On Wed, 04 Sep 2024 11:07:30 +0200, Chancel Liu wrote:
Hi Takashi,
Thanks for your reply and suggestions. Finally we have found the root cause. Seems it's related to both drivers and alsa-lib.
When two dmix clients run in parallel we get two direct dmix instances. 1st dmix instance: snd_pcm_dmix_open() snd_pcm_direct_initialize_slave() save_slave_setting() Since the driver we are using has SND_PCM_INFO_RESUME flag, dmix->spcm->info has this flag. Then this flag is cleared in dmix->shmptr->s.info. 2nd dmix instance: snd_pcm_dmix_open() snd_pcm_direct_open_secondary_client() copy_slave_setting() 2nd dmix->spcm->info is copied from dmix->shmptr->s.info so it doesn' has this flag.
If 1st dmix instance resumes firstly it should implement recovery of slave pcm in snd_pcm_direct_slave_recover(). Because 1st dmix->spcm->info has SND_PCM_INFO_RESUME，snd_pcm_resume(direct->spcm) can be called correctly to resume slave pcm.
... and immediately stop the stream, then prepare and restart as a usual restart.
However if 2nd dmix instance resumes firstly, snd_pcm_resume(direct->spcm) will not be called because it's spcm->info doesn't has SND_PCM_INFO_RESUME flag. The 1st dmix instance assumes someone else already did recovery so snd_pcm_resume(direct->spcm) won't be called neither. In result the slave pcm fails to resume.
Something wrong happening here, then.
In dmix, there is no hardware resume at all, but it's always a restart of the stream. The call of snd_pcm_resume() is only temporarily for inconsistencies that can be a problem on some drivers (IIRC dmaengine stuff). That said, dmix does a kind of fake resume, stops and restarts the stream cleanly on the first instance. On the second instance, it's already recovered, hence it bails out.
If poll() hangs on the second instance, there can be some other problem. Maybe the resume -> stop -> restart sequence doesn't work with your driver well?
SND_PCM_INFO_RESUME flag has impact on the flow of dmix resume. In my opinion the first resumed dmix instance should make sure slave pcm can be recovered properly no matter it's the first opened instance or secondary opened instance
.
The snd_pcm_resume() gets called no matter which instance, just the first one who tries to recover the suspended state. (And it's called internally at updating the various state, not necessarily an explicit recovery call.)
Do you know why the secondary opened instance clear the SND_PCM_INFO_RESUME flag? Can we do the following modification?
diff --git a/src/pcm/pcm_direct.c b/src/pcm/pcm_direct.c @@ -1183,8 +1226,6 @@ static void save_slave_setting(snd_pcm_direct_t *dmix, snd_pcm_t *spcm) COPY_SLAVE(buffer_time); COPY_SLAVE(sample_bits); COPY_SLAVE(frame_bits);
dmix->shmptr->s.info &= ~SND_PCM_INFO_RESUME;
I don't think so. The clearance of the RESUME flag here is correct. dmix doesn't support the hardware resume feature. It does its own. (And this flag is merely a info for apps, which isn't really evaluated except for the code in dmix workaround there.)
Takashi
Regards, Chancel Liu
[ it seems that my previous post didn't go out properly, so resent; if you've seen already the same, please disregard ]
On Tue, 27 Aug 2024 09:06:39 +0200, Chancel Liu wrote:
Hi Takashi Iwai, Jaroslav Kysela
We found an issue on dmix in alsa-lib when do suspend and resume. It can be easily reproduced by following steps:
- Run two dmix clients in parallel. (Only one client doesnʼt has such
issue)
~# aplay xxx1.wav &
~# aplay xxx2.wav &
Here I attach the asound.conf we're using.
~# cat /etc/asound.conf
defaults.pcm.rate_converter "linear"
pcm.dmix_44100{
type dmix ipc_key 5678293 ipc_key_add_uid yes slave{ pcm "hw:0,0" period_time 40000 format S16_LE rate 44100 }}
pcm.asymed{
type asym playback.pcm "dmix_44100" capture.pcm "dsnoop_44100"}
pcm.!default{
type plug route_policy "average" slave.pcm "asymed"}
- Let linux enter into suspend and then resume(Repeat this step if
not reproduced)
- After resume, aplay will get stuck in snd_pcm_wait(). The GDB shows:
(gdb) bt
#0 0x0000fffff7da9264 in __GI___poll (fds=fds@entry=0xfffffffff480, nfds= nfds@entry=1, timeout=timeout@entry=240)
at /usr/src/debug/glibc/2.39+git/sysdeps/unix/sysv/linux/poll.c:41#1 0x0000fffff7edf468 in poll (__timeout=240, __nfds=1, __fds=0xfffffffff480)
#2 snd1_pcm_wait_nocheck (pcm=pcm@entry=0xaaaaaaad2cb0,
timeout=240,
timeout@entry=-10001) at pcm.c:2993
#3 0x0000fffff7ee54a0 in snd1_pcm_write_areas (pcm=pcm@entry=0xaaaaaaad2cb0, areas=areas@entry=0xfffffffff560, offset=<optimized out>, offset@entry=0, size =<optimized out>,
size@entry=1768, func=func@entry=0xfffff7ef5190<snd_pcm_plugin_write_areas>) at pcm.c:7699
#4 0x0000fffff7ef5020 in snd_pcm_plugin_writei (pcm=0xaaaaaaad2cb0, buffer= <optimized out>, size=1768) at pcm_plugin.c:354
It seems that sometimes after suspend and resume there's no available space for data written into buffer. Then aplay keeps stuck in snd_pcm_wait(). I checked the hw_ptr of dmix and found that hw_ptr is
always 0 after resume.
I don't have a solution now so I turn to you for help. The version of alsa-lib is v1.2.11. Could you please help check it?
I tried your setup but I couldn't reproduce the issue locally with my laptop and HD-audio device. Possibly depending on the kernel driver?
In the case of dmix, it's a poll() against the PCM slave timer. So it doesn't take care of suspend/resume state unlike the real PCM device. OTOH, the timer device should send notification events at suspend/resume, and it should trigger the poll wakeup, too.
Does poll() return after the suspend/resume once but falls into a loop due to revents being unset? Or it's stuck and never returns at suspend/resume?
thanks,
Takashi