Skip to contentSkip to conversation

Do not always expect multishot recv to stop posting events

PATCH Applied Dylan Yudaken 2 participants 2 messages Nov 7, 2022 io_uring

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Do not always expect multishot recv to stop posting events

Later kernels can have a fix that does not stop multishot from posting events, and would just continue in overflow mode.

---
 test/recv-multishot.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
test/recv-multishot.c +10 -2
index 2cfe6898de4c..ed26a5f78759 100644
@@ -264,11 +264,19 @@ static int test(struct args *args)
 
 		bool const is_last = i == recv_cqes - 1;
 
+		/*
+		 * Older kernels could terminate multishot early due to overflow,
+		 * but later ones will not. So discriminate based on the MORE flag.
+		 */
+		bool const early_last = args->early_error == ERROR_EARLY_OVERFLOW &&
+					!args->wait_each &&
+					i == N_CQE_OVERFLOW &&
+					!(cqe->flags & IORING_CQE_F_MORE);
+
 		bool const should_be_last =
 			(cqe->res <= 0) ||
 			(args->stream && is_last) ||
-			(args->early_error == ERROR_EARLY_OVERFLOW &&
-			 !args->wait_each && i == N_CQE_OVERFLOW);
+			early_last;
 		int *this_recv;
 		int orig_payload_size = cqe->res;
 

base-commit: 754bc068ec482c5338a07dd74b7d3892729bb847

Signature · 1 line
2.30.2
maintainer Applied
#1 lore raw
Quoted 4 lines · Dylan Yudaken

Later kernels can have a fix that does not stop multishot from posting events, and would just continue in overflow mode.

Applied, thanks!

[1/1] Do not always expect multishot recv to stop posting events

      commit: 0d4fdb416718a70a4a90c5c4722b38cf44849195

Best regards,

Signature · 1 line
Jens Axboe