dmesg flood with fail 'cur < head || cur > kring->nr_hwtail' while running click with netmap

[mwaldiya]

Hi All,

I am running userlevel click with netmap(with igb patched driver).
I am able to get the packets but in dmesg i am continuously getting the following messages:

[15306.056809] 595.463893 [1662] nm_rxsync_prologue eth7 RX0: fail 'cur < head || cur > kring->nr_hwtail' h 0 c 255 t 0 rh 0 rc 255 rt 0 hc 254 ht 0
[15306.121517] 595.528551 [1655] nm_rxsync_prologue eth7 RX0: fail 'cur < head || cur > kring->nr_hwtail' h 71 c 70 t 71 rh 71 rc 70 rt 71 hc 68 ht 71
[15306.121601] 595.528637 [1698] netmap_ring_reinit called for eth7 RX0
[15306.129491] 595.536519 [1655] nm_rxsync_prologue eth7 RX0: fail 'cur < head || cur > kring->nr_hwtail' h 217 c 216 t 217 rh 217 rc 216 rt 217 hc 215 ht 217
[15306.129597] 595.536627 [1698] netmap_ring_reinit called for eth7 RX0
[15306.131698] 595.538724 [1655] nm_rxsync_prologue eth7 RX0: fail 'cur < head || cur > kring->nr_hwtail' h 38 c 37 t 38 rh 38 rc 37 rt 38 hc 35 ht 38
[15306.131790] 595.538818 [1698] netmap_ring_reinit called for eth7 RX0
[15306.183361] 595.590346 [1655] nm_rxsync_prologue eth7 RX0: fail 'cur < head || cur > kring->nr_hwtail' h 249 c 248 t 249 rh 249 rc 248 rt 249 hc 247 ht 249
[15306.183467] 595.590455 [1698] netmap_ring_reinit called for eth7 RX0
[15306.188469] 595.595451 [1655] nm_rxsync_prologue eth7 RX0: fail 'cur < head || cur > kring->nr_hwtail' h 243 c 242 t 243 rh 243 rc 242 rt 243 hc 241 ht 243
[15306.188576] 595.595559 [1698] netmap_ring_reinit called for eth7 RX0
[15306.190770] 595.597752 [1698] netmap_ring_reinit called for eth7 RX0
[15306.276476] 595.683390 [1698] netmap_ring_reinit called for eth7 RX0
[15306.291895] 595.698798 [1698] netmap_ring_reinit called for eth7 RX0
[15306.380849] 595.787681 [1698] netmap_ring_reinit called for eth7 RX0
[15306.544454] 595.951158 [1698] netmap_ring_reinit called for eth7 RX0

Is this normal or am i doing something wrong while running the netmap with click.

following is my click script element that i am using for from device and to device using netmap:

from_eth :: FromDevice(netmap:eth6*, BURST 16,SNIFFER false, METHOD NETMAP);
intf1 ::ToDevice(netmap:eth7, BURST 16, METHOD NETMAP);

Thanks

[tbarbette]

Hi,
It is normal that Netmap is verbose.
However, the current implementation is not efficient at all and was tested only with an old version of Netmap. There is a pending pull request (#225 ) that makes it a little better (zero copy support at least), or you can use FastClick's one.
DPDK support is much more up-to-date, if you're not particularly stuck with Netmap.

Tom

[mwaldiya]

Thanks tom for clarifying on this.
As i have got an "rcu_sched self-detected stall on CPU" while running FromDevice in 2 simultaneous threads using netmap devices.

I am currently not looking for the DPDK option.

[tbarbette]

Ouch, that basically means Netmap blocked one CPU core... That happens when you misbehave with the Netmap API. Can you try the pull-request version?

[mwaldiya]

Will run with the patch and share the result.

Thanks

[mwaldiya]

Hi,
The patch work well for me, from last 3 days click with the patch running fine, and it also suppresses the above mentioned logs.

Thanks @tbarbette .