How pidstat calculates CPU usage on each core and in total?

[wuzengshun]

Hi everyone,
I've been using pidstat recently, but I'm having some problems with the interpretation of the data, I'd like to get some answers, thanks!

1. What is the method about calculating the total CPU usage? I added up the usage on each core and found that it is not equal to the total usage shown by pidstat?

The table below:
The total %usr of pidstat is 43.14, but the %usr of each core adds up to 45.09
pidstat total %system is 15.69, but %system per core adds up to 15.68
pidstat total %CPU is 58.82, but %CPU per core adds up to 60.77

adb shell '/data/local/tmp/sysstat/pidstat 1 -p pidof -s %perf_proc_name% -r -d -t -u' > pidstat.txt

Linux 5.10.81-gki-g5cc049ba3e0a (localhost) 09/07/22 aarch64 (8 CPU)

	UID	TGID	TID	%usr	%system	%guest	%wait	%CPU	CPU	Command
16:09:23	10250	26897	-	43.14	15.69	0.00	2.94	58.82	6	droid.ugc.aweme
16:09:23	10250	-	26897	8.82	0.98	0	2.94	9.8	6	|__droid.ugc.aweme
16:09:23	10250	-	26912	0.98	0.98	0	0.98	1.96	3	|__Binder:26897_2
16:09:23	10250	-	26980	3.92	1.96	0	0	5.88	4	|__pool-3-thread-1
16:09:23	10250	-	27067	0.98	0	0	0	0.98	2	|__looper_monitor
16:09:23	10250	-	27068	0	0.98	0	0	0.98	1	|__looper_monitor
16:09:23	10250	-	27074	12.75	2.94	0	2.94	15.69	4	|__RenderThread
16:09:23	10250	-	27163	0.98	0.98	0	0	1.96	3	|__GPU completion
16:09:23	10250	-	27184	0.98	0	0	0	0.98	2	|__Binder:26897_5
16:09:23	10250	-	27289	0.98	0.98	0	0	96	2	|__motion_familiar
16:09:23	10250	-	27297	0	0.98	0	0	98	0	|__Reader-V1
16:09:23	10250	-	27304	2.94	0.98	0	1.96	3.92	4	|__MediaCodec_loop
16:09:23	10250	-	27305	0.98	0	0	0	0.98	3	|__CodecLooper
16:09:23	10250	-	27309	0.98	1.96	0	1.96	2.94	4	|_HwBinder:26897
16:09:23	10250	-	27330	0.98	0.98	0	0.98	1.96	3	|__VDecode-V1
16:09:23	10250	-	27331	2.94	0	0	0	2.94	2	|__ADecode-V1
16:09:23	10250	-	27332	0.98	0	0	0	0.98	3	|__AGraph-V1
16:09:23	10250	-	27333	1.96	0.98	0	0.98	2.94	3	|__VOutlet-V1
16:09:23	10250	-	27334	2.94	0	0	0.98	2.94	0	|__AOutlet-V1
16:09:23	10250	-	27984	0	0.98	0	0	0.98	1	|__Reader-V2
				45.09	15.68			60.77

2. What does the CPU id represent in the line of data about total usage (CPU 6 as shown below)? I'm guessing it doesn't mean the process is executing on this core.

	UID	TGID	TID	%usr	%system	%guest	%wait	%CPU	CPU	Command
16:09:23	10250	26897	-	43.14	15.69	0.00	2.94	58.82	6	droid.ugc.aweme

[wuzengshun]

3. In combination with mpstat, there is another rather strange phenomenon.
I see from mpstat that the CPU usage of core 4 at 16:09:23, %usr is 11.22.
But I use pidstat to count the CPU usage of process 'droid.ugc.aweme' in different threads on core 4 and add up 20.59. This figure is much higher than 11.2. Is this a bug?

adb shell "/data/local/tmp/sysstat/mpstat 1 -P ALL" > mpstat.txt
Linux 5.10.81-gki-g5cc049ba3e0a (localhost) 09/07/22 aarch64 (8 CPU)

16:09:22	CPU	%usr	%nice	%sys	%iowait	%irq	%soft	%steal	%guest	%gnice	%idle
16:09:23	all	9.53	2.16	8.01	0.00	1.78	0.13	0.00	0.00	0.00	78.40
16:09:23	0	14.74	0.00	12.63	0.00	4.21	0.00	0.00	0.00	0.00	68.42
16:09:23	1	10.00	0.00	16.00	0.00	3.00	1.00	0.00	0.00	0.00	70.00
16:09:23	2	15.62	0.00	11.46	0.00	2.08	0.00	0.00	0.00	0.00	70.83
16:09:23	3	11.96	0.00	8.70	0.00	3.26	0.00	0.00	0.00	0.00	76.09
16:09:23	4	11.22	0.00	6.12	0.00	0.00	0.00	0.00	0.00	0.00	82.65
16:09:23	5	4.85	4.85	0.97	0.00	0.97	0.00	0.00	0.00	0.00	88.35
16:09:23	6	8.65	6.73	7.69	0.00	0.96	0.00	0.00	0.00	0.00	75.96
16:09:23	7	0.00	5.05	1.01	0.00	0.00	0.00	0.00	0.00	0.00	93.94

[brofu]

This maybe helpful, for the 1st question. #344 (comment)
@wuzengshun