Hi Dave,
On Fri, Oct 21, 2016 at 09:33:08AM +1100, Dave Chinner wrote:
> On Wed, Oct 19, 2016 at 01:24:26PM -0400, Johannes Weiner wrote:
> > With this in place, the next patch can restore refault detection in
> > single-page files. It will also move the shadow count from the upper
> > bits of count to the new special counter, and then shrink count to a
> > char as well; the growth of struct radix_tree_node is temporary.
[...]
> > @@ -90,9 +91,10 @@ enum radix_tree_tags {
> > #define RADIX_TREE_COUNT_MASK ((1UL << RADIX_TREE_COUNT_SHIFT) - 1)
> >
> > struct radix_tree_node {
> > - unsigned char shift; /* Bits remaining in each slot */
> > - unsigned char offset; /* Slot offset in parent */
> > - unsigned int count;
> > + unsigned char shift; /* Bits remaining in each slot */
> > + unsigned char offset; /* Slot offset in parent */
> > + unsigned int count; /* Total entry count */
> > + unsigned char special; /* Special entry count */
>
> How about putting the new char field into the implicit hole between
> offset and count? pahole is your friend here:
>
> struct radix_tree_node {
> unsigned char shift; /* 0 1 */
> unsigned char offset; /* 1 1 */
>
> /* XXX 2 bytes hole, try to pack */
>
> unsigned int count; /* 4 4 */
> .....
The next patch turns `count' into an unsigned char again, so the hole
is only temporary.
Thanks
