Re: [PATCH v2 4/9] x86/spec-ctrl: Don't use spec_ctrl_{enter,exit}_idle() for S3

[Andrew Cooper <Andrew.Cooper3@xxxxxxxxxx>]

On 31/01/2022 10:15, Jan Beulich wrote:
> On 28.01.2022 14:29, Andrew Cooper wrote:
>> 'idle' here refers to hlt/mwait.  The S3 path isn't an idle path - it is a
>> platform reset.
>>
>> Conditionally clearing IBRS and flushing the store buffers on the way down is
>> a waste of time.
> I can buy this for the flushing aspect; I'm less certain about the clearing
> of IBRS: Whether the act of clearing is slower than the performance price
> of running with it enabled is unknown, I suppose?

There are a handful of instructions from now until the core is powered
down.  The cost of the WRMSR is going to dominate everything else, but
we're still only talking microseconds.

But honestly, the perf aspect isn't relevant.  It's wrong to use the
enter/exit idle helpers here.

>> Furthermore, we want to load default_xen_mcu_opt_ctrl unilaterally on the way
>> back up.  Currently it happens as a side effect of X86_FEATURE_SC_MSR_IDLE or
>> the next return-to-guest, but that's fragile behaviour.
> I'll assume from your reply that you've adjusted the description as well.

I have, yes.

>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> Preferably with the statement above softened a little:
> Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>

Thanks.

~Andrew