2013-05-31 05:12:42

by Chris David

[permalink] [raw]
Subject: incoming packets seemed to be delayed with rtl8723ae


I have an rtl8723ae card. I'm noticing a "minature" version of the
bug reported with this subject line:
rtl8723ae losing IP connectivity (3.8.2 + 3.9-rc2)

My description would be: When I ping the machine this card is
installed in from another machine, some packets seemed to be delayed.
Here is a sample ping output:

64 bytes from icmp_req=194 ttl=64 time=2002 ms
64 bytes from icmp_req=195 ttl=64 time=996 ms
64 bytes from icmp_req=196 ttl=64 time=1.15 ms
64 bytes from icmp_req=197 ttl=64 time=1.25 ms
64 bytes from icmp_req=198 ttl=64 time=2002 ms
64 bytes from icmp_req=199 ttl=64 time=1002 ms
64 bytes from icmp_req=200 ttl=64 time=3.66 ms
64 bytes from icmp_req=201 ttl=64 time=5.46 ms
64 bytes from icmp_req=202 ttl=64 time=2.21 ms
64 bytes from icmp_req=203 ttl=64 time=1005 ms
64 bytes from icmp_req=204 ttl=64 time=5.73 ms
64 bytes from icmp_req=205 ttl=64 time=1.25 ms
64 bytes from icmp_req=206 ttl=64 time=1.72 ms
64 bytes from icmp_req=207 ttl=64 time=1011 ms

As you can see, sometimes there is a second or two delay, other times
its very quick. And the delayed packets seem to happen every second
or two, very consistently.

This happens to be on kernel 3.9.4, with the backports-3.10-rc1
drivers installed. But I've seen this behavior on a 3.8.13 kernel
also with the same backports, and also on 3.10-rc3. On 3.9.4 and
3.8.13 without backports, I observe the same issue except its 50
seconds, instead of 2.

I've seen this behaviour with wpa_supplicant using both the wext
driver on 3.8.18, and the nl80211 driver on 3.9.4.

If the pings are sourced from the box with the 8723ae installed, I
don't observe this behavior. Its only when the pings come from some
external box. Also, when I ssh into the box with the 8723ae from an
external machine, I notice the same sort of delays with the commands
that I type. (its making me a better typist!)

It's "as-if" the buffer for incoming packets (at least TCP and UDP),
likes to hang on to the data for roughly mod(gettime(), 2) amount of

I'm definitely willing to try patches, and even might have the skill
to fix it. But if anybody has hints or advice, or perhaps a reference
to the patch that reeled this bug in from 50 seconds to 2 seconds,
that would be good information for me to know. Also, I really don't
know its a 8723ae problem. But, I can say I don't observe this
behavior on the wired ethernet ports on the same box. So it seems at
least wireless related.

There's not much in dmesg that looks out of place. And iw event -t
doesn't print anything while this issue happens.
[ 4.724126] rtl8723ae: Using firmware rtlwifi/rtl8723fw_B.bin

BTW, I'm subscribed to this email list.

Thank you,