Hi Takashi,
We found another issue with non atomic ops.
This occurs when we have snd_pcm_do_prepare being invoked which acquires the lock snd_pcm_link_rwsem twice, first in snd_pcm_common_ioctl1, and then again in dpcm_dapm_stream_event. Normally this is fine, but if we have another stream get closed in between two read accesses we get stuck.
First thread:
[ 419.657259] AudioOut_2 D ffff8800704a3a60 0 3691 1 0x20020004 [ 419.665946] ffff8800704a3a60 00000000704a3a60 ffff88006d1fddd0 ffffffff8252ffd8 [ 419.674678] ffff8800704a4000 ffffffff8252ffc0 ffffffffffffffff ffffffff8252ffd8 [ 419.683513] 0000000000000000 ffff8800704a3a80 ffffffff81ca2987 ffffffffffffffff [ 419.683574] Call Trace: [ 419.692290] [<ffffffff81ca2987>] schedule+0x37/0x90 [ 419.692306] [<ffffffff81ca6f6d>] rwsem_down_read_failed+0xdd/0x130 [ 419.701050] [<ffffffff814df454>] call_rwsem_down_read_failed+0x14/0x30 [ 419.709758] [<ffffffff810e793b>] ? down_read_nested+0x5b/0x70 [ 419.709805] [<ffffffff81adfda9>] ? snd_pcm_stream_lock+0xa9/0x120 [ 419.723012] [<ffffffff81adfda9>] snd_pcm_stream_lock+0xa9/0x120
< ============ 2nd Read lock is acquired here, without releasing previous one.
[ 419.723026] [<ffffffff81adfe4c>] snd_pcm_stream_lock_irq+0x2c/0x30 [ 419.731801] [<ffffffff81b01f5e>] ? dpcm_dapm_stream_event+0xbe/0xd0 [ 419.740565] [<ffffffff81b0079c>] dpcm_set_fe_update_state+0x3c/0xb0 [ 419.749252] [<ffffffff81b04514>] dpcm_fe_dai_prepare+0xc4/0x150 [ 419.749301] [<ffffffff81adf7eb>] snd_pcm_do_prepare+0x1b/0x30 [ 419.758083] [<ffffffff81adf3bf>] snd_pcm_action_single+0x2f/0x70 [ 419.766897] [<ffffffff81ca6827>] ? down_read+0x47/0x60 [ 419.766962] [<ffffffff81adf577>] ? snd_pcm_action_nonatomic+0x27/0x80 [ 419.775565] [<ffffffff81adf5c6>] snd_pcm_action_nonatomic+0x76/0x80
< ============ 1st Read lock is acquired here
[ 419.784419] [<ffffffff81ae2c62>] snd_pcm_common_ioctl1+0x802/0xd30 [ 419.784495] [<ffffffff81ae332b>] snd_pcm_playback_ioctl1+0x19b/0x280 [ 419.793106] [<ffffffff81230915>] ? __fget+0x5/0x210 [ 419.801943] [<ffffffff81ae3dca>] snd_pcm_ioctl_compat+0x41a/0x770 [ 419.801959] [<ffffffff81230915>] ? __fget+0x5/0x210 [ 419.810793] [<ffffffff8126d670>] compat_SyS_ioctl+0xd0/0x13b0 [ 419.819491] [<ffffffff81ca7bc4>] ? do_nanosleep+0x94/0x190 [ 419.819535] [<ffffffff810eb696>] ? trace_hardirqs_on_caller+0x16/0x210 [ 419.828306] [<ffffffff814df4f8>] ? trace_hardirqs_on_thunk+0x17/0x19 [ 419.837104] [<ffffffff81cabf0e>] sysenter_dispatch+0x7/0x1f
Second thread: [ 419.543062] mediaserver D ffff880170957cd8 0 3690 1 0x20020004 [ 419.543191] ffff880170957cd8 0000000070957cd8 ffff88006de49ad0 ffffffff8252ffd8 [ 419.551920] ffff880170958000 ffff88006de49ad0 ffffffff8252ffc0 ffffffff00000001 [ 419.560555] ffffffff8252ffd8 ffff880170957cf8 ffffffff81ca2987 ffffffff8252ffc0 [ 419.569368] Call Trace: [ 419.569392] [<ffffffff81ca2987>] schedule+0x37/0x90 [ 419.578026] [<ffffffff81ca71f5>] rwsem_down_write_failed+0x235/0x450 [ 419.586884] [<ffffffff81ca7005>] ? rwsem_down_write_failed+0x45/0x450 [ 419.586941] [<ffffffff814df483>] call_rwsem_down_write_failed+0x13/0x20 [ 419.595620] [<ffffffff81ca689f>] ? down_write+0x5f/0x80
< ============ Write lock acquired here. If this occurs between two read lock's then it's hangs
[ 419.604428] [<ffffffff81aded58>] ? snd_pcm_unlink+0x28/0x120 [ 419.604465] [<ffffffff81aded58>] snd_pcm_unlink+0x28/0x120 [ 419.613215] [<ffffffff81adee5e>] pcm_release_private+0xe/0x10 [ 419.622058] [<ffffffff81ae0d57>] snd_pcm_release_substream.part.34+0x67/0x90 [ 419.622143] [<ffffffff81ae0e68>] snd_pcm_release+0xb8/0xc0 [ 419.630889] [<ffffffff8120f1aa>] __fput+0xea/0x200 [ 419.639680] [<ffffffff8120f30e>] ____fput+0xe/0x10 [ 419.639744] [<ffffffff810ba43c>] task_work_run+0xcc/0x100 [ 419.648435] [<ffffffff81002e41>] do_notify_resume+0x51/0x70 [ 419.657205] [<ffffffff81ca986c>] int_signal+0x12/0x17
Thanks