| Message ID | 1469189981-19000-15-git-send-email-jack@suse.cz (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Fri, Jul 22, 2016 at 02:19:40PM +0200, Jan Kara wrote: > Currently PTE gets updated in wp_pfn_shared() after dax_pfn_mkwrite() > has released corresponding radix tree entry lock. When we want to > writeprotect PTE on cache flush, we need PTE modification to happen > under radix tree entry lock to ensure consisten updates of PTE and radix > tree (standard faults use page lock to ensure this consistency). So move > update of PTE bit into dax_pfn_mkwrite(). > > Signed-off-by: Jan Kara <jack@suse.cz> After applying the whole series to a v4.7 baseline I was hitting a deadlock in my testing, and it bisected to this commit. This deadlock happens in my QEMU guest with generic/068, ext4 and DAX. It reproduces 100% of the time after this commit. Here is the lockdep info, passed through kasan_symbolize.py: run fstests generic/068 at 2016-07-25 15:29:10 EXT4-fs (pmem0p2): DAX enabled. Warning: EXPERIMENTAL, use at your own risk EXT4-fs (pmem0p2): mounted filesystem with ordered data mode. Opts: dax ====================================================== [ INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected ] 4.7.0+ #1 Not tainted ------------------------------------------------------ fstest/1856 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire: (&(ptlock_ptr(page))->rlock#2){+.+...}, at: [< inline >] spin_lock include/linux/spinlock.h:302 (&(ptlock_ptr(page))->rlock#2){+.+...}, at: [<ffffffff8121c0b7>] finish_mkwrite_fault+0xa7/0x120 mm/memory.c:2286 and this task is already holding: (&(&mapping->tree_lock)->rlock){-.-...}, at: [< inline >] spin_lock_irq include/linux/spinlock.h:332 (&(&mapping->tree_lock)->rlock){-.-...}, at: [<ffffffff812d5086>] dax_pfn_mkwrite+0x36/0x90 fs/dax.c:1280 which would create a new lock dependency: (&(&mapping->tree_lock)->rlock){-.-...} -> (&(ptlock_ptr(page))->rlock#2){+.+...} but this new dependency connects a HARDIRQ-irq-safe lock: (&(&mapping->tree_lock)->rlock){-.-...} ... which became HARDIRQ-irq-safe at: [< inline >] mark_irqflags kernel/locking/lockdep.c:2912 [<ffffffff8110a236>] __lock_acquire+0x706/0x14b0 kernel/locking/lockdep.c:3287 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:112 [<ffffffff81ad02ff>] _raw_spin_lock_irqsave+0x4f/0x90 kernel/locking/spinlock.c:159 [<ffffffff811f2a27>] test_clear_page_writeback+0x67/0x2a0 mm/page-writeback.c:2737 [<ffffffff811de81f>] end_page_writeback+0x1f/0xa0 mm/filemap.c:858 [<ffffffff812b9495>] end_buffer_async_write+0xc5/0x180 fs/buffer.c:375 [<ffffffff812b8338>] end_bio_bh_io_sync+0x28/0x40 fs/buffer.c:2936 [<ffffffff81575907>] bio_endio+0x57/0x60 block/bio.c:1758 [<ffffffff818915fc>] dec_pending+0x21c/0x340 drivers/md/dm.c:1015 [<ffffffff818922a6>] clone_endio+0x76/0xe0 drivers/md/dm.c:1059 [<ffffffff81575907>] bio_endio+0x57/0x60 block/bio.c:1758 [< inline >] req_bio_endio block/blk-core.c:155 [<ffffffff8157f072>] blk_update_request+0xa2/0x3c0 block/blk-core.c:2644 [<ffffffff8158998a>] blk_mq_end_request+0x1a/0x70 block/blk-mq.c:320 [<ffffffff8177bcdf>] virtblk_request_done+0x3f/0x70 drivers/block/virtio_blk.c:131 [<ffffffff81588983>] __blk_mq_complete_request_remote+0x13/0x20 block/blk-mq.c:330 [<ffffffff8114bebf>] flush_smp_call_function_queue+0x5f/0x150 kernel/smp.c:249 [<ffffffff8114c903>] generic_smp_call_function_single_interrupt+0x13/0x60 kernel/smp.c:194 [< inline >] __smp_call_function_single_interrupt arch/x86/kernel/smp.c:311 [<ffffffff810575a7>] smp_call_function_single_interrupt+0x27/0x40 arch/x86/kernel/smp.c:318 [<ffffffff81ad1926>] call_function_single_interrupt+0x96/0xa0 arch/x86/entry/entry_64.S:639 [< inline >] raw_spin_unlock_irq_rcu_node kernel/rcu/tree.h:718 [< inline >] rcu_gp_init kernel/rcu/tree.c:1934 [<ffffffff8112a5f7>] rcu_gp_kthread+0x157/0x8f0 kernel/rcu/tree.c:2175 [<ffffffff810d59e6>] kthread+0xf6/0x110 kernel/kthread.c:209 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 to a HARDIRQ-irq-unsafe lock: (&(ptlock_ptr(page))->rlock#2){+.+...} ... which became HARDIRQ-irq-unsafe at: ... [< inline >] mark_irqflags kernel/locking/lockdep.c:2929 ... [<ffffffff8110a0f1>] __lock_acquire+0x5c1/0x14b0 kernel/locking/lockdep.c:3287 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [< inline >] do_anonymous_page mm/memory.c:2823 [< inline >] handle_pte_fault mm/memory.c:3378 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121ea5c>] handle_mm_fault+0x196c/0x1d60 mm/memory.c:3534 [< inline >] faultin_page mm/gup.c:378 [<ffffffff8121829a>] __get_user_pages+0x18a/0x760 mm/gup.c:577 [< inline >] __get_user_pages_locked mm/gup.c:754 [<ffffffff81218c84>] get_user_pages_remote+0x54/0x60 mm/gup.c:962 [< inline >] get_arg_page fs/exec.c:206 [<ffffffff81280bcf>] copy_strings.isra.21+0x15f/0x3e0 fs/exec.c:521 [<ffffffff81280e84>] copy_strings_kernel+0x34/0x40 fs/exec.c:566 [<ffffffff812815fa>] do_execveat_common.isra.36+0x57a/0x970 fs/exec.c:1690 [<ffffffff81281a1c>] do_execve+0x2c/0x30 fs/exec.c:1747 [<ffffffff810c9f80>] call_usermodehelper_exec_async+0xf0/0x140 kernel/kmod.c:252 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 other info that might help us debug this: Possible interrupt unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&(ptlock_ptr(page))->rlock#2); local_irq_disable(); lock(&(&mapping->tree_lock)->rlock); lock(&(ptlock_ptr(page))->rlock#2); <Interrupt> lock(&(&mapping->tree_lock)->rlock); *** DEADLOCK *** 4 locks held by fstest/1856: #0: (&mm->mmap_sem){++++++}, at: [<ffffffff81070ad2>] __do_page_fault+0x152/0x4c0 arch/x86/mm/fault.c:1295 #1: (sb_pagefaults){++++..}, at: [<ffffffff8127c854>] __sb_start_write+0xb4/0xf0 fs/super.c:1197 #2: (&ei->i_mmap_sem){++++.+}, at: [<ffffffff8131cd04>] ext4_dax_pfn_mkwrite+0x54/0xa0 fs/ext4/file.c:273 #3: (&(&mapping->tree_lock)->rlock){-.-...}, at: [< inline >] spin_lock_irq include/linux/spinlock.h:332 #3: (&(&mapping->tree_lock)->rlock){-.-...}, at: [<ffffffff812d5086>] dax_pfn_mkwrite+0x36/0x90 fs/dax.c:1280 the dependencies between HARDIRQ-irq-safe lock and the holding lock: -> (&(&mapping->tree_lock)->rlock){-.-...