Bug 440095

We are working on enabling PMDK for Arm64, which is the persist memory development kit.

Running the test with valgrind, I experienced a valgrind error:
==256557== Memcheck, a memory error detector
==256557== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==256557== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==256557== Command: ./obj_basic_integration /mnt/mem0//test_obj_basic_integration5__________________PMDK_________/testfile1
==256557== Parent PID: 256503
==256557==
ARM64 front end: branch_etc
disInstr(arm64): unhandled instruction 0xD50B7A20
disInstr(arm64): 1101'0101 0000'1011 0111'1010 0010'0000
==256557== valgrind: Unrecognised instruction at address 0x493a148.
==256557==    at 0x493A148: arm_clean_va_to_poc (arm_cacheops.h:55)
==256557==    by 0x493A3BB: flush_poc_nolog (flush.h:27)
==256557==    by 0x493A3BB: flush_poc (init.c:33)
==256557==    by 0x4934C2F: pmem_deep_flush (pmem.c:214)
==256557==    by 0x48A8037: os_part_deep_common (os_deep_linux.c:136)
==256557==    by 0x48B672B: shutdown_state_checksum (shutdown_state.c:32)
==256557==    by 0x48B678B: shutdown_state_init (shutdown_state.c:47)
==256557==    by 0x48AFE23: util_header_create (set.c:2313)
==256557==    by 0x48B15A3: util_replica_init_headers_local (set.c:2777)
==256557==    by 0x48B17BB: util_replica_create_local (set.c:2823)
==256557==    by 0x48B2FEF: util_pool_create_uuids (set.c:3313)
==256557==    by 0x48B31C3: util_pool_create (set.c:3356)
==256557==    by 0x48D49DB: pmemobj_createU (obj.c:1351)
==256557== Your program just tried to execute an instruction that Valgrind
==256557== did not recognise.  There are two possible reasons for this.
==256557== 1. Your program has a bug and erroneously jumped to a non-code
==256557==    location.  If you are running Memcheck and you just saw a
==256557==    warning about a bad jump, it's probably your program's fault.
==256557== 2. The instruction is legitimate but Valgrind doesn't handle it,
==256557==    i.e. it's Valgrind's fault.  If you think this is the case or
==256557==    you are not sure, please let us know and we'll try to fix it.
==256557== Either way, Valgrind will now raise a SIGILL signal which will
==256557== probably kill your program.
==256557==
==256557== HEAP SUMMARY:
==256557==     in use at exit: 3,039 bytes in 39 blocks
==256557==   total heap usage: 1,220 allocs, 1,181 frees, 1,653,499 bytes allocated
==256557==
==256557== LEAK SUMMARY:
==256557==    definitely lost: 0 bytes in 0 blocks
==256557==    indirectly lost: 0 bytes in 0 blocks
==256557==      possibly lost: 0 bytes in 0 blocks
==256557==    still reachable: 3,039 bytes in 39 blocks
==256557==         suppressed: 0 bytes in 0 blocks
==256557== Reachable blocks (those to which a pointer was found) are not shown.
==256557== To see them, rerun with: --leak-check=full --show-leak-kinds=all
==256557==
==256557== For lists of detected and suppressed errors, rerun with: -s
==256557== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

The one above will call: https://github.com/pmem/pmdk/blob/master/src/libpmem2/aarch64/arm_cacheops.h#L53, but Valgrind can not handle it.

Also, the other one as below with "dc cvap" also got the same consequence. The dc cvap was called at: https://github.com/pmem/pmdk/blob/master/src/libpmem2/aarch64/arm_cacheops.h#L59

[root@fedora obj_basic_integration]# cat memcheck5.log
==261736== Memcheck, a memory error detector
==261736== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==261736== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==261736== Command: ./obj_basic_integration /mnt/mem0//test_obj_basic_integration5__________________PMDK_________/testfile1
==261736== Parent PID: 261682
==261736==
ARM64 front end: branch_etc
disInstr(arm64): unhandled instruction 0xD50B7C20
disInstr(arm64): 1101'0101 0000'1011 0111'1100 0010'0000
==261736== valgrind: Unrecognised instruction at address 0x493a164.
==261736==    at 0x493A164: arm_clean_va_to_pop (arm_cacheops.h:61)
==261736==    by 0x493A453: flush_pop_nolog (flush.h:45)
==261736==    by 0x493A453: flush_pop (init.c:49)
==261736==    by 0x4934C2F: pmem_deep_flush (pmem.c:214)
==261736==    by 0x48A8037: os_part_deep_common (os_deep_linux.c:136)
==261736==    by 0x48B672B: shutdown_state_checksum (shutdown_state.c:32)
==261736==    by 0x48B678B: shutdown_state_init (shutdown_state.c:47)
==261736==    by 0x48AFE23: util_header_create (set.c:2313)
==261736==    by 0x48B15A3: util_replica_init_headers_local (set.c:2777)
==261736==    by 0x48B17BB: util_replica_create_local (set.c:2823)
==261736==    by 0x48B2FEF: util_pool_create_uuids (set.c:3313)
==261736==    by 0x48B31C3: util_pool_create (set.c:3356)
==261736==    by 0x48D49DB: pmemobj_createU (obj.c:1351)
==261736== Your program just tried to execute an instruction that Valgrind
==261736== did not recognise.  There are two possible reasons for this.
==261736== 1. Your program has a bug and erroneously jumped to a non-code
==261736==    location.  If you are running Memcheck and you just saw a
==261736==    warning about a bad jump, it's probably your program's fault.
==261736== 2. The instruction is legitimate but Valgrind doesn't handle it,
==261736==    i.e. it's Valgrind's fault.  If you think this is the case or
==261736==    you are not sure, please let us know and we'll try to fix it.
==261736== Either way, Valgrind will now raise a SIGILL signal which will
==261736== probably kill your program.
==261736==
==261736== HEAP SUMMARY:
==261736==     in use at exit: 3,039 bytes in 39 blocks
==261736==   total heap usage: 1,220 allocs, 1,181 frees, 1,653,499 bytes allocated
==261736==
==261736== LEAK SUMMARY:
==261736==    definitely lost: 0 bytes in 0 blocks
==261736==    indirectly lost: 0 bytes in 0 blocks
==261736==      possibly lost: 0 bytes in 0 blocks
==261736==    still reachable: 3,039 bytes in 39 blocks
==261736==         suppressed: 0 bytes in 0 blocks
==261736== Reachable blocks (those to which a pointer was found) are not shown.
==261736== To see them, rerun with: --leak-check=full --show-leak-kinds=all
==261736==
==261736== For lists of detected and suppressed errors, rerun with: -s
==261736== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

OBSERVED RESULT


EXPECTED RESULT


Linux/KDE Plasma: Fedora 34, linux 5.12