[alsa-devel] [PATCH - dmix v5 0/1] pcm: dmix: Add option to allow alignment of slave pointers
On 11/12/18 17:20, Takashi Iwai wrote:
Please provide your own sign-off when you submitting a patch, too.
The subject line should be adjusted better to match with the change; it's about adding an option to allow alignment of hwptr reporting.
About the changes: the term "hw_ptr_filter" might sound confusing. IMO, this is about the (hw)ptr alignment, not about filtering. Also, a too long option string can easily lead to a misspell :)
After all, we do either "no" alignment, "roundup" or "rounddown", or "auto" choice.
thanks,
Takashi
Subject changed. Sign-off added. Option changed to hw_ptr_alignment. Option values changed to no, roundup, rounddown, auto
Best regards
Timo
From: Laxmi Devi Laxmi.Devi@in.bosch.com
These changes are required due to the kernel commit 07b7acb51d283d8469696c906b91f1882696a4d4 ("ASoC: rsnd: update pointer more accurate")
Issue is that snd_pcm_wait() goes back to waiting because the hw_ptr is not period aligned. Therefore snd_pcm_wait() will block for a longer time as required.
With these rcar driver changes the exact position of the dma is returned. During snd_pcm_start they read hw_ptr as reference, and this hw_ptr is now not period aligned, and is a little ahead over the period while it is read. Therefore when the avail is calculated during snd_pcm_wait(), it is missing the avail_min by a few frames.
An additional option hw_ptr_alignment is provided to dmix configuration, to allow the user to configure the slave application and hw pointer alignment at startup
Signed-off-by: Laxmi Devi Laxmi.Devi@in.bosch.com Signed-off-by: Timo Wischer twischer@de.adit-jv.com
diff --git a/src/pcm/pcm_direct.c b/src/pcm/pcm_direct.c index 091fd78..a264197 100644 --- a/src/pcm/pcm_direct.c +++ b/src/pcm/pcm_direct.c @@ -1883,6 +1883,7 @@ int snd_pcm_direct_parse_open_conf(snd_config_t *root, snd_config_t *conf, rec->max_periods = 0; rec->var_periodsize = 0; rec->direct_memory_access = 1; + rec->hw_ptr_alignment = SND_PCM_HW_PTR_ALIGNMENT_AUTO;
/* read defaults */ if (snd_config_search(root, "defaults.pcm.dmix_max_periods", &n) >= 0) { @@ -1924,6 +1925,28 @@ int snd_pcm_direct_parse_open_conf(snd_config_t *root, snd_config_t *conf, rec->ipc_perm = perm; continue; } + if (strcmp(id, "hw_ptr_alignment") == 0) { + const char *str; + err = snd_config_get_string(n, &str); + if (err < 0) { + SNDERR("Invalid type for %s", id); + return -EINVAL; + } + if (strcmp(str, "no") == 0) + rec->hw_ptr_alignment = SND_PCM_HW_PTR_ALIGNMENT_NO; + else if (strcmp(str, "roundup") == 0) + rec->hw_ptr_alignment = SND_PCM_HW_PTR_ALIGNMENT_ROUNDUP; + else if (strcmp(str, "rounddown") == 0) + rec->hw_ptr_alignment = SND_PCM_HW_PTR_ALIGNMENT_ROUNDDOWN; + else if (strcmp(str, "auto") == 0) + rec->hw_ptr_alignment = SND_PCM_HW_PTR_ALIGNMENT_AUTO; + else { + SNDERR("The field hw_ptr_alignment is invalid : %s", str); + return -EINVAL; + } + + continue; + } if (strcmp(id, "ipc_gid") == 0) { char *group; char *endp; diff --git a/src/pcm/pcm_direct.h b/src/pcm/pcm_direct.h index 582f055..a487616 100644 --- a/src/pcm/pcm_direct.h +++ b/src/pcm/pcm_direct.h @@ -49,6 +49,13 @@ typedef void (mix_areas_u8_t)(unsigned int size, volatile signed int *sum, size_t dst_step, size_t src_step, size_t sum_step);
+typedef enum snd_pcm_direct_hw_ptr_alignment { + SND_PCM_HW_PTR_ALIGNMENT_NO = 0, /* use the hw_ptr as is and do no rounding */ + SND_PCM_HW_PTR_ALIGNMENT_ROUNDUP = 1, /* round the slave_appl_ptr up to slave_period */ + SND_PCM_HW_PTR_ALIGNMENT_ROUNDDOWN = 2, /* round slave_hw_ptr and slave_appl_ptr down to slave_period */ + SND_PCM_HW_PTR_ALIGNMENT_AUTO = 3 /* automatic selection */ +} snd_pcm_direct_hw_ptr_alignment_t; + struct slave_params { snd_pcm_format_t format; int rate; @@ -159,6 +166,7 @@ struct snd_pcm_direct { unsigned int *bindings; unsigned int recoveries; /* mirror of executed recoveries on slave */ int direct_memory_access; /* use arch-optimized buffer RW */ + snd_pcm_direct_hw_ptr_alignment_t hw_ptr_alignment; union { struct { int shmid_sum; /* IPC global sum ring buffer memory identification */ @@ -341,6 +349,7 @@ struct snd_pcm_direct_open_conf { int max_periods; int var_periodsize; int direct_memory_access; + snd_pcm_direct_hw_ptr_alignment_t hw_ptr_alignment; snd_config_t *slave; snd_config_t *bindings; }; diff --git a/src/pcm/pcm_dmix.c b/src/pcm/pcm_dmix.c index 881154e..567aad3 100644 --- a/src/pcm/pcm_dmix.c +++ b/src/pcm/pcm_dmix.c @@ -55,6 +55,9 @@ const char *_snd_module_pcm_dmix = ""; #define STATE_RUN_PENDING 1024 #endif
