Intel CPUs design flaw OS fix, effects on emulators?

[Chocolates]

I'm sorry if this is not on the correct board, I'm curious how this would affect dolphin (and emulators in general)

Articles in question :

https://www.tweaktown.com/news/60357/massive-design-flaw-intel-cpus-found-reduces-performance/index.html
https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/
https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

[Helios]

meh, all those articles suck.

Here's a more technical writeup.

https://pythonsweetness.tumblr.com/post/169166980422/the-mysterious-case-of-the-linux-page-table

[MonadoBoy]

I'm curious as to how this will affect emulation and gaming performance. Like how it will affect dolphin. Should we expect noticeable slowdowns now?

[JonnyH]

(01-04-2018, 12:40 AM)MonadoBoy Wrote: I'm curious as to how this will affect emulation and gaming performance. Like how it will affect dolphin. Should we expect noticeable slowdowns now?

Likely not a lot - the proposed Linux solution will only cause a fixed-time slowdown for each ioctl user/kernel transition, so it will hit apps that do a *lot* of kernel calls (read/write filesystem data, networking stuff, memory mapping, contended mutexes etc.). I don't think emulators are are quite so reliant on ioctl speed.

[Helios]

I don't see why Dolphin would be super hit by this.

However

oops. looks like some AMD CPUs (and ARM, uh oh) are hit too

womp womp

https://spectreattack.com/

[JonnyH]

(01-04-2018, 08:50 AM)Helios Wrote: I don't see why Dolphin would be super hit by this.

However

oops. looks like some AMD CPUs (and ARM, uh oh) are hit too

womp womp

https://spectreattack.com/

At least the AMD side is only 'vulnerable' if you're in the same protection ring - it requires you to be in kernel-space to read kernel-space data, and in the user-space app to read it's data, and (BIG DEAL) in the same VM to read that VM's memory. The issue is that it can be implemented even with "secured" interpreters/jits, such as the eBPF example for an in-kernel version in Linux. [0]

[0] These are "not shown to be vulnerable" on ARM/AMD - that is *not* the same as saying they're not vulnerable to this (or similar attacks). It's hard proving a negative - especially with something as crazily complex as morden CPUs.

[DrHouse64]

I heard Spectre attack is unfixable without massive perf loss, even more than Meltdown.
Still, most of this is speculation so let's wait and see.
Everybody panicked about the 30% Meltdown fix slowdown, but actually it seems that it only really hits server software.

[Admentus]

So the Windows update is supposedly released. Post-update I literally see no difference in performance at all, be it Dolphin, CEMU (yeah, 60 FPS BoTW with the FPS++ mod still obtains 60 FPS in the previous tested open world area's) or even The Witcher 3, all while maintaining 4K. I am still sure the harm the exploit it can do is immense and that large companies dealing with servers can suffer major performance from it, but I am not inclined to believe your daily use of a computer as the average user is impacted. Perhaps that older CPU models are affected more. I recently upgraded (begin-december 2017) to a i5-8600k processor (Intel of course...). While it is not reassuring my new hardware is already having a design flaw, design flaws come all the time and so far I am not impacted yet in my use. Of course, a design flaw this major is not discovered every day.

[Shonumi]

(01-05-2018, 05:42 AM)Admentus Wrote: Perhaps that older CPU models are affected more. I recently upgraded (begin-december 2017) to a i5-8600k processor (Intel of course...).

Anything Haswell and up has PCID baked into it, which is supposed to lessen the performance drain from all of these patches. You probably won't see any big performance dip until you have a program that makes a bunch of syscalls or otherwise invokes the kernel a lot.