ipq40xx: switch QCA4019 firmware to -ct #2541
Conversation
|
Should we also switch to the ct kernel module? Maybe it would be a good idea to keep the default driver and module for all ath10k variants where ct supports 11s? |
|
@NeoRaider Verified mesh working for QCA9884 and QCA9886. Regarding switching to -ct kernel-module: For sake of upstream consistency, I'd say this might be a good idea. The rationale for keeping the non-ct flavor was the lacking mesh-feature back in the day (which is only the case on Wave-1 nowadays). |
blocktrron
force-pushed
the
ipq4019-ath10k-ct
branch
from
0d23c6c
to
f639175
Compare
|
For ath9k/minstrel there are known issues that they become exceedingly aggressive when a neighbor vanishes without dissociation. Especially when there is data without congestion control from the network stack which keeps on pushing. In that case Minstrel goes down to 1 MBit/s, uses many retries and even uses CTS/RTS. @blocktrron just out ouf curiousity, if ath10k firmware's internal rate control might do something similar to Minstrel-HT, did you have lingering neighbors in "iw dev wlanX station dump" with a high inactive timer? Or did you also see these many retries with only active neighbors/clients, too? |
targets/ipq40xx-generic
Outdated
| '-ath10k-firmware-qca4019', | ||
| 'ath10k-firmware-qca4019-ct', | ||
| '-ath10k-firmware-qca9888', | ||
| 'ath10k-firmware-qca9888-ct', |
There was a problem hiding this comment.
I think we should just drop these variables altogether where they don't change anything.
There was a problem hiding this comment.
Thought the same, but i won't rule out we need to change it in the future, so I would keep them.
There was a problem hiding this comment.
Hmm, or keep the variable, but make it an empty list? That would also give us the smallbuffers configuration where OpenWrt defaults to it.
There was a problem hiding this comment.
Empty list sounds like a good compromise.
|
@T-X we see this already with 1 station associated and only in case the link is established with an 802.11ac device. The effect is very noticeable, throughput reduces from ~60 Mbit/s to less than 1 Mbit/s in a 1 clients single VAP scenario. I still have the pcaps in case you are interested. To me this looks like the rate control does not correctly drive down the rate (3x retransmitted with the same rate). |
Use the candelatech firmware for the QCA Wave-2 firmware. The Qualcomm firmware used for the IPQ401x chip in OpenWrt in 22.03 is experiencing heavily degraded performance due to excessive retransmits when using A-MSDU. Disabling VHT modes or switching to the candelatech firmware circumvents this issue. Apply the same to other Wave-2 platforms in order to keep consistency with upstream. Wave-1 chips do not support mesh modes with the -ct firmware, so keep using the QCA firmware in their case. Signed-off-by: David Bauer <[email protected]>
blocktrron
force-pushed
the
ipq4019-ath10k-ct
branch
from
f639175
to
9335897
Compare
It seems like the Aruba devices have problems with the eLNA control and the the official firmware. But many other devices have packet loss with the ct firmware. This reverts commit 15ef885.
|
there are issues with that change, see #2692 |
This is a temporary measure that fixes freifunk-gluon#2692. This reverts commit 15ef885.
Revert "ipq40xx: switch Wave2 firmware to -ct (#2541)"
This is a temporary measure that fixes freifunk-gluon#2692. This reverts commit 15ef885. (cherry picked from commit 22c47df)
[Backport v2022.1.x] Revert "ipq40xx: switch Wave2 firmware to -ct (#2541)"
…funk-gluon#2541)"" This partially reverts commit 22c47df. Devices in ath79-generic like the TP-Link EAP225-Outdoor v1 are really unstable with the non -ct Wave2 firmware and regulary crash with 100% memory consumption when only a handful devices are connected.
Use the candelatech firmware for the QCA Wave-2 firmware. The Qualcomm firmware used for the IPQ401x chip in OpenWrt in 22.03 is experiencing heavily degraded performance due to excessive retransmits when using A-MSDU. Disabling VHT modes or switching to the candelatech firmware circumvents this issue. Apply the same to other Wave-2 platforms in order to keep consistency with upstream. Wave-1 chips do not support mesh modes with the -ct firmware, so keep using the QCA firmware in their case. Signed-off-by: David Bauer <[email protected]>
This is a temporary measure that fixes freifunk-gluon#2692. This reverts commit 15ef885.
Use the candelatech firmware for the QCA4019 WiSoC.
The Qualcomm firmware used for this specific chip in OpenWrt in 22.03
is experiencing heavily degraded performance due to excessive
retransmits when using A-MSDU. Disabling VHT modes or switching to the
candelatech firmware circumvents this issue.
Signed-off-by: David Bauer [email protected]