+#define SEC_TO_MS 1000 /* Seconds representing in Milli seconds */ +#define LOW_LATENCY_PERIOD_TIME 10 /* slave_period time for low latency requirements in ms */ + /* * */ @@ -560,14 +563,17 @@ static int snd_pcm_dmix_hwsync(snd_pcm_t *pcm) static void reset_slave_ptr(snd_pcm_t *pcm, snd_pcm_direct_t *dmix) { dmix->slave_appl_ptr = dmix->slave_hw_ptr = *dmix->spcm->hw.ptr; - if (pcm->buffer_size > pcm->period_size * 2) - return; - /* If we have too litte periods, better to align the start position - * to the period boundary so that the interrupt can be handled properly - * at the right time. - */ - dmix->slave_appl_ptr = ((dmix->slave_appl_ptr + dmix->slave_period_size - 1) + + if (dmix->hw_ptr_alignment == SND_PCM_HW_PTR_ALIGNMENT_ROUNDUP || + (dmix->hw_ptr_alignment == SND_PCM_HW_PTR_ALIGNMENT_AUTO && + pcm->buffer_size <= pcm->period_size * 2)) + dmix->slave_appl_ptr = ((dmix->slave_appl_ptr + dmix->slave_period_size - 1) / dmix->slave_period_size) * dmix->slave_period_size; + else if (dmix->hw_ptr_alignment == SND_PCM_HW_PTR_ALIGNMENT_ROUNDDOWN || + (dmix->hw_ptr_alignment == SND_PCM_HW_PTR_ALIGNMENT_AUTO && + (((dmix->slave_period_size * SEC_TO_MS) / pcm->rate) < LOW_LATENCY_PERIOD_TIME))) + dmix->slave_appl_ptr = dmix->slave_hw_ptr = ((dmix->slave_hw_ptr + / dmix->slave_period_size) * dmix->slave_period_size); }
static int snd_pcm_dmix_reset(snd_pcm_t *pcm) @@ -1086,6 +1092,7 @@ int snd_pcm_dmix_open(snd_pcm_t **pcmp, const char *name, dmix->slowptr = opts->slowptr; dmix->max_periods = opts->max_periods; dmix->var_periodsize = opts->var_periodsize; + dmix->hw_ptr_alignment = opts->hw_ptr_alignment; dmix->sync_ptr = snd_pcm_dmix_sync_ptr; dmix->direct_memory_access = opts->direct_memory_access;
@@ -1241,6 +1248,12 @@ pcm.name { ipc_key INT # unique IPC key ipc_key_add_uid BOOL # add current uid to unique IPC key ipc_perm INT # IPC permissions (octal, default 0600) + hw_ptr_alignment STR # Slave application and hw pointer alignment type + # STR can be one of the below strings : + # no + # roundup + # rounddown + # auto (default) slave STR # or slave { # Slave definition @@ -1273,6 +1286,24 @@ added to the value set in <code>ipc_key</code>. This will avoid the confliction of the same IPC key with different users concurrently.
+<code>hw_ptr_alignment</code> specifies slave application and hw pointer alignment +type. By default hw_ptr_alignment is auto. Below are the possible configurations : +- no : minimal latency with minimal frames dropped at startup. But wakeup of + application (return from snd_pcm_wait() or poll()) can take up to 2 * period. +- roundup : It is guaranteed that all frames will be played at startup. But the + latency will increase upto period-1 frames. +- rounddown : It is guaranteed that a wakeup will happen for each period and + frames can be written from application. But on startup upto period-1 frames + will be dropped. +- auto : Selects the best approach depending on the used period and buffer size. + If the application buffer size is < 2 * application period, "roundup" + will be selected to avoid under runs. If the slave_period is < 10ms we + could expect that there are low latency requirements. Therefore "rounddown" + will be chosen to avoid long wakeup times. Such wakeup delay could otherwise + end up with Xruns in case of a dependency to another sound device + (e.g. forwarding of microphone to speaker). Else "no" + will be chosen. + Note that the dmix plugin itself supports only a single configuration. That is, it supports only the fixed rate (default 48000), format (\c S16), channels (2), and period_time (125000).
On Tue, 13 Nov 2018 08:43:00 +0100, twischer@de.adit-jv.com wrote:
From: Laxmi Devi Laxmi.Devi@in.bosch.com
These changes are required due to the kernel commit 07b7acb51d283d8469696c906b91f1882696a4d4 ("ASoC: rsnd: update pointer more accurate")
Issue is that snd_pcm_wait() goes back to waiting because the hw_ptr is not period aligned. Therefore snd_pcm_wait() will block for a longer time as required.
With these rcar driver changes the exact position of the dma is returned. During snd_pcm_start they read hw_ptr as reference, and this hw_ptr is now not period aligned, and is a little ahead over the period while it is read. Therefore when the avail is calculated during snd_pcm_wait(), it is missing the avail_min by a few frames.
An additional option hw_ptr_alignment is provided to dmix configuration, to allow the user to configure the slave application and hw pointer alignment at startup
Signed-off-by: Laxmi Devi Laxmi.Devi@in.bosch.com Signed-off-by: Timo Wischer twischer@de.adit-jv.com
Thanks, applied now with minor fixes (mostly spaces and line breaks).
Takashi
participants (2)
-
Takashi Iwai
-
twischer@de.adit-jv.com