Skip to contentSkip to conversation

6.0 updates

PATCH Applied Dylan Yudaken 3 participants 7 messages Sep 27, 2022 io_uring

lore
Share & tools
lore
#0 lore raw

v2 cover 6.0 updates

liburing updates for 6.0:

Patch 1 updates to account for the single issuer ring being assigned at ring creation time.
Patch 2 updates man pages from 5.20 -> 6.0
Patch 3 reduces test flakiness

v2:
- handle IORING_SETUP_R_DISABLED semantics
- fix unique path in open-direct-pick test

Dylan Yudaken (3):
handle single issuer task registration at ring creation update documentation to reflect no 5.20 kernel give open-direct-pick.c a unique path

 man/io_uring_prep_recv.3    |  2 +-
 man/io_uring_prep_recvmsg.3 |  2 +-
 man/io_uring_setup.2        | 18 +++++++++---------
 test/defer-taskrun.c        |  8 ++++++--
 test/open-direct-pick.c     |  2 +-
 test/single-issuer.c        | 34 ++++++++++++++++++++++++++--------
 6 files changed, 44 insertions(+), 22 deletions(-)

base-commit: 1cba5d6c9e41e2f55ac4c3f93f5e05f9b5082a3e

Signature · 1 line
2.30.2
#1 lore raw

PATCH v2 1/3 handle single issuer task registration at ring creation

IORING_SETUP_SINGLE_ISSUER now registers the task at ring creation, so update the tests and documentation to reflect this.

---
 man/io_uring_setup.2 | 18 +++++++++---------
 test/defer-taskrun.c |  8 ++++++--
 test/single-issuer.c | 34 ++++++++++++++++++++++++++--------
 3 files changed, 41 insertions(+), 19 deletions(-)
man/io_uring_setup.2 +9 -9
index fb1f90dc2ebc..502a803d2308 100644
@@ -241,22 +241,22 @@ only the
 passthrough command for NVMe passthrough needs this. Available since 5.19.
 .TP
 .B IORING_SETUP_SINGLE_ISSUER
-A hint to the kernel that only a single task can submit requests, which is used
-for internal optimisations. The kernel enforces the rule, which only affects
-.I
-io_uring_enter(2)
-calls submitting requests and will fail them with
+A hint to the kernel that only a single task (or thread) will submit requests, which is
+used for internal optimisations. The submission task is either the task that created the
+ring, or if
+.B IORING_SETUP_R_DISABLED
+is specified then it is the task that enables the ring through
+.BR io_uring_register (2) .
+The kernel enforces this rule, failing requests with
 .B -EEXIST
 if the restriction is violated.
-The submitter task may differ from the task that created the ring.
 Note that when
 .B IORING_SETUP_SQPOLL
 is set it is considered that the polling task is doing all submissions
 on behalf of the userspace and so it always complies with the rule disregarding
 how many userspace tasks do
-.I
-io_uring_enter(2).
-Available since 5.20.
+.BR io_uring_enter(2).
+Available since 6.0.
 .TP
 .B IORING_SETUP_DEFER_TASKRUN
 By default, io_uring will process all outstanding work at the end of any system
test/defer-taskrun.c +6 -2
index c6e0ea0d6c99..9283f2866c6b 100644
@@ -123,6 +123,7 @@ void *thread(void *t)
 {
 	struct thread_data *td = t;
 
+	io_uring_enable_rings(&td->ring);
 	io_uring_prep_read(io_uring_get_sqe(&td->ring), td->efd, td->buff, sizeof(td->buff), 0);
 	io_uring_submit(&td->ring);
 
@@ -138,11 +139,12 @@ static int test_thread_shutdown(void)
 	uint64_t val = 1;
 
 	ret = io_uring_queue_init(8, &td.ring, IORING_SETUP_SINGLE_ISSUER |
-					       IORING_SETUP_DEFER_TASKRUN);
+					       IORING_SETUP_DEFER_TASKRUN |
+					       IORING_SETUP_R_DISABLED);
 	if (ret)
 		return ret;
 
-	CHECK(io_uring_get_events(&td.ring) == -EEXIST);
+	CHECK(io_uring_get_events(&td.ring) == -EBADFD);
 
 	td.efd = eventfd(0, 0);
 	CHECK(td.efd >= 0);
@@ -150,6 +152,8 @@ static int test_thread_shutdown(void)
 	CHECK(pthread_create(&t1, NULL, thread, &td) == 0);
 	CHECK(pthread_join(t1, NULL) == 0);
 
+	CHECK(io_uring_get_events(&td.ring) == -EEXIST);
+
 	CHECK(write(td.efd, &val, sizeof(val)) == sizeof(val));
 	CHECK(io_uring_wait_cqe(&td.ring, &cqe) == -EEXIST);
 
test/single-issuer.c +26 -8
index 29830f1af998..1d13f47202f8 100644
@@ -96,30 +96,48 @@ int main(int argc, char *argv[])
 	if (!fork_t()) {
 		ret = try_submit(&ring);
 		if (ret != -EEXIST)
-			fprintf(stderr, "not owner child could submit %i\n", ret);
+			fprintf(stderr, "1: not owner child could submit %i\n", ret);
 		return ret != -EEXIST;
 	}
 	wait_child_t();
 	io_uring_queue_exit(&ring);
 
