1

I was assigned to a pointer and it was free immediately, but sigsegv occurs depending on some devices and xlc options.

xlC_r -o cc.o c.c -lhm has sigsegv occurred, but xlC_r -o cc.o c.c works.

( lhm option is linking libhm.a libhu.a. )
( malloc and free works. Only posix_memalign and free sigsegv occurred. )

Source code: 

#include<stdlib.h>
#include<stdio.h>

int main()
{
    void * sPtr = NULL;

    size_t sAlign = 8388608;
    size_t sSize = 8388648;
    int sRet = 0;

    sRet = posix_memalign(&sPtr, sAlign, sSize);
    printf( "after malloc : %p return : %d\n", sPtr, sRet );
    free(sPtr);

    printf( "after free\n" );
    return 0;
}

It works:

aix7.1 tl3 IBM XL C/C++ for AIX, V12.1 (5765-J02, 5725-C72) Version: 12.01.0000.0000
aix6.1 tl9 IBM XL C/C++ for AIX, V10.1 Version: 10.01.0000.0008

Result is:

after malloc : 22000000 return : 0
after free

but

aix6.1 tl3 IBM XL C/C++ for AIX, V10.1 Version: 10.01.0000.0008
aix5.3 tl9 IBM XL C/C++ for AIX, V10.1 Version: 10.01.0000.0000
aix5.3 tl1 C for AIX version 6.0.0.0 Version: 10.01.0000.0000

Result is

after malloc : 22000000 return : 0
Segmentation fault (core dumped)

% dbx cc.o core
Type 'help' for help.
[using memory image in core]
reading symbolic information ...

Segmentation fault in _ufree at 0xd3c0fcc4 ($t1)
0xd3c0fcc4 (_ufree+0x44) 80be0000         lwz   r5,0x0(r30)
(dbx) where
_ufree(??) at 0xd3c0fcc4
cbase.free(??) at 0xd2b7c048
main(), line 16 in "c.c"

I have not solved this problem although I've been trying for over a week. I searched for a bug report from AIX, but I couldn't find one.

It is not possible to delete both posix_memalign and lhm option.
Do you know a workaround or how to solve it?

karel
  • 5,489
  • 46
  • 45
  • 50
martin.m.kang
  • 13
  • 2
  • `%d` is not the correct `printf` directive for a `size_t`. Use `%zu`, instead. It is possible that this error is misleading you into thinking that your allocation succeeded when in fact it did not. – John Bollinger Feb 12 '19 at 02:39
  • 1
    Regardless, I'm curious: what purpose does it serve to align the allocated object so strongly? – John Bollinger Feb 12 '19 at 02:43
  • It seems `libhm`.`free` cannot handle the pointer returned by `libc`.`posix_memalloc`. The solution is easy: drop `libhm` – Lorinczy Zsigmond Feb 12 '19 at 03:38
  • It seems to me `posix_memalign` does not allocate from the same arena in the memory managment. Your code works for musl libc. Perhaps, you can read the document of libhm, and find out what function can free the memory allocated by `post_memalign`. – stensal Feb 12 '19 at 04:09
  • to John Bollinger : - in source code %d is print for return value. - posix_mealing is necessary for efficiency of memory management. thank you for answer. – martin.m.kang Feb 12 '19 at 04:21
  • to Lorinczy Zsigmond: - after drop libhm, this code is work. but i need libhm for using heapmin. TT thank you for answer. – martin.m.kang Feb 12 '19 at 04:24
  • to stensal: i have searched for libhm on the ibm support many times, but ihave not found the answer. thank you for answer. – martin.m.kang Feb 12 '19 at 04:28
  • There might be a chance that `realloc(p,0)` would work. Or you could use dlopen+dlsym to access `libc`.`free`. – Lorinczy Zsigmond Feb 12 '19 at 07:36
  • 1
    Could you convert the alignment and size to hex please? `0x800000` is a lot easier to read than `8388608`. – Martin Bonner supports Monica Feb 12 '19 at 08:06
  • to Lorinczy Zsigmond : thank for your suggestion. i am considering yout method ( dlopen + dlsym ). thankyou – martin.m.kang Feb 13 '19 at 01:48

1 Answers1

2

libhm and libhmd are debug memory management libraries. They detect errors by keeping extra information after malloc and then checking before free that the block was not previously freed and that it was allocated via malloc. These libraries don't handle posix_memalign and don't keep extra information for it. So when you call free after posix_memalign, they think you're trying to free memory that was not properly allocated. That said, I don't recall that causing a segmentation fault, but a false positive error message. (I'm more used to libhmd than libhm.)

These libraries are no longer maintained. See for example the XLF compiler manual page for them. The official recommendation is to use AIX's native debug malloc tool instead.

Rafik Zurob
  • 361
  • 2
  • 6