[elrepo] kernel-lt and kernel-ml updates for Meltdown and Spectre
Dave Chiluk
dchiluk at indeed.com
Tue Mar 13 15:50:16 EDT 2018
Those articles are all well and good, but the direction that upstream
took for enabling/disabling ibpb and ibrs are available in
Documentation/admin-guide/kernel-parameters.txt. The RHEL sysfs
mechanisms do not exist in the mainline kernels.
Here's the relevant portion of the admin guide.
"
spectre_v2= [X86] Control mitigation of Spectre variant 2
(indirect branch speculation) vulnerability.
on - unconditionally enable
off - unconditionally disable
auto - kernel detects whether your CPU model is
vulnerable
Selecting 'on' will, and 'auto' may, choose a
mitigation method at run time according to the
CPU, the available microcode, the setting of the
CONFIG_RETPOLINE configuration option, and the
compiler with which the kernel was built.
Specific mitigations can also be selected manually:
retpoline - replace indirect branches
retpoline,generic - google's original retpoline
retpoline,amd - AMD-specific minimal thunk
Not specifying this option is equivalent to
spectre_v2=auto.
"
Let us know if/when you get benchmarks as I think we'd all like to know.
Happy benching.
Dave.
On Sat, Mar 10, 2018 at 3:39 AM Phil Perry <phil at elrepo.org> wrote:
>
> On 10/03/18 01:23, David Ranch wrote:
> >
> > Thanks for the news!
> >
> > Now the next big question, taking a kernel compile as the use case: does
> > anyone have any benchmarks on the built time pre and post BIOS, CPU
> > Instruction, and Compiler fixes say for a non-cutting edge CPU from Intel?
> >
> > --David
> >
> >
> > > STATUS: NOT VULNERABLE (Mitigation: Full generic retpoline)
> >
>
> Hi David,
>
> We have not done any specific performance measurements, but I note a
> couple of Red Hat articles on the subject:
>
> https://access.redhat.com/articles/3307751
> https://access.redhat.com/articles/3311301
>
> and there could be further differences between RHEL and elrepo kernels
> which may or may not be Meltdown/Spectre/retpoline related.
>
> Google will no doubt find you more articles, and you now have the tools
> to make your own assessment for the throughputs and benchmarks that
> matter to you.
>
> The only benchmark I routinely run is glmark2 for our nvidia driver
> releases and I haven't observed any measurable differences pre/post
> Meltdown/Spectre that do not fall within normal tolerances of the tests.
>
> _______________________________________________
> elrepo mailing list
> elrepo at lists.elrepo.org
> http://lists.elrepo.org/mailman/listinfo/elrepo
More information about the elrepo
mailing list