Linux coresight arm64 : Incorrect data in cstrace.bin

[James Clark]

On 05/07/2023 10:56, Pandey, Radhey Shyam wrote:
>> -----Original Message-----
>> From: Suzuki K Poulose <suzuki.poulose at arm.com>
>> Sent: Wednesday, July 5, 2023 1:57 PM
>> To: Pandey, Radhey Shyam <radhey.shyam.pandey at amd.com>;
>> mathieu.poirier at linaro.org; mike.leach at linaro.org; leo.yan at linaro.org;
>> alexander.shishkin at linux.intel.com
>> Cc: coresight at lists.linaro.org; linux-arm-kernel at lists.infradead.org
>> Subject: Re: Linux coresight arm64 : Incorrect data in cstrace.bin
>>
>> Hi Radhe Shyam
>>
>> On 05/07/2023 05:38, Pandey, Radhey Shyam wrote:
>>> Hi,
>>>
>>> I am using linux 6.1 kernel coresight framework to capture ETM trace.
>>> Enabled coresight driver and added coresight component device node in
>>> DTS. With that, we could probe all coresight components.
>>>
>>> But when we do capture using sysfs and then read the trace.bin using
>>> ptm2human we see invalid trace data.  Trace content changes on each
>>> capture. Any pointers to help narrow down the issue?
>>>
>>> Tried: sysfs capture and perf (with OpenCSD enabled)
>>>
>>> Development board: VCK190 :
>>>   https://docs.xilinx.com/r/en-US/ug1366-vck190-eval-bd
>>> https://www.xilinx.com/support/documents/architecture-
>> manuals/am011-ve
>>> rsal-acap-trm.pdf
>>>
>>> xilinx-vck190-20231:/home/petalinux# dmesg | grep -i amba
>>> [    0.301054] Serial: AMBA PL011 UART driver
>>> [    0.306226] amba f0c20000.funnel: Fixing up cyclic dependency with
>> f0d70000.etm
>>> [    0.313624] amba f0c20000.funnel: Fixing up cyclic dependency with
>> f0d30000.etm
>>> [    0.321110] amba f0c30000.etf: Fixing up cyclic dependency with
>> f0c20000.funnel
>>> [    1.470247] Serial: AMBA driver
>>> xilinx-vck190-20231:/home/petalinux# dmesg | grep -i coresight
>>> [    1.865006]  cs_system_cfg: CoreSight Configuration manager initialised
>>> [    1.876858] coresight etm0: CPU0: etm v4.0 initialized
>>> [    1.885666] coresight etm1: CPU1: etm v4.0 initialized
>>> [    1.894435] coresight-cpu-debug f0d00000.debug1: Coresight debug-CPU0
>> initialized
>>> [    1.902072] coresight-cpu-debug f0d40000.debug1: Coresight debug-CPU1
>> initialized
>>>
>>> xilinx-vck190-20231:/home/petalinux# ls /sys/bus/coresight/devices/
>>> etm0      etm1      funnel0   tmc_etf0
>>>
>>>   cd /sys/bus/coresight/devices/
>>> echo 1 > tmc_etf0/enable_sink
>>> echo 1 > etm0/enable_source
>>> echo 0 > etm0/enable_source
>>> echo 0 > tmc_etf0/enable_sink
>>> cd /root/ dd if=/dev/tmc_etf0 of=cstrace_28Jun.bin
>>>
>>> ./ptm2human/ptm2human -e -i cstrace.bin -d
>>
>> Please note that ptm2human is for PTM trace decoding and
>> ETMv4 uses a different format and thus is not compatible.
>>
>>>
>>> I also tried OpenCSD integration with PERF.
>>
>> Have you made sure the perf is "linked" to the opencsd ?
> 
> Earlier I statically linked but after adding CORESIGHT=1.
> 
> 
> linux-xlnx$ make ARCH=arm64 NO_LIBELF=1 NO_JVMTI=1 VF=1 CORESIGHT=1 -C tools/perf/
> make: Entering directory 'linux-xlnx/tools/perf'
>   BUILD:   Doing 'make -j24' parallel build
> 
> Makefile.config:520: *** Error: No libopencsd library found or the version is not up-to-date. 
> Please install recent libopencsd to build with CORESIGHT=1.  Stop.
> 
> 
> echo $CSINCLUDES
> /scratch/development/coresight/my-opencsd/decoder/include
> radheys at xhdradheys41:/scratch/development/linux-xlnx$ ls /scratch/development/coresight/my-opencsd/decoder/include
> common  i_dec  interfaces  mem_acc  opencsd  opencsd.h  pkt_printers
> 
> echo $CSLIBS
> /scratch/development/coresight/my-opencsd/decoder/lib/builddir
> radheys at xhdradheys41:/scratch/development/linux-xlnx$ ls /scratch/development/coresight/my-opencsd/decoder/lib/builddir
> libopencsd.a        libopencsd_c_api.so    libopencsd_c_api.so.1.4.0  libopencsd.so.1
> libopencsd_c_api.a  libopencsd_c_api.so.1  libopencsd.so              libopencsd.so.1.4.0
> 
> Anything I am missing to fix this opencsd lib not found?

There is a minimum version of 1.1.1 but it looks like you are using
higher than that so that part should be fine.

Personally I "make install" OpenCSD to the system path. I had a play
around with $CSLIBS it and maybe there is also some stickyness to the
feature detection so a make clean might help to make sure the errors you
are seeing are real.

Lastly you can print the output of why the feature test compilation
failed which might help. I see some linking issues in there if
I delete my system installed version of OpenCSD, so there might be a bug
with just using CSINCLUDES and CSLIBS:

 $ cat  tools/build/feature/test-libopencsd.make.output

  /usr/bin/ld: warning: libopencsd.so.1, needed by
opencsd-local/decoder/lib/builddir//libopencsd_c_api.so, not found (try
using -rpath or -rpath-link)
  /usr/bin/ld: opencsd-local/decoder/lib/builddir//libopencsd_c_api.so:
undefined reference to `vtable for StmTrcPacket'
  /usr/bin/ld: opencsd-local/decoder/lib/builddir//libopencsd_c_api.so:
undefined reference to `DecodeTree::getDecoderStats(unsigned char,
_ocsd_decode_stats**)'
  /usr/bin/ld: opencsd-local/decoder/lib/builddir//libopencsd_c_api.so:
undefined reference to
`DecodeTree::addRawFramePrinter(RawFramePrinter**, unsigned int)'

Can you try the make install and see if that works? If that works but
the other way doesn't I can try looking into why its not and make a fix.

Thanks
James