Re: [Qemu-devel] [PATCH] linux-user: fix mremap for 64bit targets on 32bit hosts

[Peter Maydell]

On 17 September 2016 at 18:20, Felix Janda <address@hidden> wrote:
> Signed-off-by: Felix Janda <address@hidden>
> ---
>  linux-user/mmap.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/linux-user/mmap.c b/linux-user/mmap.c
> index c4371d9..4882816 100644
> --- a/linux-user/mmap.c
> +++ b/linux-user/mmap.c
> @@ -682,7 +682,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong 
> old_size,
>
>      if (flags & MREMAP_FIXED) {
>          host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
> -                                     old_size, new_size,
> +                                     (size_t) old_size, (size_t) new_size,
>                                       flags,
>                                       g2h(new_addr));
>
> @@ -701,7 +701,7 @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong 
> old_size,
>              host_addr = MAP_FAILED;
>          } else {
>              host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
> -                                         old_size, new_size,
> +                                         (size_t) old_size, (size_t) 
> new_size,
>                                           flags | MREMAP_FIXED,
>                                           g2h(mmap_start));
>              if (reserved_va) {
> --
> 2.7.3

Rather than this, I think it would be better to switch to
using the mremap() library call rather than direct syscall
here, which then matches the other mremap()s later in the
function. (That will work right because mremap()'s prototype
says it takes size_t arguments, whereas syscall() is a
generic thing which doesn't, and so the C default promotions
do the wrong thing with the abi_ulongs.)

The use of syscall(__NR_mremap, ...) originally dates back to 2008:
https://lists.gnu.org/archive/html/qemu-devel/2008-12/msg01087.html
https://lists.gnu.org/archive/html/qemu-devel/2008-12/msg00480.html

and was to permit compilation with glibc 2.4 which didn't
support the 5-argument mremap() or define MREMAP_FIXED.

Since glibc 2.4 dates back to a decade ago now, we no longer
need to carry this ugly (and buggy) workaround for it.

thanks
-- PMM