valgrind on armhf fails with with many errors

[Daniel van Vugt]

Valgrind is now failing on armhf. Both in CI and when tested manually.

This is the first and most obvious error, but there are more...

[ RUN ] UdevWrapperDeathTest.DereferencingEndReturnsInvalidObject
==10542== Invalid write of size 4
==10542== at 0x4ECA00C: ??? (in /lib/arm-linux-gnueabihf/libc-2.21.so)
==10542== Address 0xbdbafa88 is on thread 1's stack
==10542== 16 bytes below stack pointer
==10542==
==10542== Conditional jump or move depends on uninitialised value(s)
==10542== at 0x4ECA118: __udivmoddi4 (in /lib/arm-linux-gnueabihf/libc-2.21.so)
==10542==
==10542== Use of uninitialised value of size 4
==10542== at 0x4ECA11A: __udivmoddi4 (in /lib/arm-linux-gnueabihf/libc-2.21.so)
==10542==
==10542== Use of uninitialised value of size 4
==10542== at 0x4ECA018: ??? (in /lib/arm-linux-gnueabihf/libc-2.21.so)
==10542==
==10611==
==10611== Process terminating with default action of signal 11 (SIGSEGV)
==10611== Bad permissions for mapped region at address 0x2CDA000
==10611== at 0x2CDA000: ??? (in /home/phablet/mir/dev/bin/.mir_unit_tests-uninstalled)
==10611==
==10611== HEAP SUMMARY:
==10611== in use at exit: 392,022 bytes in 7,695 blocks
==10611== total heap usage: 19,288 allocs, 11,593 frees, 1,117,616 bytes allocated
==10611==
==10611== LEAK SUMMARY:
==10611== definitely lost: 0 bytes in 0 blocks
==10611== indirectly lost: 0 bytes in 0 blocks
==10611== possibly lost: 137,447 bytes in 3,483 blocks
==10611== still reachable: 254,507 bytes in 4,211 blocks
==10611== suppressed: 0 bytes in 0 blocks
==10611== Reachable blocks (those to which a pointer was found) are not shown.
==10611== To see them, rerun with: --leak-check=full --show-leak-kinds=all
==10611==
==10611== For counts of detected and suppressed errors, rerun with: -v
==10611== Use --track-origins=yes to see where uninitialised values come from
==10611== ERROR SUMMARY: 55 errors from 4 contexts (suppressed: 6 from 6)

[Daniel van Vugt (vanvugt)]

Logs from valgrind of unit tests on mako.

[Daniel van Vugt (vanvugt)]

Logs from valgrind of integration tests on mako.

[Daniel van Vugt (vanvugt)]

Logs from valgrind of acceptance tests on mako.

[Launchpad Janitor (janitor)]

This bug was fixed in the package mir - 0.12.1+15.04.20150324-0ubuntu1

---------------
mir (0.12.1+15.04.20150324-0ubuntu1) vivid; urgency=medium

  [ Alexandros Frantzis ]
  * New upstream release 0.12.1 (https://launchpad.net/mir/+milestone/0.12.1)
    - Bug fixes:
      . [regression] mali, powervr locks up with around the introduction or
        removal of a third overlay (LP: #1413211)
      . USC - mouse cursor on AMD graphics is drawing incorrectly
        (LP: #1417581)
      . mir_demo_server doesn't emit hover_exit events (LP: #1418569)
      . SessionMediator locks mutexes in one thread and unlocks them in
        another (LP: #1427976)
      . ProtobufResponder::send_response_result race (LP: #1428402)
      . Some protobuf Closure objects can access dead objects (LP: #1433330)
      . DisplayConfigurationOutput.physical_size_mm is undefined/zero
        (LP: #1430315)
      . vivid fails to build Mir as of 2015-03-22: error: #warning
        "_BSD_SOURCE and _SVID_SOURCE are deprecated, use _DEFAULT_SOURCE"
        [-Werror=cpp] (LP: #1435127)
      . valgrind on armhf fails with with many errors (LP: #1435186)
 -- CI Train Bot <email address hidden> Tue, 24 Mar 2015 16:09:54 +0000

[StephanBeal (sgbeal)]

This, or something indistinguishable from it, is still happening in Ubunut 15.10:

[odroid@host:~/fossil/cwal/s2]$ cat /etc/issue
Ubuntu 15.10 \n \l

[odroid@host:~/fossil/cwal/s2]$ uname -a
Linux odroid 3.8.13.30 #1 SMP PREEMPT Fri Feb 20 14:34:08 BRST 2015 armv7l armv7l armv7l GNU/Linux

[odroid@host:~/fossil/cwal/s2]$ valgrind --version
valgrind-3.11.0

[odroid@host:~/fossil/cwal/s2]$ gcc --version
gcc-5.real (Ubuntu 5.2.1-22ubuntu2) 5.2.1 20151010
...

but my repro case is not trivial. The errors are very similar, though:

==5160== HEAP SUMMARY:
==5160== in use at exit: 0 bytes in 0 blocks
==5160== total heap usage: 596 allocs, 596 frees, 22,345 bytes allocated
==5160==
==5160== All heap blocks were freed -- no leaks are possible
==5160==
==5160== ERROR SUMMARY: 2240 errors from 4 contexts (suppressed: 3 from 3)
==5160==
==5160== 448 errors in context 1 of 4:
==5160== Use of uninitialised value of size 4
==5160== at 0x6A850: ??? (in /home/odroid/fossil/cwal/s2/s2sh)
==5160== Uninitialised value was created by a stack allocation
==5160== at 0x6A844: ??? (in /home/odroid/fossil/cwal/s2/s2sh)
==5160==
==5160==
==5160== 448 errors in context 2 of 4:
==5160== Use of uninitialised value of size 4
==5160== at 0x6A956: __udivmoddi4 (in /home/odroid/fossil/cwal/s2/s2sh)
==5160== Uninitialised value was created by a stack allocation
==5160== at 0x6A844: ??? (in /home/odroid/fossil/cwal/s2/s2sh)
==5160==
==5160==
==5160== 448 errors in context 3 of 4:
==5160== Conditional jump or move depends on uninitialised value(s)
==5160== at 0x6A954: __udivmoddi4 (in /home/odroid/fossil/cwal/s2/s2sh)
==5160== Uninitialised value was created by a stack allocation
==5160== at 0x6A844: ??? (in /home/odroid/fossil/cwal/s2/s2sh)
==5160==
==5160==
==5160== 896 errors in context 4 of 4:
==5160== Invalid write of size 4
==5160== at 0x6A844: ??? (in /home/odroid/fossil/cwal/s2/s2sh)
==5160== Address 0xbdaf5890 is on thread 1's stack
==5160== 16 bytes below stack pointer

none of these show up on x86/64, and the lack of symbol names/locations is inexplicable, as all code is built with -g (debugging).

[Daniel van Vugt (vanvugt)]

This bug is about running valgrind on Mir specifically :)