lkml.org 
[lkml]   [2020]   [Jul]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH -v2.1] x86/msr: Filter MSR writes
From
Date
On Tue, 2020-07-14 at 18:56 +0200, Borislav Petkov wrote:
> On Tue, Jul 14, 2020 at 05:04:48PM +0100, Chris Down wrote:
> > Since the issue involves DPTF which is only supported via binary
> > blobs, I
> > can't say for certain what the issue is. As I understand it, when
> > the
> > throttling behaviour isn't explicitly configured by the OS kernel,
> > the
> > default policy is extremely overeager. Matthew also had a look at
> > it[0], but
> > I don't know if anything eventually happened there. I've cc'ed him.
> >
> > Either way, again, this isn't really the point. :-) The point is
> > that there
> > _are_ currently widespread cases involving poking MSRs from
> > userspace,
> > however sacrilegious or ugly (which I agree with!), and while
> > people should
> > be told about that, it's excessive to have the potential to take up
> > 80% of
> > kmsg in the default configuration. It doesn't take thousands of
> > messages to
> > get the message across, that's what a custom printk ratelimit is
> > for.
>
> Ok, feel free to suggest a fix, better yet send a patch. Otherwise,
> you'd have to wait for my vacation to end first. :-)
>
> > 0: https://twitter.com/mjg59/status/1034132444201582596
>
> As to the power issue, lemme CC some Intel folks I found in
> MAINTAINERS.
>
> Intel folks, pls check the link above and upthread: Why TF do people
> need to use some luserspace daemon which pokes at MSRs which the
> kernel
> writes to too, in order to bypass some apparently too conservative
> throttling, AFAIU?
For issues related to thermal or power, we don't expect to poke MSRs
from user space by any daemon. We have sysfs interfaces for the
required controls. This is also true for controls via MMIO space.
Anytime if it is safe to add, we are adding controls via sysfs.

The tool in question from the link (not from Intel), when developed may
not have TCC or RAPL-MMIO controls via sysfs. We have sysfs interfaces
for a while. They can send email to me to justify other controls if
any.

Only time direct MSR access is required is for debug tools like
turbostat.

>
> And why does this work on windoze reportedly?
This is not related to MSR or MMIO. This is related to some ACPI
tables. In Linux, thermald will adjust these knobs like Windows. It was
missing some ACPI details, which Matthew Garrett submitted patches to
kernel and getting merged with 5.8 series.

Thanks,
Srinivas


\
 
 \ /
  Last update: 2020-07-14 21:07    [W:0.117 / U:0.088 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site