Linux KPTI Tests Using Linux 4.14 vs. 4.9 vs. 4.4

[andyprough]

Really? Up to 35% lower performance on the latest kernel is nothing for you?

Not a single one of these tests showed anywhere close to that. About 4% in one test is the most.

The only test I did was compile firefox beta. It took around the same time on linux 4.9.75

[    0.000000] Kernel/User page tables isolation: enabled

[elapsed]

Does it work for you if you now refresh the page?

Yes I did enable it but it works fine now, thanks. What was the solution?

[Kayote]

Congrats Michael your benchmarks are being featured all over, just seen your benchmarks in a Spanish tech channel.
They guy on youtube said that this would take like 10 years to solve in a hardware level? is that correct? if that's the case you can't blame Intel that much because there's not much they could actually have done by delaying their last gen processors.

[Xan's]

More KPTI benchmarks by CybDex: https://forum.xanmod.org/thread-221-...0.html#pid2730

[chimpy]

I've already been posting benchmarks from them the other day.

Oh thanks Michael. I didn't see it

[R00KIE]

Hey Michael you might want to do some testing with wine, one of the patches which I think is related to KPTI mentions Wine specifically.

commit f55f0501cbf65ec41cca5058513031b711730b1d
Author: Andy Lutomirski <[email protected]>
Date: Tue Dec 12 07:56:45 2017 -0800

x86/pti: Put the LDT in its own PGD if PTI is on

(snip)

This will significantly slow down LDT users, but that shouldn't matter for
important workloads -- the LDT is only used by DOSEMU(2), Wine, and very
old libc implementations.

[Tomin]

As mentioned in the article and shown in the screenshots, the very latest patches do.

And I missed that. Thanks. I usually don't read the texts in screenshots because they are so small and unreadable (unless I open them on new page of course). Funnily I did notice the cpu_insecure stuff on the text, but somehow skipped the sentence before that. I'm just too busy and tried to have a quick look at the benchmarks while doing other things.

This "being busy all the time" seems like an excuse for me to do (less important) things poorly, so maybe I should do something about that. Like not checking Phoronix during the day while I'm supposed to be doing something else...

[ryszardzonk]

Do you have the new site isolation feature enabled? Have heard it's breaking the graph rendering but haven't had time to investigate yet.

Earlier in the day I was unable to see any graphs in the article either while I didn't have that problem any time since new web page layout is used on Phoronix. Running 4.14.11 kernel here with PTI enabled on Firefox with quite a number of extensions. For some reason they work now without the problem while I did not make any changes to my setup.

PS Either it is me or suricata running on the server now takes much more cpu time (twice more it seems) on Ironlake than previously with kernel 3.14.79...

[Michael]

The results on page 2 are totally broken on Chrome. Each result is a separate frame with a vertical and horizontal scroll bar, showing a quarter of the result.
Shows fine in firefox.

Does it work for you if you now refresh the page?