Modify the defect that the timer is not initialized during IO transfer
when passthrough is used with hybrid polling to ensure that the program
can run normally.

Signed-off-by: hexue <xue01.he@samsung.com>
---
 io_uring/uring_cmd.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
index e6701b7aa147..678a2f7d14ff 100644
--- a/io_uring/uring_cmd.c
+++ b/io_uring/uring_cmd.c
@@ -244,6 +244,11 @@ int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
 			return -EOPNOTSUPP;
 		issue_flags |= IO_URING_F_IOPOLL;
 		req->iopoll_completed = 0;
+		if (ctx->flags & IORING_SETUP_HYBRID_IOPOLL) {
+			/* make sure every req only blocks once */
+			req->flags &= ~REQ_F_IOPOLL_STATE;
+			req->iopoll_start = ktime_get_ns();
+		}
 	}
 
 	ret = file->f_op->uring_cmd(ioucmd, issue_flags);
-- 
2.43.0

On 5/7/25 8:18 PM, hexue wrote:
> Modify the defect that the timer is not initialized during IO transfer
> when passthrough is used with hybrid polling to ensure that the program
> can run normally.

Patch title should be:

io_uring/uring_cmd: fix hybrid polling initialization issue

and then you need a Fixes line and a stable tag, if appropriate.

-- 
Jens Axboe