#include <pulse/xmalloc.h>
#include <pulse/util.h>
#include <pulse/timeval.h>
+#include <pulse/i18n.h>
#include <pulsecore/core-error.h>
#include <pulsecore/core.h>
#include "alsa-util.h"
#include "alsa-source.h"
+/* #define DEBUG_TIMING */
+
#define DEFAULT_DEVICE "default"
#define DEFAULT_TSCHED_BUFFER_USEC (2*PA_USEC_PER_SEC) /* 2s */
#define DEFAULT_TSCHED_WATERMARK_USEC (20*PA_USEC_PER_MSEC) /* 20ms */
-#define TSCHED_MIN_SLEEP_USEC (3*PA_USEC_PER_MSEC) /* 3ms */
-#define TSCHED_MIN_WAKEUP_USEC (3*PA_USEC_PER_MSEC) /* 3ms */
+#define TSCHED_MIN_SLEEP_USEC (10*PA_USEC_PER_MSEC) /* 10ms */
+#define TSCHED_MIN_WAKEUP_USEC (4*PA_USEC_PER_MSEC) /* 4ms */
struct userdata {
pa_core *core;
snd_mixer_elem_t *mixer_elem;
long hw_volume_max, hw_volume_min;
long hw_dB_max, hw_dB_min;
- pa_bool_t hw_dB_supported;
- pa_bool_t mixer_seperate_channels;
+ pa_bool_t hw_dB_supported:1;
+ pa_bool_t mixer_seperate_channels:1;
pa_cvolume hardware_volume;
- size_t frame_size, fragment_size, hwbuf_size, tsched_watermark;
+ size_t frame_size, fragment_size, hwbuf_size, tsched_watermark, hwbuf_unused, min_sleep, min_wakeup;
unsigned nfragments;
char *device_name;
- pa_bool_t use_mmap, use_tsched;
+ pa_bool_t use_mmap:1, use_tsched:1;
pa_rtpoll_item *alsa_rtpoll_item;
snd_mixer_selem_channel_id_t mixer_map[SND_MIXER_SCHN_LAST];
pa_smoother *smoother;
- int64_t frame_index;
-
- snd_pcm_sframes_t hwbuf_unused_frames;
+ uint64_t read_count;
};
static void userdata_free(struct userdata *u);
+static void fix_min_sleep_wakeup(struct userdata *u) {
+ size_t max_use, max_use_2;
+ pa_assert(u);
+
+ max_use = u->hwbuf_size - u->hwbuf_unused;
+ max_use_2 = pa_frame_align(max_use/2, &u->source->sample_spec);
+
+ u->min_sleep = pa_usec_to_bytes(TSCHED_MIN_SLEEP_USEC, &u->source->sample_spec);
+ u->min_sleep = PA_CLAMP(u->min_sleep, u->frame_size, max_use_2);
+
+ u->min_wakeup = pa_usec_to_bytes(TSCHED_MIN_WAKEUP_USEC, &u->source->sample_spec);
+ u->min_wakeup = PA_CLAMP(u->min_wakeup, u->frame_size, max_use_2);
+}
+
static void fix_tsched_watermark(struct userdata *u) {
size_t max_use;
- size_t min_sleep, min_wakeup;
pa_assert(u);
- max_use = u->hwbuf_size - (size_t) u->hwbuf_unused_frames * u->frame_size;
+ max_use = u->hwbuf_size - u->hwbuf_unused;
+
+ if (u->tsched_watermark > max_use - u->min_sleep)
+ u->tsched_watermark = max_use - u->min_sleep;
- min_sleep = pa_usec_to_bytes(TSCHED_MIN_SLEEP_USEC, &u->source->sample_spec);
- min_wakeup = pa_usec_to_bytes(TSCHED_MIN_WAKEUP_USEC, &u->source->sample_spec);
+ if (u->tsched_watermark < u->min_wakeup)
+ u->tsched_watermark = u->min_wakeup;
+}
+
+static void adjust_after_overrun(struct userdata *u) {
+ size_t old_watermark;
+ pa_usec_t old_min_latency, new_min_latency;
+
+ pa_assert(u);
+
+ /* First, just try to increase the watermark */
+ old_watermark = u->tsched_watermark;
+ u->tsched_watermark *= 2;
+ fix_tsched_watermark(u);
+
+ if (old_watermark != u->tsched_watermark) {
+ pa_log_notice("Increasing wakeup watermark to %0.2f ms",
