Skip to contentSkip to conversation

arm64: Add initial support for FEAT_WFxT

PATCH In review Marc Zyngier 2 participants 21 messages Apr 19, 2022 – Apr 20, 2022

Subsystems (4)
lore
Share & tools
lore

Marc Zyngier

author maintainer
#0 lore raw

v2 cover arm64: Add initial support for FEAT_WFxT

The ARMv8.7 WFxT feature is a new take on the good old WFI/WFE instructions as they behave the same way, only taking an extra timeout parameter.

This small series aims at adding the minimal support for this feature, enabling it for both the kernel and KVM.

A potential addition to this series would be to remove the event generation from the counters, and rely on the timeout where it matters (spinlocks?). Feedback welcome.

Patches on top of 5.18-rc2, tested of the FVP AEM.

* From v1 [1]:
- Properly generate traces even if the deadline has already expired
- Collect RBs, with thanks.

[1] https://lore.kernel.org/r/[email protected]

Marc Zyngier (10):
arm64: Expand ESR_ELx_WFx_ISS_TI to match its ARMv8.7 definition
arm64: Add RV and RN fields for ESR_ELx_WFx_ISS
KVM: arm64: Simplify kvm_cpu_has_pending_timer()
KVM: arm64: Introduce kvm_counter_compute_delta() helper
KVM: arm64: Handle blocking WFIT instruction
KVM: arm64: Offer early resume for non-blocking WFxT instructions
KVM: arm64: Expose the WFXT feature to guests
arm64: Add HWCAP advertising FEAT_WFXT
arm64: Add wfet()/wfit() helpers
arm64: Use WFxT for __delay() when possible

 Documentation/arm64/cpu-feature-registers.rst |  2 +
 Documentation/arm64/elf_hwcaps.rst            |  4 ++
 arch/arm64/include/asm/barrier.h              |  4 ++
 arch/arm64/include/asm/esr.h                  |  8 +++-
 arch/arm64/include/asm/hwcap.h                |  1 +
 arch/arm64/include/asm/kvm_host.h             |  1 +
 arch/arm64/include/uapi/asm/hwcap.h           |  1 +
 arch/arm64/kernel/cpufeature.c                | 13 +++++
 arch/arm64/kernel/cpuinfo.c                   |  1 +
 arch/arm64/kvm/arch_timer.c                   | 47 ++++++++++++-------
 arch/arm64/kvm/arm.c                          |  6 +--
 arch/arm64/kvm/handle_exit.c                  | 35 ++++++++++++--
 arch/arm64/kvm/sys_regs.c                     |  2 +
 arch/arm64/lib/delay.c                        | 12 ++++-
 arch/arm64/tools/cpucaps                      |  1 +
 include/kvm/arm_arch_timer.h                  |  2 -
 16 files changed, 110 insertions(+), 30 deletions(-)
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#1 lore raw

PATCH v2 1/10 arm64: Expand ESR_ELx_WFx_ISS_TI to match its ARMv8.7 definition

Starting with FEAT_WFXT in ARMv8.7, the TI field in the ISS that is reported on a WFx trap is expanded by one bit to allow the description of WFET and WFIT.

Special care is taken to exclude the WFxT bit from the mask used to match WFI so that it also matches WFIT when trapped from EL0.

---
 arch/arm64/include/asm/esr.h | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
arch/arm64/include/asm/esr.h +4 -2
index d52a0b269ee8..65c2201b11b2 100644
@@ -133,7 +133,8 @@
 #define ESR_ELx_CV		(UL(1) << 24)
 #define ESR_ELx_COND_SHIFT	(20)
 #define ESR_ELx_COND_MASK	(UL(0xF) << ESR_ELx_COND_SHIFT)
-#define ESR_ELx_WFx_ISS_TI	(UL(1) << 0)
+#define ESR_ELx_WFx_ISS_TI	(UL(3) << 0)
+#define ESR_ELx_WFx_ISS_WFxT	(UL(2) << 0)
 #define ESR_ELx_WFx_ISS_WFI	(UL(0) << 0)
 #define ESR_ELx_WFx_ISS_WFE	(UL(1) << 0)
 #define ESR_ELx_xVC_IMM_MASK	((1UL << 16) - 1)
@@ -146,7 +147,8 @@
 #define DISR_EL1_ESR_MASK	(ESR_ELx_AET | ESR_ELx_EA | ESR_ELx_FSC)
 
