[PATCH v5 0/7] arm64: Add ftrace support

[AKASHI Takahiro]

Hi,

On 03/11/2014 11:35 PM, Will Deacon wrote:
> Hi Akashi,
>
> On Fri, Feb 28, 2014 at 05:18:37AM +0000, AKASHI Takahiro wrote:
>> This patchset implements a function tracer on arm64.
>> There was another implementation from Cavium network, but both of us agreed
>> to use my patchset as future base. He is supposed to review this code, too.
>>
>> The only issue that I had some concern on was "fault protection" code
>> in prepare_ftrace_return(). With discussions with Steven and Tim (as author
>> of arm ftrace), I removed that code since I'm not quite sure about possibility
>> of "fault" occurrences in this function.
>>
>> The code is tested on ARMv8 Fast Model with the following tracers & events:
>>       function tracer with dynamic ftrace
>>       function graph tracer with dynamic ftrace
>>       syscall tracepoint
>>       irqsoff & preemptirqsoff (which use CALLER_ADDRx)
>> and also verified with in-kernel tests, FTRACE_SELFTEST, FTRACE_STARTUP_TEST
>> and EVENT_TRACE_TEST_SYSCALLS.
>>
>> Prerequisites are:
>>   * "arm64: Add regs_return_value() in syscall.h"
>>   * "arm64: make a single hook to syscall_trace() for all syscall features" patch
>
> It looks like there might be some more dependencies that that. Do you have a
> branch anywhere containing this series, along with all the dependencies so I
> can have a play?

I think you saw the following messages:

 >   HOSTCC  scripts/recordmcount
 > /.../linux-aarch64/scripts/recordmcount.c: In function 'do_file':
 > /.../linux-aarch64/scripts/recordmcount.c:350:7: error:
 > 'EM_AARCH64' undeclared (first use in this function)
 > /.../linux-aarch64/scripts/recordmcount.c:350:7: note:
 > each undeclared identifier is reported only once for each function it appears in
 > /.../linux-aarch64/scripts/recordmcount.c:351:15: error:
 > 'R_AARCH64_ABS64' undeclared (first > use in this function)
 > make[2]: *** [scripts/recordmcount] Error 1
 > make[1]: *** [scripts] Error 2
 > make: *** [sub-make] Error 2

This happens when a header file, elf.h, on "your host machine" does not have
definitions of EM_AARCH64 nor R_AARCH64_ABS64 because "recordmcount" is a binary
utility on host(x86), not target. It is very likely for most distros.
(I mentioned this in the cover letter, but the description might not be clear.)

Possible solutions are:
1) Define both macros directly in scripts/recordmcount.c
2) Use perl version of recordmcount

Currently I take 1), but it is a workaround.
2) should work with my current patch, too. But you need to remove HAVE_C_RECORDCOUNT
from arm64/Kconfig.

Which one do you prefer? I will include a fix for 1) in the next revision, anyway.

Thanks,
-Takahiro AKASHI

> Cheers,
>
> Will
>