 	/* test that the first submitter but not creator can submit */
-	ret = io_uring_queue_init(8, &ring, IORING_SETUP_SINGLE_ISSUER);
+	ret = io_uring_queue_init(8, &ring, IORING_SETUP_SINGLE_ISSUER |
+					    IORING_SETUP_R_DISABLED);
 	if (ret)
 		error(1, ret, "ring init (2) %i", ret);
 
 	if (!fork_t()) {
+		io_uring_enable_rings(&ring);
 		ret = try_submit(&ring);
 		if (ret)
-			fprintf(stderr, "not owner child could submit %i\n", ret);
+			fprintf(stderr, "2: not owner child could submit %i\n", ret);
 		return !!ret;
 	}
 	wait_child_t();
 	io_uring_queue_exit(&ring);
 
+	/* test that only the first enabler can submit */
+	ret = io_uring_queue_init(8, &ring, IORING_SETUP_SINGLE_ISSUER |
+					    IORING_SETUP_R_DISABLED);
+	if (ret)
+		error(1, ret, "ring init (3) %i", ret);
+
+	io_uring_enable_rings(&ring);
+	if (!fork_t()) {
+		ret = try_submit(&ring);
+		if (ret != -EEXIST)
+			fprintf(stderr, "3: not owner child could submit %i\n", ret);
+		return ret != -EEXIST;
+	}
+	wait_child_t();
+	io_uring_queue_exit(&ring);
+
 	/* test that anyone can submit to a SQPOLL|SINGLE_ISSUER ring */
 	ret = io_uring_queue_init(8, &ring, IORING_SETUP_SINGLE_ISSUER|IORING_SETUP_SQPOLL);
 	if (ret)
-		error(1, ret, "ring init (3) %i", ret);
+		error(1, ret, "ring init (4) %i", ret);
 
 	ret = try_submit(&ring);
 	if (ret) {
@@ -139,13 +157,13 @@ int main(int argc, char *argv[])
 	/* test that IORING_ENTER_REGISTERED_RING doesn't break anything */
 	ret = io_uring_queue_init(8, &ring, IORING_SETUP_SINGLE_ISSUER);
 	if (ret)
-		error(1, ret, "ring init (4) %i", ret);
+		error(1, ret, "ring init (5) %i", ret);
 
 	if (!fork_t()) {
 		ret = try_submit(&ring);
-		if (ret)
-			fprintf(stderr, "not owner child could submit %i\n", ret);
-		return !!ret;
+		if (ret != -EEXIST)
+			fprintf(stderr, "4: not owner child could submit %i\n", ret);
+		return ret != -EEXIST;
 	}
 	wait_child_t();
 	io_uring_queue_exit(&ring);
Signature · 1 line
2.30.2
#2 lore raw

PATCH v2 2/3 update documentation to reflect no 5.20 kernel

In reply to Dylan Yudaken

The documentation referenced the wrong kernel version.

---
 man/io_uring_prep_recv.3    | 2 +-
 man/io_uring_prep_recvmsg.3 | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
man/io_uring_prep_recv.3 +1 -1
index 60a77fc241b5..b3862369affa 100644
@@ -55,7 +55,7 @@ If a posted CQE does not have the
 .B IORING_CQE_F_MORE
 flag set then the multishot receive will be done and the application should issue a
 new request.
-Multishot variants are available since kernel 5.20.
+Multishot variants are available since kernel 6.0.
 
 
 After calling this function, additional io_uring internal modifier flags
man/io_uring_prep_recvmsg.3 +1 -1
index 07096ee4826c..65f324dfef9b 100644
@@ -65,7 +65,7 @@ submitted with the request. See
 .B io_uring_recvmsg_out (3)
 for more information on accessing the data.
 
-Multishot variants are available since kernel 5.20.
+Multishot variants are available since kernel 6.0.
 
 After calling this function, additional io_uring internal modifier flags
 may be set in the SQE
Signature · 1 line
2.30.2
#4 lore raw

PATCH v2 3/3 give open-direct-pick.c a unique path

In reply to Dylan Yudaken

This was making make runtest-parallel flaky

---
 test/open-direct-pick.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
test/open-direct-pick.c +1 -1
index b1597e73231e..a1edf8464dc8 100644
@@ -158,7 +158,7 @@ int main(int argc, char *argv[])
 		return 0;
 	}
 
-	path = "/tmp/.open.close";
+	path = "/tmp/.open.direct.pick";
 	t_create_file(path, 4096);
 
 	ret = test_openat(&ring, path);
Signature · 1 line
2.30.2
Reviewed-by on v2 3/3
#5 lore raw

Re: [PATCH liburing v2 3/3] give open-direct-pick.c a unique path

Quoted 6 lines · Dylan Yudaken

This was making make runtest-parallel flaky

---
  test/open-direct-pick.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

Yeah, it clashes with test/open-close.c's.

Signature · 1 line
Ammar Faizi
maintainer Applied
#6 lore raw
In reply to Dylan Yudaken
Quoted 8 lines · Dylan Yudaken

liburing updates for 6.0:

Patch 1 updates to account for the single issuer ring being assigned at ring creation time.
Patch 2 updates man pages from 5.20 -> 6.0
Patch 3 reduces test flakiness

[...]

Applied, thanks!

[1/3] handle single issuer task registration at ring creation

      commit: f37012787e5cde63fb30ae92c9ac25153298dc5b
[2/3] update documentation to reflect no 5.20 kernel
      commit: b08210967b53c339a1cb983e176e8b53b5c8e0db
[3/3] give open-direct-pick.c a unique path
      commit: eb90f4229c0526bace370175eeec1329bf72311b

Best regards,

Signature · 1 line
Jens Axboe