+ (double) pa_bytes_to_usec(u->tsched_watermark, &u->source->sample_spec) / PA_USEC_PER_MSEC);
+ return;
+ }
- if (min_sleep > max_use/2)
- min_sleep = pa_frame_align(max_use/2, &u->source->sample_spec);
- if (min_sleep < u->frame_size)
- min_sleep = u->frame_size;
+ /* Hmm, we cannot increase the watermark any further, hence let's raise the latency */
+ old_min_latency = u->source->thread_info.min_latency;
+ new_min_latency = PA_MIN(old_min_latency * 2, u->source->thread_info.max_latency);
- if (min_wakeup > max_use/2)
- min_wakeup = pa_frame_align(max_use/2, &u->source->sample_spec);
- if (min_wakeup < u->frame_size)
- min_wakeup = u->frame_size;
+ if (old_min_latency != new_min_latency) {
+ pa_log_notice("Increasing minimal latency to %0.2f ms",
+ (double) new_min_latency / PA_USEC_PER_MSEC);
- if (u->tsched_watermark > max_use-min_sleep)
- u->tsched_watermark = max_use-min_sleep;
+ pa_source_update_latency_range(u->source, new_min_latency, u->source->thread_info.max_latency);
+ return;
+ }
- if (u->tsched_watermark < min_wakeup)
- u->tsched_watermark = min_wakeup;
+ /* When we reach this we're officialy fucked! */
}
static pa_usec_t hw_sleep_time(struct userdata *u, pa_usec_t *sleep_usec, pa_usec_t*process_usec) {
if (usec == (pa_usec_t) -1)
usec = pa_bytes_to_usec(u->hwbuf_size, &u->source->sample_spec);
-/* pa_log_debug("hw buffer time: %u ms", (unsigned) (usec / PA_USEC_PER_MSEC)); */
-
wm = pa_bytes_to_usec(u->tsched_watermark, &u->source->sample_spec);
- if (usec >= wm) {
- *sleep_usec = usec - wm;
- *process_usec = wm;
- } else
- *process_usec = *sleep_usec = usec /= 2;
+ if (wm > usec)
+ wm = usec/2;
-/* pa_log_debug("after watermark: %u ms", (unsigned) (*sleep_usec / PA_USEC_PER_MSEC)); */
+ *sleep_usec = usec - wm;
+ *process_usec = wm;
+
+#ifdef DEBUG_TIMING
+ pa_log_debug("Buffer time: %lu ms; Sleep time: %lu ms; Process time: %lu ms",
+ (unsigned long) (usec / PA_USEC_PER_MSEC),
+ (unsigned long) (*sleep_usec / PA_USEC_PER_MSEC),
+ (unsigned long) (*process_usec / PA_USEC_PER_MSEC));
+#endif
return usec;
}
if (err == -EPIPE)
pa_log_debug("%s: Buffer overrun!", call);
- if ((err = snd_pcm_recover(u->pcm_handle, err, 1)) == 0) {
- snd_pcm_start(u->pcm_handle);
- return 0;
+ if ((err = snd_pcm_recover(u->pcm_handle, err, 1)) < 0) {
+ pa_log("%s: %s", call, snd_strerror(err));
+ return -1;
}
- pa_log("%s: %s", call, snd_strerror(err));
- return -1;
+ snd_pcm_start(u->pcm_handle);
+ return 0;
}
-static size_t check_left_to_record(struct userdata *u, snd_pcm_sframes_t n) {
+static size_t check_left_to_record(struct userdata *u, size_t n_bytes) {
size_t left_to_record;
- size_t rec_space = u->hwbuf_size - (size_t) u->hwbuf_unused_frames*u->frame_size;
+ size_t rec_space = u->hwbuf_size - u->hwbuf_unused;
- if ((size_t) n*u->frame_size < rec_space)
- left_to_record = rec_space - ((size_t) n*u->frame_size);
- else
- left_to_record = 0;
+ /* We use <= instead of < for this check here because an overrun
