[PATCH v6 27/48] KVM: PPC: Book3S HV P9: Reduce irq_work vs guest decrementer races

[Nicholas Piggin]

irq_work's use of the DEC SPR is racy with guest<->host switch and guest
entry which flips the DEC interrupt to guest, which could lose a host
work interrupt.

This patch closes one race, and attempts to comment another class of
races.

Signed-off-by: Nicholas Piggin <npiggin at gmail.com>
---
 arch/powerpc/kvm/book3s_hv.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index f4e5a64457e6..dae59f05ef50 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -3761,6 +3761,18 @@ static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
 	if (!(vcpu->arch.ctrl & 1))
 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
 
+	/*
+	 * When setting DEC, we must always deal with irq_work_raise via NMI vs
+	 * setting DEC. The problem occurs right as we switch into guest mode
+	 * if a NMI hits and sets pending work and sets DEC, then that will
+	 * apply to the guest and not bring us back to the host.
+	 *
+	 * irq_work_raise could check a flag (or possibly LPCR[HDICE] for
+	 * example) and set HDEC to 1? That wouldn't solve the nested hv
+	 * case which needs to abort the hcall or zero the time limit.
+	 *
+	 * XXX: Another day's problem.
+	 */
 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - tb);
 
 	if (kvmhv_on_pseries()) {
@@ -3897,6 +3909,9 @@ static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
 
 	next_timer = timer_get_next_tb();
 	mtspr(SPRN_DEC, next_timer - tb);
+	/* We may have raced with new irq work */
+	if (test_irq_work_pending())
+		set_dec(1);
 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
 
 	kvmhv_load_host_pmu();
-- 
2.23.0