Skip to contentSkip to conversation

dm: Fix lock dependency warning for request based dm

PATCH In review Christof Schmitt 2 participants 3 messages Feb 18, 2009 – Feb 19, 2009 Device mapper

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dm: Fix lock dependency warning for request based dm

From: Christof Schmitt <[email protected]>

Testing with request based dm multipathing and lock dependency checking revealed this problem. Fix this by disabling interrupts when acquiring the map_lock from the ioctl call in __bind and __unbind.

It seems that the problem has been introduced with this patch:
http://lkml.indiana.edu/hypermail/linux/kernel/0810.0/1067.html

======================================================
[ INFO: soft-safe -> soft-unsafe lock order detected ] 2.6.27.13-1.1.mz11dbg-default #1
------------------------------------------------------
multipath/3857 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
(&md->map_lock){----}, at: [<000003e000573a0e>] dm_get_table+0x32/0x68 [dm_mod]

and this task is already holding:
(&q->__queue_lock){.+..}, at: [<000003e000573b04>] dm_resume+0xc0/0x1b8 [dm_mod]
which would create a new lock dependency:
(&q->__queue_lock){.+..} -> (&md->map_lock){----}

but this new dependency connects a soft-irq-safe lock:
(&q->__queue_lock){.+..}
... which became soft-irq-safe at:

  [<00000000001633fa>] mark_irqflags+0xa6/0x154
  [<0000000000165486>] __lock_acquire+0x5ea/0x780
  [<0000000000165686>] lock_acquire+0x6a/0x90
  [<00000000003b5f8a>] _spin_lock+0x4a/0x64
  [<000003e000786fd2>] scsi_device_unbusy+0xb2/0xe4 [scsi_mod]
  [<000003e00077def4>] scsi_finish_command+0x4c/0x11c [scsi_mod]
  [<000003e000787278>] scsi_softirq_done+0x170/0x180 [scsi_mod]
  [<00000000002a4c2a>] blk_done_softirq+0xc6/0xe4
  [<000000000013c9d2>] __do_softirq+0xee/0x1ec
  [<000000000010fcc8>] do_softirq+0x9c/0xd8
  [<000000000013c3e6>] irq_exit+0x82/0xe0
  [<00000000002fda4c>] do_IRQ+0x170/0x190
  [<0000000000113c36>] io_return+0x0/0x8
  [<0000000000109c10>] default_idle+0x198/0x1a8

to a soft-irq-unsafe lock:
(&md->map_lock){----}
... which became soft-irq-unsafe at:
... [<0000000000163478>] mark_irqflags+0x124/0x154

  [<0000000000165486>] __lock_acquire+0x5ea/0x780
  [<0000000000165686>] lock_acquire+0x6a/0x90
  [<00000000003b6146>] _write_lock+0x4a/0x64
  [<000003e000572a1a>] __bind+0x116/0x184 [dm_mod]
  [<000003e000572b18>] dm_swap_table+0x90/0xb8 [dm_mod]
  [<000003e0005791e4>] do_resume+0xe4/0x17c [dm_mod]
  [<000003e00057932e>] dev_suspend+0xb2/0xc8 [dm_mod]
  [<000003e000579f46>] ctl_ioctl+0x236/0x2ac [dm_mod]
  [<000003e000579fea>] dm_ctl_ioctl+0x2e/0x40 [dm_mod]
  [<00000000001e7764>] vfs_ioctl+0x50/0xbc
  [<00000000001e7c64>] do_vfs_ioctl+0x26c/0x280
  [<00000000001e7d4e>] SyS_ioctl+0xd6/0xfc
  [<0000000000113696>] sysc_noemu+0x10/0x16
  [<000002000019269e>] 0x2000019269e

[...]

stack backtrace:
CPU: 3 Not tainted 2.6.27.13-1.1.mz11dbg-default #1
Process multipath (pid: 3857, task: 0000000139e06038, ksp: 000000013416b858)
0000000139e067d0 000000013416b708 0000000000000002 0000000000000000 
       000000013416b7a8 000000013416b720 000000013416b720 0000000000104c3e 
       0000000000000003 0000000000000000 0000000000000000 000000010000000b 
       0000000000000060 0000000000000008 000000013416b708 000000013416b778 
       00000000003ba8b8 0000000000104c3e 000000013416b708 000000013416b760 
Call Trace:
([<0000000000104b86>] show_trace+0xb2/0xd0)
 [<0000000000104c5c>] show_stack+0xb8/0xc8
 [<00000000003b2392>] dump_stack+0xae/0xbc
 [<00000000001640ec>] print_bad_irq_dependency+0x320/0x334
 [<0000000000164226>] check_usage+0x126/0x138
 [<0000000000164368>] check_prev_add+0x130/0x530
 [<0000000000164d3a>] validate_chain+0x5d2/0x734
 [<00000000001655aa>] __lock_acquire+0x70e/0x780
 [<0000000000165686>] lock_acquire+0x6a/0x90
 [<00000000003b645a>] _read_lock+0x4a/0x64
 [<000003e000573a0e>] dm_get_table+0x32/0x68 [dm_mod]
 [<000003e000574b9a>] dm_request_fn+0x3e/0x1a4 [dm_mod]
 [<000000000029f1ac>] blk_invoke_request_fn+0x94/0x170
 [<000000000029f304>] blk_start_queue+0x7c/0x9c
 [<000003e000573b20>] dm_resume+0xdc/0x1b8 [dm_mod]
 [<000003e00057924a>] do_resume+0x14a/0x17c [dm_mod]
 [<000003e00057932e>] dev_suspend+0xb2/0xc8 [dm_mod]
 [<000003e000579f46>] ctl_ioctl+0x236/0x2ac [dm_mod]
 [<000003e000579fea>] dm_ctl_ioctl+0x2e/0x40 [dm_mod]
 [<00000000001e7764>] vfs_ioctl+0x50/0xbc
 [<00000000001e7c64>] do_vfs_ioctl+0x26c/0x280
 [<00000000001e7d4e>] SyS_ioctl+0xd6/0xfc
 [<0000000000113696>] sysc_noemu+0x10/0x16
 [<000002000019269e>] 0x2000019269e

INFO: lockdep is turned off.

drivers/md/dm.c +4 -4
@@ -1943,7 +1943,7 @@ static int __bind(struct mapped_device *
 	if (dm_table_request_based(t) && !blk_queue_stopped(q))
 		stop_queue(q);
 
-	write_lock(&md->map_lock);
+	write_lock_irq(&md->map_lock);
 	md->map = t;
 	dm_table_set_restrictions(t, q);
 	dm_table_set_integrity(t, md);
@@ -1952,7 +1952,7 @@ static int __bind(struct mapped_device *
 	} else {
 		set_disk_ro(md->disk, 0);
 	}
-	write_unlock(&md->map_lock);
+	write_unlock_irq(&md->map_lock);
 
 	return 0;
 }
@@ -1965,9 +1965,9 @@ static void __unbind(struct mapped_devic
 		return;
 
 	dm_table_event_callback(map, NULL, NULL);
-	write_lock(&md->map_lock);
+	write_lock_irq(&md->map_lock);
 	md->map = NULL;
-	write_unlock(&md->map_lock);
+	write_unlock_irq(&md->map_lock);
 	dm_table_destroy(map);
 }
 
#1 lore raw

Hi Christof,

Quoted 8 lines · Christof Schmitt
Quoted 8 lines

From: Christof Schmitt <[email protected]>

Testing with request based dm multipathing and lock dependency checking revealed this problem. Fix this by disabling interrupts when acquiring the map_lock from the ioctl call in __bind and __unbind.

It seems that the problem has been introduced with this patch:
http://lkml.indiana.edu/hypermail/linux/kernel/0810.0/1067.html

Thank you for your testing request-based dm-multipath and the patch.

Attached is a patch to fix it.
Since request-based dm gets map_lock after taking queue_lock with interrupt disabled, we have to use save/restore variant.
(By the way, although lockdep warns the deadlock possibility, currently
there should be no such code path in request-based dm where request_fn is called from the interrupt context.)

I have done simple build and boot testings, but haven't done other testings (e.g. stress testing) yet.
I will include this patch to the next update after such testings.

Thanks,
Kiyoshi Ueda

drivers/md/dm.c +9 -6
Index: linux-2.6.29-rc2/drivers/md/dm.c
===================================================================
@@ -504,12 +504,13 @@ static int queue_io(struct mapped_device
 struct dm_table *dm_get_table(struct mapped_device *md)
 {
 	struct dm_table *t;
+	unsigned long flags;
 
-	read_lock(&md->map_lock);
+	read_lock_irqsave(&md->map_lock, flags);
 	t = md->map;
 	if (t)
 		dm_table_get(t);
-	read_unlock(&md->map_lock);
+	read_unlock_irqrestore(&md->map_lock, flags);
 
 	return t;
 }
@@ -1892,6 +1893,7 @@ static int __bind(struct mapped_device *
 {
 	struct request_queue *q = md->queue;
 	sector_t size;
+	unsigned long flags;
 
 	size = dm_table_get_size(t);
 
@@ -1921,10 +1923,10 @@ static int __bind(struct mapped_device *
 	if (dm_table_request_based(t) && !blk_queue_stopped(q))
 		stop_queue(q);
 
-	write_lock(&md->map_lock);
+	write_lock_irqsave(&md->map_lock, flags);
 	md->map = t;
 	dm_table_set_restrictions(t, q);
-	write_unlock(&md->map_lock);
+	write_unlock_irqrestore(&md->map_lock, flags);
 
 	return 0;
 }
@@ -1932,14 +1934,15 @@ static int __bind(struct mapped_device *
 static void __unbind(struct mapped_device *md)
 {
 	struct dm_table *map = md->map;
+	unsigned long flags;
 
 	if (!map)
 		return;
 
 	dm_table_event_callback(map, NULL, NULL);
-	write_lock(&md->map_lock);
+	write_lock_irqsave(&md->map_lock, flags);
 	md->map = NULL;
-	write_unlock(&md->map_lock);
+	write_unlock_irqrestore(&md->map_lock, flags);
 	dm_table_destroy(map);
 }
 
#2 lore raw
Quoted 12 lines · Kiyoshi Ueda

Hi Christof,

Quoted 8 lines · Christof Schmitt
Quoted 8 lines

From: Christof Schmitt <[email protected]>

Testing with request based dm multipathing and lock dependency checking revealed this problem. Fix this by disabling interrupts when acquiring the map_lock from the ioctl call in __bind and __unbind.

It seems that the problem has been introduced with this patch:
http://lkml.indiana.edu/hypermail/linux/kernel/0810.0/1067.html

Quoted 8 lines

Thank you for your testing request-based dm-multipath and the patch.

Attached is a patch to fix it.
Since request-based dm gets map_lock after taking queue_lock with interrupt disabled, we have to use save/restore variant.
(By the way, although lockdep warns the deadlock possibility, currently
there should be no such code path in request-based dm where request_fn is called from the interrupt context.)

I was checking for other locking problems when this warning from lockdep showed up. While i agree that it is only a problem in theory, it deserves fixing to continue using lockdep.

Quoted 3 lines

I have done simple build and boot testings, but haven't done other testings (e.g. stress testing) yet.
I will include this patch to the next update after such testings.

Thanks for the updated patch. I will include this one in future tests.

--
Christof Schmitt