 /* ESR value templates for specific events */
-#define ESR_ELx_WFx_MASK	(ESR_ELx_EC_MASK | ESR_ELx_WFx_ISS_TI)
+#define ESR_ELx_WFx_MASK	(ESR_ELx_EC_MASK |			\
+				 (ESR_ELx_WFx_ISS_TI & ~ESR_ELx_WFx_ISS_WFxT))
 #define ESR_ELx_WFx_WFI_VAL	((ESR_ELx_EC_WFx << ESR_ELx_EC_SHIFT) |	\
 				 ESR_ELx_WFx_ISS_WFI)
 
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
maintainer Acked-by on v2 1/10
#2 lore raw

Re: [PATCH v2 01/10] arm64: Expand ESR_ELx_WFx_ISS_TI to match its ARMv8.7 definition

Quoted 10 lines · Marc Zyngier

Starting with FEAT_WFXT in ARMv8.7, the TI field in the ISS that is reported on a WFx trap is expanded by one bit to allow the description of WFET and WFIT.

Special care is taken to exclude the WFxT bit from the mask used to match WFI so that it also matches WFIT when trapped from EL0.

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

author maintainer
#3 lore raw

PATCH v2 2/10 arm64: Add RV and RN fields for ESR_ELx_WFx_ISS

In reply to Marc Zyngier

The ISS field exposed by ESR_ELx contain two additional subfields with FEAT_WFxT:

- RN, the register number containing the timeout
- RV, indicating if the register number is valid

Describe these two fields according to the arch spec.

No functional change.

---
 arch/arm64/include/asm/esr.h | 2 ++
 1 file changed, 2 insertions(+)
arch/arm64/include/asm/esr.h +2 -0
index 65c2201b11b2..15156c478054 100644
@@ -133,6 +133,8 @@
 #define ESR_ELx_CV		(UL(1) << 24)
 #define ESR_ELx_COND_SHIFT	(20)
 #define ESR_ELx_COND_MASK	(UL(0xF) << ESR_ELx_COND_SHIFT)
+#define ESR_ELx_WFx_ISS_RN	(UL(0x1F) << 5)
+#define ESR_ELx_WFx_ISS_RV	(UL(1) << 2)
 #define ESR_ELx_WFx_ISS_TI	(UL(3) << 0)
 #define ESR_ELx_WFx_ISS_WFxT	(UL(2) << 0)
 #define ESR_ELx_WFx_ISS_WFI	(UL(0) << 0)
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
maintainer Acked-by on v2 2/10
#4 lore raw

Re: [PATCH v2 02/10] arm64: Add RV and RN fields for ESR_ELx_WFx_ISS

Quoted 6 lines · Marc Zyngier

The ISS field exposed by ESR_ELx contain two additional subfields with FEAT_WFxT:

- RN, the register number containing the timeout
- RV, indicating if the register number is valid

Quoted 6 lines

Describe these two fields according to the arch spec.

No functional change.

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

author maintainer
#5 lore raw

PATCH v2 3/10 KVM: arm64: Simplify kvm_cpu_has_pending_timer()

In reply to Marc Zyngier

kvm_cpu_has_pending_timer() ends up checking all the possible timers for a wake-up cause. However, we already check for pending interrupts whenever we try to wake-up a vcpu, including the timer interrupts.

Obviously, doing the same work twice is once too many. Reduce this helper to almost nothing, but keep it around, as we are going to make use of it soon.

---
 arch/arm64/kvm/arch_timer.c  | 10 ++--------
 arch/arm64/kvm/arm.c         |  5 -----
 include/kvm/arm_arch_timer.h |  2 --
 3 files changed, 2 insertions(+), 15 deletions(-)
arch/arm64/kvm/arch_timer.c +2 -8
index 6e542e2eae32..16dda1a383a6 100644
@@ -350,15 +350,9 @@ static bool kvm_timer_should_fire(struct arch_timer_context *timer_ctx)
 	return cval <= now;
 }
 
