[alsa-devel] [PATCH v3] ALSA: firewire-lib: Fix stall of process context at packet error
At Linux v3.5, packet processing can be done in process context of ALSA PCM application as well as software IRQ context for OHCI 1394. Below is an example of the callgraph (some calls are omitted).
ioctl(2) with e.g. HWSYNC (sound/core/pcm_native.c) ->snd_pcm_common_ioctl1() ->snd_pcm_hwsync() ->snd_pcm_stream_lock_irq (sound/core/pcm_lib.c) ->snd_pcm_update_hw_ptr() ->snd_pcm_udpate_hw_ptr0() ->struct snd_pcm_ops.pointer() (sound/firewire/*) = Each handler on drivers in ALSA firewire stack (sound/firewire/amdtp-stream.c) ->amdtp_stream_pcm_pointer() (drivers/firewire/core-iso.c) ->fw_iso_context_flush_completions() ->struct fw_card_driver.flush_iso_completion() (drivers/firewire/ohci.c) = flush_iso_completions() ->struct fw_iso_context.callback.sc (sound/firewire/amdtp-stream.c) = in_stream_callback() or out_stream_callback() ->... ->snd_pcm_stream_unlock_irq
When packet queueing error occurs or detecting invalid packets in 'in_stream_callback()' or 'out_stream_callback()', 'snd_pcm_stop_xrun()' is called on local CPU with disabled IRQ.
(sound/firewire/amdtp-stream.c) in_stream_callback() or out_stream_callback() ->amdtp_stream_pcm_abort() ->snd_pcm_stop_xrun() ->snd_pcm_stream_lock_irqsave() ->snd_pcm_stop() ->snd_pcm_stream_unlock_irqrestore()
The process is stalled on the CPU due to attempt to acquire recursive lock.
[ 562.630853] INFO: rcu_sched detected stalls on CPUs/tasks: [ 562.630861] 2-...: (1 GPs behind) idle=37d/140000000000000/0 softirq=38323/38323 fqs=7140 [ 562.630862] (detected by 3, t=15002 jiffies, g=21036, c=21035, q=5933) [ 562.630866] Task dump for CPU 2: [ 562.630867] alsa-source-OXF R running task 0 6619 1 0x00000008 [ 562.630870] Call Trace: [ 562.630876] ? vt_console_print+0x79/0x3e0 [ 562.630880] ? msg_print_text+0x9d/0x100 [ 562.630883] ? up+0x32/0x50 [ 562.630885] ? irq_work_queue+0x8d/0xa0 [ 562.630886] ? console_unlock+0x2b6/0x4b0 [ 562.630888] ? vprintk_emit+0x312/0x4a0 [ 562.630892] ? dev_vprintk_emit+0xbf/0x230 [ 562.630895] ? do_sys_poll+0x37a/0x550 [ 562.630897] ? dev_printk_emit+0x4e/0x70 [ 562.630900] ? __dev_printk+0x3c/0x80 [ 562.630903] ? _raw_spin_lock+0x20/0x30 [ 562.630909] ? snd_pcm_stream_lock+0x31/0x50 [snd_pcm] [ 562.630914] ? _snd_pcm_stream_lock_irqsave+0x2e/0x40 [snd_pcm] [ 562.630918] ? snd_pcm_stop_xrun+0x16/0x70 [snd_pcm] [ 562.630922] ? in_stream_callback+0x3e6/0x450 [snd_firewire_lib] [ 562.630925] ? handle_ir_packet_per_buffer+0x8e/0x1a0 [firewire_ohci] [ 562.630928] ? ohci_flush_iso_completions+0xa3/0x130 [firewire_ohci] [ 562.630932] ? fw_iso_context_flush_completions+0x15/0x20 [firewire_core] [ 562.630935] ? amdtp_stream_pcm_pointer+0x2d/0x40 [snd_firewire_lib] [ 562.630938] ? pcm_capture_pointer+0x19/0x20 [snd_oxfw] [ 562.630943] ? snd_pcm_update_hw_ptr0+0x47/0x3d0 [snd_pcm] [ 562.630945] ? poll_select_copy_remaining+0x150/0x150 [ 562.630947] ? poll_select_copy_remaining+0x150/0x150 [ 562.630952] ? snd_pcm_update_hw_ptr+0x10/0x20 [snd_pcm] [ 562.630956] ? snd_pcm_hwsync+0x45/0xb0 [snd_pcm] [ 562.630960] ? snd_pcm_common_ioctl1+0x1ff/0xc90 [snd_pcm] [ 562.630962] ? futex_wake+0x90/0x170 [ 562.630966] ? snd_pcm_capture_ioctl1+0x136/0x260 [snd_pcm] [ 562.630970] ? snd_pcm_capture_ioctl+0x27/0x40 [snd_pcm] [ 562.630972] ? do_vfs_ioctl+0xa3/0x610 [ 562.630974] ? vfs_read+0x11b/0x130 [ 562.630976] ? SyS_ioctl+0x79/0x90 [ 562.630978] ? entry_SYSCALL_64_fastpath+0x1e/0xad
This commit fixes the above bug. This assumes two cases: 1. Any error is detected in software IRQ context of OHCI 1394 context. In this case, PCM substream should be aborted in packet handler. On the other hand, it should not be done in any process context. TO distinguish these two context, use 'in_interrupt()' macro. 2. Any error is detect in process context of ALSA PCM application. In this case, PCM substream should not be aborted in packet handler because PCM substream lock is acquired. The task to abort PCM substream should be done in ALSA PCM core. For this purpose, SNDRV_PCM_POS_XRUN is returned at 'struct snd_pcm_ops.pointer()'.