} ops: 605595 { IN-HARDIRQ-W at: [< inline >] mark_irqflags kernel/locking/lockdep.c:2912 [<ffffffff8110a236>] __lock_acquire+0x706/0x14b0 kernel/locking/lockdep.c:3287 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:112 [<ffffffff81ad02ff>] _raw_spin_lock_irqsave+0x4f/0x90 kernel/locking/spinlock.c:159 [<ffffffff811f2a27>] test_clear_page_writeback+0x67/0x2a0 mm/page-writeback.c:2737 [<ffffffff811de81f>] end_page_writeback+0x1f/0xa0 mm/filemap.c:858 [<ffffffff812b9495>] end_buffer_async_write+0xc5/0x180 fs/buffer.c:375 [<ffffffff812b8338>] end_bio_bh_io_sync+0x28/0x40 fs/buffer.c:2936 [<ffffffff81575907>] bio_endio+0x57/0x60 block/bio.c:1758 [<ffffffff818915fc>] dec_pending+0x21c/0x340 drivers/md/dm.c:1015 [<ffffffff818922a6>] clone_endio+0x76/0xe0 drivers/md/dm.c:1059 [<ffffffff81575907>] bio_endio+0x57/0x60 block/bio.c:1758 [< inline >] req_bio_endio block/blk-core.c:155 [<ffffffff8157f072>] blk_update_request+0xa2/0x3c0 block/blk-core.c:2644 [<ffffffff8158998a>] blk_mq_end_request+0x1a/0x70 block/blk-mq.c:320 [<ffffffff8177bcdf>] virtblk_request_done+0x3f/0x70 drivers/block/virtio_blk.c:131 [<ffffffff81588983>] __blk_mq_complete_request_remote+0x13/0x20 block/blk-mq.c:330 [<ffffffff8114bebf>] flush_smp_call_function_queue+0x5f/0x150 kernel/smp.c:249 [<ffffffff8114c903>] generic_smp_call_function_single_interrupt+0x13/0x60 kernel/smp.c:194 [< inline >] __smp_call_function_single_interrupt arch/x86/kernel/smp.c:311 [<ffffffff810575a7>] smp_call_function_single_interrupt+0x27/0x40 arch/x86/kernel/smp.c:318 [<ffffffff81ad1926>] call_function_single_interrupt+0x96/0xa0 arch/x86/entry/entry_64.S:639 [< inline >] raw_spin_unlock_irq_rcu_node kernel/rcu/tree.h:718 [< inline >] rcu_gp_init kernel/rcu/tree.c:1934 [<ffffffff8112a5f7>] rcu_gp_kthread+0x157/0x8f0 kernel/rcu/tree.c:2175 [<ffffffff810d59e6>] kthread+0xf6/0x110 kernel/kthread.c:209 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 IN-SOFTIRQ-W at: [< inline >] mark_irqflags kernel/locking/lockdep.c:2915 [<ffffffff8110a0cf>] __lock_acquire+0x59f/0x14b0 kernel/locking/lockdep.c:3287 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:112 [<ffffffff81ad02ff>] _raw_spin_lock_irqsave+0x4f/0x90 kernel/locking/spinlock.c:159 [<ffffffff811f2a27>] test_clear_page_writeback+0x67/0x2a0 mm/page-writeback.c:2737 [<ffffffff811de81f>] end_page_writeback+0x1f/0xa0 mm/filemap.c:858 [<ffffffff8132cc89>] ext4_finish_bio+0x159/0x1e0 fs/ext4/page-io.c:119 [<ffffffff8132d09f>] ext4_end_bio+0x4f/0x100 fs/ext4/page-io.c:333 [<ffffffff81575907>] bio_endio+0x57/0x60 block/bio.c:1758 [<ffffffff818915fc>] dec_pending+0x21c/0x340 drivers/md/dm.c:1015 [<ffffffff818922a6>] clone_endio+0x76/0xe0 drivers/md/dm.c:1059 [<ffffffff81575907>] bio_endio+0x57/0x60 block/bio.c:1758 [< inline >] req_bio_endio block/blk-core.c:155 [<ffffffff8157f072>] blk_update_request+0xa2/0x3c0 block/blk-core.c:2644 [<ffffffff8158998a>] blk_mq_end_request+0x1a/0x70 block/blk-mq.c:320 [<ffffffff8177bcdf>] virtblk_request_done+0x3f/0x70 drivers/block/virtio_blk.c:131 [< inline >] blk_mq_ipi_complete_request block/blk-mq.c:354 [<ffffffff81589b48>] __blk_mq_complete_request+0x78/0xf0 block/blk-mq.c:366 [<ffffffff81589bdc>] blk_mq_complete_request+0x1c/0x20 