There are 2 optimisations:
- Now, io_worker_handler_work() do io_assign_current_work() twice per
request, and each one adds lock/unlock(worker->lock) pair. The first is
to reset worker->cur_work to NULL, and the second to set a real work
shortly after. If there is a dependant work, set it immediately, that
effectively removes the extra NULL'ing.
- And there is no use in taking wqe->lock for linked works, as they are
not hashed now. Optimise it out.
Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
---
fs/io-wq.c | 15 +++++++--------
1 file changed, 7 insertions(+), 8 deletions(-)
diff --git a/fs/io-wq.c b/fs/io-wq.c
index e438dc4d7cb3..473af080470a 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -476,7 +476,7 @@ static void io_worker_handle_work(struct io_worker *worker)
struct io_wq *wq = wqe->wq;
do {
- struct io_wq_work *work, *old_work;
+ struct io_wq_work *work;
unsigned hash = -1U;
/*
@@ -495,12 +495,13 @@ static void io_worker_handle_work(struct io_worker *worker)
spin_unlock_irq(&wqe->lock);
if (!work)
break;
+ io_assign_current_work(worker, work);
/* handle a whole dependent link */
do {
- io_assign_current_work(worker, work);
- io_impersonate_work(worker, work);
+ struct io_wq_work *old_work;
+ io_impersonate_work(worker, work);
/*
* OK to set IO_WQ_WORK_CANCEL even for uncancellable
* work, the worker function will do the right thing.
@@ -513,10 +514,8 @@ static void io_worker_handle_work(struct io_worker *worker)
old_work = work;
work->func(&work);
-
- spin_lock_irq(&worker->lock);
- worker->cur_work = NULL;
- spin_unlock_irq(&worker->lock);
+ work = (old_work == work) ? NULL : work;
+ io_assign_current_work(worker, work);
if (wq->put_work)
wq->put_work(old_work);
@@ -529,7 +528,7 @@ static void io_worker_handle_work(struct io_worker *worker)
/* dependent work is not hashed */
hash = -1U;
}
- } while (work && work != old_work);
+ } while (work);
spin_lock_irq(&wqe->lock);
} while (1);
--
2.24.0
