[alsa-devel] [PATCH 0/3] ALSA: pcm: add tracepoints for PCM params refinement
Hi,
This patchset is a part of my previous RFC and go for upstream. [alsa-devel] [PATCH RFC 00/21] ALSA: pcm: add tracepoints for PCM params operation http://mailman.alsa-project.org/pipermail/alsa-devel/2017-May/120548.html
In ALSA PCM interface, applications can get hardware capability by ioctl(2) with SNDRV_PCM_IOCTL_HW_REFINE/SNDRV_PCM_IOCT_HW_PARAMS in a shape of 'struct snd_pcm_hw_params'. In kernel side, relevant processing is somewhat complicated and developers sometimes have hard time to debug drivers for PCM constraints and rules. This patchset adds tracepoints to assist their work.
Changes: - change print format - rename events - refactoring in the RFC will be done in my later work.
This patchset adds two events; 'hw_interval_param' and 'hw_mask_param' in 'snd_pcm' subsystem. When these events are probed, tracer outputs below lines:
hw_interval_param: 0,0,0,0 000/023 BUFFER_SIZE 0 0 [0 4294967295] 0 1 [0 4294967295] hw_interval_param: 0,0,0,0 000/023 BUFFER_BYTES 0 0 [0 4294967295] 0 1 [128 65536] hw_interval_param: 0,0,0,0 000/023 TICK_TIME 0 0 [0 4294967295] 0 0 [0 4294967295] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044 hw_interval_param: 0,0,0,0 002/023 SAMPLE_BITS 0 1 [0 4294967295] 0 1 [16 32] hw_interval_param: 0,0,0,0 003/023 SAMPLE_BITS 0 1 [16 32] 0 1 [16 32]
The first field represents PCM character device and subdevice for probed runtime of PCM substream. In the above case, '/dev/snd/pcmC0D0p' and first subdevice are used for runtime of PCM substream.
The second field represents id of each rule applied to the runtime, and the total number of rules added to the runtime. Basically, when the rule is applied to the runtime, these events are probed, but lines with id 0 are exceptions. They're application of constraints to the runtime.
The third field is name of parameter in the runtime. The rest fields depends on type of the parameter (mask/interval).
In ALSA PCM core, runtimes get 22 (or 21) rules as a default. In a below sample, the 21st rule is added by driver (snd-soc-imx-ssi.ko). The rest is added by snd-pcm.ko. In detail, please see 'snd_pcm_hw_constraints_init()' and 'snd_pcm_hw_constraints_complete()' in 'sound/core/pcm_native.c'.
hw_interval_param: 0,0,0,0 019/023 PERIOD_TIME 0 0 [0 4294967295] 0 0 (166 4095875] hw_interval_param: 0,0,0,0 020/023 BUFFER_TIME 0 0 [0 4294967295] 0 0 (333 4096000] hw_interval_param: 0,0,0,0 021/023 PERIOD_SIZE 0 1 [16 32767] 0 1 [16 32766] hw_interval_param: 0,0,0,0 022/023 BUFFER_BYTES 0 1 [128 65536] 0 1 [128 65536] hw_interval_param: 0,0,0,0 023/023 RATE 0 0 [8000 96000] 0 0 [8000 96000] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044
As I noted in second patch, this patchset increases size of relocatable object. In general, kernel configuration for tracepoints should govern whether the features are enbled/disabled. However, for convenience to distributor, this patchset utilizes CONFIG_SND_DEBUG to switch the feature is enabled/disabled.
$ size sound/core/pcm_native.o.before text data bss dec hex filename 28951 40 8 28999 7147 sound/core/pcm_native.o.before $ size sound/core/pcm_native.o.after text data bss dec hex filename 34942 2264 8 37214 915e sound/core/pcm_native.o.after
Takashi Sakamoto (3): ALSA: pcm: tracepoints for refining PCM parameters ALSA: pcm: enable parameter tracepoints only when CONFIG_SND_DEBUG is enabled ALSA: pcm: obsolete RULES_DEBUG local macro
sound/core/Makefile | 1 + sound/core/pcm_native.c | 125 ++++++++++++++----------------------- sound/core/pcm_param_trace.h | 142 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 188 insertions(+), 80 deletions(-) create mode 100644 sound/core/pcm_param_trace.h
When working for devices which support configurable modes for its data transmission or which consists of several components, developers are likely to use rules of parameters of PCM substream. However, there's no infrastructure to assist their work.
In old days, ALSA PCM core got a local 'RULES_DEBUG' macro to debug refinement of parameters for PCM substream. Although this is merely a makeshift. With some modifications, we get the infrastructure.
