Re: 2.6.23-rc2-mm1 -- INFO: possible circular locking dependency detected

[Alexey Starikovskiy <aystarik@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3823 bytes --]

Andrew Morton wrote:
> On Thu, 9 Aug 2007 16:24:48 -0400
> "Miles Lane" <miles.lane@gmail.com> wrote:
> 
>> [ INFO: possible circular locking dependency detected ]
>> 2.6.23-rc2-mm1 #7
>> -------------------------------------------------------
>> kacpid/53 is trying to acquire lock:
>>  (&ec->lock){--..}, at: [<c03031a7>] mutex_lock+0x1c/0x1f
>>
>> but task is already holding lock:
>>  (&dpc->work){--..}, at: [<c012689d>] run_workqueue+0xa0/0x182
>>
>> which lock already depends on the new lock.
>>
>>
>> the existing dependency chain (in reverse order) is:
>>
>> -> #2 (&dpc->work){--..}:
>>        [<c0133d24>] __lock_acquire+0x9a6/0xb6f
>>        [<c0133f4e>] lock_acquire+0x61/0x7d
>>        [<c01268b2>] run_workqueue+0xb5/0x182
>>        [<c01271a9>] worker_thread+0xb7/0xc2
>>        [<c01296c4>] kthread+0x39/0x61
>>        [<c0104913>] kernel_thread_helper+0x7/0x10
>>        [<ffffffff>] 0xffffffff
>>
>> -> #1 (kacpid){--..}:
>>        [<c0133d24>] __lock_acquire+0x9a6/0xb6f
>>        [<c0133f4e>] lock_acquire+0x61/0x7d
>>        [<c0126f62>] flush_workqueue+0x2d/0x4f
>>        [<c01e85e0>] acpi_os_wait_events_complete+0xd/0xf
>>        [<c01ef605>] acpi_remove_gpe_handler+0x7b/0xdd
>>        [<c0205981>] ec_remove_handlers+0x26/0x29
>>        [<c02062b4>] acpi_ec_add+0x8f/0x13e
>>        [<c0205477>] acpi_device_probe+0x3e/0xdb
>>        [<c023c4c8>] driver_probe_device+0xd7/0x14d
>>        [<c023c652>] __driver_attach+0x6a/0xa1
>>        [<c023baaa>] bus_for_each_dev+0x36/0x5b
>>        [<c023c32e>] driver_attach+0x14/0x16
>>        [<c023bd7e>] bus_add_driver+0x70/0x16c
>>        [<c023c82d>] driver_register+0x60/0x65
>>        [<c020577b>] acpi_bus_register_driver+0x3a/0x3c
>>        [<c04292e4>] acpi_ec_init+0x36/0x55
>>        [<c0416650>] kernel_init+0xc5/0x20f
>>        [<c0104913>] kernel_thread_helper+0x7/0x10
>>        [<ffffffff>] 0xffffffff
>>
>> -> #0 (&ec->lock){--..}:
>>        [<c0133c44>] __lock_acquire+0x8c6/0xb6f
>>        [<c0133f4e>] lock_acquire+0x61/0x7d
>>        [<c0303006>] __mutex_lock_slowpath+0xbc/0x241
>>        [<c03031a7>] mutex_lock+0x1c/0x1f
>>        [<c0205bbd>] acpi_ec_transaction+0x65/0x1c1
>>        [<c0205d44>] acpi_ec_gpe_query+0x2b/0xab
>>        [<c01e8602>] acpi_os_execute_deferred+0x20/0x31
>>        [<c01268b7>] run_workqueue+0xba/0x182
>>        [<c01271a9>] worker_thread+0xb7/0xc2
>>        [<c01296c4>] kthread+0x39/0x61
>>        [<c0104913>] kernel_thread_helper+0x7/0x10
>>        [<ffffffff>] 0xffffffff
>>
>> other info that might help us debug this:
>>
>> 2 locks held by kacpid/53:
>>  #0:  (kacpid){--..}, at: [<c0126882>] run_workqueue+0x85/0x182
>>  #1:  (&dpc->work){--..}, at: [<c012689d>] run_workqueue+0xa0/0x182
>>
>> stack backtrace:
>>  [<c0104c6a>] show_trace_log_lvl+0x12/0x25
>>  [<c0105552>] show_trace+0xd/0x10
>>  [<c0105656>] dump_stack+0x15/0x17
>>  [<c0132580>] print_circular_bug_tail+0x5a/0x65
>>  [<c0133c44>] __lock_acquire+0x8c6/0xb6f
>>  [<c0133f4e>] lock_acquire+0x61/0x7d
>>  [<c0303006>] __mutex_lock_slowpath+0xbc/0x241
>>  [<c03031a7>] mutex_lock+0x1c/0x1f
>>  [<c0205bbd>] acpi_ec_transaction+0x65/0x1c1
>>  [<c0205d44>] acpi_ec_gpe_query+0x2b/0xab
>>  [<c01e8602>] acpi_os_execute_deferred+0x20/0x31
>>  [<c01268b7>] run_workqueue+0xba/0x182
>>  [<c01271a9>] worker_thread+0xb7/0xc2
>>  [<c01296c4>] kthread+0x39/0x61
>>  [<c0104913>] kernel_thread_helper+0x7/0x10
>>  =======================
> 
> Presumably the new debugging patches in -mm
> (workqueue-debug-flushing-deadlocks-with-lockdep.patch and
> workqueue-debug-work-related-deadlocks-with-lockdep.patch) think they have
> found a potential deadlock in ACPI.  I don't have time to pick through the
> code to confirm that, but boy I'm good at adding cc's ;)
Yep, it indeed may lock up... Here is a patch to avoid it

Thanks,
Alex.



[-- Attachment #2: remove_potential_deadlock_from_ec.patch --]
[-- Type: text/x-patch, Size: 714 bytes --]

ACPI EC: remove potential deadlock from EC.

From: Alexey Starikovskiy <astarikivskiy@suse.de>

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
---

 drivers/acpi/ec.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/drivers/acpi/ec.c b/drivers/acpi/ec.c
index ceb7c3f..4b299fd 100644
--- a/drivers/acpi/ec.c
+++ b/drivers/acpi/ec.c
@@ -723,9 +723,7 @@ static int acpi_ec_add(struct acpi_device *device)
 	/* Check if we found the boot EC */
 	if (boot_ec) {
 		if (boot_ec->gpe == ec->gpe) {
-			mutex_lock(&boot_ec->lock);
 			ec_remove_handlers(boot_ec);
-			mutex_unlock(&boot_ec->lock);
 			mutex_destroy(&boot_ec->lock);
 			kfree(boot_ec);
 			first_ec = boot_ec = NULL;