os/pm: Add PM Thread Metrics #6286

gSahitya-samsung · 2024-07-10T09:55:53Z

This commit extends the functionality of pm_metrics. Now PM Metrics show the thread specific pm metrics. It shows which process causes minimum actual sleep time (in ms) and how frequently the thread wakes up the board.

gSahitya-samsung · 2024-07-10T09:58:42Z

TASH>>pm_metrics 60000
pm_print_metrics: 
pm_print_metrics: TOTAL METRICS TIME [1] = 60001ms
pm_print_metrics: TOTAL SLEEP TRY TIME [2] = 26150ms
pm_print_metrics: *[2] = represents total duration of time, PM tries to make board sleep.
pm_print_metrics: 
pm_print_metrics: 
pm_print_metrics:               DOMAIN              | TOTAL PM SUSPEND TIME [3] | TOTAL SLEEP BLOCKING TIME [4] 
pm_print_metrics: ----------------------------------|---------------------------|-------------------------------
pm_print_metrics:                              IDLE |         13940ms ( 23.23%) |             13940ms ( 53.31%) 
pm_print_metrics:                            SCREEN |             0ms (  0.00%) |                 0ms (  0.00%) 
pm_print_metrics:                              UART |             0ms (  0.00%) |                 0ms (  0.00%) 
pm_print_metrics:                               SPI |             0ms (  0.00%) |                 0ms (  0.00%) 
pm_print_metrics:                              TASH |          8770ms ( 14.62%) |              8770ms ( 33.54%) 
pm_print_metrics: 
pm_print_metrics: *[3] = total time pm domain was suspended.
pm_print_metrics: *[4] = represents the time board was not able to sleep due to suspended domain.
pm_print_metrics: *[3] (%) = [3] / [1] * 100
pm_print_metrics: *[4] (%) = [4] / [2] * 100
pm_print_metrics: 
pm_print_metrics: 
pm_print_metrics:  WAKEUP SOURCES | COUNTS 
pm_print_metrics: ----------------|--------
pm_print_metrics:         UNKNOWN |      0 
pm_print_metrics:             BLE |      0 
pm_print_metrics:            WIFI |      0 
pm_print_metrics:    UART CONSOLE |     11 
pm_print_metrics:      UART TTYS2 |      0 
pm_print_metrics:            GPIO |      0 
pm_print_metrics:        HW TIMER |     53 
pm_print_metrics: 
pm_print_metrics: 
pm_print_metrics:  PROCESS NAME | WAKEUP COUNTS | MIN SLEEP TIME 
pm_print_metrics: --------------|---------------|----------------
pm_print_metrics:      pm_sleep |            29 |           42ms 
pm_print_metrics:      pm_sleep |            10 |           32ms 
pm_print_metrics:      pm_sleep |             9 |          596ms 
pm_print_metrics:      pm_sleep |             3 |           57ms 
pm_print_metrics:      pm_sleep |             2 |          665ms 
pm_print_metrics: 
pm_print_metrics: 
pm_print_metrics:  BOARD STATE | PM STATE |          TIME          
pm_print_metrics: -------------|----------|------------------------
pm_print_metrics:       WAKEUP |   NORMAL |      13941ms ( 23.23%) 
pm_print_metrics:              |     IDLE |          0ms (  0.00%) 
pm_print_metrics:              |  STANDBY |      12211ms ( 20.35%) 
pm_print_metrics: -------------|----------|------------------------
pm_print_metrics:        SLEEP |    SLEEP |      33849ms ( 56.41%)

os/kernel/task/task_recover.c

os/include/tinyara/pm/pm.h

os/kernel/task/task_recover.c

os/pm/pm.h

os/pm/pm_metrics.c

os/pm/pm_sleep.c

edwakuwaku · 2024-07-10T10:43:10Z

os/pm/pm_metrics.c

+		}
+	}
+	pmdbg("\n");
+	pmdbg("\n");
 	pmdbg(" BOARD STATE | PM STATE |          TIME          \n");
 	pmdbg("-------------|----------|------------------------\n");
 	pmdbg(" %11s | %8s | %10dms (%6.2f%%) \n", "WAKEUP", "NORMAL", TICK2MSEC(g_pm_metrics->state_metrics.state_accum_ticks[0]),\


