The coming patchset is another take on z3fold page layout optimization problem. The previous solution [1] used shrinker to solve the issue of in-page space fragmentation but after some discussions the decision was made to rewrite background page layout optimization using good old work queues.
The patchset thus implements in-page compaction worker for z3fold, preceded by some code optimizations and preparations which, again, deserved to be separate patches. Main changes compared to v1: - per-page locking is removed due to size problems (z3fold header becomes greater than one chunk on x86_64 with gcc 6.0) and non-obvious performance benefits - instead, per-pool spinlock is substituted with rwlock. Signed-off-by: Vitaly Wool <[email protected]> [1] https://lkml.org/lkml/2016/10/15/31
