At Wed, 7 Oct 2009 22:51:34 +0200, Krzysztof Helt wrote:
From: Krzysztof Helt krzysztof.h1@wp.pl
Fix following circular locking in the opl3 driver.
======================================================= [ INFO: possible circular locking dependency detected ] 2.6.32-rc3 #87
swapper/0 is trying to acquire lock: (&opl3->voice_lock){..-...}, at: [<cca748fe>] snd_opl3_note_off+0x1e/0xe0 [snd_opl3_synth]
but task is already holding lock: (&opl3->sys_timer_lock){..-...}, at: [<cca75169>] snd_opl3_timer_func+0x19/0xc0 [snd_opl3_synth]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (&opl3->sys_timer_lock){..-...}: [<c02461d5>] validate_chain+0xa25/0x1040 [<c0246aca>] __lock_acquire+0x2da/0xab0 [<c024731a>] lock_acquire+0x7a/0xa0 [<c044c300>] _spin_lock_irqsave+0x40/0x60 [<cca75046>] snd_opl3_note_on+0x686/0x790 [snd_opl3_synth] [<cca68912>] snd_midi_process_event+0x322/0x590 [snd_seq_midi_emul] [<cca74245>] snd_opl3_synth_event_input+0x15/0x20 [snd_opl3_synth] [<cca4dcc0>] snd_seq_deliver_single_event+0x100/0x200 [snd_seq] [<cca4de07>] snd_seq_deliver_event+0x47/0x1f0 [snd_seq] [<cca4e50b>] snd_seq_dispatch_event+0x3b/0x140 [snd_seq] [<cca5008c>] snd_seq_check_queue+0x10c/0x120 [snd_seq] [<cca5037b>] snd_seq_enqueue_event+0x6b/0xe0 [snd_seq] [<cca4e0fd>] snd_seq_client_enqueue_event+0xdd/0x100 [snd_seq] [<cca4eb7a>] snd_seq_write+0xea/0x190 [snd_seq] [<c02827b6>] vfs_write+0x96/0x160 [<c0282c9d>] sys_write+0x3d/0x70 [<c0202c45>] syscall_call+0x7/0xb
-> #0 (&opl3->voice_lock){..-...}: [<c02467e6>] validate_chain+0x1036/0x1040 [<c0246aca>] __lock_acquire+0x2da/0xab0 [<c024731a>] lock_acquire+0x7a/0xa0 [<c044c300>] _spin_lock_irqsave+0x40/0x60 [<cca748fe>] snd_opl3_note_off+0x1e/0xe0 [snd_opl3_synth] [<cca751f0>] snd_opl3_timer_func+0xa0/0xc0 [snd_opl3_synth] [<c022ac46>] run_timer_softirq+0x166/0x1e0 [<c02269e8>] __do_softirq+0x78/0x110 [<c0226ac6>] do_softirq+0x46/0x50 [<c0226e26>] irq_exit+0x36/0x40 [<c0204bd2>] do_IRQ+0x42/0xb0 [<c020328e>] common_interrupt+0x2e/0x40 [<c021092f>] apm_cpu_idle+0x10f/0x290 [<c0201b11>] cpu_idle+0x21/0x40 [<c04443cd>] rest_init+0x4d/0x60 [<c055c835>] start_kernel+0x235/0x280 [<c055c066>] i386_start_kernel+0x66/0x70
other info that might help us debug this:
2 locks held by swapper/0: #0: (&opl3->tlist){+.-...}, at: [<c022abd0>] run_timer_softirq+0xf0/0x1e0 #1: (&opl3->sys_timer_lock){..-...}, at: [<cca75169>] snd_opl3_timer_func+0x19/0xc0 [snd_opl3_synth]
stack backtrace: Pid: 0, comm: swapper Not tainted 2.6.32-rc3 #87 Call Trace: [<c0245188>] print_circular_bug+0xc8/0xd0 [<c02467e6>] validate_chain+0x1036/0x1040 [<c0247f14>] ? check_usage_forwards+0x54/0xd0 [<c0246aca>] __lock_acquire+0x2da/0xab0 [<c024731a>] lock_acquire+0x7a/0xa0 [<cca748fe>] ? snd_opl3_note_off+0x1e/0xe0 [snd_opl3_synth] [<c044c300>] _spin_lock_irqsave+0x40/0x60 [<cca748fe>] ? snd_opl3_note_off+0x1e/0xe0 [snd_opl3_synth] [<cca748fe>] snd_opl3_note_off+0x1e/0xe0 [snd_opl3_synth] [<c044c307>] ? _spin_lock_irqsave+0x47/0x60 [<cca751f0>] snd_opl3_timer_func+0xa0/0xc0 [snd_opl3_synth] [<c022ac46>] run_timer_softirq+0x166/0x1e0 [<c022abd0>] ? run_timer_softirq+0xf0/0x1e0 [<cca75150>] ? snd_opl3_timer_func+0x0/0xc0 [snd_opl3_synth] [<c02269e8>] __do_softirq+0x78/0x110 [<c044c0fd>] ? _spin_unlock+0x1d/0x20 [<c025915f>] ? handle_level_irq+0xaf/0xe0 [<c0226ac6>] do_softirq+0x46/0x50 [<c0226e26>] irq_exit+0x36/0x40 [<c0204bd2>] do_IRQ+0x42/0xb0 [<c024463c>] ? trace_hardirqs_on_caller+0x12c/0x180 [<c020328e>] common_interrupt+0x2e/0x40 [<c0208d88>] ? default_idle+0x38/0x50 [<c021092f>] apm_cpu_idle+0x10f/0x290 [<c0201b11>] cpu_idle+0x21/0x40 [<c04443cd>] rest_init+0x4d/0x60 [<c055c835>] start_kernel+0x235/0x280 [<c055c210>] ? unknown_bootoption+0x0/0x210 [<c055c066>] i386_start_kernel+0x66/0x70
Oh, it looks really like a ABBA deadlock.
Signed-off-by: Krzysztof Helt krzysztof.h1@wp.pl
I am not sure if this is the only possible way to fix my the problem. My idea is to change locking in the snd_opl3_timer_func() so the opl3 voices' part is locked with voice_lock and the timer's part is locked with sys_timer_lock.
Your fix looks correct. I applied it with minor fixes (spaces, keep changes minimal).
thanks,
Takashi