On 04/23/2018 01:51 PM, speck for Konrad Rzeszutek Wilk wrote:
>>
>> Additional: I spoke with Intel a few minutes ago to impress upon them
>> that we'll (RH) have to leave MD disabled by default unless they drive
>> the prctrl/other solutions for vulnerable processes they'd like to see.
> 
> I believe (and Linus, please correct me here) that the question of
> toggling on/off SPEC_CTRL MSR on user-space entrance is a no-go.

To clarify, it's not a userspace toggling of an MSR. It would be a prctl
(e.g. a new "SPECULATION_VULNERABILITY" major with some minor variants)
that would allow processes to say "I'm vulnerable to X" or somesuch.

On x86, that might then allow us to have the mechanics of globally
enabling MDD, disabling it on entry to the kernel (due to stack attack,
to be debated), and selectively disabling it for known vulnerable
processes. Yea, it's a lot more complicated.

Anyway, I'm personally ok with a global knob. It's just that we're
getting a lot of pressure from Intel and AMD to not do that. I've asked
Intel to talk with Thomas and Linus and represent their opinions here
because I don't want it to seem like this is my asking for knobs! ;)

Jon.

-- 
Computer Architect | Sent from my Fedora powered laptop