+ * only happens after the last sample was processed, not already when
+ * it is removed from the buffer. This is particularly important
+ * when block transfer is used. */
+
+ if (n_bytes <= rec_space) {
+ left_to_record = rec_space - n_bytes;
+
+#ifdef DEBUG_TIMING
+ pa_log_debug("%0.2f ms left to record", (double) pa_bytes_to_usec(left_to_record, &u->source->sample_spec) / PA_USEC_PER_MSEC);
+#endif
- if (left_to_record > 0) {
-/* pa_log_debug("%0.2f ms left to record", (double) pa_bytes_to_usec(left_to_record, &u->source->sample_spec) / PA_USEC_PER_MSEC); */
} else {
- if (pa_log_ratelimit())
- pa_log_info("Overrun!");
+ left_to_record = 0;
- if (u->use_tsched) {
- size_t old_watermark = u->tsched_watermark;
+#ifdef DEBUG_TIMING
+ PA_DEBUG_TRAP;
+#endif
- u->tsched_watermark *= 2;
- fix_tsched_watermark(u);
+ if (pa_log_ratelimit())
+ pa_log_info("Overrun!");
- if (old_watermark != u->tsched_watermark)
- pa_log_notice("Increasing wakeup watermark to %0.2f ms",
- (double) pa_bytes_to_usec(u->tsched_watermark, &u->source->sample_spec) / PA_USEC_PER_MSEC);
- }
+ if (u->use_tsched)
+ adjust_after_overrun(u);
}
return left_to_record;
}
static int mmap_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled) {
- int work_done = 0;
+ pa_bool_t work_done = FALSE;
pa_usec_t max_sleep_usec = 0, process_usec = 0;
size_t left_to_record;
for (;;) {
snd_pcm_sframes_t n;
+ size_t n_bytes;
int r;
- snd_pcm_hwsync(u->pcm_handle);
-
- if (PA_UNLIKELY((n = pa_alsa_safe_avail_update(u->pcm_handle, u->hwbuf_size, &u->source->sample_spec)) < 0)) {
+ if (PA_UNLIKELY((n = pa_alsa_safe_avail(u->pcm_handle, u->hwbuf_size, &u->source->sample_spec)) < 0)) {
- if ((r = try_recover(u, "snd_pcm_avail_update", (int) n)) == 0)
+ if ((r = try_recover(u, "snd_pcm_avail", (int) n)) == 0)
continue;
return r;
}
- left_to_record = check_left_to_record(u, n);
+ n_bytes = (size_t) n * u->frame_size;
+
+#ifdef DEBUG_TIMING
+ pa_log_debug("avail: %lu", (unsigned long) n_bytes);
+#endif
+
+ left_to_record = check_left_to_record(u, n_bytes);
if (u->use_tsched)
if (!polled &&
- pa_bytes_to_usec(left_to_record, &u->source->sample_spec) > process_usec+max_sleep_usec/2)
+ pa_bytes_to_usec(left_to_record, &u->source->sample_spec) > process_usec+max_sleep_usec/2) {
+#ifdef DEBUG_TIMING
+ pa_log_debug("Not reading, because too early.");
+#endif
break;
+ }
- if (PA_UNLIKELY(n <= 0)) {
+ if (PA_UNLIKELY(n_bytes <= 0)) {
if (polled && pa_log_ratelimit())
- pa_log("ALSA woke us up to read new data from the device, but there was actually nothing to read! "
- "Most likely this is an ALSA driver bug. Please report this issue to the ALSA developers. "
- "We were woken up with POLLIN set -- however a subsequent snd_pcm_avail_update() returned 0.");
+ pa_log(_("ALSA woke us up to read new data from the device, but there was actually nothing to read! "
+ "Most likely this is an ALSA driver bug. Please report this issue to the ALSA developers. "
