Il 29/07/2013 14:28, Paolo Bonzini ha scritto:
> The problem is introduced by commit 2332616 (exec: Support 64-bit
> operations in address_space_rw, 2013-07-08). Before that commit,
> memory_access_size would only return 1/2/4.
>
> Since alignment is already handled above, reduce l to the largest
> power of two that is smaller than l.
>
> Reported-by: Oleksii Shevchuk <alx...@gmail.com>
> Tested-by: Oleksii Shevchuk <alx...@gmail.com>
> Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
> ---
> exec.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/exec.c b/exec.c
> index c4f2894..e122c81 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -1925,6 +1925,9 @@ static int memory_access_size(MemoryRegion *mr,
> unsigned l, hwaddr addr)
> if (l > access_size_max) {
> l = access_size_max;
> }
> + if (l & (l - 1)) {
> + l = 1 << (qemu_fls(l) - 1);
> + }
>
> return l;
> }
>
Ping, and adding qemu-stable.
Paolo