Suggest to change "WAKEUP" -> "ACTIVE"

pmdbg(" %11s | %8s | %10dms (%6.2f%%) \n", "SLEEP", "SLEEP", TICK2MSEC(g_pm_metrics->board_sleep_ticks),\ ((double)g_pm_metrics->board_sleep_ticks) * 100.0 / total_time);

Suggest to change "SLEEP" -> "INACTIVE"

So, system active corresponds one of 3 PM states.
System inactive corresponds to PM_SLEEP

r-prabu · 2024-07-10T10:48:38Z

os/include/tinyara/pm/pm.h

+ *
+ * Inputs:
+ *   tcb - The TCB of the terminated task or thread
+ *


@gSahitya-samsung use code-i feature to generate the doxzgen documentation for the interfaces and update the description if it not suiting better.(/docstring) which is very productive.

r-prabu · 2024-07-10T10:53:23Z

os/pm/pm_wakehandler.c

@@ -56,7 +56,7 @@ void pm_wakehandler(clock_t missing_tick, pm_wakeup_reason_code_t wakeup_src)
 {
 	irqstate_t flags = enter_critical_section();
 #ifdef CONFIG_PM_METRICS
-	pm_metrics_update_wakehandler(missing_tick, wakeup_src);
+	pm_metrics_update_wakehandler((uint32_t)missing_tick, wakeup_src);


why do missing_tick is typecased to uint32_t?

The reason is missing_tick is of type uint64 while pm_metrics support only uint32.

We need to organize the type in pm_metrics.

please check and let me know the maximum time supported by pm_metrics.

You can limit all actions to the maximum time and change the data type correctly.

@ewoodev We can scale pm_metrics to UINT64_MAX

I mean, currently our pm_metrics use pm_sleep. This is the maximum int max of delay.
Therefore, the pm_metrics input is int max tick.

is it necessary for the count and ticks of pm_metrics to maintain int max tick?

Yes, but it is highly unlikely board will sleep continuously for INT32_MAX i.e. 24 days.Since we have UART and Wifi Interrupts, it guarantee the board will wakeup at least one time within 24 days sleep. So we can scale our pm_metrics to UINT64_MAX.

os/kernel/task/task_recover.c

ewoodev · 2024-07-11T01:53:22Z

os/pm/pm_wakehandler.c

@@ -56,7 +56,7 @@ void pm_wakehandler(clock_t missing_tick, pm_wakeup_reason_code_t wakeup_src)
 {
 	irqstate_t flags = enter_critical_section();
 #ifdef CONFIG_PM_METRICS
-	pm_metrics_update_wakehandler(missing_tick, wakeup_src);
+	pm_metrics_update_wakehandler((uint32_t)missing_tick, wakeup_src);


We need to organize the type in pm_metrics.

please check and let me know the maximum time supported by pm_metrics.

You can limit all actions to the maximum time and change the data type correctly.

os/pm/pm_metrics.c

ewoodev · 2024-07-11T01:57:54Z

os/pm/pm_recover.c

+void pm_recover(FAR struct tcb_s *tcb)
+{
+#ifdef CONFIG_PM_METRICS
+    pm_metrics_update_recover(tcb->pid);


we can task exited using PIDHASH is same

For only pm_metrics, do we need to add operation in task exit?

Yes, but in future we can use pm_recover for other pm cleanup activities which may be require after task termination.

I think, when add new future, add this also is better.

for pm_metrics side, It seems better to keep the data until the next thread is created.

If we do this way, then we may need to keep thread_name in memory. Can we do this?