+ "We were woken up with POLLIN set -- however a subsequent snd_pcm_avail_update() returned 0."));
+#ifdef DEBUG_TIMING
+ pa_log_debug("Not reading, because not necessary.");
+#endif
break;
}
polled = FALSE;
+#ifdef DEBUG_TIMING
+ pa_log_debug("Reading");
+#endif
+
for (;;) {
int err;
const snd_pcm_channel_area_t *areas;
- snd_pcm_uframes_t offset, frames = (snd_pcm_uframes_t) n;
+ snd_pcm_uframes_t offset, frames;
pa_memchunk chunk;
void *p;
snd_pcm_sframes_t sframes;
+ frames = (snd_pcm_uframes_t) (n_bytes / u->frame_size);
+
/* pa_log_debug("%lu frames to read", (unsigned long) frames); */
if (PA_UNLIKELY((err = pa_alsa_safe_mmap_begin(u->pcm_handle, &areas, &offset, &frames, u->hwbuf_size, &u->source->sample_spec)) < 0)) {
return r;
}
- work_done = 1;
+ work_done = TRUE;
- u->frame_index += (int64_t) frames;
+ u->read_count += frames * u->frame_size;
-/* pa_log_debug("read %lu frames", (unsigned long) frames); */
+#ifdef DEBUG_TIMING
+ pa_log_debug("Read %lu bytes", (unsigned long) (frames * u->frame_size));
+#endif
- if (frames >= (snd_pcm_uframes_t) n)
+ if ((size_t) frames * u->frame_size >= n_bytes)
break;
- n -= (snd_pcm_sframes_t) frames;
+ n_bytes -= (size_t) frames * u->frame_size;
}
}
*sleep_usec = pa_bytes_to_usec(left_to_record, &u->source->sample_spec) - process_usec;
- return work_done;
+ return work_done ? 1 : 0;
}
static int unix_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled) {
- int work_done = 0;
+ int work_done = FALSE;
pa_usec_t max_sleep_usec = 0, process_usec = 0;
size_t left_to_record;
for (;;) {
snd_pcm_sframes_t n;
+ size_t n_bytes;
int r;
- snd_pcm_hwsync(u->pcm_handle);
-
- if (PA_UNLIKELY((n = pa_alsa_safe_avail_update(u->pcm_handle, u->hwbuf_size, &u->source->sample_spec)) < 0)) {
+ if (PA_UNLIKELY((n = pa_alsa_safe_avail(u->pcm_handle, u->hwbuf_size, &u->source->sample_spec)) < 0)) {
- if ((r = try_recover(u, "snd_pcm_avail_update", (int) n)) == 0)
+ if ((r = try_recover(u, "snd_pcm_avail", (int) n)) == 0)
continue;
return r;
}
- left_to_record = check_left_to_record(u, n);
+ n_bytes = (size_t) n * u->frame_size;
+ left_to_record = check_left_to_record(u, n_bytes);
if (u->use_tsched)
if (!polled &&
pa_bytes_to_usec(left_to_record, &u->source->sample_spec) > process_usec+max_sleep_usec/2)
break;
- if (PA_UNLIKELY(n <= 0)) {
+ if (PA_UNLIKELY(n_bytes <= 0)) {
if (polled && pa_log_ratelimit())
- pa_log("ALSA woke us up to read new data from the device, but there was actually nothing to read! "
- "Most likely this is an ALSA driver bug. Please report this issue to the ALSA developers. "
- "We were woken up with POLLIN set -- however a subsequent snd_pcm_avail_update() returned 0.");
+ pa_log(_("ALSA woke us up to read new data from the device, but there was actually nothing to read! "
+ "Most likely this is an ALSA driver bug. Please report this issue to the ALSA developers. "
+ "We were woken up with POLLIN set -- however a subsequent snd_pcm_avail_update() returned 0."));
- return work_done;
+ break;
}
polled = FALSE;
