Skip to contentSkip to conversation

How is tx-flushing supposed to work now?

DISCUSSION Ben Greear 2 participants 5 messages Mar 29, 2016 – Mar 31, 2016 Qualcomm Atheros ATH10K wireless driver

lore
Share & tools
lore

Ben Greear

author
#0 lore raw

When we have multiple vdevs, and the 10.4.3 firmware with the tx-push logic, how is flushing supposed to work?

First, when we get a flush, should we first purge any frames not yet pushed to the firmware?

And second, what keeps other vdevs from continuing to transmit frames while one vdev is trying to be flushed?

Thanks,
Ben

Signature · 8 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com


_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k
#1 lore raw
Quoted 2 lines · Ben Greear

When we have multiple vdevs, and the 10.4.3 firmware with the tx-push logic, how is flushing supposed to work?

This isn't really as much about pull-push logic in 10.4.3 as it is about wake_tx_queue() in general, is it?

Quoted 2 lines

First, when we get a flush, should we first purge any frames not yet pushed to the firmware?

I would expect mac80211 to purge txqs.

Quoted 3 lines

And second, what keeps other vdevs from continuing to transmit frames while one
vdev is trying to be flushed?

Hmm.. I guess ath10k flush() implementation doesn't really do what it should. It should wait just for frames for given vif but it waits for everything.

Michał

_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k

author
#2 lore raw
Quoted 6 lines · Michal Kazior
Quoted 2 lines · Ben Greear

When we have multiple vdevs, and the 10.4.3 firmware with the tx-push logic, how is flushing supposed to work?

This isn't really as much about pull-push logic in 10.4.3 as it is about wake_tx_queue() in general, is it?

Maybe, I'm a bit confused about how it all works.

Quoted 6 lines
Quoted 2 lines

First, when we get a flush, should we first purge any frames not yet pushed to the firmware?

I would expect mac80211 to purge txqs.

Quoted 7 lines
Quoted 3 lines

And second, what keeps other vdevs from continuing to transmit frames while one
vdev is trying to be flushed?

Hmm.. I guess ath10k flush() implementation doesn't really do what it should. It should wait just for frames for given vif but it waits for everything.

Yes, I think so. If mac80211 is trying to flush a single vif, then probably it is letting other vifs on the same radio still transmit as well, so likely ath10k tx-queues are not going to go to zero in that case.

Thanks,
Ben

Signature · 7 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com

_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k
author
#3 lore raw
In reply to Michal Kazior
Quoted 6 lines · Michal Kazior
Quoted 2 lines · Ben Greear

When we have multiple vdevs, and the 10.4.3 firmware with the tx-push logic, how is flushing supposed to work?

This isn't really as much about pull-push logic in 10.4.3 as it is about wake_tx_queue() in general, is it?

In particular, I am curious about the ar->txqs, and the ath10k_mac_tx_push_pending method. Should we purge the ar->txqs as part of the flush logic?

Thanks,
Ben

Quoted 19 lines
Quoted 4 lines

First, when we get a flush, should we first purge any frames not yet pushed to the firmware?

I would expect mac80211 to purge txqs.

Quoted 3 lines

And second, what keeps other vdevs from continuing to transmit frames while one
vdev is trying to be flushed?

Hmm.. I guess ath10k flush() implementation doesn't really do what it should. It should wait just for frames for given vif but it waits for everything.

Michał

Signature · 8 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com


_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k
#4 lore raw
Quoted 9 lines · Ben Greear
Quoted 8 lines · Michal Kazior
Quoted 4 lines · Ben Greear

When we have multiple vdevs, and the 10.4.3 firmware with the tx-push logic,
how is flushing supposed to work?

Quoted 7 lines
Quoted 2 lines

This isn't really as much about pull-push logic in 10.4.3 as it is about wake_tx_queue() in general, is it?

In particular, I am curious about the ar->txqs, and the ath10k_mac_tx_push_pending
method. Should we purge the ar->txqs as part of the flush logic?

No need to. If mac80211 purges some txqs ath10k will eventually visit an empty txq and remove it (push/threshold mode). If ath10k FW is in pull-push it'll prompt fetch-ind event which will 1) submit 0 frames
2) update hostq state.

Michał

_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k