push-pull stalls (QCA 9888)

push-pull stalls (QCA 9888)

[Guilherme Weidle]

Hi,

I'm suffering from a bug where in AP mode, the TX queue stalls.

If I connect 5 stations and force some traffic (bandwidth test for
instance), the AP's TX queue seems to stall (no data frames are
getting out).

I can reply this issue by using another device (ath9k) with 5 virtual
interfaces (stations) connected to my AP (ath10k). Then, I start iperf
server on that device and start one iperf client (from AP) per
associated station. And just like that, AP stops sending data frames
and I no longer be able to ping any connected station.

If I do the same test, but with 4 stations instead of 5, it works just fine.

As far as I could see, it is related to PUSH-PULL mode. As soon as the
WLAN firmware switches to this mode, the issue happens.

Just as a test, I change the ath10k_mac_tx_can_push() to always return
true and the issue goes away.

Does anyone also came across this issue or have any idea how to solve it?

PS: This product is using kernel 4.4.60 and backports 4.19.137-1. I
also updated to backports 20210222_001 and the problem still there.

Regards,

Guilherme Weidle

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: push-pull stalls (QCA 9888)

[Maxime Bizon]

On Tue, 2022-02-08 at 16:31 -0300, Guilherme Weidle wrote:

Hello,

> 
> Does anyone also came across this issue or have any idea how to solve
> it?
> 

I do and noticed the exact same thing with the push-pull switch, but I
did not have a clean way to reproduce like you have.

Can you please try to revert this:

"mac80211: Switch to a virtual time-based airtime scheduler"

and tell me if it fixes your problem ?

-- 
Maxime




_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: push-pull stalls (QCA 9888)

[Guilherme Weidle]

On Thu, Feb 10, 2022 at 4:48 PM Maxime Bizon <mbizon@freebox.fr> wrote:
>
>
> On Tue, 2022-02-08 at 16:31 -0300, Guilherme Weidle wrote:
>
> Hello,
>
> >
> > Does anyone also came across this issue or have any idea how to solve
> > it?
> >
>
> I do and noticed the exact same thing with the push-pull switch, but I
> did not have a clean way to reproduce like you have.
>
> Can you please try to revert this:
>
> "mac80211: Switch to a virtual time-based airtime scheduler"
>
> and tell me if it fixes your problem ?
>
> --
> Maxime
>
>
>
The kernel I'm using here (v4.4.60) does not have this patch. I
believe this issue has been around for a long time now and is still
there...

--
Guilherme Weidle

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: push-pull stalls (QCA 9888)

[Maxime Bizon]

On Fri, 2022-02-11 at 11:51 -0300, Guilherme Weidle wrote:

> 
> The kernel I'm using here (v4.4.60) does not have this patch. I
> believe this issue has been around for a long time now and is still
> there...

Sorry I thought you would have it since you use backport, but you run a
version that predates this patch

-- 
Maxime




_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: push-pull stalls (QCA 9888)

[Guilherme Weidle]

On Fri, Feb 11, 2022 at 1:03 PM Maxime Bizon <mbizon@freebox.fr> wrote:
>
>
> On Fri, 2022-02-11 at 11:51 -0300, Guilherme Weidle wrote:
>
> >
> > The kernel I'm using here (v4.4.60) does not have this patch. I
> > believe this issue has been around for a long time now and is still
> > there...
>
> Sorry I thought you would have it since you use backport, but you run a
> version that predates this patch
>
> --
> Maxime

That's alright. Thanks anyway!

Let's see if anyone has another clue about this issue.

BTW: Did you fix your problem by removing that patch?

--
Guilherme Weidle

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: push-pull stalls (QCA 9888)

[Maxime Bizon]

On Fri, 2022-02-11 at 14:00 -0300, Guilherme Weidle wrote:

> BTW: Did you fix your problem by removing that patch?

Yes

I exchanged a bit with Toke and maybe the ath10k push-pull code is
somewhat incompatible with the new scheduler, he gave me some ideas to
try but I did not have time to try them yet.

Could you share your exact reproducer ?

-- 
Maxime




_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: push-pull stalls (QCA 9888)

[Guilherme Weidle]

> On Fri, 2022-02-11 at 14:00 -0300, Guilherme Weidle wrote:
>
> > BTW: Did you fix your problem by removing that patch?
>
> Yes
>
> I exchanged a bit with Toke and maybe the ath10k push-pull code is
> somewhat incompatible with the new scheduler, he gave me some ideas to
> try but I did not have time to try them yet.
>
> Could you share your exact reproducer ?

Nice to hear that you have some ideas.

Here is what I've done to reproduce the issue:

#### Station Script - ath9k device ####
# Create virtual stations
SSID="ath10k_test"
i=1
while [ $i -le 5 ]; do
    iw dev wlan0 interface add viface$i type station
    ip link set viface$i address 70:00:00:00:00:0$i
    ifconfig viface$i 10.0.0.$i netmask 255.255.255.0 up
    iw dev viface$i connect $SSID
    i=$((i+1))
done

# Sometimes just couple of interfaces connect, so we need to repeat
i=1
while [ $i -le 5 ]; do
    iw dev viface$i connect $SSID
    i=$((i+1))
done

# Start iperf server
iperf -s
###############################


#### AP Script - ath10k device ####
IPS="1 2 3 4 5"
for ip in $IPS; do iperf -c 10.0.0.$ip & done
###################################

--
Guilherme Weidle

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k