frames = (snd_pcm_sframes_t) (pa_memblock_get_length(chunk.memblock) / u->frame_size);
- if (frames > n)
- frames = n;
+ if (frames > (snd_pcm_sframes_t) (n_bytes/u->frame_size))
+ frames = (snd_pcm_sframes_t) (n_bytes/u->frame_size);
/* pa_log_debug("%lu frames to read", (unsigned long) n); */
pa_source_post(u->source, &chunk);
pa_memblock_unref(chunk.memblock);
- work_done = 1;
+ work_done = TRUE;
- u->frame_index += frames;
+ u->read_count += frames * u->frame_size;
/* pa_log_debug("read %lu frames", (unsigned long) frames); */
- if (frames >= n)
+ if ((size_t) frames * u->frame_size >= n_bytes)
break;
- n -= frames;
+ n_bytes -= (size_t) frames * u->frame_size;
}
}
*sleep_usec = pa_bytes_to_usec(left_to_record, &u->source->sample_spec) - process_usec;
- return work_done;
+ return work_done ? 1 : 0;
}
static void update_smoother(struct userdata *u) {
snd_pcm_sframes_t delay = 0;
- int64_t frames;
+ uint64_t position;
int err;
- pa_usec_t now1, now2;
+ pa_usec_t now1 = 0, now2;
+ snd_pcm_status_t *status;
+
+ snd_pcm_status_alloca(&status);
pa_assert(u);
pa_assert(u->pcm_handle);
/* Let's update the time smoother */
- snd_pcm_hwsync(u->pcm_handle);
- snd_pcm_avail_update(u->pcm_handle);
-
if (PA_UNLIKELY((err = snd_pcm_delay(u->pcm_handle, &delay)) < 0)) {
pa_log_warn("Failed to get delay: %s", snd_strerror(err));
return;
}
- frames = u->frame_index + delay;
+ if (PA_UNLIKELY((err = snd_pcm_status(u->pcm_handle, status)) < 0))
+ pa_log_warn("Failed to get timestamp: %s", snd_strerror(err));
+ else {
+ snd_htimestamp_t htstamp = { 0, 0 };
+ snd_pcm_status_get_htstamp(status, &htstamp);
+ now1 = pa_timespec_load(&htstamp);
+ }
- now1 = pa_rtclock_usec();
- now2 = pa_bytes_to_usec((uint64_t) frames * u->frame_size, &u->source->sample_spec);
+ position = u->read_count + ((uint64_t) delay * (uint64_t) u->frame_size);
+
+ /* Hmm, if the timestamp is 0, then it wasn't set and we take the current time */
+ if (now1 <= 0)
+ now1 = pa_rtclock_usec();
+
+ now2 = pa_bytes_to_usec(position, &u->source->sample_spec);
pa_smoother_put(u->smoother, now1, now2);
}
static pa_usec_t source_get_latency(struct userdata *u) {
- pa_usec_t r = 0;
- int64_t delay;
+ int64_t delay;
pa_usec_t now1, now2;
pa_assert(u);
now1 = pa_rtclock_usec();
now2 = pa_smoother_get(u->smoother, now1);
- delay = (int64_t) now2 - (int64_t) pa_bytes_to_usec((uint64_t) u->frame_index * u->frame_size, &u->source->sample_spec);
-
- if (delay > 0)
- r = (pa_usec_t) delay;
+ delay = (int64_t) now2 - (int64_t) pa_bytes_to_usec(u->read_count, &u->source->sample_spec);
- return r;
+ return delay >= 0 ? (pa_usec_t) delay : 0;
}
static int build_pollfd(struct userdata *u) {
pa_assert(u);
/* Use the full buffer if noone asked us for anything specific */
- u->hwbuf_unused_frames = 0;
+ u->hwbuf_unused = 0;
if (u->use_tsched) {
pa_usec_t latency;
if (PA_UNLIKELY(b < u->frame_size))
b = u->frame_size;
- u->hwbuf_unused_frames = (snd_pcm_sframes_t)
- (PA_LIKELY(b < u->hwbuf_size) ?
