Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758603AbdCVCcF (ORCPT ); Tue, 21 Mar 2017 22:32:05 -0400 Received: from mail-pg0-f67.google.com ([74.125.83.67]:36445 "EHLO mail-pg0-f67.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758025AbdCVCcD (ORCPT ); Tue, 21 Mar 2017 22:32:03 -0400 From: Naoya Horiguchi To: linux-mm@kvack.org Cc: Andrew Morton , Hugh Dickins , Michal Hocko , "Kirill A. Shutemov" , Hillf Danton , linux-kernel@vger.kernel.org, Naoya Horiguchi , Naoya Horiguchi Subject: [PATCH v1] mm, hugetlb: use pte_present() instead of pmd_present() in follow_huge_pmd() Date: Wed, 22 Mar 2017 11:31:38 +0900 Message-Id: <1490149898-20231-1-git-send-email-n-horiguchi@ah.jp.nec.com> X-Mailer: git-send-email 2.7.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4880 Lines: 90 I found the race condition which triggers the following bug when move_pages() and soft offline are called on a single hugetlb page concurrently. [61163.578957] Soft offlining page 0x119400 at 0x700000000000 [61163.580062] BUG: unable to handle kernel paging request at ffffea0011943820 [61163.580791] IP: follow_huge_pmd+0x143/0x190 [61163.581203] PGD 7ffd2067 [61163.581204] PUD 7ffd1067 [61163.581471] PMD 0 [61163.581723] [61163.582052] Oops: 0000 [#1] SMP [61163.582349] Modules linked in: binfmt_misc ppdev virtio_balloon parport_pc pcspkr i2c_piix4 parport i2c_core acpi_cpufreq ip_tables xfs libcrc32c ata_generic pata_acpi virtio_blk 8139too crc32c_intel ata_piix serio_raw libata virtio_pci 8139cp virtio_ring virtio mii floppy dm_mirror dm_region_hash dm_log dm_mod [last unloaded: cap_check] [61163.585130] CPU: 0 PID: 22573 Comm: iterate_numa_mo Tainted: P OE 4.11.0-rc2-mm1+ #2 [61163.586055] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 [61163.586627] task: ffff88007c951680 task.stack: ffffc90004bd8000 [61163.587181] RIP: 0010:follow_huge_pmd+0x143/0x190 [61163.587622] RSP: 0018:ffffc90004bdbcd0 EFLAGS: 00010202 [61163.588096] RAX: 0000000465003e80 RBX: ffffea0004e34d30 RCX: 00003ffffffff000 [61163.588818] RDX: 0000000011943800 RSI: 0000000000080001 RDI: 0000000465003e80 [61163.589486] RBP: ffffc90004bdbd18 R08: 0000000000000000 R09: ffff880138d34000 [61163.590097] R10: ffffea0004650000 R11: 0000000000c363b0 R12: ffffea0011943800 [61163.590751] R13: ffff8801b8d34000 R14: ffffea0000000000 R15: 000077ff80000000 [61163.591375] FS: 00007fc977710740(0000) GS:ffff88007dc00000(0000) knlGS:0000000000000000 [61163.592068] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [61163.592627] CR2: ffffea0011943820 CR3: 000000007a746000 CR4: 00000000001406f0 [61163.593330] Call Trace: [61163.593556] follow_page_mask+0x270/0x550 [61163.593908] SYSC_move_pages+0x4ea/0x8f0 [61163.594253] ? lru_cache_add_active_or_unevictable+0x4b/0xd0 [61163.594798] SyS_move_pages+0xe/0x10 [61163.595113] do_syscall_64+0x67/0x180 [61163.595434] entry_SYSCALL64_slow_path+0x25/0x25 [61163.595837] RIP: 0033:0x7fc976e03949 [61163.596148] RSP: 002b:00007ffe72221d88 EFLAGS: 00000246 ORIG_RAX: 0000000000000117 [61163.596940] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fc976e03949 [61163.597567] RDX: 0000000000c22390 RSI: 0000000000001400 RDI: 0000000000005827 [61163.598177] RBP: 00007ffe72221e00 R08: 0000000000c2c3a0 R09: 0000000000000004 [61163.598842] R10: 0000000000c363b0 R11: 0000000000000246 R12: 0000000000400650 [61163.599456] R13: 00007ffe72221ee0 R14: 0000000000000000 R15: 0000000000000000 [61163.600067] Code: 81 e4 ff ff 1f 00 48 21 c2 49 c1 ec 0c 48 c1 ea 0c 4c 01 e2 49 bc 00 00 00 00 00 ea ff ff 48 c1 e2 06 49 01 d4 f6 45 bc 04 74 90 <49> 8b 7c 24 20 40 f6 c7 01 75 2b 4c 89 e7 8b 47 1c 85 c0 7e 2a [61163.601845] RIP: follow_huge_pmd+0x143/0x190 RSP: ffffc90004bdbcd0 [61163.602376] CR2: ffffea0011943820 [61163.602767] ---[ end trace e4f81353a2d23232 ]--- [61163.603236] Kernel panic - not syncing: Fatal exception [61163.603706] Kernel Offset: disabled This bug is triggered when pmd_present() returns true for non-present hugetlb, so fixing the present check in follow_huge_pmd() prevents it. Using pmd_present() to determine present/non-present for hugetlb is not correct, because pmd_present() checks multiple bits (not only _PAGE_PRESENT) for historical reason and it can misjudge hugetlb state. Fixes: e66f17ff7177 ("mm/hugetlb: take page table lock in follow_huge_pmd()") Signed-off-by: Naoya Horiguchi Cc: [4.0+] --- mm/hugetlb.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git v4.11-rc2-mmotm-2017-03-17-15-26/mm/hugetlb.c v4.11-rc2-mmotm-2017-03-17-15-26_patched/mm/hugetlb.c index 3d0aab9..f501f14 100644 --- v4.11-rc2-mmotm-2017-03-17-15-26/mm/hugetlb.c +++ v4.11-rc2-mmotm-2017-03-17-15-26_patched/mm/hugetlb.c @@ -4651,6 +4651,7 @@ follow_huge_pmd(struct mm_struct *mm, unsigned long address, { struct page *page = NULL; spinlock_t *ptl; + pte_t pte; retry: ptl = pmd_lockptr(mm, pmd); spin_lock(ptl); @@ -4660,12 +4661,13 @@ follow_huge_pmd(struct mm_struct *mm, unsigned long address, */ if (!pmd_huge(*pmd)) goto out; - if (pmd_present(*pmd)) { + pte = huge_ptep_get((pte_t *)pmd); + if (pte_present(pte)) { page = pmd_page(*pmd) + ((address & ~PMD_MASK) >> PAGE_SHIFT); if (flags & FOLL_GET) get_page(page); } else { - if (is_hugetlb_entry_migration(huge_ptep_get((pte_t *)pmd))) { + if (is_hugetlb_entry_migration(pte)) { spin_unlock(ptl); __migration_entry_wait(mm, (pte_t *)pmd, ptl); goto retry; -- 2.7.0