percpu allocator vs reclaim
Hi Tejun,
Thomas hit the below on recent -tip kernels.
[ 371.513742] ======================================================
[ 371.514672] [ INFO: RECLAIM_FS-safe -> RECLAIM_FS-unsafe lock order detected ]
[ 371.514672] 2.6.29-rc6-tip-02441-g4239438-dirty #36
[ 371.514672] ------------------------------------------------------
[ 371.514672] umount/19574 [HC0[0]:SC0[0]:HE1:SE1] is trying to acquire:
[ 371.514672] (pcpu_mutex){+.+.+.}, at: [<ffffffff810c589e>] free_percpu+0x2f/0x17e
[ 371.514672]
[ 371.514672] and this task is already holding:
[ 371.514672] (&type->s_lock_key#7){+.+...}, at: [<ffffffff810c9367>] lock_super+0x29/0x2b
[ 371.514672] which would create a new lock dependency:
[ 371.514672] (&type->s_lock_key#7){+.+...} -> (pcpu_mutex){+.+.+.}
[ 371.514672]
[ 371.514672] but this new dependency connects a RECLAIM_FS-irq-safe lock:
[ 371.514672] (jbd_handle){+.+.-.}
[ 371.514672] ... which became RECLAIM_FS-irq-safe at:
[ 371.514672] [<ffffffff8105f36a>] __lock_acquire+0x79a/0x160a
[ 371.514672] [<ffffffff8106022f>] lock_acquire+0x55/0x71
[ 371.514672] [<ffffffffa002ceb8>] journal_start+0x10e/0x11a [jbd]
[ 371.514672] [<ffffffffa004aa39>] ext3_journal_start_sb+0x4a/0x4c [ext3]
[ 371.514672] [<ffffffffa0040db2>] ext3_ordered_writepage+0x56/0x140 [ext3]
[ 371.514672] [<ffffffff8109f0f5>] shrink_page_list+0x37d/0x62d
[ 371.514672] [<ffffffff8109f9c0>] shrink_list+0x2a3/0x5a5
[ 371.514672] [<ffffffff8109ff41>] shrink_zone+0x27f/0x329
[ 371.514672] [<ffffffff810a086c>] kswapd+0x49e/0x684
[ 371.514672] [<ffffffff810502c5>] kthread+0x49/0x76
[ 371.514672] [<ffffffff8100cbfa>] child_rip+0xa/0x20
[ 371.514672] [<ffffffffffffffff>] 0xffffffffffffffff
[ 371.514672]
[ 371.514672] to a RECLAIM_FS-irq-unsafe lock:
[ 371.514672] (pcpu_mutex){+.+.+.}
[ 371.514672] ... which became RECLAIM_FS-irq-unsafe at:
[ 371.514672] ... [<ffffffff8105dced>] mark_held_locks+0x4d/0x69
[ 371.514672] [<ffffffff8105dd7e>] lockdep_trace_alloc+0x75/0x77
[ 371.514672] [<ffffffff810c27b8>] __kmalloc+0x61/0xf8
[ 371.514672] [<ffffffff810c57cd>] pcpu_realloc+0x2e/0x91
[ 371.514672] [<ffffffff810c5a64>] pcpu_alloc_area+0x77/0x376
[ 371.514672] [<ffffffff810c5e25>] __alloc_percpu+0xc2/0x395
[ 371.514672] [<ffffffff8116dc04>] __percpu_counter_init+0x51/0x9c
[ 371.514672] [<ffffffff815579f9>] files_init+0x74/0x78
[ 371.514672] [<ffffffff81557bd6>] vfs_caches_init+0x10b/0x121
[ 371.514672] [<ffffffff8153dbf3>] start_kernel+0x340/0x383
[ 371.514672] [<ffffffff8153d29a>] x86_64_start_reservations+0xaa/0xae
[ 371.514672] [<ffffffff8153d36e>] x86_64_start_kernel+0xd0/0xd7
[ 371.514672] [<ffffffffffffffff>] 0xffffffffffffffff<snip>
[ 373.367127] stack backtrace: [ 373.367127] Pid: 19574, comm: umount Not tainted 2.6.29-rc6-tip-02441-g4239438-dirty #36 [ 373.367127] Call Trace: [ 373.367127] [<ffffffff8105eafa>] check_usage+0x3ca/0x3db [ 373.367127] [<ffffffff8105eb6c>] check_irq_usage+0x61/0xc5 [ 373.367127] [<ffffffff8105fbe8>] __lock_acquire+0x1018/0x160a [ 373.367127] [<ffffffff812df330>] ? io_schedule+0x82/0xa5 [ 373.367127] [<ffffffff8106022f>] lock_acquire+0x55/0x71 [ 373.367127] [<ffffffff810c589e>] ? free_percpu+0x2f/0x17e [ 373.367127] [<ffffffff812dff40>] mutex_lock_nested+0x45/0x29e [ 373.367127] [<ffffffff810c589e>] ? free_percpu+0x2f/0x17e [ 373.367127] [<ffffffff812dfa79>] ? __mutex_unlock_slowpath+0x115/0x121 [ 373.367127] [<ffffffff8105df29>] ? trace_hardirqs_on_caller+0x114/0x138 [ 373.367127] [<ffffffff810c589e>] free_percpu+0x2f/0x17e [ 373.367127] [<ffffffff8116dba7>] percpu_counter_destroy+0x3f/0x4b [ 373.367127] [<ffffffffa0049056>] ext3_put_super+0xc7/0x21e [ext3] [ 373.367127] [<ffffffff810ca04c>] generic_shutdown_super+0x73/0xe8 [ 373.367127] [<ffffffff810ca0e3>] kill_block_super+0x22/0x3a [ 373.367127] [<ffffffff810ca1ca>] deactivate_super+0x68/0x7d [ 373.367127] [<ffffffff810ddbeb>] mntput_no_expire+0x106/0x147 [ 373.367127] [<ffffffff810de1b9>] sys_umount+0x2dd/0x30c [ 373.367127] [<ffffffff8100bb1b>] system_call_fastpath+0x16/0x1b
Which basically states that we could deadlock due to reclaim lock recursion.
Looking at the code I don't see a quick solution, other than using GFP_NOFS, which is a bit of a bother (as I suspect it might easily grow __GFP_IO inversion too, if it doesn't already have it).
Quoted 10 lines · Peter Zijlstra
Hi Tejun,
Thomas hit the below on recent -tip kernels.
Which basically states that we could deadlock due to reclaim lock recursion.
Looking at the code I don't see a quick solution, other than using
GFP_NOFS, which is a bit of a bother (as I suspect it might easily grow__GFP_IOinversion too, if it doesn't already have it).
Ah... maybe percpu allocator should just swallow @gfp. Any better ideas? :-(
Signature · 1 line
tejun
Quoted 8 lines · Tejun Heo
Quoted 7 lines · Peter Zijlstra
Hi Tejun,
Thomas hit the below on recent -tip kernels.
Which basically states that we could deadlock due to reclaim lock recursion.
Quoted 6 lines
Quoted 3 lines
Looking at the code I don't see a quick solution, other than using
GFP_NOFS, which is a bit of a bother (as I suspect it might easily grow__GFP_IOinversion too, if it doesn't already have it).Ah... maybe percpu allocator should just swallow @gfp. Any better ideas? :-(
Could you somehow break that lock so that you get something like:
alloc_lock
kmalloc(GFP_KERNEL)
free_lock
Where
percpu_free()
lock(free_lock)
put area on free list
percpu_alloc()
lock(free_list)
collect free list
kmalloc()
Then the free code can be used from reclaim, because there's never an allocation done while holding it, and the alloc path can first check/cleanup whatever mess the last free left behind before trying an allocation.
Quoted 32 lines · Peter Zijlstra
Quoted 5 lines · Tejun Heo
Quoted 4 lines · Peter Zijlstra
Hi Tejun,
Thomas hit the below on recent -tip kernels.
Quoted 8 lines
Quoted 6 lines
Which basically states that we could deadlock due to reclaim lock recursion.
Looking at the code I don't see a quick solution, other than using
GFP_NOFS, which is a bit of a bother (as I suspect it might easily grow__GFP_IOinversion too, if it doesn't already have it).Ah... maybe percpu allocator should just swallow @gfp. Any better ideas? :-(
Could you somehow break that lock so that you get something like:
alloc_lock
kmalloc(GFP_KERNEL)
free_lockWhere
percpu_free()
lock(free_lock)
put area on free list
percpu_alloc()
lock(free_list)
collect free list
kmalloc()
Quoted 4 lines
Then the free code can be used from reclaim, because there's never an allocation done while holding it, and the alloc path can first check/cleanup whatever mess the last free left behind before trying an allocation.
Yeap, I think I can do that. Will work on it.
Thanks.
Signature · 1 line
tejun