[v2,0/8] x86/irq: KVM: Optimize KVM's PIR harvesting

Message

Sean Christopherson April 1, 2025, 4:34 p.m. UTC

Optimizing KVM's PIR harvesting using the same techniques as posted MSIs,
most notably to use 8-byte accesses on 64-bit kernels (/facepalm).

Fix a few warts along the way, and finish up by adding a helper to dedup
the PIR harvesting code between KVM and posted MSIs.

v2:
 - Collect a review. [tglx]
 - Use an "unsigned long" with a bitwise-OR to gather PIR. [tglx]

v1: https://lore.kernel.org/all/[e-mail]

Sean Christopherson (8):
  x86/irq: Ensure initial PIR loads are performed exactly once
  x86/irq: Track if IRQ was found in PIR during initial loop (to load
    PIR vals)
  KVM: VMX: Ensure vIRR isn't reloaded at odd times when sync'ing PIR
  x86/irq: KVM: Track PIR bitmap as an "unsigned long" array
  KVM: VMX: Process PIR using 64-bit accesses on 64-bit kernels
  KVM: VMX: Isolate pure loads from atomic XCHG when processing PIR
  KVM: VMX: Use arch_xchg() when processing PIR to avoid instrumentation
  x86/irq: KVM: Add helper for harvesting PIR to deduplicate KVM and
    posted MSIs

 arch/x86/include/asm/posted_intr.h | 78 +++++++++++++++++++++++++++---
 arch/x86/kernel/irq.c              | 63 ++++--------------------
 arch/x86/kvm/lapic.c               | 20 ++++----
 arch/x86/kvm/lapic.h               |  4 +-
 arch/x86/kvm/vmx/posted_intr.h     |  2 +-
 5 files changed, 95 insertions(+), 72 deletions(-)


base-commit: 782f9feaa9517caf33186dcdd6b50a8f770ed29b