Hi,
We are trying to us the Open CSD for decoding a onchip trace in our ETB. The trace was enabled and is captured in the ETB. We read the trace back and dumped it into a text file. I am attaching it.
How can I use the CSD tool to decode it?
Thanks Ajith
Hi Ajith,
OpenCSD is a trace decode library that is designed to be integrated into a tool to decode trace. It is not a trace decode tool itself.
You have two choices:-
1) Write a program to use the library and decode the trace. The 'trc_pkt_lister' test program and source [decoder/tests/source/trc_pkt_lister.cpp] is a good example of how to use the library. Use the decode tree object / api to build a decoder for your configured system.
2) Limited decode can be obtained if your data is presented as a trace snapshot to the 'trc_pkt_lister' program - this will then do some decoding of your trace. The format specification is defined in decoder/docs/specs/"ARM Trace and Debug Sanpshot file format 0v2.pdf" Example snapshots are seen in decoder/tests/snapshots. The decode from this program is limited to outputting either raw trace packets, or executed instruction ranges. Additional work will be required to provide disassembly / source code attribution of the decoded trace.
Either way you will need the following:- a) Trace data in a binary file. b) configuration information for the ETM hardware at the time of trace capture. c) Any memory images of programs executed while capturing the trace data.
I advise building the documentation for the library using the doxygen file in decoder/docs for additional information.
Regards
Mike
On 15 February 2018 at 00:51, Ajith Kurian Issac ajith-kurian.issac@broadcom.com wrote:
Hi Mike,
We tried using Open CSD and we got something like this after decoding the trace . Does the output makes sense for you? The branch address mentioned in the trace is actually valid. Could you please help
Thanks Ajith
On Thu, Feb 15, 2018 at 7:55 AM, Mike Leach mike.leach@linaro.org wrote:
Hi Ajith,
Can you give me a bit more context on your system - core in use, trace protocol type.
Also can you upload the raw trace binary, and the snapshot config files you used - I'd like to see what the CS frames look like.
The trace looks unusual in that you seem to have repeating E/N atoms - which could be the result of running a loop. What does the traced application do? Is it a T32 app as suggested by the address outputs? I sometimes expect multiple atoms in a single packet - but in this case the cycle counts are forcing a single atom output per packet. Have you tried the -decode option and providing the memory image of the application to the decode?
This is consistent with valid trace - but without knowing what the app is doing it is not possible to say for certain.
Regards
Mike On Thu, 20 Sep 2018 at 23:58, Ajith Kurian Issac ajith-kurian.issac@broadcom.com wrote:
Hi Vinu, Mike can help us to clear our basic doubts on this. Could you please answer his queries?
On Fri, Sep 21, 2018, 1:18 PM Mike Leach mike.leach@linaro.org wrote:
Hi Mike,
We rea using an A15 dual-core processor with trace enabled only for Core0 We use PTM protocol from ARM. We rea running bare metal code no OS . We are trying to validate the trace by inducing an abort in the code .
The config details Vinu can share
Thanks Ajith
On Fri, Sep 21, 2018 at 2:19 PM, Ajith Kurian Issac < ajith-kurian.issac@broadcom.com> wrote:
Hi,
Attached 7z file has all the .ini files we use with OpenCSD. I have attached a sample “tracebin.bin” file collected from the controller.
Thanks,
Vinu
From: Ajith Kurian Issac ajith-kurian.issac@broadcom.com Date: Monday, September 24, 2018 at 10:25 AM To: Mike Leach mike.leach@linaro.org, Vinu Velayudhan vinu.velayudhan@broadcom.com Cc: coresight@lists.linaro.org Subject: Re: Open CSD
Hi Mike,
We rea using an A15 dual-core processor with trace enabled only for Core0
We use PTM protocol from ARM.
We rea running bare metal code no OS .
We are trying to validate the trace by inducing an abort in the code .
The config details Vinu can share
Thanks
Ajith
On Fri, Sep 21, 2018 at 2:19 PM, Ajith Kurian Issac ajith-kurian.issac@broadcom.com wrote:
Hi Vinu,
Mike can help us to clear our basic doubts on this. Could you please answer his queries?
On Fri, Sep 21, 2018, 1:18 PM Mike Leach mike.leach@linaro.org wrote:
Hi Ajith,
Can you give me a bit more context on your system - core in use, trace protocol type.
Also can you upload the raw trace binary, and the snapshot config files you used - I'd like to see what the CS frames look like.
