Hi, In https://github.com/GNS3/dynamips-community/issues/9 I found that the reason for a crash was the compiler assuming that the pointer was 4-byte aligned in one of the O3 optimizations. Using gcc-4.4.7-3.el6.x86_64 in a Cent6.4-x86_64 box. I don't understand why it made that assumption, so I'd like to know how to tell the compiler not to assume an alignment. In detail: /* Byte-swap a memory block */ void mem_bswap32(void *ptr,size_t len) { m_uint32_t *p = ptr; size_t count = len >> 2; int i; for(i=0;i<count;i++,p++) *p = swap32(*p); } gcc turned the original loop into 3 loops: * one loop for the initial 0-3 swaps (before we reach a 16-byte aligned address) * one loop that processes 4 swaps at a time using MOVDQA, which needs addresses to be 16-byte aligned or it triggers a protection fault * one loop for the remaining 0-3 swaps Now here's the catch... it assumes that the address is already 4-byte aligned, so it discards the 2 lower bits when checking how many to run in the first loop: 0x0000000000406725 <+21>: 49 c1 ea 02 shr $0x2,%r10 unfortunately our buffer is 2-byte aligned: ptr = 0x7fffe81b3ff2 and it crashes when it tries to copy 16 bytes with MOVDQA from the start of the buffer (not 16-byte aligned): => 0x00000000004067d0 <+192>: 66 0f 6f 0c 17 movdqa (%rdi,%rdx,1),%xmm1
