[Request] Merge some linux-tkg's patches

[blackteahamburger]

I'm switching from linux-tkg to linux-cachyos, but I find some tweaks missing, such as:

• "Cake" network queue management system
• Avoid configuring old governors as default with intel_pstate
• Set vm.max_map_count to 262144 by default
• Some other clear patches
• ...

These are in:

• 0002-clear-patches.patch
• 0003-glitched-base.patch
• 0003-glitched-eevdf-additions.patch
• 0012-misc-additions.patch

I think the kernel would benefit from these tweaks.

[ptr1337]

"Cake" network queue management system

Does not really provide a benefit for desktops.

Avoid configuring old governors as default with intel_pstate

Why? What are old governors?

Set vm.max_map_count to 262144 by default

Not needed. Archlinux increased max_map_counts already OOB, and putting the value higher is somewhat useless

Some other clear patches

Already in the cachy branch

[1Naim]

"Cake" network queue management system

Does not really provide a benefit for desktops.

The patch actually just allows for it to be set by default via Kconfig, which is just extra noise. You can already do this via sysctl file.

Avoid configuring old governors as default with intel_pstate

Why? What are old governors?

It's a revert of that commit because TKG tunes the ondemand governor aggressively.

[blackteahamburger]

Thank you for your due response! What about these clear patches:

• 0104-pci-pme-wakeups.patch
• 0106-intel_idle-tweak-cpuidle-cstates.patch
• 0111-ipv4-tcp-allow-the-memory-tuning-for-tcp-to-go-a-lit.patch

I'm not sure whether they have benefits.

[1Naim]

• 0104-pci-pme-wakeups.patch

It's a workaround for older (?) laptops. I don't think it supposed to bring any performance benefits, nor do I know if the workaround is still needed today.

• 0106-intel_idle-tweak-cpuidle-cstates.patch

Unsuitable for CachyOS

• 0111-ipv4-tcp-allow-the-memory-tuning-for-tcp-to-go-a-lit.patch

Doesn't provide any benefits.

[blackteahamburger]

Thank you! It seems linux-tkg is already outdated...

One last question, does linux-tkg's Revert "cpufreq: Avoid configuring old governors as default with intel_pstate" provide any benefits?

[1Naim]

Not by itself, please see my other comment.

[blackteahamburger]

So forgive my ignorance, does "aggressive ondemand" provide any benefits (weirdly only enabled for bmq) ?

[1Naim]

It should but users would have to go out of their way to use it.

1. Majority of hardware nowadays supports active CPU scaling drivers and use it by default. This only exposes 2 CPU governors, performance and powersave (similar to schedutil, uses EPP for scheduling hints)
2. Using ondemand on these kinds of hardware would need: a). using acpi-cpufreq or the passive CPU scaling driver which exposes ondemand, and switch to ondemand from the default schedutil

Zen seems to patch this zen-kernel/zen-kernel@f654ea1 but the active scaling drivers are preferable over ondemand/schedutil. I don't mind adding it, but it's not gonna be used on possibly 99% of setups so there has to be more of a motivation behind adding it.

[blackteahamburger]

Close this then.