This commit is for the purpose. Refinement of mask/interval type of PCM parameters is probed as tracepoint events as 'hw_mask_param' and 'hw_interval_param' on existent 'snd_pcm' subsystem.
Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp --- sound/core/Makefile | 1 + sound/core/pcm_native.c | 33 ++++++++++ sound/core/pcm_param_trace.h | 142 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 176 insertions(+) create mode 100644 sound/core/pcm_param_trace.h
diff --git a/sound/core/Makefile b/sound/core/Makefile index e85d9dd12c2d..a8514b313a89 100644 --- a/sound/core/Makefile +++ b/sound/core/Makefile @@ -22,6 +22,7 @@ snd-pcm-$(CONFIG_SND_PCM_IEC958) += pcm_iec958.o
# for trace-points CFLAGS_pcm_lib.o := -I$(src) +CFLAGS_pcm_native.o := -I$(src)
snd-pcm-dmaengine-objs := pcm_dmaengine.o
diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c index bf5d0f2acfb9..b98b3ccde4f0 100644 --- a/sound/core/pcm_native.c +++ b/sound/core/pcm_native.c @@ -39,6 +39,9 @@
#include "pcm_local.h"
+#define CREATE_TRACE_POINTS +#include "pcm_param_trace.h" + /* * Compatibility */ @@ -279,6 +282,9 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, unsigned int stamp = 2; int changed, again;
+ struct snd_mask __maybe_unused old_mask; + struct snd_interval __maybe_unused old_interval; + params->info = 0; params->fifo_size = 0; if (params->rmask & (1 << SNDRV_PCM_HW_PARAM_SAMPLE_BITS)) @@ -294,6 +300,9 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, return -EINVAL; if (!(params->rmask & (1 << k))) continue; + + if (trace_hw_mask_param_enabled()) + old_mask = *m; #ifdef RULES_DEBUG pr_debug("%s = ", snd_pcm_hw_param_names[k]); pr_cont("%04x%04x%04x%04x -> ", m->bits[3], m->bits[2], m->bits[1], m->bits[0]); @@ -302,6 +311,8 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, #ifdef RULES_DEBUG pr_cont("%04x%04x%04x%04x\n", m->bits[3], m->bits[2], m->bits[1], m->bits[0]); #endif + trace_hw_mask_param(substream, k, 0, &old_mask, m); + if (changed) params->cmask |= 1 << k; if (changed < 0) @@ -314,6 +325,9 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, return -EINVAL; if (!(params->rmask & (1 << k))) continue; + + if (trace_hw_interval_param_enabled()) + old_interval = *i; #ifdef RULES_DEBUG pr_debug("%s = ", snd_pcm_hw_param_names[k]); if (i->empty) @@ -333,6 +347,8 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, i->openmin ? '(' : '[', i->min, i->max, i->openmax ? ')' : ']'); #endif + trace_hw_interval_param(substream, k, 0, &old_interval, i); + if (changed) params->cmask |= 1 << k; if (changed < 0) @@ -359,6 +375,15 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, } if (!doit) continue; + + if (trace_hw_mask_param_enabled()) { + if (hw_is_mask(r->var)) + old_mask = *hw_param_mask(params, r->var); + } + if (trace_hw_interval_param_enabled()) { + if (hw_is_interval(r->var)) + old_interval = *hw_param_interval(params, r->var); + } #ifdef RULES_DEBUG pr_debug("Rule %d [%p]: ", k, r->func); if (r->var >= 0) { @@ -394,6 +419,14 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, } pr_cont("\n"); #endif + if (hw_is_mask(r->var)) { + trace_hw_mask_param(substream, r->var, k + 1, + &old_mask, hw_param_mask(params, r->var)); + } + if (hw_is_interval(r->var)) { + trace_hw_interval_param(substream, r->var, k + 1, + &old_interval, hw_param_interval(params, r->var)); + } rstamps[k] = stamp; if (changed && r->var >= 0) { params->cmask |= (1 << r->var); diff --git a/sound/core/pcm_param_trace.h b/sound/core/pcm_param_trace.h new file mode 100644 index 000000000000..872922326b38 --- /dev/null +++ b/sound/core/pcm_param_trace.h @@ -0,0 +1,142 @@ +#undef TRACE_SYSTEM +#define TRACE_SYSTEM snd_pcm + +#if !defined(_PCM_PARAMS_TRACE_H) || defined(TRACE_HEADER_MULTI_READ) +#define _PCM_PARAMS_TRACE_H + +#include <linux/tracepoint.h> + +#define HW_PARAM_ENTRY(param) {SNDRV_PCM_HW_PARAM_##param, #param} +#define hw_param_labels \ + HW_PARAM_ENTRY(ACCESS), \ + HW_PARAM_ENTRY(FORMAT), \ + HW_PARAM_ENTRY(SUBFORMAT), \ + HW_PARAM_ENTRY(SAMPLE_BITS), \ + HW_PARAM_ENTRY(FRAME_BITS), \ + HW_PARAM_ENTRY(CHANNELS), \ + HW_PARAM_ENTRY(RATE), \ + HW_PARAM_ENTRY(PERIOD_TIME), \ + HW_PARAM_ENTRY(PERIOD_SIZE), \ + HW_PARAM_ENTRY(PERIOD_BYTES), \ + HW_PARAM_ENTRY(PERIODS), \ + HW_PARAM_ENTRY(BUFFER_TIME), \ + HW_PARAM_ENTRY(BUFFER_SIZE), \ + HW_PARAM_ENTRY(BUFFER_BYTES), \ + HW_PARAM_ENTRY(TICK_TIME) + +TRACE_EVENT(hw_mask_param, + TP_PROTO(struct snd_pcm_substream *substream, snd_pcm_hw_param_t type, int index, const struct snd_mask *prev, const struct snd_mask *curr), + TP_ARGS(substream, type, index, prev, curr), + TP_STRUCT__entry( + __field(int, card) + __field(int, device) + __field(int, subdevice) + __field(int, direction) + __field(snd_pcm_hw_param_t, type) + __field(int, index) + __field(int, total) + __array(__u32, prev_bits, 8) + __array(__u32, curr_bits, 8) + ), + TP_fast_assign( + __entry->card = substream->pcm->card->number; + __entry->device = substream->pcm->device; + __entry->subdevice = substream->number; + __entry->direction = substream->stream; + __entry->type = type; + __entry->index = index; + __entry->total = substream->runtime->hw_constraints.rules_num; + memcpy(__entry->prev_bits, prev->bits, sizeof(__u32) * 8); + memcpy(__entry->curr_bits, curr->bits, sizeof(__u32) * 8); + ), + TP_printk("%d,%d,%d,%d %03d/%03d %s %08x%08x%08x%08x %08x%08x%08x%08x", + __entry->card, + __entry->device, + __entry->subdevice, + __entry->direction, + __entry->index, + __entry->total, + __print_symbolic(__entry->type, hw_param_labels), + __entry->prev_bits[3], __entry->prev_bits[2], + __entry->prev_bits[1], __entry->prev_bits[0], + __entry->curr_bits[3], __entry->curr_bits[2], + __entry->curr_bits[1], __entry->curr_bits[0] + ) +); + +TRACE_EVENT(hw_interval_param, + TP_PROTO(struct snd_pcm_substream *substream, snd_pcm_hw_param_t type, int index, const struct snd_interval *prev, const struct snd_interval *curr), + TP_ARGS(substream, type, index, prev, curr), + TP_STRUCT__entry( + __field(int, card) + __field(int, device) + __field(int, subdevice) + __field(int, direction) + __field(snd_pcm_hw_param_t, type) + __field(int, index) + __field(int, total) + __field(unsigned int, prev_min) + __field(unsigned int, prev_max) + __field(unsigned int, prev_openmin) + __field(unsigned int, prev_openmax) + __field(unsigned int, prev_integer) + __field(unsigned int, prev_empty) + __field(unsigned int, curr_min) + __field(unsigned int, curr_max) + __field(unsigned int, curr_openmin) + __field(unsigned int, curr_openmax) + __field(unsigned int, curr_integer) + __field(unsigned int, curr_empty) + ), + TP_fast_assign( + __entry->card = substream->pcm->card->number; + __entry->device = substream->pcm->device; + __entry->subdevice = substream->number; + __entry->direction = substream->stream; + __entry->type = type; + __entry->index = index; + __entry->total = substream->runtime->hw_constraints.rules_num; + __entry->prev_min = prev->min; + __entry->prev_max = prev->max; + __entry->prev_openmin = prev->openmin; + __entry->prev_openmax = prev->openmax; + __entry->prev_integer = prev->integer; + __entry->prev_empty = prev->empty; + __entry->curr_min = curr->min; + __entry->curr_max = curr->max; + __entry->curr_openmin = curr->openmin; + __entry->curr_openmax = curr->openmax; + __entry->curr_integer = curr->integer; + __entry->curr_empty = curr->empty; + ), + TP_printk("%d,%d,%d,%d %03d/%03d %s %d %d %s%u %u%s %d %d %s%u %u%s", + __entry->card, + __entry->device, + __entry->subdevice, + __entry->direction, + __entry->index, + __entry->total, + __print_symbolic(__entry->type, hw_param_labels), + __entry->prev_empty, + __entry->prev_integer, + __entry->prev_openmin ? "(" : "[", + __entry->prev_min, + __entry->prev_max, + __entry->prev_openmax ? ")" : "]", + __entry->curr_empty, + __entry->curr_integer, + __entry->curr_openmin ? "(" : "[", + __entry->curr_min, + __entry->curr_max, + __entry->curr_openmax ? ")" : "]" + ) +); + +#endif /* _PCM_PARAMS_TRACE_H */ + +/* This part must be outside protection */ +#undef TRACE_INCLUDE_PATH +#define TRACE_INCLUDE_PATH . +#undef TRACE_INCLUDE_FILE +#define TRACE_INCLUDE_FILE pcm_param_trace +#include <trace/define_trace.h>
In a previous commit, tracepoints are added for PCM parameter processing. As long as I know, this implementation increases size of relocatable object by 35%. For vendors who are conscious of memory footprint, it brings apparent disadvantage.
