the original motivation for this patch was that qemu (and possibly
other syscall emulators) nop out madvise, resulting in an infinite
loop. however, there is another benefit to this change: madvise may
actually undo an explicit madvise the application intended for its
stack, whereas the mremap operation is a true nop. the logic here is
that mremap must fail if it cannot resize the mapping in-place, and
the caller knows that it cannot resize in-place because it knows the
next page of virtual memory is already occupied.
+#define _GNU_SOURCE
#include "pthread_impl.h"
#include <sys/mman.h>
size_t l = PAGE_SIZE;
p += -(uintptr_t)p & PAGE_SIZE-1;
a->_a_stackaddr = (uintptr_t)p;
- while (!posix_madvise(p-l-PAGE_SIZE, PAGE_SIZE, POSIX_MADV_NORMAL))
+ while (mremap(p-l-PAGE_SIZE, PAGE_SIZE, 2*PAGE_SIZE, 0)==MAP_FAILED && errno==ENOMEM)
l += PAGE_SIZE;
a->_a_stacksize = l - DEFAULT_STACK_SIZE;
}