Messages in this thread Patches in this message | | | Subject | [PATCH] prevent deadlock in __unmap_hugepage_range() when alloc_huge_page() fails. | From | Larry Woodman <> | Date | Tue, 10 Nov 2009 12:00:56 -0500 |
| |
hugetlb_fault() takes the mm->page_table_lock spinlock then calls hugetlb_cow(). If the alloc_huge_page() in hugetlb_cow() fails due to an insufficient huge page pool it calls unmap_ref_private() with the mm->page_table_lock held. unmap_ref_private() then calls unmap_hugepage_range() which tries to acquire the mm->page_table_lock.
[<ffffffff810928c3>] print_circular_bug_tail+0x80/0x9f [<ffffffff8109280b>] ? check_noncircular+0xb0/0xe8 [<ffffffff810935e0>] __lock_acquire+0x956/0xc0e [<ffffffff81093986>] lock_acquire+0xee/0x12e [<ffffffff8111a7a6>] ? unmap_hugepage_range+0x3e/0x84 [<ffffffff8111a7a6>] ? unmap_hugepage_range+0x3e/0x84 [<ffffffff814c348d>] _spin_lock+0x40/0x89 [<ffffffff8111a7a6>] ? unmap_hugepage_range+0x3e/0x84 [<ffffffff8111afee>] ? alloc_huge_page+0x218/0x318 [<ffffffff8111a7a6>] unmap_hugepage_range+0x3e/0x84 [<ffffffff8111b2d0>] hugetlb_cow+0x1e2/0x3f4 [<ffffffff8111b935>] ? hugetlb_fault+0x453/0x4f6 [<ffffffff8111b962>] hugetlb_fault+0x480/0x4f6 [<ffffffff8111baee>] follow_hugetlb_page+0x116/0x2d9 [<ffffffff814c31a7>] ? _spin_unlock_irq+0x3a/0x5c [<ffffffff81107b4d>] __get_user_pages+0x2a3/0x427 [<ffffffff81107d0f>] get_user_pages+0x3e/0x54 [<ffffffff81040b8b>] get_user_pages_fast+0x170/0x1b5 [<ffffffff81160352>] dio_get_page+0x64/0x14a [<ffffffff8116112a>] __blockdev_direct_IO+0x4b7/0xb31 [<ffffffff8115ef91>] blkdev_direct_IO+0x58/0x6e [<ffffffff8115e0a4>] ? blkdev_get_blocks+0x0/0xb8 [<ffffffff810ed2c5>] generic_file_aio_read+0xdd/0x528 [<ffffffff81219da3>] ? avc_has_perm+0x66/0x8c [<ffffffff81132842>] do_sync_read+0xf5/0x146 [<ffffffff8107da00>] ? autoremove_wake_function+0x0/0x5a [<ffffffff81211857>] ? security_file_permission+0x24/0x3a [<ffffffff81132fd8>] vfs_read+0xb5/0x126 [<ffffffff81133f6b>] ? fget_light+0x5e/0xf8 [<ffffffff81133131>] sys_read+0x54/0x8c [<ffffffff81011e42>] system_call_fastpath+0x16/0x1b
This can be fixed by dropping the mm->page_table_lock around the call to unmap_ref_private() if alloc_huge_page() fails, its dropped right below in the normal path anyway:
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 5d7601b..f4daef4 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1973,12 +1973,15 @@ retry_avoidcopy: */ if (outside_reserve) { BUG_ON(huge_pte_none(pte)); + spin_unlock(&mm->page_table_lock); if (unmap_ref_private(mm, vma, old_page, address)) { + spin_lock(&mm->page_table_lock); BUG_ON(page_count(old_page) != 1); BUG_ON(huge_pte_none(pte)); goto retry_avoidcopy; } WARN_ON_ONCE(1); + spin_lock(&mm->page_table_lock); }
return -PTR_ERR(new_page);
Signed-off-by: Larry Woodman <lwooman@redhat.com> diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 5d7601b..f4daef4 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1973,12 +1973,15 @@ retry_avoidcopy: */ if (outside_reserve) { BUG_ON(huge_pte_none(pte)); + spin_unlock(&mm->page_table_lock); if (unmap_ref_private(mm, vma, old_page, address)) { + spin_lock(&mm->page_table_lock); BUG_ON(page_count(old_page) != 1); BUG_ON(huge_pte_none(pte)); goto retry_avoidcopy; } WARN_ON_ONCE(1); + spin_lock(&mm->page_table_lock); } return -PTR_ERR(new_page);
| |