The trace looks unusual in that you seem to have repeating E/N atoms - which could be the result of running a loop. What does the traced application do? Is it a T32 app as suggested by the address outputs? I sometimes expect multiple atoms in a single packet - but in this case the cycle counts are forcing a single atom output per packet. Have you tried the -decode option and providing the memory image of the application to the decode?
This is consistent with valid trace - but without knowing what the app is doing it is not possible to say for certain.
Regards
Mike On Thu, 20 Sep 2018 at 23:58, Ajith Kurian Issac ajith-kurian.issac@broadcom.com wrote:
Hi Vinu,
As Mike says, to decode to an instruction trace, you will need to give the decoder the memory image(s) of the code being executed.
I have run this trace through another PTM decoder and it’s not flagging up anything that looks problematic. The code appears to be looping over code around 0x1200000, 0x7f800000 and 0xf4000000. To show details of instructions you’d need to provide the decoder with the code located in these areas. If you could send the image (e.g. ELF file) we could check the decode here.
Al
From: CoreSight coresight-bounces@lists.linaro.org On Behalf Of Vinu Velayudhan Sent: 24 September 2018 18:48 To: Ajith Kurian Issac ajith-kurian.issac@broadcom.com; Mike Leach mike.leach@linaro.org Cc: coresight@lists.linaro.org Subject: Re: Open CSD
Attaching the files this time. zip file attachment was blocked by the mail server last time.
Thanks, Vinu
On Mon, Sep 24, 2018 at 10:42 AM Vinu Velayudhan <vinu.velayudhan@broadcom.commailto:vinu.velayudhan@broadcom.com> wrote: Hi,
Attached 7z file has all the .ini files we use with OpenCSD. I have attached a sample “tracebin.bin” file collected from the controller.
Thanks, Vinu
From: Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com> Date: Monday, September 24, 2018 at 10:25 AM To: Mike Leach <mike.leach@linaro.orgmailto:mike.leach@linaro.org>, Vinu Velayudhan <vinu.velayudhan@broadcom.commailto:vinu.velayudhan@broadcom.com> Cc: <coresight@lists.linaro.orgmailto:coresight@lists.linaro.org> Subject: Re: Open CSD
Hi Mike,
We rea using an A15 dual-core processor with trace enabled only for Core0 We use PTM protocol from ARM. We rea running bare metal code no OS . We are trying to validate the trace by inducing an abort in the code .
The config details Vinu can share
Thanks Ajith
On Fri, Sep 21, 2018 at 2:19 PM, Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com> wrote: Hi Vinu, Mike can help us to clear our basic doubts on this. Could you please answer his queries?
On Fri, Sep 21, 2018, 1:18 PM Mike Leach <mike.leach@linaro.orgmailto:mike.leach@linaro.org> wrote: Hi Ajith,
Can you give me a bit more context on your system - core in use, trace protocol type.
Also can you upload the raw trace binary, and the snapshot config files you used - I'd like to see what the CS frames look like.
The trace looks unusual in that you seem to have repeating E/N atoms - which could be the result of running a loop. What does the traced application do? Is it a T32 app as suggested by the address outputs? I sometimes expect multiple atoms in a single packet - but in this case the cycle counts are forcing a single atom output per packet. Have you tried the -decode option and providing the memory image of the application to the decode?
This is consistent with valid trace - but without knowing what the app is doing it is not possible to say for certain.
Regards
Mike On Thu, 20 Sep 2018 at 23:58, Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com> wrote:
-- Mike Leach Principal Engineer, ARM Ltd. Manchester Design Centre. UK
Thanks AI,
We will try it out we saw a lot of below lines getting repeated intially in the trace. Do you also see the same? Idx:822; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:824; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:826; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:828; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:830; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:833; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:835; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:837; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:839; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:841; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:843; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:845; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:848; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:850; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:852; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:854; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:856; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:858; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:860; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:864; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:866; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:868; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:870; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:872; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:874; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:876; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:878; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:881; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:883; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:885; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:887; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:889; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:891; ID:1a; ATOM : Atom packet; N; Cycles=114;
Thanks Ajith
On Mon, Sep 24, 2018 at 2:28 PM, Al Grant Al.Grant@arm.com wrote:
Yes I see the same. Normally, I would not expect to see trace this regular, but for an artificial test case, run bare-metal from reset, it may be correct.
** It’s impossible to know whether this trace is correct or not without seeing the code image. **
If you could disassemble the code around 0x1200766, it will be possible to see whether this is the trace we would expect.
