In converting some existing spinlocks to rwlock, it was found that the write lock slowpath performance isn't as good as the qspinlock. With a workload that added a large number of inodes to the superblock and was rate-limited by the inode_sb_list_lock, converting that spinlock into a write lock slowed down its performance from about 22s to 36s. This patch series tries to squeeze out as much performance as possible to close the performance gap between qspinlock and qrwlock. With all that patches applies, the workload performance improves to about 24-25s which is much better than before, though still a bit slower than the spinlock. With this patch series in place, we can start converting some spinlocks back to rwlocks where it makes sense and the lock size increase isn't a concern. Waiman Long (4): locking/qrwlock: Better optimization for interrupt context readers locking/qrwlock: Reduce reader/writer to reader lock transfer latency locking/qrwlock: Reduce writer to writer lock transfer latency locking/qrwlock: Use direct MCS lock/unlock in slowpath arch/x86/include/asm/qrwlock.h | 4 + include/asm-generic/qrwlock.h | 4 +- include/asm-generic/qrwlock_types.h | 26 ++++- kernel/locking/qrwlock.c | 185 ++++++++++++++++++++++++++-------- kernel/locking/qspinlock.c | 9 +- 5 files changed, 174 insertions(+), 54 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html