block/blk-mq.c:385 [<ffffffff8177b543>] virtblk_done+0x73/0x100 drivers/block/virtio_blk.c:147 [<ffffffff816800ac>] vring_interrupt+0x3c/0x90 drivers/virtio/virtio_ring.c:892 [<ffffffff8111fa81>] handle_irq_event_percpu+0x41/0x330 kernel/irq/handle.c:145 [<ffffffff8111fda9>] handle_irq_event+0x39/0x60 kernel/irq/handle.c:192 [<ffffffff811232c4>] handle_edge_irq+0x74/0x130 kernel/irq/chip.c:623 [< inline >] generic_handle_irq_desc include/linux/irqdesc.h:147 [<ffffffff81036103>] handle_irq+0x73/0x120 arch/x86/kernel/irq_64.c:78 [<ffffffff81ad2f81>] do_IRQ+0x61/0x120 arch/x86/kernel/irq.c:240 [<ffffffff81ad0e16>] ret_from_intr+0x0/0x20 arch/x86/entry/entry_64.S:482 [< inline >] invoke_softirq kernel/softirq.c:350 [<ffffffff810b6dff>] irq_exit+0x10f/0x120 kernel/softirq.c:391 [< inline >] exiting_irq ./arch/x86/include/asm/apic.h:658 [<ffffffff81ad3082>] smp_apic_timer_interrupt+0x42/0x50 arch/x86/kernel/apic/apic.c:932 [<ffffffff81ad11a6>] apic_timer_interrupt+0x96/0xa0 arch/x86/entry/entry_64.S:618 [< inline >] rcu_lock_acquire include/linux/rcupdate.h:486 [< inline >] rcu_read_lock_sched include/linux/rcupdate.h:971 [< inline >] percpu_ref_get_many include/linux/percpu-refcount.h:174 [< inline >] percpu_ref_get include/linux/percpu-refcount.h:194 [< inline >] blk_queue_enter_live block/blk.h:85 [<ffffffff8158b06a>] blk_mq_map_request+0x5a/0x440 block/blk-mq.c:1175 [<ffffffff8158c275>] blk_sq_make_request+0xa5/0x500 block/blk-mq.c:1364 [<ffffffff8157e816>] generic_make_request+0xf6/0x2a0 block/blk-core.c:2076 [<ffffffff8157ea36>] submit_bio+0x76/0x170 block/blk-core.c:2139 [<ffffffff8132d25f>] ext4_io_submit+0x2f/0x40 fs/ext4/page-io.c:345 [< inline >] io_submit_add_bh fs/ext4/page-io.c:385 [<ffffffff8132d428>] ext4_bio_write_page+0x198/0x3c0 fs/ext4/page-io.c:495 [<ffffffff8132205d>] mpage_submit_page+0x5d/0x80 fs/ext4/inode.c:2091 [<ffffffff8132217b>] mpage_process_page_bufs+0xfb/0x110 fs/ext4/inode.c:2196 [<ffffffff81323662>] mpage_prepare_extent_to_map+0x202/0x300 fs/ext4/inode.c:2575 [<ffffffff81327988>] ext4_writepages+0x618/0x1020 fs/ext4/inode.c:2736 [<ffffffff811f2661>] do_writepages+0x21/0x30 mm/page-writeback.c:2364 [<ffffffff811e12a6>] __filemap_fdatawrite_range+0xc6/0x100 mm/filemap.c:300 [<ffffffff811e1424>] filemap_write_and_wait_range+0x44/0x90 mm/filemap.c:490 [<ffffffff8131dc1e>] ext4_sync_file+0x9e/0x4a0 fs/ext4/fsync.c:115 [<ffffffff812b545b>] vfs_fsync_range+0x4b/0xb0 fs/sync.c:195 [< inline >] vfs_fsync fs/sync.c:209 [<ffffffff812b551d>] do_fsync+0x3d/0x70 fs/sync.c:219 [< inline >] SYSC_fsync fs/sync.c:227 [<ffffffff812b57d0>] SyS_fsync+0x10/0x20 fs/sync.c:225 [<ffffffff81003fa7>] do_syscall_64+0x67/0x190 arch/x86/entry/common.c:350 [<ffffffff81ad053f>] return_from_SYSCALL_64+0x0/0x7a arch/x86/entry/entry_64.S:248 INITIAL USE at: [<ffffffff81109daf>] __lock_acquire+0x27f/0x14b0 kernel/locking/lockdep.c:3291 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock_irq include/linux/spinlock_api_smp.h:130 [<ffffffff81acfa03>] _raw_spin_lock_irq+0x43/0x80 kernel/locking/spinlock.c:167 [< inline >] spin_lock_irq include/linux/spinlock.h:332 [<ffffffff811dfb9e>] __add_to_page_cache_locked+0x13e/0x500 mm/filemap.c:653 [<ffffffff811dffce>] add_to_page_cache_lru+0x4e/0xe0 mm/filemap.c:702 [<ffffffff811e012e>] pagecache_get_page+0xce/0x300 mm/filemap.c:1208 [<ffffffff811e0389>] grab_cache_page_write_begin+0x29/0x40 mm/filemap.c:2581 [<ffffffff812a7ee8>] simple_write_begin+0x28/0x1b0 fs/libfs.c:428 [<ffffffff811dd80f>] pagecache_write_begin+0x1f/0x30 mm/filemap.c:2484 [<ffffffff81284220>] __page_symlink+0xc0/0x100 fs/namei.c:4720 [<ffffffff81284282>] page_symlink+0x22/0x30 fs/namei.c:4743 [<ffffffff8138e71a>] ramfs_symlink+0x4a/0xc0 fs/ramfs/inode.c:129 [<ffffffff812859bc>] vfs_symlink+0xac/0x110 fs/namei.c:4071 [< inline >] SYSC_symlinkat fs/namei.c:4098 [< inline >] SyS_symlinkat fs/namei.c:4078 [< inline >] SYSC_symlink fs/namei.c:4111 [<ffffffff8128bbc0>] SyS_symlink+0x80/0xf0 fs/namei.c:4109 [<ffffffff825b4e53>] do_symlink+0x4d/0x90 init/initramfs.c:393 [<ffffffff825b4c5f>] write_buffer+0x23/0x34 init/initramfs.c:417 [<ffffffff825b4c9b>] flush_buffer+0x2b/0x85 init/initramfs.c:429 [<ffffffff825ff590>] __gunzip+0x27e/0x322 lib/decompress_inflate.c:147 [<ffffffff825ff645>] gunzip+0x11/0x13 lib/decompress_inflate.c:193 [<ffffffff825b560b>] unpack_to_rootfs+0x17e/0x294 init/initramfs.c:485 [<ffffffff825b58f7>] populate_rootfs+0x5c/0xfc init/initramfs.c:617 [<ffffffff81002190>] do_one_initcall+0x50/0x190 init/main.c:772 [< inline >] do_initcall_level init/main.c:837 [< inline >] do_initcalls init/main.c:845 [< inline >] do_basic_setup init/main.c:863 [<ffffffff825b320f>] kernel_init_freeable+0x1f6/0x290 init/main.c:1010 [<ffffffff81ac0f1e>] kernel_init+0xe/0x100 init/main.c:936 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 } ... key at: [<ffffffff8355c9a0>] __key.44708+0x0/0x8 ??:? ... acquired at: [<ffffffff81108bcb>] check_irq_usage+0x4b/0xb0 kernel/locking/lockdep.c:1620 [< inline >] check_prev_add_irq kernel/locking/lockdep_states.h:7 [< inline >] check_prev_add kernel/locking/lockdep.c:1828 [< inline >] check_prevs_add kernel/locking/lockdep.c:1934 [< inline >] validate_chain kernel/locking/lockdep.c:2261 [<ffffffff8110a972>] __lock_acquire+0xe42/0x14b0 kernel/locking/lockdep.c:3330 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [<ffffffff8121c0b7>] finish_mkwrite_fault+0xa7/0x120 mm/memory.c:2286 [<ffffffff812d50b5>] dax_pfn_mkwrite+0x65/0x90 fs/dax.c:1290 [<ffffffff8131cd4b>] ext4_dax_pfn_mkwrite+0x9b/0xa0 fs/ext4/file.c:278 [< inline >] wp_pfn_shared mm/memory.c:2317 [<ffffffff8121c643>] do_wp_page+0x513/0x760 mm/memory.c:2403 [< inline >] handle_pte_fault mm/memory.c:3397 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121e102>] handle_mm_fault+0x1012/0x1d60 mm/memory.c:3534 [<ffffffff81070b5e>] __do_page_fault+0x1de/0x4c0 arch/x86/mm/fault.c:1356 [<ffffffff81070f1c>] trace_do_page_fault+0x5c/0x280 arch/x86/mm/fault.c:1449 [<ffffffff8106af7a>] do_async_page_fault+0x1a/0xa0 arch/x86/kernel/kvm.c:265 [<ffffffff81ad2708>] async_page_fault+0x28/0x30 arch/x86/entry/entry_64.S:923 the dependencies between the lock to be acquired and HARDIRQ-irq-unsafe lock: -> (&(ptlock_ptr(page))->rlock#2){+.+...