This commit utilizes CONFIG_SND_DEBUG configuration to enable/disable the tracepoints.
Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp --- sound/core/pcm_native.c | 7 +++++++ 1 file changed, 7 insertions(+)
diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c index b98b3ccde4f0..2ce3c98a1418 100644 --- a/sound/core/pcm_native.c +++ b/sound/core/pcm_native.c @@ -39,8 +39,15 @@
#include "pcm_local.h"
+#ifdef CONFIG_SND_DEBUG #define CREATE_TRACE_POINTS #include "pcm_param_trace.h" +#else +#define trace_hw_mask_param_enabled() 0 +#define trace_hw_interval_param_enabled() 0 +#define trace_hw_mask_param(substream, type, index, prev, curr) +#define trace_hw_interval_param(substream, type, index, prev, curr) +#endif
/* * Compatibility
Added tracepoints obsoleted RULES_DEBUG local macro and relevant codes.
Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp --- sound/core/pcm_native.c | 89 ++++--------------------------------------------- 1 file changed, 7 insertions(+), 82 deletions(-)
diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c index 2ce3c98a1418..2bde07a4a87f 100644 --- a/sound/core/pcm_native.c +++ b/sound/core/pcm_native.c @@ -253,29 +253,6 @@ static bool hw_support_mmap(struct snd_pcm_substream *substream) return true; }
-#undef RULES_DEBUG - -#ifdef RULES_DEBUG -#define HW_PARAM(v) [SNDRV_PCM_HW_PARAM_##v] = #v -static const char * const snd_pcm_hw_param_names[] = { - HW_PARAM(ACCESS), - HW_PARAM(FORMAT), - HW_PARAM(SUBFORMAT), - HW_PARAM(SAMPLE_BITS), - HW_PARAM(FRAME_BITS), - HW_PARAM(CHANNELS), - HW_PARAM(RATE), - HW_PARAM(PERIOD_TIME), - HW_PARAM(PERIOD_SIZE), - HW_PARAM(PERIOD_BYTES), - HW_PARAM(PERIODS), - HW_PARAM(BUFFER_TIME), - HW_PARAM(BUFFER_SIZE), - HW_PARAM(BUFFER_BYTES), - HW_PARAM(TICK_TIME), -}; -#endif - int snd_pcm_hw_refine(struct snd_pcm_substream *substream, struct snd_pcm_hw_params *params) { @@ -310,14 +287,9 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream,
if (trace_hw_mask_param_enabled()) old_mask = *m; -#ifdef RULES_DEBUG - pr_debug("%s = ", snd_pcm_hw_param_names[k]); - pr_cont("%04x%04x%04x%04x -> ", m->bits[3], m->bits[2], m->bits[1], m->bits[0]); -#endif + changed = snd_mask_refine(m, constrs_mask(constrs, k)); -#ifdef RULES_DEBUG - pr_cont("%04x%04x%04x%04x\n", m->bits[3], m->bits[2], m->bits[1], m->bits[0]); -#endif + trace_hw_mask_param(substream, k, 0, &old_mask, m);
if (changed) @@ -335,25 +307,9 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream,
if (trace_hw_interval_param_enabled()) old_interval = *i; -#ifdef RULES_DEBUG - pr_debug("%s = ", snd_pcm_hw_param_names[k]); - if (i->empty) - pr_cont("empty"); - else - pr_cont("%c%u %u%c", - i->openmin ? '(' : '[', i->min, - i->max, i->openmax ? ')' : ']'); - pr_cont(" -> "); -#endif + changed = snd_interval_refine(i, constrs_interval(constrs, k)); -#ifdef RULES_DEBUG - if (i->empty) - pr_cont("empty\n"); - else - pr_cont("%c%u %u%c\n", - i->openmin ? '(' : '[', i->min, - i->max, i->openmax ? ')' : ']'); -#endif + trace_hw_interval_param(substream, k, 0, &old_interval, i);
