How to compile master branch kernel 5.4 with DSA driver? In target/linux/ramips there is only the kernel 4.14 directory and the 5.4 patch directory. I don't see 5.4 kernel files or DSA driver. And KERNEL_PATCHVER: = 4.14 appears in the Makefile file.
Just make clean, git pull and compile it for the latest master. It will compile with kernel 5.4 for ramips.
1 Like
Thanks. Compiling it.
Running the latest master powered with Linux 5.4 kernel now. According to the output:
CPU0 CPU1 CPU2 CPU3
8: 225655 225453 225761 225733 MIPS GIC Local 1 timer
9: 6781 0 0 0 MIPS GIC 63 IPI call
10: 0 2403 0 0 MIPS GIC 64 IPI call
11: 0 0 7466 0 MIPS GIC 65 IPI call
12: 0 0 0 2508 MIPS GIC 66 IPI call
13: 6960 0 0 0 MIPS GIC 67 IPI resched
14: 0 55429 0 0 MIPS GIC 68 IPI resched
15: 0 0 18217 0 MIPS GIC 69 IPI resched
16: 0 0 0 52894 MIPS GIC 70 IPI resched
17: 0 0 0 0 MIPS GIC 19 1e000600.gpio-bank0, 1e000600.gpio-bank1, 1e000600.gpio-bank2
19: 12 0 0 0 MIPS GIC 33 ttyS0
20: 0 0 0 0 MIPS GIC 29 xhci-hcd:usb1
21: 469880 0 0 0 MIPS GIC 10 1e100000.ethernet
23: 0 0 0 0 1e000600.gpio 7 keys
24: 0 0 0 0 1e000600.gpio 18 keys
25: 2 0 74989 0 MIPS GIC 11 mt76x2e
26: 97702 0 0 0 MIPS GIC 31 mt76x2e
ERR: 1
it seems the WiFi interrupt 31 (mt76x2e) is using cpu0, which used cpu 1 with 4.14 IIRC.
Edit: Much less ERRs, which is a good sign. I thought it can be 0 ERR until the first one emerged.
CPU1 is a virtual core, only CPU0 and CPU2 are real. I believe it is correct to tie the two wifi cards to the two actually existing cores.
On my side there are 250 ERRs, but no actual issues. Even on my x86 NAS (HP microserver) there are a few errors in /proc/interrupts, so this may even be completely normal.
There's a real issue here, the ed2k server I was connected will always lost connection every 15 minutes, which is normal before with 4.14 kernel.
This is with soft offloading on, now I am testing how's it going with soft offloading off.
Edit: I remember I have had similar issue with 4.14 and hard offloading, but that issue has been resolved after some time.
Edit2: Without soft offloading, the 860L only can handle 200Mbps download or upload. I started to miss the full-functioned 4.14.
Mostly correct. Assuming the other configs don't care about raw interfaces.
This is snapshot and not guaranteed to be stable. Give it time for the kinks to be worked out.
Sure, those that use interfaces will modify it
New kernel only occupy 22 MB ram, much less than around 30 MB with 4.14 kernel. Isn’t it fascinating.
Hello,
I have netgear r6220 and has same problem with
mtk_soc_eth 1e100000.ethernet eth0: transmit timed out
I want to test software based on kernel 5.4 but unfortunately is not available here:
http://downloads.openwrt.org/snapshots/targets/ramips/mt7621/
because of build failure:
could someone help, please?
Today, just now, there's a bunch of commits to the master, in which I believe are hw nat related. So it looks like the hw nat is back on horse again. 
Had a quick test with Ookla speed test, with no offloading the 860L only can reach a little bit more than 200Mbps, with soft offloading it can reach around 370Mbps, and with hw offload it became a little bit more than 400Mbps.
It seems hw offloading works, in some way. But the test consumes one core, which should not happen once the real hw offload in place.
Hope there's a working hw offload 5.4 version in the next weeks, perhaps?
No, that is for the mediatek traget, ours is the ramips target. There is no HW offload in sight so far.
On 5.4 kernel, you can use SW offload and you need to set the RPS to "e":
echo e > /sys/class/net/eth0/queues/rx-0/rps_cpus
to get proper WAN speed.
5 Likes
The latest build has succeeded and is now on the snapshots page.
1 Like
The new 5.4 kernel will cause ed2k server not be able to keep-alive, whether soft offloading is on or off.
It didn't happen with former 4.14.x kernels.
Edit: Sorry for the rush, the issue only with soft offloading is on. Need to restart the client after back to off, it appeared okay after that.
Some good news from IRC:
00:26 < Rene__> blogic: Thanks for the hwnat! Can it also be used on mt7621?
00:28 < blogic> rmilecki: and on its way upstrema
00:28 < blogic> Rene__: i will rebase it on ramips when its bumped to v5.4
00:28 < rmilecki> blogic: like patches sent?
00:29 < blogic> no but grabbed a job to send it upstream
00:29 < blogic> rmilecki: problem is that flow offload is part of the ethernet driver
00:29 < blogic> but with mt7622/9 we can also do wifi offload
00:29 < blogic> so we need an extra pdev to desacribe the offload engine
00:29 < blogic> and then the drivers can check if they share this
00:29 < rmilecki> oh, fun
00:30 < blogic> right now the hw need to check iof the in/egress device both are on the same hwnat
00:30 < blogic> gets more complicated when we do qca80 offload where it is only offloading inside DSA
00:30 < blogic> but I am on it ;)
00:30 < blogic> i am able to shift wirespeed with 256byte frames
00:31 < blogic> at 0% cpu load
00:31 < blogic> at 128byte frames it drops to ~780mbit
00:32 < Rene__> blogic: openwrt/master is bumped to v5.4 or are you waiting that it is in v19?
00:32 < blogic> Rene__: ramips is on v5.4 ?
00:32 < blogic> what is v19 ?
00:33 < Rene__> v19 = v19.0xx openwrt branch
00:33 < gch981213> blogic: I've pushed the patches for mt7621 last Saturday.
00:33 < blogic> gch981213: ah ok, let me bump the driver in that case
00:33 < blogic> is it on DSA ?
00:33 < gch981213> blogic: Yes.
00:34 < blogic> ok, let me go looking for a mt7621 unit
00:35 < gch981213> image for devices with w25q256 flash can't reboot atm. I'm working on it.
00:35 < Rene__> mt7621 has also a trick to wifi offload. What I understand is that they just push the wifi packed to packet engine.
5 Likes
Is there anywhere I can find this conversation in full? 
Idle on IRC like me
Or just go here: http://logs.nslu2-linux.org/livelogs/openwrt-devel/openwrt-devel.20200407.txt
5 Likes
Thank you very much! It's a very useful link!