- ((u->hwbuf_size - b) / u->frame_size) : 0);
+ u->hwbuf_unused = PA_LIKELY(b < u->hwbuf_size) ? (u->hwbuf_size - b) : 0;
}
+ fix_min_sleep_wakeup(u);
fix_tsched_watermark(u);
}
- pa_log_debug("hwbuf_unused_frames=%lu", (unsigned long) u->hwbuf_unused_frames);
+ pa_log_debug("hwbuf_unused=%lu", (unsigned long) u->hwbuf_unused);
avail_min = 1;
for (;;) {
int ret;
-/* pa_log_debug("loop"); */
+#ifdef DEBUG_TIMING
+ pa_log_debug("Loop");
+#endif
/* Read some data and pass it to the sources */
if (PA_SOURCE_IS_OPENED(u->source->thread_info.state)) {
- int work_done = 0;
+ int work_done;
pa_usec_t sleep_usec = 0;
if (u->use_mmap)
goto fail;
}
- if (revents & (POLLOUT|POLLERR|POLLNVAL|POLLHUP|POLLPRI)) {
+ if (revents & ~POLLIN) {
if (pa_alsa_recover_from_poll(u->pcm_handle, revents) < 0)
goto fail;
snd_pcm_start(u->pcm_handle);
- }
+ } else if (revents && u->use_tsched && pa_log_ratelimit())
+ pa_log_debug("Wakeup from ALSA!");
- if (revents && u->use_tsched && pa_log_ratelimit())
- pa_log_debug("Wakeup from ALSA!%s%s", (revents & POLLIN) ? " INPUT" : "", (revents & POLLOUT) ? " OUTPUT" : "");
} else
revents = 0;
}
pa_assert(ma);
ss = m->core->default_sample_spec;
+ map = m->core->default_channel_map;
if (pa_modargs_get_sample_spec_and_channel_map(ma, &ss, &map, PA_CHANNEL_MAP_ALSA) < 0) {
pa_log("Failed to parse sample specification");
goto fail;
pa_thread_mq_init(&u->thread_mq, m->core->mainloop, u->rtpoll);
u->alsa_rtpoll_item = NULL;
- u->smoother = pa_smoother_new(DEFAULT_TSCHED_WATERMARK_USEC, DEFAULT_TSCHED_WATERMARK_USEC, TRUE, 5);
+ u->smoother = pa_smoother_new(DEFAULT_TSCHED_WATERMARK_USEC*2, DEFAULT_TSCHED_WATERMARK_USEC*2, TRUE, 5);
pa_smoother_set_time_offset(u->smoother, pa_rtclock_usec());
b = use_mmap;
u->fragment_size = frag_size = (uint32_t) (period_frames * frame_size);
u->nfragments = nfrags;
u->hwbuf_size = u->fragment_size * nfrags;
- u->hwbuf_unused_frames = 0;
u->tsched_watermark = tsched_watermark;
- u->frame_index = 0;
- u->hw_dB_supported = FALSE;
- u->hw_dB_min = u->hw_dB_max = 0;
- u->hw_volume_min = u->hw_volume_max = 0;
- u->mixer_seperate_channels = FALSE;
pa_cvolume_mute(&u->hardware_volume, u->source->sample_spec.channels);
- if (use_tsched)
+ if (use_tsched) {
+ fix_min_sleep_wakeup(u);
fix_tsched_watermark(u);
+ }
pa_source_set_latency_range(u->source,
!use_tsched ? pa_bytes_to_usec(u->hwbuf_size, &ss) : (pa_usec_t) -1,