SUMMARY

This is somewhat connected to https://bugs.kde.org/show_bug.cgi?id=474332, because if I do a workaround for this here: https://sourceware.org/git/?p=valgrind.git;a=blob;f=coregrind/m_mallocfree.c;h=44beb3d8b57cc3d7859a2966b4f5824e255fb6d4;hb=23250889de4e2079ad1ede6874cc824bc9dd92db#l2258, by removing the panic and setting req_alignB to VG_MIN_MALLOC_SZB, I hit the other issue.

I'm using Valgrind on master, at commit: d4c9a98527afdbd67b25a1300339a42b9e0a24c2

STEPS TO REPRODUCE
1. Compile and run the following C snippet:

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

int main() {
    char *p = aligned_alloc(4, 4);

    if(p == NULL) {
        printf("Allocation failed!\n");     
        return 1;
    }

    printf("Allocation succeeded!\n");

    return 0;
}

2. Then run it again under Valgrind

OBSERVED RESULT

The allocation succeeds when run on its own, but with Valgrind fails:

VG_(arena_memalign)(0x58889810, 4, 4)
bad alignment value 4
(it is too small, below the lower limit of 16)
valgrind: the 'impossible' happened:
   VG_(arena_memalign)
 
EXPECTED RESULT

Both succeeds.

ADDITIONAL INFORMATION

The issue seems to be similar to the other bug I linked above, but here the requirement on the alignment seems to be more strict and actually connected to what's required from malloc, but not aligned_alloc, which again is implementation dependent and on glibc the only limit seems to be that it has to be a power of 2.