if (changed) @@ -391,41 +347,9 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, if (hw_is_interval(r->var)) old_interval = *hw_param_interval(params, r->var); } -#ifdef RULES_DEBUG - pr_debug("Rule %d [%p]: ", k, r->func); - if (r->var >= 0) { - pr_cont("%s = ", snd_pcm_hw_param_names[r->var]); - if (hw_is_mask(r->var)) { - m = hw_param_mask(params, r->var); - pr_cont("%x", *m->bits); - } else { - i = hw_param_interval(params, r->var); - if (i->empty) - pr_cont("empty"); - else - pr_cont("%c%u %u%c", - i->openmin ? '(' : '[', i->min, - i->max, i->openmax ? ')' : ']'); - } - } -#endif + changed = r->func(params, r); -#ifdef RULES_DEBUG - if (r->var >= 0) { - pr_cont(" -> "); - if (hw_is_mask(r->var)) - pr_cont("%x", *m->bits); - else { - if (i->empty) - pr_cont("empty"); - else - pr_cont("%c%u %u%c", - i->openmin ? '(' : '[', i->min, - i->max, i->openmax ? ')' : ']'); - } - } - pr_cont("\n"); -#endif + if (hw_is_mask(r->var)) { trace_hw_mask_param(substream, r->var, k + 1, &old_mask, hw_param_mask(params, r->var)); @@ -434,6 +358,7 @@ int snd_pcm_hw_refine(struct snd_pcm_substream *substream, trace_hw_interval_param(substream, r->var, k + 1, &old_interval, hw_param_interval(params, r->var)); } + rstamps[k] = stamp; if (changed && r->var >= 0) { params->cmask |= (1 << r->var);
On Wed, 07 Jun 2017 01:46:42 +0200, Takashi Sakamoto wrote:
Hi,
This patchset is a part of my previous RFC and go for upstream. [alsa-devel] [PATCH RFC 00/21] ALSA: pcm: add tracepoints for PCM params operation http://mailman.alsa-project.org/pipermail/alsa-devel/2017-May/120548.html
In ALSA PCM interface, applications can get hardware capability by ioctl(2) with SNDRV_PCM_IOCTL_HW_REFINE/SNDRV_PCM_IOCT_HW_PARAMS in a shape of 'struct snd_pcm_hw_params'. In kernel side, relevant processing is somewhat complicated and developers sometimes have hard time to debug drivers for PCM constraints and rules. This patchset adds tracepoints to assist their work.
Changes:
- change print format
- rename events
- refactoring in the RFC will be done in my later work.
This patchset adds two events; 'hw_interval_param' and 'hw_mask_param' in 'snd_pcm' subsystem. When these events are probed, tracer outputs below lines:
hw_interval_param: 0,0,0,0 000/023 BUFFER_SIZE 0 0 [0 4294967295] 0 1 [0 4294967295] hw_interval_param: 0,0,0,0 000/023 BUFFER_BYTES 0 0 [0 4294967295] 0 1 [128 65536] hw_interval_param: 0,0,0,0 000/023 TICK_TIME 0 0 [0 4294967295] 0 0 [0 4294967295] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044 hw_interval_param: 0,0,0,0 002/023 SAMPLE_BITS 0 1 [0 4294967295] 0 1 [16 32] hw_interval_param: 0,0,0,0 003/023 SAMPLE_BITS 0 1 [16 32] 0 1 [16 32]
The first field represents PCM character device and subdevice for probed runtime of PCM substream. In the above case, '/dev/snd/pcmC0D0p' and first subdevice are used for runtime of PCM substream.
How about using snd_pcm_debug_name()? It'll be like "pcmC0D1c:2" and slightly more understandable than "0,1,1,2".
The second field represents id of each rule applied to the runtime, and the total number of rules added to the runtime. Basically, when the rule is applied to the runtime, these events are probed, but lines with id 0 are exceptions. They're application of constraints to the runtime.
The third field is name of parameter in the runtime. The rest fields depends on type of the parameter (mask/interval).
