unit-conversions SIGSEGV on armv7hl

[dkopecek]

Hello,
when trying to build and run the test suite on armv7hl architecture, I get the following results:

make[1]: Leaving directory '/builddir/build/BUILD/json-2.0.3/test'
test/json_unit "*"
json_unit is a Catch v1.5.6 host application.

Run with -? for options

value conversion
  get an object (explicit)

src/unit-conversions.cpp:41
src/unit-conversions.cpp:43: FAILED:
due to a fatal error condition:
  SIGSEGV - Segmentation violation signal

test cases:   14 |   13 passed | 1 failed
assertions: 4004 | 4003 passed | 1 failed
Makefile:28: recipe for target 'check' failed
RPM build errors:
make: *** [check] Error 245

[nlohmann]

Thanks for reporting. Unfortunately, I do not have access to an armv7hl architecture, so I cannot debug into the problem. Can you please give me details on the compiler you are using?

[dkopecek]

The compiler is gcc-c++ 6.2.1. Here is the full build.log: https://kojipkgs.fedoraproject.org//work/tasks/7056/15497056/build.log

[dkopecek]

I've managed to run the conversions unit test via valgrind on the armv7hl machine, this is the output:

+ valgrind ./test/json_unit '*conversion*'
==25420== Memcheck, a memory error detector
==25420== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al.
==25420== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info
==25420== Command: ./test/json_unit *conversion*
==25420== 
==25420== Invalid read of size 4
==25420==    at 0x4975034: std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::basic_string(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) (in /usr/lib/libstdc++.so.6.0.22)
==25420==    by 0x92BEF: std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, nlohmann::basic_json<std::map, std::vector, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, long long, unsigned long long, double, std::allocator> >::pair(std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, nlohmann::basic_json<std::map, std::vector, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, long long, unsigned long long, double, std::allocator> > const&) (in /builddir/build/BUILD/json-2.0.5/test/json_unit)
==25420==  Address 0x6 is not stack'd, malloc'd or (recently) free'd
==25420== 

[nlohmann]

Meanwhile, the build log at https://kojipkgs.fedoraproject.org//work/tasks/7056/15497056/build.log is no longer available. Do you have a copy?

[nlohmann]

It may not be 100% comparable, but the tests (current develop version) succeed on a Raspberry Pi:

• gcc (Raspbian 4.9.2-10) 4.9.2
• Linux raspberrypi 4.1.19+ 🔧 Fix up a few more effc++ items #858 Tue Mar 15 15:52:03 GMT 2016 armv6l GNU/Linux

Valgrind reports:

==27237== Memcheck, a memory error detector
==27237== Copyright (C) 2002-2011, and GNU GPL'd, by Julian Seward et al.
==27237== Using Valgrind-3.7.0 and LibVEX; rerun with -h for copyright info
==27237== Command: test/json_unit
==27237== 
disInstr(arm): unhandled instruction: 0xF1010200
                 cond=15(0xF) 27:20=16(0x10) 4:4=0 3:0=0(0x0)
==27237== valgrind: Unrecognised instruction at address 0x4842588.
==27237==    at 0x4842588: ??? (in /usr/lib/arm-linux-gnueabihf/libcofi_rpi.so)
==27237== Your program just tried to execute an instruction that Valgrind
==27237== did not recognise.  There are two possible reasons for this.
==27237== 1. Your program has a bug and erroneously jumped to a non-code
==27237==    location.  If you are running Memcheck and you just saw a
==27237==    warning about a bad jump, it's probably your program's fault.
==27237== 2. The instruction is legitimate but Valgrind doesn't handle it,
==27237==    i.e. it's Valgrind's fault.  If you think this is the case or
==27237==    you are not sure, please let us know and we'll try to fix it.
==27237== Either way, Valgrind will now raise a SIGILL signal which will
==27237== probably kill your program.
==27237== 
==27237== Process terminating with default action of signal 4 (SIGILL)
==27237==  Illegal opcode at address 0x4842588
==27237==    at 0x4842588: ??? (in /usr/lib/arm-linux-gnueabihf/libcofi_rpi.so)
==27237== 
==27237== HEAP SUMMARY:
==27237==     in use at exit: 194 bytes in 7 blocks
==27237==   total heap usage: 7 allocs, 0 frees, 194 bytes allocated
==27237== 
==27237== 50 bytes in 3 blocks are possibly lost in loss record 4 of 5
==27237==    at 0x4833F2C: operator new(unsigned int) (vg_replace_malloc.c:282)
==27237==    by 0x49089E7: std::string::_Rep::_S_create(unsigned int, unsigned int, std::allocator<char> const&) (in /usr/lib/arm-linux-gnueabihf/libstdc++.so.6.0.20)
==27237== 
==27237== LEAK SUMMARY:
==27237==    definitely lost: 0 bytes in 0 blocks
==27237==    indirectly lost: 0 bytes in 0 blocks
==27237==      possibly lost: 50 bytes in 3 blocks
==27237==    still reachable: 144 bytes in 4 blocks
==27237==         suppressed: 0 bytes in 0 blocks
==27237== Reachable blocks (those to which a pointer was found) are not shown.
==27237== To see them, rerun with: --leak-check=full --show-reachable=yes
==27237== 
==27237== For counts of detected and suppressed errors, rerun with: -v
==27237== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
Illegal instruction

That said - I cannot reproduce the error with a Raspberry Pi. I am still lacking access to a armv7hl system.

[nlohmann]

Thanks @navybk for helping debugging!

[dkopecek]

@nlohmann Here's the build log as an attachment.

json-2.0.5-armv7-build.txt

[nlohmann]

I fear I have to mark this issue as "won't fix"...

[dkopecek]

@nlohmann ok, I'll have to exclude amrv7 builds in Fedora then.

[nlohmann]

Thanks for the quick response. Sorry I could not fix this, but I lack the hardware to debug into it.

[mwittgen]

I was running into the same problem on armv7l. Using 32 bit data types instead of the default bsaic_json<> template fixed the problem. There seems to be a problem with memory alignment of structures on this platform.

[nlohmann]

@mwittgen Thanks for the hint! Can you provide the concrete instantiation of the templates so I can update the README?

[mwittgen]

basic_json<std::map, std::vector, std::string, bool, std::int32_t, std::uint32_t, float>

I have only run our application code and switching to 32 bit values fixed it. This would need some detailed investigation to confirm what is actually going wrong. I might get to it later.

[nlohmann]

Thanks a lot! This is really strange...

[mwittgen]

I'll see if I can provide a simple test case with QEMU.

[mwittgen]

Forgot to mention gcc 5.2 on ZYNQ (armv7l) running ArchLinux. Compiled natively or cross compiled with crosstool-ng. Both executables crashed on invalid memory pointers in different places. Switching off optimization made the crash go away.