On Wed, Oct 20, 2021 at 07:35:35PM -0500, Michael Roth wrote:

Fortunately, all the code makes use of sev_status to get at the SEV MSR
bits, so breaking the appropriate bits out of sme_enable() into an earlier
sev_init() routine that's the exclusive writer of sev_status sounds like a
promising approach.

Ack.

It makes sense to do it immediately after the first #VC handler is set
up, so CPUID is available, and since that's where SNP CPUID table
initialization would need to happen if it's to be made available in
#VC handler.

Right, and you can do all your init/CPUID prep there.

It may even be similar enough between boot/compressed and run-time kernel
that it could be a shared routine in sev-shared.c.

Uuh, bonus points! :-)

But then again it also sounds like the appropriate place to move the
snp_cpuid_init*() calls, and locating the cc_blob, and since there's
differences there it might make sense to keep the boot/compressed and
kernel proper sev_init() routines separate to avoid #ifdeffery).

Not to get ahead of myself though. Just seems like a good starting point
for how to consolidate the various users.

I like how you're thinking. :)

Got it, and my apologies if I've given you that impression as it's
certainly not my intent. (though I'm sure you've heard that before.)

Nothing to apologize - all good.

Agreed, if we need to check SEV MSR early for the purposes of SNP it makes
sense to move the overall SEV feature detection code earlier as well. I
should have looked into that aspect more closely before introducing the
changes.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette