Comment 5 for bug 861296

Note that all this is decided by the kernel's mmap area allocation strategy; see the various "get_unmapped_area" routines in platform-specific kernel code (e.g arch/arm/mm/mmap.c:arch_get_unmapped_area vs. x86/kernel/sys_x86_64.c:arch_get_unmapped_area_topdown).

If any change is required here (and this behaviour *was* in fact changed on x86; they used to behave the same way arm still does), it needs to be done on the kernel side.