Suggested-by: Clemens Ladisch clemens@ladisch.de Fixes: e9148dddc3c7("ALSA: firewire-lib: flush completed packets when reading PCM position") Cc: stable@vger.kernel.org # 4.9+ Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp --- sound/firewire/amdtp-stream.c | 8 ++++++-- sound/firewire/amdtp-stream.h | 2 +- 2 files changed, 7 insertions(+), 3 deletions(-)
diff --git a/sound/firewire/amdtp-stream.c b/sound/firewire/amdtp-stream.c index 9678bc75dc5b..3fc581a5ad62 100644 --- a/sound/firewire/amdtp-stream.c +++ b/sound/firewire/amdtp-stream.c @@ -701,7 +701,9 @@ static void out_stream_callback(struct fw_iso_context *context, u32 tstamp, cycle = increment_cycle_count(cycle, 1); if (s->handle_packet(s, 0, cycle, i) < 0) { s->packet_index = -1; - amdtp_stream_pcm_abort(s); + if (in_interrupt()) + amdtp_stream_pcm_abort(s); + WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN); return; } } @@ -753,7 +755,9 @@ static void in_stream_callback(struct fw_iso_context *context, u32 tstamp, /* Queueing error or detecting invalid payload. */ if (i < packets) { s->packet_index = -1; - amdtp_stream_pcm_abort(s); + if (in_interrupt()) + amdtp_stream_pcm_abort(s); + WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN); return; }
diff --git a/sound/firewire/amdtp-stream.h b/sound/firewire/amdtp-stream.h index 6d613f2eb612..ed6eafd10992 100644 --- a/sound/firewire/amdtp-stream.h +++ b/sound/firewire/amdtp-stream.h @@ -135,7 +135,7 @@ struct amdtp_stream { /* For a PCM substream processing. */ struct snd_pcm_substream *pcm; struct tasklet_struct period_tasklet; - unsigned int pcm_buffer_pointer; + snd_pcm_uframes_t pcm_buffer_pointer; unsigned int pcm_period_pointer;
/* To wait for first packet. */
On Sun, 11 Jun 2017 09:08:21 +0200, Takashi Sakamoto wrote:
At Linux v3.5, packet processing can be done in process context of ALSA PCM application as well as software IRQ context for OHCI 1394. Below is an example of the callgraph (some calls are omitted).
ioctl(2) with e.g. HWSYNC (sound/core/pcm_native.c) ->snd_pcm_common_ioctl1() ->snd_pcm_hwsync() ->snd_pcm_stream_lock_irq (sound/core/pcm_lib.c) ->snd_pcm_update_hw_ptr() ->snd_pcm_udpate_hw_ptr0() ->struct snd_pcm_ops.pointer() (sound/firewire/*) = Each handler on drivers in ALSA firewire stack (sound/firewire/amdtp-stream.c) ->amdtp_stream_pcm_pointer() (drivers/firewire/core-iso.c) ->fw_iso_context_flush_completions() ->struct fw_card_driver.flush_iso_completion() (drivers/firewire/ohci.c) = flush_iso_completions() ->struct fw_iso_context.callback.sc (sound/firewire/amdtp-stream.c) = in_stream_callback() or out_stream_callback() ->... ->snd_pcm_stream_unlock_irq
When packet queueing error occurs or detecting invalid packets in 'in_stream_callback()' or 'out_stream_callback()', 'snd_pcm_stop_xrun()' is called on local CPU with disabled IRQ.
