rockchip: disable kernel preemption

[kien-truong]

This setting is more suitable for device running OpenWRT. Most OpenWRT targets are already default to this configuration, and it has shown better performance in VPN (wireguard).

Fix: #17454

[wurzerj]

The change looks good to me. Tested on my NanoPi R4S production router. I can confirm that the system is booting. It is stable since I updated (about an hour ago). I only have a 250 Mbit line which the R4S was able to fully utilize before, so I probably won't feel the performance increase. Nevertheless I ran a speed test and the device still maxes out my line. So all good.

[1715173329]

LGTM, thanks!

[robimarko]

Thanks! Rebased on top of main and merged!

[samara1531]

"kernel": "6.6.63",
"hostname": "OpenWrt",
"system": "ARMv8 Processor rev 0",
"model": "FriendlyElec NanoPi R3S",
"board_name": "friendlyarm,nanopi-r3s",
"rootfs_type": "ext4",
"release": {
"distribution": "OpenWrt",
"version": "24.10.0-rc2",
"revision": "r28161-ea17e958b9",
"target": "rockchip/armv8",
"description": "OpenWrt 24.10.0-rc2 r28161-ea17e958b9",
"builddate": "1733226068"
}
}
Connecting to host 169.254.200.2, port 4242
[ 5] local 169.254.200.1 port 49556 connected to 169.254.200.2 port 4242
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-1.00 sec 45.8 MBytes 383 Mbits/sec 0 423 KBytes
[ 5] 1.00-2.00 sec 43.9 MBytes 368 Mbits/sec 0 423 KBytes
[ 5] 2.00-3.00 sec 44.1 MBytes 370 Mbits/sec 0 423 KBytes
[ 5] 3.00-4.00 sec 44.1 MBytes 370 Mbits/sec 0 423 KBytes
[ 5] 4.00-5.00 sec 43.8 MBytes 367 Mbits/sec 0 423 KBytes
[ 5] 5.00-6.00 sec 44.4 MBytes 372 Mbits/sec 0 453 KBytes
[ 5] 6.00-7.00 sec 43.6 MBytes 366 Mbits/sec 0 453 KBytes
[ 5] 7.00-8.00 sec 44.5 MBytes 373 Mbits/sec 0 484 KBytes
[ 5] 8.00-9.00 sec 44.6 MBytes 374 Mbits/sec 0 484 KBytes
[ 5] 9.00-10.00 sec 44.5 MBytes 373 Mbits/sec 0 603 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 443 MBytes 372 Mbits/sec 0 sender
[ 5] 0.00-10.01 sec 442 MBytes 371 Mbits/sec receiver

nft flush ruleset →

[chinese-maintainer]

PREEMPT provides a balance between performance and throughput, and it's optimal for routers. While PREEMPT_NONE is only provides the throughput, but gives worse performance. Don't copypaste from uncertain maintainers and revert this commit.

[chinese-maintainer]

kien-truong You have not provided any convincing arguments for disable kernel preemption. Thread #17454 based on VPN test only, and does not provide a full assessment of the potential loss productivity of all other process.

Your have only argument is that other routers have disable kernel preemption, but all devices on OpenWrt it's more than just a router and can be used to execute multiple services, from which you want to take away resources in favor of bandwidth. This is not right. Especially, I tested the firmware with it disabled kernel preemption and did not get any positive results.

The problem with poor network performance on the NanoPi devices is that OpenWrt only can use open source Realtek network drivers, which have bad implementation. This has been discussed for many years.

[issuex]

PREEMPT provides a balance between performance and throughput, and it's optimal for routers. While PREEMPT_NONE is only provides the throughput, but gives worse performance. Don't copypaste from uncertain maintainers and revert this commit.

show me some benchmark that relate what you reply.