Adding OpenWrt support for Xiaomi AX3600 (Part 1)

kirdes · 2022-08-18T21:10:46.217Z

Done, Dynalink is part of robimarkos PR branch

robimarko · 2022-08-19T08:58:56.659Z

Can somebody share the IRQ balancing script, I want to include it as part of the target.

rmandrad · 2022-08-19T09:40:43.636Z

I believe this is the script that @dchard posted

#assign 4 rx interrupts to each cores
echo 8 > /proc/irq/50/smp_affinity
echo 4 > /proc/irq/51/smp_affinity
echo 2 > /proc/irq/52/smp_affinity
echo 1 > /proc/irq/53/smp_affinity

#assign 3 tcl completions to 3 CPUs
echo 4 > /proc/irq/73/smp_affinity
echo 2 > /proc/irq/74/smp_affinity
echo 1 > /proc/irq/75/smp_affinity

Ansuel · 2022-08-19T09:52:48.049Z

@robimarko news for the single Fw on ax3600?

robimarko · 2022-08-19T09:55:20.984Z

None, havent looked at it, was busy rebasing and sending all of the stuff that was partially reviewed again to try and get it into 6.1

sumo · 2022-08-19T10:03:00.430Z

robimarko:

Damn, had no idea that QCA6391 cards can be this fast

So you didn't watch my ELC talk?

robimarko · 2022-08-19T10:10:13.288Z

Nope, barely got any time to watch the conferences.
Is it public?

sumo · 2022-08-19T10:12:22.401Z

Full ELC Playlist:

robimarko · 2022-08-19T10:14:19.029Z

Thanks, will give it a go over the weekend.

robimarko · 2022-08-20T09:45:01.573Z

I have opened a tracking issue for PR blockers:

github.com/robimarko/openwrt

[TODO] Tracking stuff required for PR

opened 09:44AM - 20 Aug 22 UTC

robimarko

This is a summary issue to track blockers from making a PR for OpenWrt inclusion…. **Remaining TODO**: - [ ] Redo the kernel tree using upstreamed patches - [ ] Move SSDK and NSS-DP out of feed into the main tree - [ ] Single FW on Xiaomi devices - [ ] SMP affinity script for the target - [ ] ath11k needs to be fixed to stop breaking sysupgrade - [ ] ath11k board-ids need to be dropped and BDF-s repackaged without them (Result of upstream conversation) - [ ] Probably ath11k inclusion cleanup as we depend on the target before it even exists and so on

If you think I missed something, please comment.

@sumo They seriously posted 360p recording

XiaoliChan · 2022-08-20T15:13:21.552Z

@robimarko Seems you didn't enable WolfSSL CPU crypto instructions in your build profile.

robimarko · 2022-08-20T15:41:32.159Z

It was the default until recently

XiaoliChan · 2022-08-20T15:47:34.156Z

Okay, I see.

The next question is how to enable the neon version of wolfssl?
(BTW, I only changed "CONFIG_WOLFSSL_HAS_CPU_CRYPTO" to "y".)

robimarko · 2022-08-20T16:26:00.521Z

That will enable the use of crypto extensions I dont know about anything else, though NEON should pretty much always be enabled as its part of ARMv8 and not optional AFAIK.

XiaoliChan · 2022-08-20T17:01:15.088Z

I have confusion on /proc/crypto, even I turn on CPU crypto, I didn't get the "neon" result.

robimarko · 2022-08-20T17:25:05.824Z

That is kernel crypto, NEON is enabled for algos that support it.

XiaoliChan · 2022-08-20T18:36:54.392Z

Thanks for your answer.

kirdes · 2022-08-21T14:20:12.137Z

Unfortunately the firmware version 2.7.0.1-01744 is still crashing under certain conditions, even on the Dynalink

Edrikk · 2022-08-21T19:43:36.900Z

@dchard did you succeed in the end with this version?

If yes, I suggest updating because the current version does have crashed as well…. So hopefully less crashes is better

dchard · 2022-08-21T19:46:59.892Z

No. I repacked the BDF to match the one expected by my AX6, but it crashes instantly on load the same way as with the original AX6. I did not have the time to analyze the cal file sent by Kirdes, if there is any difference there.

Non the less the fact it works at least the same according to Kirdes, points us towards either the BDF or the CAL file, as the driver is the same.