In ALSA PCM core, runtimes get 22 (or 21) rules as a default. In a below sample, the 21st rule is added by driver (snd-soc-imx-ssi.ko). The rest is added by snd-pcm.ko. In detail, please see 'snd_pcm_hw_constraints_init()' and 'snd_pcm_hw_constraints_complete()' in 'sound/core/pcm_native.c'.
hw_interval_param: 0,0,0,0 019/023 PERIOD_TIME 0 0 [0 4294967295] 0 0 (166 4095875] hw_interval_param: 0,0,0,0 020/023 BUFFER_TIME 0 0 [0 4294967295] 0 0 (333 4096000] hw_interval_param: 0,0,0,0 021/023 PERIOD_SIZE 0 1 [16 32767] 0 1 [16 32766] hw_interval_param: 0,0,0,0 022/023 BUFFER_BYTES 0 1 [128 65536] 0 1 [128 65536] hw_interval_param: 0,0,0,0 023/023 RATE 0 0 [8000 96000] 0 0 [8000 96000] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044
Could you add these explanations in Documentation? The cover letter is gone at merging.
thanks,
Takashi
Hi,
On Jun 7 2017 14:42, Takashi Iwai wrote:
On Wed, 07 Jun 2017 01:46:42 +0200, Takashi Sakamoto wrote:
Hi,
This patchset is a part of my previous RFC and go for upstream. [alsa-devel] [PATCH RFC 00/21] ALSA: pcm: add tracepoints for PCM params operation http://mailman.alsa-project.org/pipermail/alsa-devel/2017-May/120548.html
In ALSA PCM interface, applications can get hardware capability by ioctl(2) with SNDRV_PCM_IOCTL_HW_REFINE/SNDRV_PCM_IOCT_HW_PARAMS in a shape of 'struct snd_pcm_hw_params'. In kernel side, relevant processing is somewhat complicated and developers sometimes have hard time to debug drivers for PCM constraints and rules. This patchset adds tracepoints to assist their work.
Changes:
- change print format
- rename events
- refactoring in the RFC will be done in my later work.
This patchset adds two events; 'hw_interval_param' and 'hw_mask_param' in 'snd_pcm' subsystem. When these events are probed, tracer outputs below lines:
hw_interval_param: 0,0,0,0 000/023 BUFFER_SIZE 0 0 [0 4294967295] 0 1 [0 4294967295] hw_interval_param: 0,0,0,0 000/023 BUFFER_BYTES 0 0 [0 4294967295] 0 1 [128 65536] hw_interval_param: 0,0,0,0 000/023 TICK_TIME 0 0 [0 4294967295] 0 0 [0 4294967295] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044 hw_interval_param: 0,0,0,0 002/023 SAMPLE_BITS 0 1 [0 4294967295] 0 1 [16 32] hw_interval_param: 0,0,0,0 003/023 SAMPLE_BITS 0 1 [16 32] 0 1 [16 32]
The first field represents PCM character device and subdevice for probed runtime of PCM substream. In the above case, '/dev/snd/pcmC0D0p' and first subdevice are used for runtime of PCM substream.
How about using snd_pcm_debug_name()? It'll be like "pcmC0D1c:2" and slightly more understandable than "0,1,1,2".
Yep. This looks a good idea, however for a reason I'm against it.
When mounting tracefs, you can see formats of the tracepoint. For example:
$ cat /sys/kernel/debug/tracing/events/snd_pcm/hw_mask_param/format ... format: field:int card: ... ... print fmt: "%d,%d,%d,%d %03d/%03d %s %08x%08x%08x%08x %08x%08x%08x%08x", ...
This means that structured data is passed to userspace in a shape of the 'format', then userspace application handle the data with the 'print fmt' for friendly printing.
If using strings generated by 'snd_pcm_debug_name()' for the printing, we cannot write it in 'TP_printk()' field, because format in this field just copied to userspace as is, as we just saw. It should be written in 'TP_fast_assign()' field in this case. This means that the structured data need to include u8 array for the string.
Of cource, we can program this tracepoint in your way. But to me this is not so convenient, because it loses flexibility for userspace applications to handle the structured data. The application should parse the u8 array as pre-formatted text to reuse its elements; e.g. subdevice. This is not so friendly in a point to write tracer program.
The second field represents id of each rule applied to the runtime, and the total number of rules added to the runtime. Basically, when the rule is applied to the runtime, these events are probed, but lines with id 0 are exceptions. They're application of constraints to the runtime.
The third field is name of parameter in the runtime. The rest fields depends on type of the parameter (mask/interval).
In ALSA PCM core, runtimes get 22 (or 21) rules as a default. In a below sample, the 21st rule is added by driver (snd-soc-imx-ssi.ko). The rest is added by snd-pcm.ko. In detail, please see 'snd_pcm_hw_constraints_init()' and 'snd_pcm_hw_constraints_complete()' in 'sound/core/pcm_native.c'.
hw_interval_param: 0,0,0,0 019/023 PERIOD_TIME 0 0 [0 4294967295] 0 0 (166 4095875] hw_interval_param: 0,0,0,0 020/023 BUFFER_TIME 0 0 [0 4294967295] 0 0 (333 4096000] hw_interval_param: 0,0,0,0 021/023 PERIOD_SIZE 0 1 [16 32767] 0 1 [16 32766] hw_interval_param: 0,0,0,0 022/023 BUFFER_BYTES 0 1 [128 65536] 0 1 [128 65536] hw_interval_param: 0,0,0,0 023/023 RATE 0 0 [8000 96000] 0 0 [8000 96000] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044
Could you add these explanations in Documentation? The cover letter is gone at merging.
