Clean up code related to setting and getting MSR_IA32_APICBASE_BASE.

E.g. it's absurdly difficult to tease out that kvm_set_apic_base() exists
purely to avoid an extra call to kvm_recalculate_apic_map() (which may or
may not be worth the code, but whatever).

Simiarly, it's quite difficult to see that kvm_lapic_set_base() doesn't
do anything useful if the incoming MSR value is the same as the current
value.

v2:
 - Collect reviews. [Kai, Paolo]
 - Add a comment in kvm_lapic_reset() to explain its usage of the inner
   __kvm_apic_set_base() helper. [Paolo]
 - Unpack "struct msr_data" before calling kvm_apic_set_base(), e.g. so
   that __set_sregs_common() doesn't half-fill a structure and rely on
   kvm_apic_set_base() to ignore "index". [Kai]
 - Tack on a patch to short circuit all of kvm_apic_set_base() when the
   MSR isn't changing, to avoid the rare slow path of triggering an APIC
   map recalc due to some other task marking the map dirty.

v1: https://lore.kernel.org/all/20241009181742.1128779-1-seanjc@google.com

Sean Christopherson (9):
  KVM: x86: Short-circuit all kvm_lapic_set_base() if MSR value isn't
    changing
  KVM: x86: Drop superfluous kvm_lapic_set_base() call when setting APIC
    state
  KVM: x86: Get vcpu->arch.apic_base directly and drop
    kvm_get_apic_base()
  KVM: x86: Inline kvm_get_apic_mode() in lapic.h
  KVM: x86: Move kvm_set_apic_base() implementation to lapic.c (from
    x86.c)
  KVM: x86: Rename APIC base setters to better capture their
    relationship
  KVM: x86: Make kvm_recalculate_apic_map() local to lapic.c
  KVM: x86: Unpack msr_data structure prior to calling
    kvm_apic_set_base()
  KVM: x86: Short-circuit all of kvm_apic_set_base() if MSR value is
    unchanged

 arch/x86/kvm/lapic.c | 39 ++++++++++++++++++++++++++++++++++----
 arch/x86/kvm/lapic.h | 11 ++++++-----
 arch/x86/kvm/x86.c   | 45 +++++---------------------------------------
 3 files changed, 46 insertions(+), 49 deletions(-)


base-commit: e466901b947d529f7b091a3b00b19d2bdee206ee
-- 
2.47.0.163.g1226f6d8fa-goog

On Fri, 01 Nov 2024 11:35:46 -0700, Sean Christopherson wrote:
> Clean up code related to setting and getting MSR_IA32_APICBASE_BASE.
> 
> E.g. it's absurdly difficult to tease out that kvm_set_apic_base() exists
> purely to avoid an extra call to kvm_recalculate_apic_map() (which may or
> may not be worth the code, but whatever).
> 
> Simiarly, it's quite difficult to see that kvm_lapic_set_base() doesn't
> do anything useful if the incoming MSR value is the same as the current
> value.
> 
> [...]

Applied to kvm-x86 misc.

Kai, please holler if you have any concerns with patch 9.  I'm more than happy
to tweak it as needed (or even drop it if necessary).  I applied it quickly
purely to get more soak time in -next.  Thanks for all the reviews!

[1/9] KVM: x86: Short-circuit all kvm_lapic_set_base() if MSR value isn't changing
      https://github.com/kvm-x86/linux/commit/d7d770bed98f
[2/9] KVM: x86: Drop superfluous kvm_lapic_set_base() call when setting APIC state
      https://github.com/kvm-x86/linux/commit/8166d2557912
[3/9] KVM: x86: Get vcpu->arch.apic_base directly and drop kvm_get_apic_base()
      https://github.com/kvm-x86/linux/commit/d91060e342a6
[4/9] KVM: x86: Inline kvm_get_apic_mode() in lapic.h
      https://github.com/kvm-x86/linux/commit/adfec1f4591c
[5/9] KVM: x86: Move kvm_set_apic_base() implementation to lapic.c (from x86.c)
      https://github.com/kvm-x86/linux/commit/c9c9acfcd573
[6/9] KVM: x86: Rename APIC base setters to better capture their relationship
      https://github.com/kvm-x86/linux/commit/7d1cb7cee94f
[7/9] KVM: x86: Make kvm_recalculate_apic_map() local to lapic.c
      https://github.com/kvm-x86/linux/commit/ff6ce56e1d88
[8/9] KVM: x86: Unpack msr_data structure prior to calling kvm_apic_set_base()
      https://github.com/kvm-x86/linux/commit/c9155eb012b9
[9/9] KVM: x86: Short-circuit all of kvm_apic_set_base() if MSR value is unchanged
      https://github.com/kvm-x86/linux/commit/a75b7bb46a83

--
https://github.com/kvm-x86/linux/tree/next