(sound/firewire/amdtp-stream.c) in_stream_callback() or out_stream_callback() ->amdtp_stream_pcm_abort() ->snd_pcm_stop_xrun() ->snd_pcm_stream_lock_irqsave() ->snd_pcm_stop() ->snd_pcm_stream_unlock_irqrestore()
The process is stalled on the CPU due to attempt to acquire recursive lock.
[ 562.630853] INFO: rcu_sched detected stalls on CPUs/tasks: [ 562.630861] 2-...: (1 GPs behind) idle=37d/140000000000000/0 softirq=38323/38323 fqs=7140 [ 562.630862] (detected by 3, t=15002 jiffies, g=21036, c=21035, q=5933) [ 562.630866] Task dump for CPU 2: [ 562.630867] alsa-source-OXF R running task 0 6619 1 0x00000008 [ 562.630870] Call Trace: [ 562.630876] ? vt_console_print+0x79/0x3e0 [ 562.630880] ? msg_print_text+0x9d/0x100 [ 562.630883] ? up+0x32/0x50 [ 562.630885] ? irq_work_queue+0x8d/0xa0 [ 562.630886] ? console_unlock+0x2b6/0x4b0 [ 562.630888] ? vprintk_emit+0x312/0x4a0 [ 562.630892] ? dev_vprintk_emit+0xbf/0x230 [ 562.630895] ? do_sys_poll+0x37a/0x550 [ 562.630897] ? dev_printk_emit+0x4e/0x70 [ 562.630900] ? __dev_printk+0x3c/0x80 [ 562.630903] ? _raw_spin_lock+0x20/0x30 [ 562.630909] ? snd_pcm_stream_lock+0x31/0x50 [snd_pcm] [ 562.630914] ? _snd_pcm_stream_lock_irqsave+0x2e/0x40 [snd_pcm] [ 562.630918] ? snd_pcm_stop_xrun+0x16/0x70 [snd_pcm] [ 562.630922] ? in_stream_callback+0x3e6/0x450 [snd_firewire_lib] [ 562.630925] ? handle_ir_packet_per_buffer+0x8e/0x1a0 [firewire_ohci] [ 562.630928] ? ohci_flush_iso_completions+0xa3/0x130 [firewire_ohci] [ 562.630932] ? fw_iso_context_flush_completions+0x15/0x20 [firewire_core] [ 562.630935] ? amdtp_stream_pcm_pointer+0x2d/0x40 [snd_firewire_lib] [ 562.630938] ? pcm_capture_pointer+0x19/0x20 [snd_oxfw] [ 562.630943] ? snd_pcm_update_hw_ptr0+0x47/0x3d0 [snd_pcm] [ 562.630945] ? poll_select_copy_remaining+0x150/0x150 [ 562.630947] ? poll_select_copy_remaining+0x150/0x150 [ 562.630952] ? snd_pcm_update_hw_ptr+0x10/0x20 [snd_pcm] [ 562.630956] ? snd_pcm_hwsync+0x45/0xb0 [snd_pcm] [ 562.630960] ? snd_pcm_common_ioctl1+0x1ff/0xc90 [snd_pcm] [ 562.630962] ? futex_wake+0x90/0x170 [ 562.630966] ? snd_pcm_capture_ioctl1+0x136/0x260 [snd_pcm] [ 562.630970] ? snd_pcm_capture_ioctl+0x27/0x40 [snd_pcm] [ 562.630972] ? do_vfs_ioctl+0xa3/0x610 [ 562.630974] ? vfs_read+0x11b/0x130 [ 562.630976] ? SyS_ioctl+0x79/0x90 [ 562.630978] ? entry_SYSCALL_64_fastpath+0x1e/0xad
This commit fixes the above bug. This assumes two cases:
- Any error is detected in software IRQ context of OHCI 1394 context.
In this case, PCM substream should be aborted in packet handler. On the other hand, it should not be done in any process context. TO distinguish these two context, use 'in_interrupt()' macro. 2. Any error is detect in process context of ALSA PCM application. In this case, PCM substream should not be aborted in packet handler because PCM substream lock is acquired. The task to abort PCM substream should be done in ALSA PCM core. For this purpose, SNDRV_PCM_POS_XRUN is returned at 'struct snd_pcm_ops.pointer()'.
