diff mbox series

[v2,05/12] KVM: arm64: nv: Accelerate EL0 timer read accesses when FEAT_ECV in use

Message ID 20241217142321.763801-6-maz@kernel.org (mailing list archive)
State New
Headers show
Series KVM: arm64: Add NV timer support | expand

Commit Message

Marc Zyngier Dec. 17, 2024, 2:23 p.m. UTC
Although FEAT_ECV allows us to correctly emulate the timers, it also
reduces performances pretty badly.

Mitigate this by emulating the CTL/CVAL register reads in the
inner run loop, without returning to the general kernel.

Signed-off-by: Marc Zyngier <maz@kernel.org>
---
 arch/arm64/kvm/arch_timer.c             | 21 +-----
 arch/arm64/kvm/hyp/include/hyp/switch.h |  5 ++
 arch/arm64/kvm/hyp/vhe/switch.c         | 99 +++++++++++++++++++++++++
 include/kvm/arm_arch_timer.h            | 15 ++++
 4 files changed, 122 insertions(+), 18 deletions(-)
diff mbox series

Patch

diff --git a/arch/arm64/kvm/arch_timer.c b/arch/arm64/kvm/arch_timer.c
index b6a06bda9e534..6f04f31c0a7f2 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;
@@ -964,10 +949,10 @@  void kvm_timer_sync_nested(struct kvm_vcpu *vcpu)
 	 * which allows trapping of the timer registers even with NV2.
 	 * Still, this is still worse than FEAT_NV on its own. Meh.
 	 */
-	if (cpus_have_final_cap(ARM64_HAS_ECV) || !is_hyp_ctxt(vcpu))
-		return;
-
 	if (!vcpu_el2_e2h_is_set(vcpu)) {
+		if (cpus_have_final_cap(ARM64_HAS_ECV))
+			return;
+
 		/*
 		 * A non-VHE guest hypervisor doesn't have any direct access
 		 * to its timers: the EL2 registers trap (and the HW is
diff --git a/arch/arm64/kvm/hyp/include/hyp/switch.h b/arch/arm64/kvm/hyp/include/hyp/switch.h
index 34f53707892df..30e572de28749 100644
--- a/arch/arm64/kvm/hyp/include/hyp/switch.h
+++ b/arch/arm64/kvm/hyp/include/hyp/switch.h
@@ -501,6 +501,11 @@  static inline bool handle_tx2_tvm(struct kvm_vcpu *vcpu)
 	return true;
 }
 
+static inline u64 compute_counter_value(struct arch_timer_context *ctxt)
+{
+	return arch_timer_read_cntpct_el0() - timer_get_offset(ctxt);
+}
+
 static bool kvm_hyp_handle_cntpct(struct kvm_vcpu *vcpu)
 {
 	struct arch_timer_context *ctxt;
diff --git a/arch/arm64/kvm/hyp/vhe/switch.c b/arch/arm64/kvm/hyp/vhe/switch.c
index 80581b1c39959..51119d58ecff8 100644
--- a/arch/arm64/kvm/hyp/vhe/switch.c
+++ b/arch/arm64/kvm/hyp/vhe/switch.c
@@ -256,6 +256,102 @@  void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu)
 	host_data_ptr(host_ctxt)->__hyp_running_vcpu = NULL;
 }
 
+static u64 compute_emulated_cntx_ctl_el0(struct kvm_vcpu *vcpu,
+					 enum vcpu_sysreg reg)
+{
+	unsigned long ctl;
+	u64 cval, cnt;
+	bool stat;
+
+	switch (reg) {
+	case CNTP_CTL_EL0:
+		cval = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
+		ctl  = __vcpu_sys_reg(vcpu, CNTP_CTL_EL0);
+		cnt  = compute_counter_value(vcpu_ptimer(vcpu));
+		break;
+	case CNTV_CTL_EL0:
+		cval = __vcpu_sys_reg(vcpu, CNTV_CVAL_EL0);
+		ctl  = __vcpu_sys_reg(vcpu, CNTV_CTL_EL0);
+		cnt  = compute_counter_value(vcpu_vtimer(vcpu));
+		break;
+	default:
+		BUG();
+	}
+
+	stat = cval <= cnt;
+	__assign_bit(__ffs(ARCH_TIMER_CTRL_IT_STAT), &ctl, stat);
+
+	return ctl;
+}
+
+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 from guest's hypervisor context 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 = compute_emulated_cntx_ctl_el0(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 = compute_emulated_cntx_ctl_el0(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 = compute_emulated_cntx_ctl_el0(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 = compute_emulated_cntx_ctl_el0(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_eret(struct kvm_vcpu *vcpu, u64 *exit_code)
 {
 	u64 esr = kvm_vcpu_get_esr(vcpu);
@@ -409,6 +505,9 @@  static bool kvm_hyp_handle_sysreg_vhe(struct kvm_vcpu *vcpu, u64 *exit_code)
 	if (kvm_hyp_handle_tlbi_el2(vcpu, exit_code))
 		return true;
 
+	if (kvm_hyp_handle_timer(vcpu, exit_code))
+		return true;
+
 	if (kvm_hyp_handle_cpacr_el1(vcpu, exit_code))
 		return true;
 
diff --git a/include/kvm/arm_arch_timer.h b/include/kvm/arm_arch_timer.h
index 6e3f6b7ff2b22..c1ba31fab6f52 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