} ops: 921722 { HARDIRQ-ON-W at: [< inline >] mark_irqflags kernel/locking/lockdep.c:2929 [<ffffffff8110a0f1>] __lock_acquire+0x5c1/0x14b0 kernel/locking/lockdep.c:3287 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [< inline >] do_anonymous_page mm/memory.c:2823 [< inline >] handle_pte_fault mm/memory.c:3378 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121ea5c>] handle_mm_fault+0x196c/0x1d60 mm/memory.c:3534 [< inline >] faultin_page mm/gup.c:378 [<ffffffff8121829a>] __get_user_pages+0x18a/0x760 mm/gup.c:577 [< inline >] __get_user_pages_locked mm/gup.c:754 [<ffffffff81218c84>] get_user_pages_remote+0x54/0x60 mm/gup.c:962 [< inline >] get_arg_page fs/exec.c:206 [<ffffffff81280bcf>] copy_strings.isra.21+0x15f/0x3e0 fs/exec.c:521 [<ffffffff81280e84>] copy_strings_kernel+0x34/0x40 fs/exec.c:566 [<ffffffff812815fa>] do_execveat_common.isra.36+0x57a/0x970 fs/exec.c:1690 [<ffffffff81281a1c>] do_execve+0x2c/0x30 fs/exec.c:1747 [<ffffffff810c9f80>] call_usermodehelper_exec_async+0xf0/0x140 kernel/kmod.c:252 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 SOFTIRQ-ON-W at: [< inline >] mark_irqflags kernel/locking/lockdep.c:2933 [<ffffffff8110a11f>] __lock_acquire+0x5ef/0x14b0 kernel/locking/lockdep.c:3287 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [< inline >] do_anonymous_page mm/memory.c:2823 [< inline >] handle_pte_fault mm/memory.c:3378 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121ea5c>] handle_mm_fault+0x196c/0x1d60 mm/memory.c:3534 [< inline >] faultin_page mm/gup.c:378 [<ffffffff8121829a>] __get_user_pages+0x18a/0x760 mm/gup.c:577 [< inline >] __get_user_pages_locked mm/gup.c:754 [<ffffffff81218c84>] get_user_pages_remote+0x54/0x60 mm/gup.c:962 [< inline >] get_arg_page fs/exec.c:206 [<ffffffff81280bcf>] copy_strings.isra.21+0x15f/0x3e0 fs/exec.c:521 [<ffffffff81280e84>] copy_strings_kernel+0x34/0x40 fs/exec.c:566 [<ffffffff812815fa>] do_execveat_common.isra.36+0x57a/0x970 fs/exec.c:1690 [<ffffffff81281a1c>] do_execve+0x2c/0x30 fs/exec.c:1747 [<ffffffff810c9f80>] call_usermodehelper_exec_async+0xf0/0x140 kernel/kmod.c:252 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 INITIAL USE at: [<ffffffff81109daf>] __lock_acquire+0x27f/0x14b0 kernel/locking/lockdep.c:3291 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [< inline >] do_anonymous_page mm/memory.c:2823 [< inline >] handle_pte_fault mm/memory.c:3378 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121ea5c>] handle_mm_fault+0x196c/0x1d60 mm/memory.c:3534 [< inline >] faultin_page mm/gup.c:378 [<ffffffff8121829a>] __get_user_pages+0x18a/0x760 mm/gup.c:577 [< inline >] __get_user_pages_locked mm/gup.c:754 [<ffffffff81218c84>] get_user_pages_remote+0x54/0x60 mm/gup.c:962 [< inline >] get_arg_page fs/exec.c:206 [<ffffffff81280bcf>] copy_strings.isra.21+0x15f/0x3e0 fs/exec.c:521 [<ffffffff81280e84>] copy_strings_kernel+0x34/0x40 fs/exec.c:566 [<ffffffff812815fa>] do_execveat_common.isra.36+0x57a/0x970 fs/exec.c:1690 [<ffffffff81281a1c>] do_execve+0x2c/0x30 fs/exec.c:1747 [<ffffffff810c9f80>] call_usermodehelper_exec_async+0xf0/0x140 kernel/kmod.c:252 [<ffffffff81ad06af>] ret_from_fork+0x1f/0x40 arch/x86/entry/entry_64.S:389 } ... key at: [<ffffffff8279cc18>] __key.17932+0x0/0x8 ??:? ... acquired at: [<ffffffff81108bcb>] check_irq_usage+0x4b/0xb0 kernel/locking/lockdep.c:1620 [< inline >] check_prev_add_irq kernel/locking/lockdep_states.h:7 [< inline >] check_prev_add kernel/locking/lockdep.c:1828 [< inline >] check_prevs_add kernel/locking/lockdep.c:1934 [< inline >] validate_chain kernel/locking/lockdep.c:2261 [<ffffffff8110a972>] __lock_acquire+0xe42/0x14b0 kernel/locking/lockdep.c:3330 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [<ffffffff8121c0b7>] finish_mkwrite_fault+0xa7/0x120 mm/memory.c:2286 [<ffffffff812d50b5>] dax_pfn_mkwrite+0x65/0x90 fs/dax.c:1290 [<ffffffff8131cd4b>] ext4_dax_pfn_mkwrite+0x9b/0xa0 fs/ext4/file.c:278 [< inline >] wp_pfn_shared mm/memory.c:2317 [<ffffffff8121c643>] do_wp_page+0x513/0x760 mm/memory.c:2403 [< inline >] handle_pte_fault mm/memory.c:3397 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121e102>] handle_mm_fault+0x1012/0x1d60 mm/memory.c:3534 [<ffffffff81070b5e>] __do_page_fault+0x1de/0x4c0 arch/x86/mm/fault.c:1356 [<ffffffff81070f1c>] trace_do_page_fault+0x5c/0x280 arch/x86/mm/fault.c:1449 [<ffffffff8106af7a>] do_async_page_fault+0x1a/0xa0 arch/x86/kernel/kvm.c:265 [<ffffffff81ad2708>] async_page_fault+0x28/0x30 arch/x86/entry/entry_64.S:923 stack backtrace: CPU: 0 PID: 1856 Comm: fstest Not tainted 4.7.0+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014 0000000000000086 0000000070a78a50 ffff8800a768f9a8 ffffffff815b20a3 ffffffff82ec47f0 0000000000000030 ffff8800a768fac0 ffffffff81108b29 0000000000000000 0000000000000000 0000000000000001 000000000d3ea148 Call Trace: [< inline >] __dump_stack lib/dump_stack.c:15 [<ffffffff815b20a3>] dump_stack+0x85/0xc2 lib/dump_stack.c:51 [< inline >] print_bad_irq_dependency kernel/locking/lockdep.c:1532 [<ffffffff81108b29>] check_usage+0x539/0x590 kernel/locking/lockdep.c:1564 [<ffffffff81108bcb>] check_irq_usage+0x4b/0xb0 kernel/locking/lockdep.c:1620 [< inline >] check_prev_add_irq kernel/locking/lockdep_states.h:7 [< inline >] check_prev_add kernel/locking/lockdep.c:1828 [< inline >] check_prevs_add kernel/locking/lockdep.c:1934 [< inline >] validate_chain kernel/locking/lockdep.c:2261 [<ffffffff8110a972>] __lock_acquire+0xe42/0x14b0 kernel/locking/lockdep.c:3330 [<ffffffff8110b490>] lock_acquire+0xf0/0x1d0 kernel/locking/lockdep.c:3741 [< inline >] __raw_spin_lock include/linux/spinlock_api_smp.h:144 [<ffffffff81acf796>] _raw_spin_lock+0x36/0x70 kernel/locking/spinlock.c:151 [< inline >] spin_lock include/linux/spinlock.h:302 [<ffffffff8121c0b7>] finish_mkwrite_fault+0xa7/0x120 mm/memory.c:2286 [<ffffffff812d50b5>] dax_pfn_mkwrite+0x65/0x90 fs/dax.c:1290 [<ffffffff8131cd4b>] ext4_dax_pfn_mkwrite+0x9b/0xa0 fs/ext4/file.c:278 [< inline >] wp_pfn_shared mm/memory.c:2317 [<ffffffff8121c643>] do_wp_page+0x513/0x760 mm/memory.c:2403 [< inline >] handle_pte_fault mm/memory.c:3397 [< inline >] __handle_mm_fault mm/memory.c:3505 [<ffffffff8121e102>] handle_mm_fault+0x1012/0x1d60 mm/memory.c:3534 [<ffffffff81070b5e>] __do_page_fault+0x1de/0x4c0 