Hm. Could I postpone the task for my later work in this development period? Text writing takes me a bit time, but I have patches for the rest of my RFC and firewire stack.
Regards
Takashi Sakamoto
On Wed, 07 Jun 2017 09:50:42 +0200, Takashi Sakamoto wrote:
Hi,
On Jun 7 2017 14:42, Takashi Iwai wrote:
On Wed, 07 Jun 2017 01:46:42 +0200, Takashi Sakamoto wrote:
Hi,
This patchset is a part of my previous RFC and go for upstream. [alsa-devel] [PATCH RFC 00/21] ALSA: pcm: add tracepoints for PCM params operation http://mailman.alsa-project.org/pipermail/alsa-devel/2017-May/120548.html
In ALSA PCM interface, applications can get hardware capability by ioctl(2) with SNDRV_PCM_IOCTL_HW_REFINE/SNDRV_PCM_IOCT_HW_PARAMS in a shape of 'struct snd_pcm_hw_params'. In kernel side, relevant processing is somewhat complicated and developers sometimes have hard time to debug drivers for PCM constraints and rules. This patchset adds tracepoints to assist their work.
Changes:
- change print format
- rename events
- refactoring in the RFC will be done in my later work.
This patchset adds two events; 'hw_interval_param' and 'hw_mask_param' in 'snd_pcm' subsystem. When these events are probed, tracer outputs below lines:
hw_interval_param: 0,0,0,0 000/023 BUFFER_SIZE 0 0 [0 4294967295] 0 1 [0 4294967295] hw_interval_param: 0,0,0,0 000/023 BUFFER_BYTES 0 0 [0 4294967295] 0 1 [128 65536] hw_interval_param: 0,0,0,0 000/023 TICK_TIME 0 0 [0 4294967295] 0 0 [0 4294967295] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044 hw_interval_param: 0,0,0,0 002/023 SAMPLE_BITS 0 1 [0 4294967295] 0 1 [16 32] hw_interval_param: 0,0,0,0 003/023 SAMPLE_BITS 0 1 [16 32] 0 1 [16 32]
The first field represents PCM character device and subdevice for probed runtime of PCM substream. In the above case, '/dev/snd/pcmC0D0p' and first subdevice are used for runtime of PCM substream.
How about using snd_pcm_debug_name()? It'll be like "pcmC0D1c:2" and slightly more understandable than "0,1,1,2".
Yep. This looks a good idea, however for a reason I'm against it.
When mounting tracefs, you can see formats of the tracepoint. For example:
$ cat /sys/kernel/debug/tracing/events/snd_pcm/hw_mask_param/format ... format: field:int card: ... ... print fmt: "%d,%d,%d,%d %03d/%03d %s %08x%08x%08x%08x %08x%08x%08x%08x", ...
This means that structured data is passed to userspace in a shape of the 'format', then userspace application handle the data with the 'print fmt' for friendly printing.
If using strings generated by 'snd_pcm_debug_name()' for the printing, we cannot write it in 'TP_printk()' field, because format in this field just copied to userspace as is, as we just saw. It should be written in 'TP_fast_assign()' field in this case. This means that the structured data need to include u8 array for the string.
Of cource, we can program this tracepoint in your way. But to me this is not so convenient, because it loses flexibility for userspace applications to handle the structured data. The application should parse the u8 array as pre-formatted text to reuse its elements; e.g. subdevice. This is not so friendly in a point to write tracer program.
Fair enough, we can keep as is then.
The second field represents id of each rule applied to the runtime, and the total number of rules added to the runtime. Basically, when the rule is applied to the runtime, these events are probed, but lines with id 0 are exceptions. They're application of constraints to the runtime.
The third field is name of parameter in the runtime. The rest fields depends on type of the parameter (mask/interval).
