[LAU] Major CPU bugs - Was: Christmas present for self.

Brent Busby brent at keycorner.org
Thu Jan 4 18:19:36 UTC 2018


Will Godfrey <willgodfrey at musically.me.uk> writes:

> On Thu, 4 Jan 2018 10:17:54 +0100
> Jeremy Jongepier <jeremy at autostatic.com> wrote:

[...]

>>I wonder how much impact the KAISER patches to mitigate Meltdown will
>>have on Intel CPU's in a Linux audio context. If those patches have a
>>significant impact I'd surely go for AMD. And performance impact aside,
>>the way Intel tries to find its way out of this mess doesn't really
>>deserve a prize....

> The people that seem to be most up-to-date on this are The Register
> https://www.theregister.co.uk/
>
> They've been going at this like a dog with a bone since Tuesday. On one of
> their topics on the subject they include links to the *finally* released
> official documentations.
> They also have a hillarious 'translation' of Intel's official response :)
>
> In brief, there are three issues that have surfaced about the same time. The
> most serious one is Intel specific, and enabled direct access to kernel memory
> space - passwords etc. patches are being distrubuted which will cause
> slowdowns on heavy I/O work - so probably won't affect us much.
>
> The other two are variations of a different issues and apear across all
> processor types. This enables access to a different user space's memory. It is
> not easy to exploit, and there is no known fix, although it seems better
> software security will (eventually) resolve it.
>
> As usual, there is no problem while you are off-line, and personally, my DAW
> only ever goes on-line when I want to update something.


I'm glad I'm not the only one who was worried about how this will affect
possible track count and plugin capacity.  :)

On the bright side, as you have mentioned, there seems to be little
impact for AMD CPU's, which is what I'm using.  However, it looks like
all of the free and closed OS vendors are treating this as a design
problem that needs to be fixed regardless of chip type, and most of what
I'm getting from what I've read has been that KPTI patches to fix this
will not discriminate CPU type (will do the same thing on AMD machines
that they do on Intel).  They are not absolutely sure that the Meltdown
vulnerability cannot be used against AMD, so they're not chancing it,
and they're doing better kernel/userspace isolation for everyone.

Also on a positive note, Phoronix, who are a site very oriented to
gaming on Linux and not plugging for Intel or AMD either one, had a
story today saying that they had found the KPTI patches to have very
little affect on fullscreen frame rate in video games:

https://www.phoronix.com/scan.php?page=article&item=nvidia-linux-kpti&num=1

It's possible that all the doom and gloom from the benchmarks that said
we'd get getting ~30% performance loss may be measuring contrived
situations, as benchmark tests so often do anyway.  Probably the only
way to really know will be to run a patched kernel and see what happens
to you when you try to run a heavy session in Ardour or QTracktor.  With
luck, it will be the same as Phoronix found for gaming -- not really
quite as bad as the benchmarks implied.

-- 
- Brent Busby	+ ===============================================
		+	With the rise of social networking
--  Studio   --	+	sites, computers are making people
--  Amadeus  --	+	easier to use every day.
----------------+ ===============================================


More information about the Linux-audio-user mailing list