backport "x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl"

backport "x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl"

[Mahmoud Adam]

Hi,

commit 2f46993d83ff upstream.

Please backport this commit to the other stable kernels, since this
patch landed on 6.1 and We've seen 30% improvement with docker
containers running heavy cpu/mem tasks


-MNAdam

Re: backport "x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl"

[Greg KH]

On Fri, Oct 13, 2023 at 04:28:37PM +0200, Mahmoud Adam wrote:
> 
> Hi,
> 
> commit 2f46993d83ff upstream.
> 
> Please backport this commit to the other stable kernels, since this
> patch landed on 6.1 and We've seen 30% improvement with docker
> containers running heavy cpu/mem tasks

Why is this required at all?  It just changes a default option, you can
set the "faster" option yourself if you really feel it is a good idea
for older kernels, why can't you do that for containers you want it
enabled on for these older kernels?

And what exact kernel versions did you test this on?

And how is this not a new feature?

thanks,

greg k-h

Re: backport "x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl"

[Mahmoud Adam]

> Why is this required at all?  It just changes a default option, you can
> set the "faster" option yourself if you really feel it is a good idea
> for older kernels, why can't you do that for containers you want it
> enabled on for these older kernels?

The patch has already motivations of why this is a better default, and
changing this would give performance improvements

> And what exact kernel versions did you test this on?

this was tested on 5.15 and 5.10

> And how is this not a new feature?

As you mentioned it changes the default option, Why is this considered a
new feature?

thanks,
MNAdam

Re: backport "x86: change default to spec_store_bypass_disable=prctl spectre_v2_user=prctl"

[Greg KH]

On Wed, Nov 01, 2023 at 04:15:24PM +0100, Mahmoud Adam wrote:
> > Why is this required at all?  It just changes a default option, you can
> > set the "faster" option yourself if you really feel it is a good idea
> > for older kernels, why can't you do that for containers you want it
> > enabled on for these older kernels?
> 
> The patch has already motivations of why this is a better default, and
> changing this would give performance improvements
> 
> > And what exact kernel versions did you test this on?
> 
> this was tested on 5.15 and 5.10
> 
> > And how is this not a new feature?
> 
> As you mentioned it changes the default option, Why is this considered a
> new feature?

Changing the default option when booting is a new feature in that it
changes what was previously happening before.

And, anyone running those old kernels, if they are comfortable with
making this feature the default, in order to get the performance
increase you say is there, can change their command line to do so, no
kernel changes needed.

So for now, I'll leave this as-is, thanks.

greg k-h