On 07/30/2018 03:13 PM, Wei Wang wrote:
This macro was ported from Linux and we've reached an aggreement there
that the corner case "nbits = 0" is not applicable to this macro, because
when "nbits = 0", which means no bits to mask, this macro is expected to
return 0, instead of 0xffffffff. This patch simply adds a comment above
the macro as a note to users about the corner case.

Signed-off-by: Wei Wang <wei.w.w...@intel.com>
CC: Dr. David Alan Gilbert <dgilb...@redhat.com>
CC: Juan Quintela <quint...@redhat.com>
CC: Peter Xu <pet...@redhat.com>
---
  include/qemu/bitmap.h | 1 +
  1 file changed, 1 insertion(+)

diff --git a/include/qemu/bitmap.h b/include/qemu/bitmap.h
index 509eedd..f53c640 100644
--- a/include/qemu/bitmap.h
+++ b/include/qemu/bitmap.h
@@ -60,6 +60,7 @@
   */
#define BITMAP_FIRST_WORD_MASK(start) (~0UL << ((start) & (BITS_PER_LONG - 1)))
+/* "nbits = 0" is not applicable to this macro. Callers should avoid that. */
  #define BITMAP_LAST_WORD_MASK(nbits) (~0UL >> (-(nbits) & (BITS_PER_LONG - 
1)))
#define DECLARE_BITMAP(name,bits) \

A better fix would be to directly change the macro to: nbits ? (~0UL >> (-(nbits) & (BITS_PER_LONG - 1))) : 0, so that we don't need to fix other callers like bitmap_full, bitmap_intersects.

So just post this out for a discussion whether it's preferred to just adding note comments as we did for linux or fixing the macro directly.

Best,
Wei

Reply via email to