netdev - Re: [BUG] pfifo_fast may cause out-of-order CAN frame transmission

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-Id: <20201007210744.8546-1-vijayendra.suman@oracle.com>
Date:   Wed,  7 Oct 2020 14:07:44 -0700
From:   Vijayendra Suman <vijayendra.suman@...cle.com>
To:     pabeni@...hat.com
Cc:     a.fatoum@...gutronix.de, kernel@...gutronix.de,
        linux-can@...r.kernel.org, netdev@...r.kernel.org,
        somasundaram.krishnasamy@...cle.com,
        ramanan.govindarajan@...cle.com,
        Vijayendra Suman <vijayendra.suman@...cle.com>
Subject: Re: [BUG] pfifo_fast may cause out-of-order CAN frame transmission

[PATCH] Patch with Network Performance Improvment qperf:tcp_lat

Check Performed for __QDISC_STATE_DEACTIVATED before checking BYPASS flag 

qperf tcp_lat 65536bytes over an ib_switch 

For 64K packet Performance improvment is around 47 % and performance deviation 
is reduced to 5 % which was 27 % prior to this patch.

As mentioned by Paolo, With  "net: dev: introduce support for sch BYPASS for lockless qdisc" commit
there may be out of order packet issue.
Is there any update to solve out of order packet issue.

qperf Counters for tcp_lat for 60 sec and packet size 64k

With Below Patch
1. 53817 
2. 54100 
3. 57016 
4. 59410 
5. 62017 
6. 54625 
7. 55770 
8. 54015 
9. 54406 
10. 53137

Without Patch [Upstream Stream]
1. 83742 
2. 107320 
3. 82807 
4. 105384 
5. 77406 
6. 132665 
7. 117566 
8. 109279 
9. 94959 
10. 82331 
11. 91614 
12. 104701 
13. 91123 
14. 93908 
15. 200485 

With UnRevert of commit 379349e9bc3b42b8b2f8f7a03f64a97623fff323 
[Revert "net: dev: introduce support for sch BYPASS for lockless qdisc"] 

1. 65550
2. 64285
3. 64110
4. 64300
5. 64645
6. 63928
7. 63574
8. 65024
9. 65153
10. 64281

Signed-off-by: Vijayendra Suman <vijayendra.suman@...cle.com>
---
 net/core/dev.c | 27 ++++++++++-----------------
 1 file changed, 10 insertions(+), 17 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 40bbb5e43f5d..6cc8e0209b20 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3384,35 +3384,27 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
 				 struct net_device *dev,
 				 struct netdev_queue *txq)
 {
 	struct sk_buff *to_free = NULL;
 	bool contended;
-	int rc;
+	int rc = NET_XMIT_SUCCESS;
 
 	qdisc_calculate_pkt_len(skb, q);
 
 	if (q->flags & TCQ_F_NOLOCK) {
-		if ((q->flags & TCQ_F_CAN_BYPASS) && READ_ONCE(q->empty) &&
-		    qdisc_run_begin(q)) {
-			if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
-					      &q->state))) {
-				__qdisc_drop(skb, &to_free);
-				rc = NET_XMIT_DROP;
-				goto end_run;
-			}
-			qdisc_bstats_cpu_update(q, skb);
-
-			rc = NET_XMIT_SUCCESS;
+		if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
+			__qdisc_drop(skb, &to_free);
+			rc = NET_XMIT_DROP;
+		} else if ((q->flags & TCQ_F_CAN_BYPASS) && READ_ONCE(q->empty) &&
+				qdisc_run_begin(q)) {
+			qdisc_bstats_update(q, skb);
 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true))
 				__qdisc_run(q);
-
-end_run:
 			qdisc_run_end(q);
 		} else {
 			rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
 			qdisc_run(q);
 		}
-
 		if (unlikely(to_free))
 			kfree_skb_list(to_free);
 		return rc;
-- 
2.27.0