This commit extends the functionality of pm_metrics. Now PM Metrics show the thread specific pm metrics. It shows which process causes minimum actual sleep time (in ms) and how frequently the thread wakes up the board.

ewoodev · 2024-07-15T14:06:53Z

os/pm/pm_metrics.c

+	int hash_pid;
+	if (g_pm_metrics_running) {
+		hash_pid = PIDHASH(pid);
+		g_pm_metrics->thread_metrics.min_sleep_ticks[hash_pid] = UINT32_MAX;


We can modify this data when the next pid comes in.
For this, do not add action to task end.

ewoodev · 2024-07-15T14:09:22Z

os/pm/pm_sleep.c

 {
+#ifdef CONFIG_PM_METRICS
+	pm_metrics_update_sleep((pid_t)pid);


please find wd_setwakeupsource()

we can use timer as wakeup source using api rather than pm_sleep.
Could you consider how to add this?

Currently whole pm_metrics logic is inside PM module, if we want to move logic into wd_setwakeupsource then we need to change pm interface and give pm_metrics API.

kishore-sn · 2024-07-17T10:38:32Z

os/pm/pm_metrics.c

@@ -80,6 +91,17 @@ static void pm_print_metrics(double total_time, int n_domains)
 	for (index = 0; index < PM_WAKEUP_SRC_COUNT; index++) {
 		pmdbg(" %14s | %6d \n", wakeup_src_name[index], g_pm_metrics->wakeup_src_counts[index]);
 	}
+	if (g_pm_metrics->wakeup_src_counts[PM_WAKEUP_HW_TIMER]) {
+		pmdbg("\n");


How about adding new definition for pmdbg_noarg (based on dbg_noarg) and use that to print these mertrics? When we are printing a table, then we dont need the api name at the beginning of each line of the table.

gSahitya-samsung force-pushed the pm_met branch from 87675bc to 61ae141 Compare July 10, 2024 10:00

abhishek-samsung reviewed Jul 10, 2024

View reviewed changes

os/kernel/task/task_recover.c Show resolved Hide resolved

kishore-sn reviewed Jul 10, 2024

View reviewed changes

os/include/tinyara/pm/pm.h Outdated Show resolved Hide resolved

os/kernel/task/task_recover.c Outdated Show resolved Hide resolved

os/pm/pm.h Outdated Show resolved Hide resolved

os/pm/pm_metrics.c Show resolved Hide resolved

os/pm/pm_sleep.c Outdated Show resolved Hide resolved

edwakuwaku reviewed Jul 10, 2024

View reviewed changes

r-prabu reviewed Jul 10, 2024

View reviewed changes

ewoodev reviewed Jul 11, 2024

View reviewed changes

os/pm: Add PM Thread Metrics

4c5fe45

This commit extends the functionality of pm_metrics. Now PM Metrics show the thread specific pm metrics. It shows which process causes minimum actual sleep time (in ms) and how frequently the thread wakes up the board.

gSahitya-samsung force-pushed the pm_met branch from 61ae141 to 4c5fe45 Compare July 11, 2024 03:13

ewoodev reviewed Jul 15, 2024

View reviewed changes

kishore-sn reviewed Jul 17, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

os/pm: Add PM Thread Metrics #6286

os/pm: Add PM Thread Metrics #6286

gSahitya-samsung commented Jul 10, 2024

gSahitya-samsung commented Jul 10, 2024

edwakuwaku Jul 10, 2024

edwakuwaku Jul 10, 2024

edwakuwaku Jul 10, 2024

r-prabu Jul 10, 2024

r-prabu Jul 10, 2024

gSahitya-samsung Jul 11, 2024

ewoodev Jul 11, 2024

gSahitya-samsung Jul 11, 2024

ewoodev Jul 15, 2024

gSahitya-samsung Jul 16, 2024

ewoodev Jul 11, 2024

ewoodev Jul 11, 2024

gSahitya-samsung Jul 11, 2024

ewoodev Jul 11, 2024

gSahitya-samsung Jul 16, 2024

ewoodev Jul 15, 2024

ewoodev Jul 15, 2024

gSahitya-samsung Jul 16, 2024

kishore-sn Jul 17, 2024

os/pm: Add PM Thread Metrics #6286

Are you sure you want to change the base?

os/pm: Add PM Thread Metrics #6286

Conversation

gSahitya-samsung commented Jul 10, 2024

gSahitya-samsung commented Jul 10, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment