j+hA2WcI7De51L4Rv9pFwzp161mvtc6eC/GxaiUGuH BNAVP0PY0fqvIC68p3rLIAW3f97uv4ce2RSQ7LbsPsimOeCo/5vgS6YQsj83E+AipPr09Caj 0hloj+hFoqiticNpmsxdWKoOsV0PftcQvBCCYuhKbZV9s5hjt9qn8CE86A5g5KqDf83Fxqm/ vXKgHNFHE5zgXGZnrmaf6resQzbvJHO0Fb0CcIohzrpPaL3YepcLDoCCgElGMGQjdCcSQ+Ci FCRl0Bvyj1YZUql+ZkptgGjikQARAQABiQIfBBgBAgAJBQJUY85qAhsMAAoJEGg1lTBwyZKw l4IQAIKHs/9po4spZDFyfDjunimEhVHqlUt7ggR1Hsl/tkvTSze8pI1P6dGp2XW6AnH1iayn yRcoyT0ZJ+Zmm4xAH1zqKjWplzqdb/dO28qk0bPso8+1oPO8oDhLm1+tY+cOvufXkBTm+whm +AyNTjaCRt6aSMnA/QHVGSJ8grrTJCoACVNhnXg/R0g90g8iV8Q+IBZyDkG0tBThaDdw1B2l asInUTeb9EiVfL/Zjdg5VWiF9LL7iS+9hTeVdR09vThQ/DhVbCNxVk+DtyBHsjOKifrVsYep WpRGBIAu3bK8eXtyvrw1igWTNs2wazJ71+0z2jMzbclKAyRHKU9JdN6Hkkgr2nPb561yjcB8 sIq1pFXKyO+nKy6SZYxOvHxCcjk2fkw6UmPU6/j/nQlj2lfOAgNVKuDLothIxzi8pndB8Jju KktE5HJqUUMXePkAYIxEQ0mMc8Po7tuXdejgPMwgP7x65xtfEqI0RuzbUioFltsp1jUaRwQZ MTsCeQDdjpgHsj+P2ZDeEKCbma4m6Ez/YWs4+zDm1X8uZDkZcfQlD9NldbKDJEXLIjYWo1PH hYepSffIWPyvBMBTW2W5FRjJ4vLRrJSUoEfJuPQ3vW9Y73foyo/qFoURHO48AinGPZ7PC7TF vUaNOTjKedrqHkaOcqB185ahG2had0xnFsDPlx5y Message-ID: <67dac226-00ca-dd0a-800e-0867e12d3ad5@intel.com> Date: Mon, 7 Jan 2019 11:25:16 -0800 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.2.1 MIME-Version: 1.0 In-Reply-To: <20190107141545.GX6310@bombadil.infradead.org> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 8bit Sender: linux-doc-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-doc@vger.kernel.org On 1/7/19 6:15 AM, Matthew Wilcox wrote: > You're going to get data corruption doing this. try_to_unmap_one() > does: > > /* Move the dirty bit to the page. Now the pte is gone. */ > if (pte_dirty(pteval)) > set_page_dirty(page); > > so PageDirty() can be false above, but made true by calling > try_to_unmap(). I don't think that PageDirty() check is _required_ for correctness. You can always safely try_to_unmap() no matter the state of the PTE. We can't lock out the hardware from setting the Dirty bit, ever. It's also just fine to unmap PageDirty() pages, as long as when the PTE is created, we move the dirty bit from the PTE into the 'struct page' (which try_to_unmap() does, as you noticed). > I also think the way you've done this is expedient at the cost of > efficiency and layering violations. I think you should first tear > down the mappings of userspace processes (which will reclaim a lot > of pages allocated to page tables), then you won't need to touch the > invalidate_inode_pages paths at all. By "tear down the mappings", do you mean something analogous to munmap() where the VMA goes away? Or madvise(MADV_DONTNEED) where the PTE is destroyed but the VMA remains? Last time I checked, we only did free_pgtables() when tearing down VMAs, but not for pure unmappings like reclaim or MADV_DONTNEED. I've thought it might be fun to make a shrinker that scanned page tables looking for zero'd pages, but I've never run into a system where empty page table pages were actually causing a noticeable problem.