Hi Jens,
On 7/19/2022 10:29 AM, Jens Axboe wrote:
> I'll poke at this tomorrow.
Just FYI. Another finding (test is based on commit 584b0180f0):
If the code block is put to different function, the fio performance result is
different:
Patch1:
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 616d857f8fc6..b0578a3d063a 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -3184,10 +3184,6 @@ static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
struct file *file = req->file;
int ret;
- if (likely(file && (file->f_mode & FMODE_WRITE)))
- if (!io_req_ffs_set(req))
- req->flags |= io_file_get_flags(file) << REQ_F_SUPPORT_NOWAIT_BIT;
-
kiocb->ki_pos = READ_ONCE(sqe->off);
ioprio = READ_ONCE(sqe->ioprio);
@@ -7852,6 +7848,10 @@ static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
return 0;
}
+ if (likely(req->file))
+ if (!io_req_ffs_set(req))
+ req->flags |= io_file_get_flags(req->file) << REQ_F_SUPPORT_NOWAIT_BIT;
+
io_queue_sqe(req);
return 0;
Patch2:
diff --git a/fs/io_uring.c b/fs/io_uring.c
index b0578a3d063a..af705e7ba8d3 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -7639,6 +7639,11 @@ static void io_queue_sqe_fallback(struct io_kiocb *req)
static inline void io_queue_sqe(struct io_kiocb *req)
__must_hold(&req->ctx->uring_lock)
{
+
+ if (likely(req->file))
+ if (!io_req_ffs_set(req))
+ req->flags |= io_file_get_flags(req->file) << REQ_F_SUPPORT_NOWAIT_BIT;
+
if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL))))
__io_queue_sqe(req);
else
@@ -7848,10 +7853,6 @@ static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
return 0;
}
- if (likely(req->file))
- if (!io_req_ffs_set(req))
- req->flags |= io_file_get_flags(req->file) << REQ_F_SUPPORT_NOWAIT_BIT;
-
io_queue_sqe(req);
return 0;
}
Patch3:
diff --git a/fs/io_uring.c b/fs/io_uring.c
index af705e7ba8d3..5771d6d0ad8a 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -7598,6 +7598,10 @@ static inline void __io_queue_sqe(struct io_kiocb *req)
struct io_kiocb *linked_timeout;
int ret;
+ if (likely(req->file))
+ if (!io_req_ffs_set(req))
+ req->flags |= io_file_get_flags(req->file) << REQ_F_SUPPORT_NOWAIT_BIT;
+
ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
if (req->flags & REQ_F_COMPLETE_INLINE) {
@@ -7640,10 +7644,6 @@ static inline void io_queue_sqe(struct io_kiocb *req)
__must_hold(&req->ctx->uring_lock)
{
- if (likely(req->file))
- if (!io_req_ffs_set(req))
- req->flags |= io_file_get_flags(req->file) << REQ_F_SUPPORT_NOWAIT_BIT;
-
if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL))))
__io_queue_sqe(req);
else
The test result (confirmed on my own test env and LKP):
patch1 and patch2 have no regression. patch3 has regression.
Regards
Yin, Fengwei
