Disable Meltdown/Spectre to regain performance

[bester69]

Hi,

Because of the current Intel CPU security hole issue, there is a patch expected which slows down the system performance.
Reading some articles, they say that in order to recover the lost performance before the security paches, You need to disable the PTI(Page Table Isolation).
https://www.stevenrombauts.be/2018/02/h ... chmarking/

So would be theses considerations ok?.:
- For Spectre patches lost performance >> uninstall firmware-intel or install version before Spectre pacthes
- For Meltdown patches lost performance >> Disable PTI (nopti pti=off)

Ive got into this solution, Add the following to the end of your kernel argument in grub:
spectre_v2=off nopti pti=off

is it right?? , any additional consideration?

Thanks.

[sunrat]

Now you're being silly. Why compromise security for an imperceptible performance boost? Have you tested how much difference it makes?
The Meltdown/Spectre patches were added months ago. Personally I noticed no difference.

[debiman]

the intel firmware has nothing to do with linux kernel's PTI.
(i think intel "promised" to "fix" the issue via firmware, but it seems linux kernel devs don't trust that so much)

i noticed a significant performance drop after PTI was first introduced into the kernel, but idk - either i got used to it, or it got better in subsequent kernel verisons.
i am now on 4.14.51.

[stevepusser]

Just wait--BSD has disabled hyperthreading in their kernels in response to a yet undisclosed Intel vulnerability scheduled to be announced at the Black Hat conference in August. That drops performance by at least 20% on affected Intel CPUs...including mine.
Intel is rumored to be trying to come with a fix that won't affect performance as much, but the bad news just keeps coming.

[Head_on_a_Stick]

Just wait--BSD has disabled hyperthreading in their kernels in response to a yet undisclosed Intel vulnerability

The OpenBSD devs believe that Intel's technique leaves the door open for an entire class of new vulnerabilities, not just one:

https://marc.info/?l=openbsd-cvs&m=152943660103446&w=2

That drops performance by at least 20% on affected Intel CPUs

I've been running without SMT in OpenBSD since the original commit and I can notice very little difference

Theo even suggests that the box may run faster without hyperthreading:

https://marc.info/?l=openbsd-misc&m=152938776411086&w=2

[bester69]

the intel firmware has nothing to do with linux kernel's PTI.
(i think intel "promised" to "fix" the issue via firmware, but it seems linux kernel devs don't trust that so much)

i noticed a significant performance drop after PTI was first introduced into the kernel, but idk - either i got used to it, or it got better in subsequent kernel verisons.
i am now on 4.14.51.

My system is very old spec's, and Im still using 4.4.39 cos performance and graphical feelings, I dont need this kind of security patches in my computer, none is gonna knock at my door..
but other day I was thinking perhaps if i disabled PTI I might be able to move on to new kernels and keep similar or improve performance.. I did it with PTI and tried with last 4.4.x and with 4.17.x and It wasnt the case, I think from some old version on, they introduced something in kernel that makes my old computer feel badly different, I feel it like if graphics were using something like VAAPI with a litle smooth graphical response, Its an annoy and artificial feeling I cant feed up with. I cant explain it, but every kernel i try i come back to 4.4.39.

[Head_on_a_Stick]

I cant explain it

It is a basic feature of human psychology called confirmation bias:

https://en.wikipedia.org/wiki/Confirmation_bias

This is why we have benchmarks

[GarryRicketson]

Great Link, I like this part:

Explanations for the observed biases include wishful thinking and the limited human capacity to process information. Another explanation is that people show confirmation bias because they are weighing up the costs of being wrong, rather than investigating in a neutral, scientific way. However, even scientists can be prone to confirmation bias.[2]
Confirmation biases contribute to overconfidence in personal beliefs and can maintain or strengthen beliefs in the face of contrary evidence. Poor decisions due to these biases have been found in political and organizational contexts.[3][4]

[stevepusser]

Just wait--BSD has disabled hyperthreading in their kernels in response to a yet undisclosed Intel vulnerability

The OpenBSD devs believe that Intel's technique leaves the door open for an entire class of new vulnerabilities, not just one:

https://marc.info/?l=openbsd-cvs&m=152943660103446&w=2

That drops performance by at least 20% on affected Intel CPUs

I've been running without SMT in OpenBSD since the original commit and I can notice very little difference

Theo even suggests that the box may run faster without hyperthreading:

https://marc.info/?l=openbsd-misc&m=152938776411086&w=2

Well, I can't turn it off on my machine, but I know it can on some computers. Is there a kernel boot flag to do this? Let's see some benchmarks! Though I'm sure Phoronix will have some, too.

[bester69]

I cant explain it

It is a basic feature of human psychology called confirmation bias:

https://en.wikipedia.org/wiki/Confirmation_bias

This is why we have benchmarks

Very intersting and complex subject, Dont think i didnt thought of that, I started to think perhaps my mind was tricking me , So I tried some kernels for a three , four days period, and there was something I didnt feel confortable.. I can swear you I dont need benchmarks to feel if it performance worse, similar or better, we're talking about and old low spec laptop where minimal differences can be felt it easily.

To me, happend the same with some interestings beliefs, I do know they're "truth" cos the model that explains them
fits an empirical reality according to my experiences life. Im talking about two different and important litle known concepts: "reactive mind" and "organic portals"..I share that "bias confimation" with anyone of you that might be interested.

[Head_on_a_Stick]

Is there a kernel boot flag to do this?

Have you tried

Code: Select all

noht

Let's see some benchmarks!

The framerate in games/openarena is unchanged, as are the render times in graphics/blender.

[stevepusser]

The noht flag doesn't have any effect, at least on backported Debian experimental 4.17.2 and Liquorix 4.16-16.3 kernels...

[Head_on_a_Stick]

The noht flag doesn't have any effect

Yes, my apologies, I should have tested that before posting

I can only find nosmt but it doesn't apply to x86_64 (amd64), unfortunately:

https://github.com/torvalds/linux/blob/ ... .txt#L2687

[Head_on_a_Stick]

Is there a kernel boot flag to do this?

This works on my ThinkPad X201 with the stock Debian stable kernel:

Code: Select all

nosmp

Performance seems restricted though