Linux 6.3 Introducing Hardware Noise "hwnoise" Tool

[phoronix]

Phoronix: Linux 6.3 Introducing Hardware Noise "hwnoise" Tool

As part of the tracing updates sent in for Linux 6.3 is the introduction of the new "hwnoise" tool within the kernel source tree for monitoring and quantifying hardware noise...

https://www.phoronix.com/news/Linux-6.3-hwnoise

[peterdk]

I can't get what hardware noise exactly means? I doubt it's random bits that the CPU spits out, or any other peripherial?

[JanC]

The normal processing on the CPU gets interrupted by hardware that needs attention (hard disks, network adapters, GPUs, USB devices, etc.), and that happens at more or less random intervals, hence you can use the timing between those interruptions to “harvest” randomness.

[Britoid]

I can't seem to get the osnoise application to work or find any real documentation on it.

Anyway, this would be useful, I imagine it can find bad PCI-E devices that are interupting all the time

[milkylainen]

I can't get what hardware noise exactly means? I doubt it's random bits that the CPU spits out, or any other peripherial?

See it like if you have a CPU but CPU time is stolen by entities you don't control or can't opt out from. Noise.
NMIs, whatever underlying crap you might have, etc.

[stormcrow]

I can't get what hardware noise exactly means? I doubt it's random bits that the CPU spits out, or any other peripherial?

You have to dig deeper than the surface and Michael's article to find out the context. Easier to just read the underlying kernel documentation itself.

https://www.kernel.org/doc/html/latest/trace/osnoise-tracer.html

This is important, because people coming from different backgrounds will consider "noise" based on their own points of view. For myself, "hardware noise" was either RFI or the randomness associated with practically non-deterministic hardware interrupts. However, that's not the case. "Hardware noise" in this context are the events triggered in the system that interfere with running threads while interrupts are disabled, which means this has very little to do with cryptographic function noise harvesting. The italicized part is important. This isn't about cryptographic harvesting. It's a performance counter. It intends to show metrics in how much the underlying hardware is interfering with compute threads. Ideally the hardware noise should be zero. The details are in that URL link.

[rogerx]

Wow, from other posts, glad to see I'm not the only one wondering the definition of osnoise!

OSNOISE Tracer

https://docs.kernel.org/trace/osnoise-tracer.html

The article seems to imply noise due to communication between applications and IRQs, while assuming everybody else knows exactly what they're talking about.

Think the first paragraph should more accurately describe the problem attempting to be solved.

One thing I've noticed with programmers writing code and programmers trying to explain what they're writing or attempting to code, if the programmer cannot successfully simply communicate what they're programming (or writing code for) to an ordinary common person, how can one then successfully write reliable safe working code?

It is really easy, to write a lot of meaningless explanations or nothing at all, and ending-up with writing something like systemd.

(I know... I somehow brought-up systemd into the mix again...)

[yump]

So it sounds like this would be useful for detecting things like that AMD fwTPM bug, or long-running firmware SMIs.

If Phoronix ever gets into motherboard reviews, it could be an interesting point of comparison. Ditto laptops.