@@ -740,6 +740,44 @@ static void kvm_timer_vcpu_load_nested_switch(struct kvm_vcpu *vcpu,
WARN_ON_ONCE(ret);
}
+ /*
+ * NV2 badly breaks the timer semantics by redirecting accesses to
+ * the EL0 timer state to memory, so let's call ECV to the rescue if
+ * available: we trap all CNT{P,V}_{CTL,CVAL,TVAL}_EL0 accesses.
+ *
+ * The treatment slightly varies depending whether we run a nVHE or
+ * VHE guest: nVHE will use the _EL0 registers directly, while VHE
+ * will use the _EL02 accessors. This translates in different trap
+ * bits.
+ *
+ * None of the trapping is engaged when running in non-HYP context,
+ * unless required by the L1 hypervisor settings once we advertise
+ * ECV+NV in the guest.
+ */
+ if (vcpu_has_nv2(vcpu) && cpus_have_const_cap(ARM64_HAS_ECV)) {
+ u64 clr = 0, set = 0;
+
+ if (vcpu_el2_e2h_is_set(vcpu)) {
+ set |= CNTHCTL_EL1NVVCT | CNTHCTL_EL1NVPCT;
+ } else {
+ /*
+ * Only trap the timers, not the counters, as they
+ * are in the "direct" state for nVHE.
+ */
+ clr |= CNTHCTL_EL1PCEN << 10;
+ set |= CNTHCTL_EL1TVT;
+ }
+
+ /*
+ * Apply these values for HYP context, and reverse them
+ * otherwise
+ */
+ if (is_hyp_ctxt(vcpu))
+ sysreg_clear_set(cntkctl_el1, clr, set);
+ else
+ sysreg_clear_set(cntkctl_el1, set, clr);
+ }
+
/*
* Apply the trapping bits that the guest hypervisor has
* requested for its own guest.
@@ -838,6 +876,16 @@ void kvm_timer_vcpu_put(struct kvm_vcpu *vcpu)
if (kvm_vcpu_is_blocking(vcpu))
kvm_timer_blocking(vcpu);
+
+ /*
+ * Restore the standard, non-nested configuration so that we're
+ * ready to run a non-NV vcpu.
+ */
+ if (vcpu_has_nv2(vcpu))
+ sysreg_clear_set(cntkctl_el1,
+ (CNTHCTL_EL1NVPCT | CNTHCTL_EL1NVVCT |
+ CNTHCTL_EL1TVT | CNTHCTL_EL1TVCT),
+ (CNTHCTL_EL1PCEN | CNTHCTL_EL1PCTEN) << 10);
}
void kvm_timer_sync_nested(struct kvm_vcpu *vcpu)
@@ -847,8 +895,12 @@ void kvm_timer_sync_nested(struct kvm_vcpu *vcpu)
* accesses redirected to the VNCR page. Any guest action taken on
* the timer is postponed until the next exit, leading to a very
* poor quality of emulation.
+ *
+ * This is an unmitigated disaster, only papered over by FEAT_ECV,
+ * which allows trapping of the timer registers even with NV2.
+ * Still, this is still worse than FEAT_NV on its own. Meh.
*/
- if (!is_hyp_ctxt(vcpu))
+ if (cpus_have_const_cap(ARM64_HAS_ECV) || !is_hyp_ctxt(vcpu))
return;
if (!vcpu_el2_e2h_is_set(vcpu)) {
@@ -21,6 +21,10 @@
#define CNTHCTL_EVNTEN (1 << 2)
#define CNTHCTL_EVNTDIR (1 << 3)
#define CNTHCTL_EVNTI (0xF << 4)
+#define CNTHCTL_EL1TVT (1 << 13)
+#define CNTHCTL_EL1TVCT (1 << 14)
+#define CNTHCTL_EL1NVPCT (1 << 15)
+#define CNTHCTL_EL1NVVCT (1 << 16)
enum arch_timer_reg {
ARCH_TIMER_REG_CTRL,
Although FEAT_NV2 makes most things fast, it also makes it impossible to correctly emulate the timers, as the sysreg accesses are redirected to memory. FEAT_ECV addresses this by giving a hypervisor the ability to trap the EL02 sysregs as well as the virtual timer. Add the required trap setting to make use of the feature, allowing us to elide the ugly resync in the middle of the run loop. Signed-off-by: Marc Zyngier <maz@kernel.org> --- arch/arm64/kvm/arch_timer.c | 54 +++++++++++++++++++++++++++- include/clocksource/arm_arch_timer.h | 4 +++ 2 files changed, 57 insertions(+), 1 deletion(-)