arch/x86/mm/fault.c:1356 [<ffffffff81070f1c>] trace_do_page_fault+0x5c/0x280 arch/x86/mm/fault.c:1449 [<ffffffff8106af7a>] do_async_page_fault+0x1a/0xa0 arch/x86/kernel/kvm.c:265 [<ffffffff81ad2708>] async_page_fault+0x28/0x30 arch/x86/entry/entry_64.S:923 -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon 25-07-16 15:30:59, Ross Zwisler wrote: > On Fri, Jul 22, 2016 at 02:19:40PM +0200, Jan Kara wrote: > > Currently PTE gets updated in wp_pfn_shared() after dax_pfn_mkwrite() > > has released corresponding radix tree entry lock. When we want to > > writeprotect PTE on cache flush, we need PTE modification to happen > > under radix tree entry lock to ensure consisten updates of PTE and radix > > tree (standard faults use page lock to ensure this consistency). So move > > update of PTE bit into dax_pfn_mkwrite(). > > > > Signed-off-by: Jan Kara <jack@suse.cz> > > After applying the whole series to a v4.7 baseline I was hitting a deadlock in > my testing, and it bisected to this commit. This deadlock happens in my QEMU > guest with generic/068, ext4 and DAX. It reproduces 100% of the time after > this commit. > > Here is the lockdep info, passed through kasan_symbolize.py: Thanks! I've checked why I didn't see this and apparently I've run last round of testing on the wrong branch. Drat. I've fixed the bug you've spotted (we need to release tree_lock earlier) but xfstests are triggering some more issues now for me so I'm debugging those. Honza
diff --git a/fs/dax.c b/fs/dax.c index 513881431be6..e8d61ac3d148 100644 --- a/fs/dax.c +++ b/fs/dax.c @@ -1218,6 +1218,12 @@ int dax_pfn_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) if (!entry || !radix_tree_exceptional_entry(entry)) goto out; radix_tree_tag_set(&mapping->page_tree, index, PAGECACHE_TAG_DIRTY); + /* + * If we race with somebody updating the PTE and finish_mkwrite_fault() + * fails, we don't care. We need to return VM_FAULT_NOPAGE and retry + * the fault in either case. + */ + finish_mkwrite_fault(vma, vmf); put_unlocked_mapping_entry(mapping, index, entry); out: spin_unlock_irq(&mapping->tree_lock); diff --git a/mm/memory.c b/mm/memory.c index 30cf7b36df48..47241c2f6178 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -2315,7 +2315,7 @@ static int wp_pfn_shared(struct mm_struct *mm, linear_page_index(vma, address), FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE, orig_pte); ret = vma->vm_ops->pfn_mkwrite(vma, &vmf); - if (ret & VM_FAULT_ERROR) + if (ret & VM_FAULT_ERROR || ret & VM_FAULT_NOPAGE) return ret; if (finish_mkwrite_fault(vma, &vmf) < 0) return 0;
Currently PTE gets updated in wp_pfn_shared() after dax_pfn_mkwrite() has released corresponding radix tree entry lock. When we want to writeprotect PTE on cache flush, we need PTE modification to happen under radix tree entry lock to ensure consisten updates of PTE and radix tree (standard faults use page lock to ensure this consistency). So move update of PTE bit into dax_pfn_mkwrite(). Signed-off-by: Jan Kara <jack@suse.cz> --- fs/dax.c | 6 ++++++ mm/memory.c | 2 +- 2 files changed, 7 insertions(+), 1 deletion(-)