[BUG] Use report_sample.py --show_tracing_data to parse a record data failed

[jhldreams]

Description

Use kill -2 send SIGINT to to stop simpleperf monitoring and get the record data file. Then try to use report_sample.py -i $inputfile --show_tracing_data to parse the record data always get following error:

File "C:\Code\CamPerfTool\CameraDataGrab\bat\simpleperf\report_sample.py", line 107, in
main()
~~~~^^
File "C:\Code\CamPerfTool\CameraDataGrab\bat\simpleperf\report_sample.py", line 97, in main
report_sample(
~~~~~~~~~~~~~^
record_file=args.record_file,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...<3 lines>...
header=args.header,
^^^^^^^^^^^^^^^^^^^
report_lib_options=args.report_lib_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Code\CamPerfTool\CameraDataGrab\bat\simpleperf\report_sample.py", line 72, in report_sample
data = lib.GetTracingDataOfCurrentSample()
File "C:\Code\CamPerfTool\CameraDataGrab\bat\simpleperf\simpleperf_report_lib.py", line 502, in GetTracingDataOfCurrentSample
result[field.name] = field.parse_value(data)
~~~~~~~~~~~~~~~~~^^^^^^
File "C:\Code\CamPerfTool\CameraDataGrab\bat\simpleperf\simpleperf_report_lib.py", line 116, in parse_value
while length < max_len and bytes_to_str(data[offset + length]) != '\x00':
~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Code\CamPerfTool\CameraDataGrab\bat\simpleperf\simpleperf_utils.py", line 69, in bytes_to_str
return bytes_value.decode('utf-8')
~~~~~~~~~~~~~~~~~~^^^^^^^^^
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xba in position 0: invalid start byte

But the record data can be parsed success while using report_html.py or stackcollapse.py.
Is there any bug while use kill -2 to stop recording ?

I am using a supported NDK

• I have checked and the NDK I'm using is currently supported

Affected versions

r28

[pirama-arumuga-nainar]

@yabinc

[jhldreams]

Hi, anyone can help with this problem ? I see the simpleperf readme.md show it supports SIGINT:

https://android.googlesource.com/platform/system/extras/+/refs/heads/main/simpleperf/doc/executable_commands_reference.md#set-the-path-to-store-profiling-data
"If you want to write a script to control how long to monitor, you can send one of SIGINT, SIGTERM, SIGHUP signals to simpleperf to stop monitoring"

by the way, my recording data has some tracing_data frome customize tracepoint. The information demo is below:

tracing data:
	common_type : 749
	common_flags : 0
	common_preempt_count : 1
	common_pid : 15826
	ino : 50522
	s_name : u
	s_exp_name : qcom,system
	size : 299008

[yabinc]

Maybe some non utf8 string in a tracepoint field (having type __data_loc char[])?

We can skip converting bytes to str in this case, with the following code modification:

         if self.is_dynamic:
             offset, max_len = struct.unpack('<HH', data[self.offset:self.offset + 4])
-            length = 0
-            while length < max_len and bytes_to_str(data[offset + length]) != '\x00':
-                length += 1
-            return bytes_to_str(data[offset: offset + length])
+            try:
+                length = 0
+                while length < max_len and bytes_to_str(data[offset + length]) != '\x00':
+                    length += 1
+                return bytes_to_str(data[offset: offset + length])
+            except:
+                return data[offset: offset + max_len]

I will make the change in the next NDK release. But you can apply the change locally.

[DanAlbert]

@yabinc next meaning r29 beta 2, which is coming up soon, or r30? Either's fine with me.

[jhldreams]

@yabinc
I have tried to apply the change locally, it's can parse without errors, But the parse result will have some unrecoginize characters.

and by the way ,I find this is a probabilistic event。 If I try to use report_sample.py to parse a same perf data several times. It might be success . But the success file content might has following error size, for example, the kgsl_mem_alloc event:

tracing data:
	common_type : 1757
	common_flags : 0
	common_preempt_count : 1
	common_pid : 318767123
	gpuaddr : 14926529573000726898
	size : 14907443494250349119
	tgid : 575
	usage : VK/others( 37)
	id : 23
	flags : 2366383360

this total tracing data content might be parsed fail. the gpuaddr or the size value is wrong. The correct data is like this :
tracing data:
common_type : 1757
common_flags : 0
common_preempt_count : 1
common_pid : 9338
gpuaddr : 275615117312
size : 4096
tgid : 9218
usage : VK/others( 37)
id : 24
flags : 2366383360

[jhldreams]

Hi @yabinc

I tried simpleperf tool in ndk version:25.1.8937393. It works very well. So the problem might be newly simpleperf tool's problem.

[yabinc]

Since it's a probabilistic event, I can't guess where is wrong. Could you upload a perf.data that can reproduce this problem?

[jhldreams]

@yabinc
here is a perf record data you could use report_sample.py with "--show_tracing_data" to parse and reproduce this problem.

perf.zip