KVM: VMX: Drop manual TLB flush when migrating vmcs.APIC_ACCESS_ADDR

Remove the superfluous flush of the current TLB in VMX's handling of
migration of the APIC-access page, as a full TLB flush on all vCPUs will
have already been performed in response to kvm_unmap_gfn_range() *if*
there were SPTEs pointing at the APIC-access page.  And if there were no
valid SPTEs, then there can't possibly be TLB entries to flush.

The extra flush was added by commit fb6c8198 ("kvm: vmx: Flush TLB
when the APIC-access address changes"), with the justification of "because
the SDM says so".  The SDM said, and still says:

 As detailed in Section xx.x.x, an access to the APIC-access page might
 not cause an APIC-access VM exit if software does not properly invalidate
 information that may be cached from the EPT paging structures. If EPT was
 in use on a logical processor at one time with EPTP X, it is recommended
 that software use the INVEPT instruction with the “single-context” INVEPT
 type and with EPTP X in the INVEPT descriptor before a VM entry on the
 same logical processor that enables EPT with EPTP X and either (a) the
 "virtualize APIC accesses" VM- execution control was changed from 0 to 1;
 or (b) the value of the APIC-access address was changed.

But the "recommendation" for (b) is predicated on there actually being
a valid EPT translation *and* possible TLB entries for the GPA (or guest
VA when using shadow paging).  It's possible that a different vCPU has
established a mapping for the new page, but the current vCPU can't have
entered the guest, i.e. can't have created a TLB entry, between flushing
the old mappings and changing its vmcs.APIC_ACCESS_ADDR.

kvm_unmap_gfn_range() waits for all vCPUs to ack KVM_REQ_APIC_PAGE_RELOAD,
and then flushes remote TLBs (which may or may not also pend a request).
Thus the vCPU is guaranteed to update vmcs.APIC_ACCESS_ADDR before
re-entering the guest and before it can possibly create new TLB entries.

In other words, KVM does flush in this case, it just does so earlier
on while handling the page migration.

Note, VMX also flushes if the vCPU is migrated to a new pCPU, i.e. if
the vCPU is migrated to a pCPU that entered the guest for a different
vCPU.

Suggested-by: Yu Zhang <yu.c.zhang@linux.intel.com>
Cc: Jim Mattson <jmattson@google.com>
Reviewed-by: Yu Zhang <yu.c.zhang@linux.intel.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Link: https://lore.kernel.org/r/20230721233858.2343941-1-seanjc@google.com


Signed-off-by: Sean Christopherson <seanjc@google.com>