On 2020/10/30 19:50, Joakim Tjernlund wrote:

quoted

On Fri, 2020-10-30 at 09:36 +0800, Yunsheng Lin wrote:

quoted

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.


On 2020/10/29 23:18, David Ahern wrote:

quoted

On 10/29/20 8:10 AM, Joakim Tjernlund wrote:

quoted

OK, bisecting (was a bit of a bother since we merge upstream releases into our tree, is there a way to just bisect that?)

Result was commit "net: sch_generic: aviod concurrent reset and enqueue op for lockless qdisc"  (749cc0b0c7f3dcdfe5842f998c0274e54987384f)

Reverting that commit on top of our tree made it work again. How to fix?

Adding the author of that patch (linyunsheng@huawei.com) to take a look.

quoted

 Jocke

On Mon, 2020-10-26 at 12:31 -0600, David Ahern wrote:

quoted

On 10/26/20 6:58 AM, Joakim Tjernlund wrote:

quoted

Ping  (maybe it should read "arping" instead :)

 Jocke

On Thu, 2020-10-22 at 17:19 +0200, Joakim Tjernlund wrote:

quoted

strace arping -q -c 1 -b -U  -I eth1 0.0.0.0
...
sendto(3, "\0\1\10\0\6\4\0\1\0\6\234\v\6 \v\v\v\v\377\377\377\377\377\377\0\0\0\0", 28, 0, {sa_family=AF_PACKET, proto=0x806, if4, pkttype=PACKET_HOST, addr(6)={1, ffffffffffff},
20) = -1 ENOBUFS (No buffer space available)
....
and then arping loops.

in 4.19.127 it was:
sendto(3, "\0\1\10\0\6\4\0\1\0\6\234\5\271\362\n\322\212E\377\377\377\377\377\377\0\0\0\0", 28, 0, {​sa_family=AF_PACKET, proto=0x806, if4, pkttype=PACKET_HOST, addr(6)={​1,
ffffffffffff}​, 20) = 28

Seems like something has changed the IP behaviour between now and then ?
eth1 is UP but not RUNNING and has an IP address.

"eth1 is UP but not RUNNING" usually mean user has configure the netdev as up,
but the hardware has not detected a linkup yet.

Also What is the output of "ethtool eth1"?

echo 1 >  /sys/class/net/eth1/carrier
cu3-jocke ~ # arping -q -c 1 -b -U  -I eth1 0.0.0.0
cu3-jocke ~ # echo 0 >  /sys/class/net/eth1/carrier
cu3-jocke ~ # arping -q -c 1 -b -U  -I eth1 0.0.0.0
^Ccu3-jocke ~ # ethtool eth1
Settings for eth1:
	Supported ports: [ MII ]
	Supported link modes:   1000baseT/Full 
	Supported pause frame use: Symmetric Receive-only
	Supports auto-negotiation: Yes
	Advertised link modes:  1000baseT/Full 
	Advertised pause frame use: Symmetric Receive-only
	Advertised auto-negotiation: Yes
	Speed: 10Mb/s
	Duplex: Half
	Port: MII
	PHYAD: 1
	Transceiver: external
	Auto-negotiation: on
	Current message level: 0x00000037 (55)
			       drv probe link ifdown ifup
	Link detected: no

We have a writeable carrier since eth device is PHY less. Maybe that path is different ?
Check drivers/net/ethernet/freescale/dpaa/dpa_eth.c

The above difference does not seems to matter.

quoted

quoted

It would be good to see the status of netdev before and after executing arping cmd
too.

hmm, how do you mean?

I was trying to find out when the netdev' state became "eth1 is UP but not RUNNING".

Anyway, when I looked at the backported patch, I did find new qdisc assignment is
missing from the upstream patch.

Please see if the below patch fix your problem, thanks:

diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index bd96fd2..4e15913 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c

@@ -1116,10 +1116,13 @@ static void dev_deactivate_queue(struct net_device *dev,

                                 void *_qdisc_default)
 {
        struct Qdisc *qdisc = rtnl_dereference(dev_queue->qdisc);
+       struct Qdisc *qdisc_default = _qdisc_default;

        if (qdisc) {
                if (!(qdisc->flags & TCQ_F_BUILTIN))
                        set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
+
+               rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
        }
 }