halz

Context: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/...

Writes to arbitrary MSR are still allowed by linux, but it'll squawk a warning with the PID from what poked the MSR. What has been preventing writing to MSR is microcode updates (loaded via Linux via initram/firmware, or via the vendor in a BIOS update), such as the one(s) in response to plundervolt https://plundervolt.com/ Writing to the plundervolt-related MSR was particularly helpful to undervolt a thermal-throttled laptop. So! Can either downgrade the BIOS and/or rollback the microcode that your distro is loading during boot.

Weird, the measurement scale they use does not jive with the scale of the knobs being tweaked. For example, `sched_cfs_bandwidth_slice_us` is adjusted by the project's `bandwidth_size` knob. The default is 5000 us, or 5 ms. But the knob they mention it being '5 us'. Same goes for scheduling latency-- its not `us`, its `ms`.

And afaik, the 'sauce' used in Liquorix is MuQSS, by Con Kolivas. But the project has been put on ice somewhat recently[0].

[0]https://ck-hack.blogspot.com/2021/08/514-and-future-of-muqss...

Tread softly, there are some reports¹ that things like PIM/RIP are not working/not implemented.

¹https://forum.mikrotik.com/viewtopic.php?t=180896

Disabling the E-cores apparently activates AVX-512. So it was planned and engineered to be there all along, but was disabled to match the marketing talking points. [https://www.phoronix.com/vr.php?view=30664]

Would you considered using intel_pstate=passive, and then use x86_energy_perf_policy/cpupower to turn the knobs on how to influence the governor? This'll let the CPU do its lower latency internal frequency changing. But of course, its not a one-size-fits-all approach for AMD/ARM.

Oracle spelled backwards, "El Caro", is Spanish for "expensive". I can never forget this tidbit when dealing with Oracle.

I wonder if part of this pricing scheme is to counter (or at least to short-term profit from and eventually change the behavior of) the provider being abused by spammers/scammers who could previously scoop up benign reputation IPv4 addresses from the far corners of the world and pull them over to Hetzner for very little $.

Pinning to the physically closest core is a bit misleading. Take a look at output from something like `lstopo` [https://www.open-mpi.org/projects/hwloc/], where you can filter pids across the NUMA topology and trace which components are routed into which nodes. Pin the network based workloads into the corresponding NUMA node and isolate processes from hitting the IRQ that drives the NIC.

There is at lease one open issue¹ with the clipboard crate that causes a high amount of wakeups (under Wayland at least). Whatmore, the wakeups scale with the number of terminals open. The project is ruthless about performance[latency] regressions, but not so much about performance[energy] overhead.

[¹https://github.com/alacritty/alacritty/issues/3108]

Be advised that there is still an issue in the QTWayland front that makes OBS in Wayland rather unpleasant in practice; https://bugreports.qt.io/browse/QTBUG-81504 (via https://github.com/obsproject/obs-studio/issues/4295)

A project to keep an eye on, to replicate 'autotype' support offered by many password managers, is wlrctl¹. A popular alternative is ydotool², but brings some additional risks the operator might not be comfortable with.

¹https://git.sr.ht/%7Ebrocellous/wlrctl

²https://github.com/ReimuNotMoe/ydotool

Chances are theres an EFI partition thats still unencrypted, too! This is where secureboot/tboot has its chance to shine.

'Purple' sounds like the 'Display options>Merge exe, comm and cmdline in Command' setting, hotkey-able with `m`. Yeah, echo'ing to the UI which hotkey was just activated could be helpful.

Supposedly this was 'fixed' in 5.9 with experimental nvidia-drivers (455.45.01)

'Cloud-space' seem to be making use of hibernate to some acclaim (e.g. https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/Hibernat...)

Yes, ACPI DSDT being utterly broken seems to be a pretty common trip point for Linux suspend/sleep. For my Dell Precision 5510, I was able to patch[1] a missing device back into the DSDT, as well as 'fix' some thunderbolt port enumeration race condition. The hackintosh community was helpful in providing ACPI dumps from similar enough Dell systems in the same generation that I was able to paste in the missing device to get very consistent S3 performance. Using `acpiexec` was very helpful to simulate invoking sleep/suspend and being able to follow the execution path of where it would get stuck.

Also want to add that random wakeups can be triggered by interrupts hitting devices in `/proc/acpi/wakeup`. These wakeup sources can be selectively disabled once the problematic ones are identified. I disable XHC (USB ports), because it seems my routine closes the lid to suspend and _then_ remove peripherals.. which can cause a wakeup!

[1] https://gist.github.com/mailhost/d5b27b247fa26d8bdc8a5d20223...

A somewhat lighter-weight approach with PulseAudio (vs running full-blown PulseEffects) can be achieved through the use of ladspa-sink plugins, as described briefly on the Arch wiki for dynamic range compression https://wiki.archlinux.org/index.php/PulseAudio#Steve_Harris...

Using "plugin=sc4_1882 label=sc4 control=1,1.5,401,-30,20,5,12", I am generally able to smooth things out pretty well.

Another technique to dynamically manipulate task isolation is through the kernel's 'cpuset' interface (https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v1...). Coupled with threadirqs, and RCU 'kthread-ification' with rcu_nocbs, can wield a large amount of flexibility in organizing real-time workloads. The userspace interface to cpusets via 'cpuset' makes it even more accessible (https://documentation.suse.com/sle-rt/15-SP1/html/SLE-RT-all...)