Although FEAT_ECV allows us to correctly enable the timers, it also
reduces performances pretty badly (a L2 guest doing a lot of virtio
emulated in L1 userspace results in a 30% degradation).
Mitigate this by emulating the CTL/CVAL register reads in the
inner run loop, without returning to the general kernel. This halves
the overhead described above.
Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
---
arch/arm64/kvm/arch_timer.c | 15 -------
arch/arm64/kvm/hyp/vhe/switch.c | 70 +++++++++++++++++++++++++++++++++
include/kvm/arm_arch_timer.h | 15 +++++++
3 files changed, 85 insertions(+), 15 deletions(-)
diff --git a/arch/arm64/kvm/arch_timer.c b/arch/arm64/kvm/arch_timer.c
index 860f6e190e63..1ee1ede23607 100644
--- a/arch/arm64/kvm/arch_timer.c
+++ b/arch/arm64/kvm/arch_timer.c
@@ -101,21 +101,6 @@ u64 timer_get_cval(struct arch_timer_context *ctxt)
}
}
-static u64 timer_get_offset(struct arch_timer_context *ctxt)
-{
- u64 offset = 0;
-
- if (!ctxt)
- return 0;
-
- if (ctxt->offset.vm_offset)
- offset += *ctxt->offset.vm_offset;
- if (ctxt->offset.vcpu_offset)
- offset += *ctxt->offset.vcpu_offset;
-
- return offset;
-}
-
static void timer_set_ctl(struct arch_timer_context *ctxt, u32 ctl)
{
struct kvm_vcpu *vcpu = ctxt->vcpu;
diff --git a/arch/arm64/kvm/hyp/vhe/switch.c b/arch/arm64/kvm/hyp/vhe/switch.c
index 360328aaaf7c..8d1e9d1adabe 100644
--- a/arch/arm64/kvm/hyp/vhe/switch.c
+++ b/arch/arm64/kvm/hyp/vhe/switch.c
@@ -258,11 +258,81 @@ static bool kvm_hyp_handle_tlbi_el1(struct kvm_vcpu *vcpu, u64 *exit_code)
return true;
}
+static bool kvm_hyp_handle_timer(struct kvm_vcpu *vcpu, u64 *exit_code)
+{
+ u64 esr, val;
+
+ /*
+ * Having FEAT_ECV allows for a better quality of timer emulation.
+ * However, this comes at a huge cost in terms of traps. Try and
+ * satisfy the reads without returning to the kernel if we can.
+ */
+ if (!is_hyp_ctxt(vcpu))
+ return false;
+
+ esr = kvm_vcpu_get_esr(vcpu);
+ if ((esr & ESR_ELx_SYS64_ISS_DIR_MASK) != ESR_ELx_SYS64_ISS_DIR_READ)
+ return false;
+
+ switch (esr_sys64_to_sysreg(esr)) {
+ case SYS_CNTP_CTL_EL02:
+ val = __vcpu_sys_reg(vcpu, CNTP_CTL_EL0);
+ break;
+ case SYS_CNTP_CTL_EL0:
+ if (vcpu_el2_e2h_is_set(vcpu))
+ val = read_sysreg_el0(SYS_CNTP_CTL);
+ else
+ val = __vcpu_sys_reg(vcpu, CNTP_CTL_EL0);
+ break;
+ case SYS_CNTP_CVAL_EL02:
+ val = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
+ break;
+ case SYS_CNTP_CVAL_EL0:
+ if (vcpu_el2_e2h_is_set(vcpu)) {
+ val = read_sysreg_el0(SYS_CNTP_CVAL);
+
+ if (!has_cntpoff())
+ val -= timer_get_offset(vcpu_hptimer(vcpu));
+ } else {
+ val = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
+ }
+ break;
+ case SYS_CNTV_CTL_EL02:
+ val = __vcpu_sys_reg(vcpu, CNTV_CTL_EL0);
+ break;
+ case SYS_CNTV_CTL_EL0:
+ if (vcpu_el2_e2h_is_set(vcpu))
+ val = read_sysreg_el0(SYS_CNTV_CTL);
+ else
+ val = __vcpu_sys_reg(vcpu, CNTV_CTL_EL0);
+ break;
+ case SYS_CNTV_CVAL_EL02:
+ val = __vcpu_sys_reg(vcpu, CNTV_CVAL_EL0);
+ break;
+ case SYS_CNTV_CVAL_EL0:
+ if (vcpu_el2_e2h_is_set(vcpu))
+ val = read_sysreg_el0(SYS_CNTV_CVAL);
+ else
+ val = __vcpu_sys_reg(vcpu, CNTV_CVAL_EL0);
+ break;
+ default:
+ return false;
+ }
+
+ vcpu_set_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu), val);
+ __kvm_skip_instr(vcpu);
+
+ return true;
+}
+
static bool kvm_hyp_handle_sysreg_vhe(struct kvm_vcpu *vcpu, u64 *exit_code)
{
if (kvm_hyp_handle_tlbi_el1(vcpu, exit_code))
return true;
+ if (kvm_hyp_handle_timer(vcpu, exit_code))
+ return true;
+
return kvm_hyp_handle_sysreg(vcpu, exit_code);
}
diff --git a/include/kvm/arm_arch_timer.h b/include/kvm/arm_arch_timer.h
index 6e3f6b7ff2b2..c1ba31fab6f5 100644
--- a/include/kvm/arm_arch_timer.h
+++ b/include/kvm/arm_arch_timer.h
@@ -156,4 +156,19 @@ static inline bool has_cntpoff(void)
return (has_vhe() && cpus_have_final_cap(ARM64_HAS_ECV_CNTPOFF));
}
+static inline u64 timer_get_offset(struct arch_timer_context *ctxt)
+{
+ u64 offset = 0;
+
+ if (!ctxt)
+ return 0;
+
+ if (ctxt->offset.vm_offset)
+ offset += *ctxt->offset.vm_offset;
+ if (ctxt->offset.vcpu_offset)
+ offset += *ctxt->offset.vcpu_offset;
+
+ return offset;
+}
+
#endif
--
2.39.2