In ALSA PCM core, runtimes get 22 (or 21) rules as a default. In a below sample, the 21st rule is added by driver (snd-soc-imx-ssi.ko). The rest is added by snd-pcm.ko. In detail, please see 'snd_pcm_hw_constraints_init()' and 'snd_pcm_hw_constraints_complete()' in 'sound/core/pcm_native.c'.
hw_interval_param: 0,0,0,0 019/023 PERIOD_TIME 0 0 [0 4294967295] 0 0 (166 4095875] hw_interval_param: 0,0,0,0 020/023 BUFFER_TIME 0 0 [0 4294967295] 0 0 (333 4096000] hw_interval_param: 0,0,0,0 021/023 PERIOD_SIZE 0 1 [16 32767] 0 1 [16 32766] hw_interval_param: 0,0,0,0 022/023 BUFFER_BYTES 0 1 [128 65536] 0 1 [128 65536] hw_interval_param: 0,0,0,0 023/023 RATE 0 0 [8000 96000] 0 0 [8000 96000] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044
Could you add these explanations in Documentation? The cover letter is gone at merging.
Hm. Could I postpone the task for my later work in this development period? Text writing takes me a bit time, but I have patches for the rest of my RFC and firewire stack.
It's fine if you can promise to submit the document patch in this development cycle :)
thanks,
Takashi
On Ju 7 2017 17:09, Takashi Iwai wrote:
The second field represents id of each rule applied to the runtime, and the total number of rules added to the runtime. Basically, when the rule is applied to the runtime, these events are probed, but lines with id 0 are exceptions. They're application of constraints to the runtime.
The third field is name of parameter in the runtime. The rest fields depends on type of the parameter (mask/interval).
In ALSA PCM core, runtimes get 22 (or 21) rules as a default. In a below sample, the 21st rule is added by driver (snd-soc-imx-ssi.ko). The rest is added by snd-pcm.ko. In detail, please see 'snd_pcm_hw_constraints_init()' and 'snd_pcm_hw_constraints_complete()' in 'sound/core/pcm_native.c'.
hw_interval_param: 0,0,0,0 019/023 PERIOD_TIME 0 0 [0 4294967295] 0 0 (166 4095875] hw_interval_param: 0,0,0,0 020/023 BUFFER_TIME 0 0 [0 4294967295] 0 0 (333 4096000] hw_interval_param: 0,0,0,0 021/023 PERIOD_SIZE 0 1 [16 32767] 0 1 [16 32766] hw_interval_param: 0,0,0,0 022/023 BUFFER_BYTES 0 1 [128 65536] 0 1 [128 65536] hw_interval_param: 0,0,0,0 023/023 RATE 0 0 [8000 96000] 0 0 [8000 96000] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044
Could you add these explanations in Documentation? The cover letter is gone at merging.
Hm. Could I postpone the task for my later work in this development period? Text writing takes me a bit time, but I have patches for the rest of my RFC and firewire stack.
It's fine if you can promise to submit the document patch in this development cycle :)
I promise.
Regards
Takashi Sakamoto
On Wed, 07 Jun 2017 10:38:24 +0200, Takashi Sakamoto wrote:
On Ju 7 2017 17:09, Takashi Iwai wrote:
The second field represents id of each rule applied to the runtime, and the total number of rules added to the runtime. Basically, when the rule is applied to the runtime, these events are probed, but lines with id 0 are exceptions. They're application of constraints to the runtime.
The third field is name of parameter in the runtime. The rest fields depends on type of the parameter (mask/interval).
In ALSA PCM core, runtimes get 22 (or 21) rules as a default. In a below sample, the 21st rule is added by driver (snd-soc-imx-ssi.ko). The rest is added by snd-pcm.ko. In detail, please see 'snd_pcm_hw_constraints_init()' and 'snd_pcm_hw_constraints_complete()' in 'sound/core/pcm_native.c'.
hw_interval_param: 0,0,0,0 019/023 PERIOD_TIME 0 0 [0 4294967295] 0 0 (166 4095875] hw_interval_param: 0,0,0,0 020/023 BUFFER_TIME 0 0 [0 4294967295] 0 0 (333 4096000] hw_interval_param: 0,0,0,0 021/023 PERIOD_SIZE 0 1 [16 32767] 0 1 [16 32766] hw_interval_param: 0,0,0,0 022/023 BUFFER_BYTES 0 1 [128 65536] 0 1 [128 65536] hw_interval_param: 0,0,0,0 023/023 RATE 0 0 [8000 96000] 0 0 [8000 96000] hw_mask_param: 0,0,0,0 001/023 FORMAT 00000000000000000000001000000044 00000000000000000000001000000044
Could you add these explanations in Documentation? The cover letter is gone at merging.
Hm. Could I postpone the task for my later work in this development period? Text writing takes me a bit time, but I have patches for the rest of my RFC and firewire stack.
It's fine if you can promise to submit the document patch in this development cycle :)
I promise.
Good to hear, I merged your three patches now. Thanks!
Takashi
participants (2)
-
Takashi Iwai -
Takashi Sakamoto