Al
From: Ajith Kurian Issac ajith-kurian.issac@broadcom.com Sent: 24 September 2018 21:33 To: Al Grant Al.Grant@arm.com Cc: Vinu Velayudhan vinu.velayudhan@broadcom.com; Mike Leach mike.leach@linaro.org; coresight@lists.linaro.org; nd nd@arm.com Subject: Re: Open CSD
Thanks AI,
We will try it out we saw a lot of below lines getting repeated intially in the trace. Do you also see the same? Idx:822; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:824; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:826; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:828; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:830; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:833; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:835; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:837; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:839; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:841; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:843; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:845; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:848; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:850; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:852; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:854; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:856; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:858; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:860; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:864; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:866; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:868; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:870; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:872; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:874; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:876; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:878; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:881; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:883; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:885; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:887; ID:1a; ATOM : Atom packet; N; Cycles=114; Idx:889; ID:1a; ATOM : Atom packet; E; Cycles=114; Idx:891; ID:1a; ATOM : Atom packet; N; Cycles=114;
Thanks Ajith
On Mon, Sep 24, 2018 at 2:28 PM, Al Grant <Al.Grant@arm.commailto:Al.Grant@arm.com> wrote: Hi Vinu,
As Mike says, to decode to an instruction trace, you will need to give the decoder the memory image(s) of the code being executed.
I have run this trace through another PTM decoder and it’s not flagging up anything that looks problematic. The code appears to be looping over code around 0x1200000, 0x7f800000 and 0xf4000000. To show details of instructions you’d need to provide the decoder with the code located in these areas. If you could send the image (e.g. ELF file) we could check the decode here.
Al
From: CoreSight <coresight-bounces@lists.linaro.orgmailto:coresight-bounces@lists.linaro.org> On Behalf Of Vinu Velayudhan Sent: 24 September 2018 18:48 To: Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com>; Mike Leach <mike.leach@linaro.orgmailto:mike.leach@linaro.org>
Cc: coresight@lists.linaro.orgmailto:coresight@lists.linaro.org Subject: Re: Open CSD
Attaching the files this time. zip file attachment was blocked by the mail server last time.
Thanks, Vinu
On Mon, Sep 24, 2018 at 10:42 AM Vinu Velayudhan <vinu.velayudhan@broadcom.commailto:vinu.velayudhan@broadcom.com> wrote: Hi,
Attached 7z file has all the .ini files we use with OpenCSD. I have attached a sample “tracebin.bin” file collected from the controller.
Thanks, Vinu
From: Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com> Date: Monday, September 24, 2018 at 10:25 AM To: Mike Leach <mike.leach@linaro.orgmailto:mike.leach@linaro.org>, Vinu Velayudhan <vinu.velayudhan@broadcom.commailto:vinu.velayudhan@broadcom.com> Cc: <coresight@lists.linaro.orgmailto:coresight@lists.linaro.org> Subject: Re: Open CSD
Hi Mike,
We rea using an A15 dual-core processor with trace enabled only for Core0 We use PTM protocol from ARM. We rea running bare metal code no OS . We are trying to validate the trace by inducing an abort in the code .
The config details Vinu can share
Thanks Ajith
On Fri, Sep 21, 2018 at 2:19 PM, Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com> wrote: Hi Vinu, Mike can help us to clear our basic doubts on this. Could you please answer his queries?
On Fri, Sep 21, 2018, 1:18 PM Mike Leach <mike.leach@linaro.orgmailto:mike.leach@linaro.org> wrote: Hi Ajith,
Can you give me a bit more context on your system - core in use, trace protocol type.
Also can you upload the raw trace binary, and the snapshot config files you used - I'd like to see what the CS frames look like.
The trace looks unusual in that you seem to have repeating E/N atoms - which could be the result of running a loop. What does the traced application do? Is it a T32 app as suggested by the address outputs? I sometimes expect multiple atoms in a single packet - but in this case the cycle counts are forcing a single atom output per packet. Have you tried the -decode option and providing the memory image of the application to the decode?
This is consistent with valid trace - but without knowing what the app is doing it is not possible to say for certain.
Regards
Mike On Thu, 20 Sep 2018 at 23:58, Ajith Kurian Issac <ajith-kurian.issac@broadcom.commailto:ajith-kurian.issac@broadcom.com> wrote:
-- Mike Leach Principal Engineer, ARM Ltd. Manchester Design Centre. UK
Hi Ajith,
I've looked at the output from the tracedump bin you provided and am seeing the same as you. I looked at the raw trace bytes too - there is nothing in the trace that would indicate the trace is incorrect or corrupted.