Suggested-by: Clemens Ladisch clemens@ladisch.de Fixes: e9148dddc3c7("ALSA: firewire-lib: flush completed packets when reading PCM position") Cc: stable@vger.kernel.org # 4.9+ Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp
sound/firewire/amdtp-stream.c | 8 ++++++-- sound/firewire/amdtp-stream.h | 2 +- 2 files changed, 7 insertions(+), 3 deletions(-)
diff --git a/sound/firewire/amdtp-stream.c b/sound/firewire/amdtp-stream.c index 9678bc75dc5b..3fc581a5ad62 100644 --- a/sound/firewire/amdtp-stream.c +++ b/sound/firewire/amdtp-stream.c @@ -701,7 +701,9 @@ static void out_stream_callback(struct fw_iso_context *context, u32 tstamp, cycle = increment_cycle_count(cycle, 1); if (s->handle_packet(s, 0, cycle, i) < 0) { s->packet_index = -1;
amdtp_stream_pcm_abort(s);
if (in_interrupt())amdtp_stream_pcm_abort(s);WRITE_ONCE(s->pcm_buffer_pointer, SNDRV_PCM_POS_XRUN); return;
[Dropped stable from Cc]
Hrm, it's only about the deadlock by snd_pcm_stop_xrun(), no? You can achieve it without locking via snd_pcm_stop(s, XRUN), too, e.g. a patch like below.
Of course, if the stop action itself causes a problem, it's not appropriate, though.
thanks,
Takashi
--- a/sound/firewire/amdtp-stream.c +++ b/sound/firewire/amdtp-stream.c @@ -60,6 +60,7 @@ #define OUT_PACKET_HEADER_SIZE 0
static void pcm_period_tasklet(unsigned long data); +static void stream_abort(struct amdtp_stream *s, bool locked);
/** * amdtp_stream_init - initialize an AMDTP stream structure @@ -701,7 +702,7 @@ static void out_stream_callback(struct fw_iso_context *context, u32 tstamp, cycle = increment_cycle_count(cycle, 1); if (s->handle_packet(s, 0, cycle, i) < 0) { s->packet_index = -1; - amdtp_stream_pcm_abort(s); + stream_abort(s, !in_interrupt()); return; } } @@ -753,7 +754,7 @@ static void in_stream_callback(struct fw_iso_context *context, u32 tstamp, /* Queueing error or detecting invalid payload. */ if (i < packets) { s->packet_index = -1; - amdtp_stream_pcm_abort(s); + stream_abort(s, !in_interrupt()); return; }
@@ -1007,6 +1008,19 @@ void amdtp_stream_stop(struct amdtp_stream *s) } EXPORT_SYMBOL(amdtp_stream_stop);
+static void stream_abort(struct amdtp_stream *s, bool locked) +{ + struct snd_pcm_substream *pcm; + + pcm = ACCESS_ONCE(s->pcm); + if (pcm) { + if (locked) + snd_pcm_stop(pcm, SNDRV_PCM_STATE_XRUN); + else + snd_pcm_stop_xrun(pcm); + } +} + /** * amdtp_stream_pcm_abort - abort the running PCM device * @s: the AMDTP stream about to be stopped @@ -1016,10 +1030,6 @@ EXPORT_SYMBOL(amdtp_stream_stop); */ void amdtp_stream_pcm_abort(struct amdtp_stream *s) { - struct snd_pcm_substream *pcm; - - pcm = ACCESS_ONCE(s->pcm); - if (pcm) - snd_pcm_stop_xrun(pcm); + stream_abort(s, false); } EXPORT_SYMBOL(amdtp_stream_pcm_abort);
On Sun, 11 Jun 2017 09:08:21 +0200, Takashi Sakamoto wrote:
At Linux v3.5, packet processing can be done in process context of ALSA PCM application as well as software IRQ context for OHCI 1394. Below is an example of the callgraph (some calls are omitted).
ioctl(2) with e.g. HWSYNC (sound/core/pcm_native.c) ->snd_pcm_common_ioctl1() ->snd_pcm_hwsync() ->snd_pcm_stream_lock_irq (sound/core/pcm_lib.c) ->snd_pcm_update_hw_ptr() ->snd_pcm_udpate_hw_ptr0() ->struct snd_pcm_ops.pointer() (sound/firewire/*) = Each handler on drivers in ALSA firewire stack (sound/firewire/amdtp-stream.c) ->amdtp_stream_pcm_pointer() (drivers/firewire/core-iso.c) ->fw_iso_context_flush_completions() ->struct fw_card_driver.flush_iso_completion() (drivers/firewire/ohci.c) = flush_iso_completions() ->struct fw_iso_context.callback.sc (sound/firewire/amdtp-stream.c) = in_stream_callback() or out_stream_callback() ->... ->snd_pcm_stream_unlock_irq
When packet queueing error occurs or detecting invalid packets in 'in_stream_callback()' or 'out_stream_callback()', 'snd_pcm_stop_xrun()' is called on local CPU with disabled IRQ.
