Change in libosmo-netif[master]: jibuf: Fix out-of-order seq queue around syncpoints

This is merely a historical archive of years 2008-2021, before the migration to mailman3.

A maintained and still updated list archive can be found at https://lists.osmocom.org/hyperkitty/list/gerrit-log@lists.osmocom.org/.

Pau Espin Pedrol gerrit-no-reply at lists.osmocom.org
Tue May 15 14:51:44 UTC 2018


Pau Espin Pedrol has uploaded this change for review. ( https://gerrit.osmocom.org/9172


Change subject: jibuf: Fix out-of-order seq queue around syncpoints
......................................................................

jibuf: Fix out-of-order seq queue around syncpoints

Fixes: OS#3262

Change-Id: Ib8c61dbe6261cf73d6efcd7873e23b7656117556
---
M src/jibuf.c
M tests/jibuf/jibuf_test.c
M tests/jibuf/jibuf_test.ok
3 files changed, 36 insertions(+), 7 deletions(-)



  git pull ssh://gerrit.osmocom.org:29418/libosmo-netif refs/changes/72/9172/1

diff --git a/src/jibuf.c b/src/jibuf.c
index 45019ae..2632a57 100644
--- a/src/jibuf.c
+++ b/src/jibuf.c
@@ -119,6 +119,35 @@
 
 }
 
+static void enqueue_pkt(struct osmo_jibuf *jb, struct msgb  *msg, bool is_syncpoint)
+{
+	struct msgb *cur;
+	struct timeval *msg_ts;
+
+	if (!is_syncpoint) {
+		llist_add_sorted(msg, &jb->msg_list);
+		return;
+	}
+
+	/* syncpoints change the reference timings, and as such they can provoke
+	   out of order enqueuing of this packet and its followups with regards
+	   to the already stored packets which may be scheduled for later times.
+	   We thus need to adapt dequeue time for the already stored pkts to be
+	   dequeued before the syncpoint pkt. See OS#3262 for related scenarios.
+	*/
+
+	msg_ts = msgb_scheduled_ts(msg);
+
+	llist_for_each_entry(cur, &jb->msg_list, list) {
+		struct timeval *cur_ts = msgb_scheduled_ts(cur);
+		if (timercmp(msg_ts, cur_ts, <))
+			*cur_ts = *msg_ts;
+	}
+	/* syncpoint goes always to the end since we moved all older packets
+	   before it */
+	llist_add_tail(&msg->list, &jb->msg_list);
+}
+
 static bool msg_get_marker(struct msgb *msg)
 {
 	/* TODO: make it more generic as a callback so that different types of
@@ -314,11 +343,13 @@
 {
 	int rel_delay, delay;
 	struct timeval delay_ts, sched_ts;
+	bool is_syncpoint;
 
 	clock_gettime_timeval(CLOCK_MONOTONIC, &jb->last_enqueue_time);
 
 	/* Check if it's time to sync, ie. start of talkspurt */
-	if (!jb->started || msg_is_syncpoint(jb, msg)) {
+	is_syncpoint = !jb->started || msg_is_syncpoint(jb, msg);
+	if (is_syncpoint) {
 		jb->started = true;
 		msg_set_as_reference(jb, msg);
 		rel_delay = 0;
@@ -365,8 +396,7 @@
 	jbcb->ts = sched_ts;
 	jbcb->old_cb = old_cb;
 
-	llist_add_sorted(msg, &jb->msg_list);
-
+	enqueue_pkt(jb, msg, is_syncpoint);
 
 	/* See if updating the timer is needed: */
 	if (!osmo_timer_pending(&jb->timer) ||
diff --git a/tests/jibuf/jibuf_test.c b/tests/jibuf/jibuf_test.c
index 7993a65..ba1bb01 100644
--- a/tests/jibuf/jibuf_test.c
+++ b/tests/jibuf/jibuf_test.c
@@ -673,8 +673,7 @@
 	clock_override_add(0, min_delay*1000);
 	osmo_select_main(0);
 
-	/* This assert shows that packets are queued out of order in this case:*/
-	OSMO_ASSERT(!osmo_jibuf_empty(jb));
+	OSMO_ASSERT(osmo_jibuf_empty(jb));
 
 	osmo_jibuf_delete(jb);
 }
diff --git a/tests/jibuf/jibuf_test.ok b/tests/jibuf/jibuf_test.ok
index 6bf52fb..b92f73c 100644
--- a/tests/jibuf/jibuf_test.ok
+++ b/tests/jibuf/jibuf_test.ok
@@ -383,10 +383,10 @@
 sys={0.140000}, mono={0.140000}: dequeue: seq=34 ts=720 INTERMEDIATE
 sys={0.140000}, mono={0.140000}: dequeue: seq=35 ts=880 INTERMEDIATE
 sys={0.140000}, mono={0.140000}: dequeue: seq=36 ts=1040 INTERMEDIATE
-sys={0.140000}, mono={0.140000}: dequeue: seq=39 ts=1520 INTERMEDIATE
 sys={0.140000}, mono={0.140000}: dequeue: seq=37 ts=1200 INTERMEDIATE
-sys={0.140000}, mono={0.140000}: dequeue: seq=40 ts=1680 LATEST
 sys={0.140000}, mono={0.140000}: dequeue: seq=38 ts=1360 INTERMEDIATE
+sys={0.140000}, mono={0.140000}: dequeue: seq=39 ts=1520 INTERMEDIATE
+sys={0.140000}, mono={0.140000}: dequeue: seq=40 ts=1680 LATEST
 ===test_rtp_out_of_sync(1600, 5, 800, 1)===
 sys={0.000000}, mono={0.000000}: clock_override_set
 sys={0.000000}, mono={0.000000}: enqueue 1st packet (seq=33, ts=560)

-- 
To view, visit https://gerrit.osmocom.org/9172
To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings

Gerrit-Project: libosmo-netif
Gerrit-Branch: master
Gerrit-MessageType: newchange
Gerrit-Change-Id: Ib8c61dbe6261cf73d6efcd7873e23b7656117556
Gerrit-Change-Number: 9172
Gerrit-PatchSet: 1
Gerrit-Owner: Pau Espin Pedrol <pespin at sysmocom.de>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osmocom.org/pipermail/gerrit-log/attachments/20180515/4fc00261/attachment.htm>


More information about the gerrit-log mailing list