One point of note - the timestamp count is not increasing - this may be due to the clock driving it not being active.
Regards
Mike On Mon, 24 Sep 2018 at 21:32, Ajith Kurian Issac ajith-kurian.issac@broadcom.com wrote:
Hi Mike,
Do you how we can feed the ELF file to the trace decoder? whats the command to do that? Is it ELF or axf file? Thanks Ajith
On Tue, Sep 25, 2018 at 12:05 PM Mike Leach mike.leach@linaro.org wrote:
Hi Ajith,
You need to add a 'dump' section into the cpu .ini file (cpu_0.ini).
e.g. [dump0] file=afile.bin address=0x8000 length=0x2000 offset=0x100
Where: file: name of file address: address in memory that the contents of file are loaded length: [optional] length of memory in file (less than or equal to file length) offset: [optional] offset within the file that the memory block starts at.
So, for an elf/axf file you need to establish the offset into the file of the loadable code section, and its length within the file and specify these as length and offset parameters above. Address is obviously the load address for the section.
See decoder/docs/specs for the snapshot file specification document.
Most of the example snapshots in decoder\tests use plain binary memory dumps where length is the length of the file, and the start of memory is the beginning of the file so do not use offset and length.
Regards
Mike On Fri, 28 Sep 2018 at 17:44, Ajith Kurian Issac ajith-kurian.issac@broadcom.com wrote:
Hi Ajith,
I was trying to create an example snapshot to demo use of the offset parameter in [dump] sections in .ini file when I discovered a bug in the snapshot reader library used by the test program than made it not use these correctly. This is now fixed in v0.9.3 of the library. There is a new test snapshot there - tests/snapshots/test-file-mem-offsets which demonstrates use of 'offset'
Regards
Mike On Mon, 1 Oct 2018 at 11:47, Mike Leach mike.leach@linaro.org wrote:
Thanks Mike!!! we will try it Thanks Ajith
On Wed, Oct 3, 2018 at 3:23 PM Mike Leach mike.leach@linaro.org wrote:
Adding Karthik,
Karthik, Could you please explain the issue you see now with the list file?
Thanks Ajith
On Wed, Oct 3, 2018 at 3:23 PM Mike Leach mike.leach@linaro.org wrote:
Hi Mike/Ajith, We are able to specify multiple dump sections and decode the trace successfully. Will let you guys know in case we encounter any further issues.
Thanks, Karthik
On Tue, Oct 16, 2018 at 1:49 PM Ajith Kurian Issac < ajith-kurian.issac@broadcom.com> wrote:
Hi Mike, We have a few additional queries regarding the output file. Copying a snippet from the decoded trace output - .... .... Idx:1189; ID:1a; [0x84 ]; ATOM : Atom packet; E; Cycles=1; Idx:1189; ID:1a; OCSD_GEN_TRC_ELEM_INSTR_RANGE(exec range=0x1201dce:[0x1201dd4] num_i(3) last_sz(2) (ISA=T32) E BR [CC=1]; ) .... ....
- Is there any way to generate the instructions itself that correspond to the above execution range? i.e. can we specify a map file/disassembly which automatically generates the instructions corresponding to the above packet? - Is there any way to stop the trace and resume the trace later without loosing the trace collected before stopping?
Thanks, Karthik
On Thu, Oct 18, 2018 at 12:11 PM Karthik Raghavan < karthik.raghavan@broadcom.com> wrote:
Hi Karthik, On Fri, 19 Oct 2018 at 00:48, Karthik Raghavan karthik.raghavan@broadcom.com wrote:
Disassembly is not a function of the opencsd library. We recommend using a disassembler supplied with your chosen compiler. For example, when using perf script under linux, we supply the address information from the INSTR_RANGE packet into objdump from the gcc toolset.
Is there any way to stop the trace and resume the trace later without loosing the trace collected before stopping?
This would depend on the mechanics and capacity of your CoreSight system. In your case, for bare metal ETB it is possible to stop trace collection - as long as the trace system is flushed on halt. Any trace would be retained in the ETB, and tracing could then restart at the current buffer pointers. However, trace can be lost if the total amount collected exceeds the buffer capacity and the buffer wraps.
Regards
Mike
-
Ajith Kurian Issac -
Al Grant -
Karthik Raghavan -
Mike Leach -
Vinu Velayudhan