On Mon, Jun 28, 2010 at 4:03 PM, Jes Sorensen <jes.soren...@redhat.com> wrote: > On 06/28/10 17:42, Blue Swirl wrote: >> On Mon, Jun 28, 2010 at 2:50 PM, Jes Sorensen <jes.soren...@redhat.com> >> wrote: >>> I figured out what was causing it. qemu-options.def has an >>> #ifdef MAP_POPULATE in it, which isn't being set without sys/mmap.h >>> being included. Pretty much every other #ifdef in qemu-options.def are >>> based on CONFIG_foo settings or things like _WIN32 which do not change >>> depending on header file inclusion. >>> >>> I think the easiest fix is to just add sys/mmap.h to the include list in >>> os-posix.c, so I just posted a patch for that. Though, in principle we >>> really shouldn't base qemu-options.def settings on defines pulled in >>> from system header files. >> >> I think more flags should be added to arch_mask field, like >> QEMU_ARCH_LINUX, QEMU_ARCH_POSIX and QEMU_ARCH_WIN32. Then the #ifdefs >> should be removed. Prealloc command line flag stuff should be >> conditional to CONFIG_LINUX only, there should be another check for >> MAP_POPULATE where mem_preallocate is set. >> >> Alternatively, we could have more arch_mask flags like QEMU_MAP_POPULATE. > > Yeah, the problem with tying it to CONFIG_LINUX is that older version of > Linux may not support it. Looking through the list, MAP_POPULATE is > really an oddball in there though, so maybe it would be cleaner to catch > it via configure and then use CONFIG_MAP_POPULATE or something like that?
There'd be 1:1 relation between MAP_POPULATE and CONFIG_MAP_POPULATE, so maybe not.