(sound/firewire/amdtp-stream.c) in_stream_callback() or out_stream_callback() ->amdtp_stream_pcm_abort() ->snd_pcm_stop_xrun() ->snd_pcm_stream_lock_irqsave() ->snd_pcm_stop() ->snd_pcm_stream_unlock_irqrestore()
The process is stalled on the CPU due to attempt to acquire recursive lock.
[ 562.630853] INFO: rcu_sched detected stalls on CPUs/tasks: [ 562.630861] 2-...: (1 GPs behind) idle=37d/140000000000000/0 softirq=38323/38323 fqs=7140 [ 562.630862] (detected by 3, t=15002 jiffies, g=21036, c=21035, q=5933) [ 562.630866] Task dump for CPU 2: [ 562.630867] alsa-source-OXF R running task 0 6619 1 0x00000008 [ 562.630870] Call Trace: [ 562.630876] ? vt_console_print+0x79/0x3e0 [ 562.630880] ? msg_print_text+0x9d/0x100 [ 562.630883] ? up+0x32/0x50 [ 562.630885] ? irq_work_queue+0x8d/0xa0 [ 562.630886] ? console_unlock+0x2b6/0x4b0 [ 562.630888] ? vprintk_emit+0x312/0x4a0 [ 562.630892] ? dev_vprintk_emit+0xbf/0x230 [ 562.630895] ? do_sys_poll+0x37a/0x550 [ 562.630897] ? dev_printk_emit+0x4e/0x70 [ 562.630900] ? __dev_printk+0x3c/0x80 [ 562.630903] ? _raw_spin_lock+0x20/0x30 [ 562.630909] ? snd_pcm_stream_lock+0x31/0x50 [snd_pcm] [ 562.630914] ? _snd_pcm_stream_lock_irqsave+0x2e/0x40 [snd_pcm] [ 562.630918] ? snd_pcm_stop_xrun+0x16/0x70 [snd_pcm] [ 562.630922] ? in_stream_callback+0x3e6/0x450 [snd_firewire_lib] [ 562.630925] ? handle_ir_packet_per_buffer+0x8e/0x1a0 [firewire_ohci] [ 562.630928] ? ohci_flush_iso_completions+0xa3/0x130 [firewire_ohci] [ 562.630932] ? fw_iso_context_flush_completions+0x15/0x20 [firewire_core] [ 562.630935] ? amdtp_stream_pcm_pointer+0x2d/0x40 [snd_firewire_lib] [ 562.630938] ? pcm_capture_pointer+0x19/0x20 [snd_oxfw] [ 562.630943] ? snd_pcm_update_hw_ptr0+0x47/0x3d0 [snd_pcm] [ 562.630945] ? poll_select_copy_remaining+0x150/0x150 [ 562.630947] ? poll_select_copy_remaining+0x150/0x150 [ 562.630952] ? snd_pcm_update_hw_ptr+0x10/0x20 [snd_pcm] [ 562.630956] ? snd_pcm_hwsync+0x45/0xb0 [snd_pcm] [ 562.630960] ? snd_pcm_common_ioctl1+0x1ff/0xc90 [snd_pcm] [ 562.630962] ? futex_wake+0x90/0x170 [ 562.630966] ? snd_pcm_capture_ioctl1+0x136/0x260 [snd_pcm] [ 562.630970] ? snd_pcm_capture_ioctl+0x27/0x40 [snd_pcm] [ 562.630972] ? do_vfs_ioctl+0xa3/0x610 [ 562.630974] ? vfs_read+0x11b/0x130 [ 562.630976] ? SyS_ioctl+0x79/0x90 [ 562.630978] ? entry_SYSCALL_64_fastpath+0x1e/0xad
This commit fixes the above bug. This assumes two cases:
- Any error is detected in software IRQ context of OHCI 1394 context.
In this case, PCM substream should be aborted in packet handler. On the other hand, it should not be done in any process context. TO distinguish these two context, use 'in_interrupt()' macro. 2. Any error is detect in process context of ALSA PCM application. In this case, PCM substream should not be aborted in packet handler because PCM substream lock is acquired. The task to abort PCM substream should be done in ALSA PCM core. For this purpose, SNDRV_PCM_POS_XRUN is returned at 'struct snd_pcm_ops.pointer()'.
Suggested-by: Clemens Ladisch clemens@ladisch.de Fixes: e9148dddc3c7("ALSA: firewire-lib: flush completed packets when reading PCM position") Cc: stable@vger.kernel.org # 4.9+ Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp
I merged this patch as is, since it apparently fixes the issue at least.
thanks,
Takashi
participants (2)
-
Takashi Iwai -
Takashi Sakamoto