From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754334Ab3JVRws (ORCPT ); Tue, 22 Oct 2013 13:52:48 -0400 Received: from mga09.intel.com ([134.134.136.24]:15930 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753054Ab3JVRwr (ORCPT ); Tue, 22 Oct 2013 13:52:47 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.93,535,1378882800"; d="scan'208";a="423036174" From: "Kirill A. Shutemov" To: Dave Hansen Cc: "Kirill A. Shutemov" , Andrew Morton , Thomas Gleixner , Ingo Molnar , "H. Peter Anvin" , Nadia Yvette Chambers , Naoya Horiguchi , x86@kernel.org, linux-kernel@vger.kernel.org, linux-mm@kvack.org, Matthew Wilcox In-Reply-To: <526697F5.7040800@intel.com> References: <1382449940-24357-1-git-send-email-kirill.shutemov@linux.intel.com> <526697F5.7040800@intel.com> Subject: Re: [PATCH] x86, mm: get ASLR work for hugetlb mappings Content-Transfer-Encoding: 7bit Message-Id: <20131022175219.BB0E3E0090@blue.fi.intel.com> Date: Tue, 22 Oct 2013 20:52:19 +0300 (EEST) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Dave Hansen wrote: > On 10/22/2013 06:52 AM, Kirill A. Shutemov wrote: > > Matthew noticed that hugetlb doesn't participate in ASLR on x86-64. > > The reason is genereic hugetlb_get_unmapped_area() which is used on > > x86-64. It doesn't support randomization and use bottom-up unmapped area > > lookup, instead of usual top-down on x86-64. > > I have to wonder if this was on purpose in order to keep the large and > small mappings separate. We don't *have* to keep them separate this, of > course, but it makes me wonder. I haven't seen any evidence that it's on purpose, but who knows... In x86-specific hugetlb_get_unmapped_area() there's explicit check what is mm->get_unmapped_area top-down or bottom-up, and doing the same. > > x86 has arch-specific hugetlb_get_unmapped_area(), but it's used only on > > x86-32. > > > > Let's use arch-specific hugetlb_get_unmapped_area() on x86-64 too. > > It fixes the issue and make hugetlb use top-down unmapped area lookup. > > Shouldn't we fix the generic code instead of further specializing the > x86 stuff? For that we need to modify info.low_limit to mm->mmap_legacy_base (which is x86 specific, no-go) or switch to top-down and set info.high_limit to mm->mmap_base. I don't know how it can affect other architectures. > In any case, you probably also want to run this through: the > libhugetlbfs tests: > > http://sourceforge.net/p/libhugetlbfs/code/ci/master/tree/tests/ I've got the same fail list for upstream and patched kernel, so no regression was found. ********** TEST SUMMARY * 2M * 32-bit 64-bit * Total testcases: 107 110 * Skipped: 0 0 * PASS: 98 108 * FAIL: 2 2 * Killed by signal: 7 0 * Bad configuration: 0 0 * Expected FAIL: 0 0 * Unexpected PASS: 0 0 * Strange test result: 0 0 ********** -- Kirill A. Shutemov