-bool kvm_timer_is_pending(struct kvm_vcpu *vcpu)
+int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 {
-	struct timer_map map;
-
-	get_timer_map(vcpu, &map);
-
-	return kvm_timer_should_fire(map.direct_vtimer) ||
-	       kvm_timer_should_fire(map.direct_ptimer) ||
-	       kvm_timer_should_fire(map.emul_ptimer);
+	return 0;
 }
 
 /*
arch/arm64/kvm/arm.c +0 -5
index 523bc934fe2f..2122c699af06 100644
@@ -356,11 +356,6 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 	kvm_arm_vcpu_destroy(vcpu);
 }
 
-int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
-{
-	return kvm_timer_is_pending(vcpu);
-}
-
 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 {
 
include/kvm/arm_arch_timer.h +0 -2
index 51c19381108c..cd6d8f260eab 100644
@@ -76,8 +76,6 @@ int kvm_arm_timer_set_attr(struct kvm_vcpu *vcpu, struct kvm_device_attr *attr);
 int kvm_arm_timer_get_attr(struct kvm_vcpu *vcpu, struct kvm_device_attr *attr);
 int kvm_arm_timer_has_attr(struct kvm_vcpu *vcpu, struct kvm_device_attr *attr);
 
-bool kvm_timer_is_pending(struct kvm_vcpu *vcpu);
-
 u64 kvm_phys_timer_read(void);
 
 void kvm_timer_vcpu_load(struct kvm_vcpu *vcpu);
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#6 lore raw

PATCH v2 4/10 KVM: arm64: Introduce kvm_counter_compute_delta() helper

In reply to Marc Zyngier

Refactor kvm_timer_compute_delta() and extract a helper that compute the delta (in ns) between a given timer and an arbitrary value.

No functional change expected.

---
 arch/arm64/kvm/arch_timer.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
arch/arm64/kvm/arch_timer.c +10 -7
index 16dda1a383a6..c92a68190f6a 100644
@@ -208,18 +208,16 @@ static irqreturn_t kvm_arch_timer_handler(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
-static u64 kvm_timer_compute_delta(struct arch_timer_context *timer_ctx)
+static u64 kvm_counter_compute_delta(struct arch_timer_context *timer_ctx,
+				     u64 val)
 {
-	u64 cval, now;
-
-	cval = timer_get_cval(timer_ctx);
-	now = kvm_phys_timer_read() - timer_get_offset(timer_ctx);
+	u64 now = kvm_phys_timer_read() - timer_get_offset(timer_ctx);
 
-	if (now < cval) {
+	if (now < val) {
 		u64 ns;
 
 		ns = cyclecounter_cyc2ns(timecounter->cc,
-					 cval - now,
+					 val - now,
 					 timecounter->mask,
 					 &timecounter->frac);
 		return ns;
@@ -228,6 +226,11 @@ static u64 kvm_timer_compute_delta(struct arch_timer_context *timer_ctx)
 	return 0;
 }
 
+static u64 kvm_timer_compute_delta(struct arch_timer_context *timer_ctx)
+{
+	return kvm_counter_compute_delta(timer_ctx, timer_get_cval(timer_ctx));
+}
+
 static bool kvm_timer_irq_can_fire(struct arch_timer_context *timer_ctx)
 {
 	WARN_ON(timer_ctx && timer_ctx->loaded);
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#7 lore raw

PATCH v2 5/10 KVM: arm64: Handle blocking WFIT instruction

In reply to Marc Zyngier

When trapping a blocking WFIT instruction, take it into account when computing the deadline of the background timer.

The state is tracked with a new vcpu flag, and is gated by a new CPU capability, which isn't currently enabled.

---
 arch/arm64/include/asm/kvm_host.h |  1 +
 arch/arm64/kvm/arch_timer.c       | 22 ++++++++++++++++++++--
 arch/arm64/kvm/arm.c              |  1 +
 arch/arm64/kvm/handle_exit.c      |  7 ++++++-
 arch/arm64/tools/cpucaps          |  1 +
 5 files changed, 29 insertions(+), 3 deletions(-)
arch/arm64/include/asm/kvm_host.h +1 -0
index e3b25dc6c367..9e6e8701933e 100644
@@ -441,6 +441,7 @@ struct kvm_vcpu_arch {
 #define KVM_ARM64_DEBUG_STATE_SAVE_TRBE	(1 << 13) /* Save TRBE context if active  */
 #define KVM_ARM64_FP_FOREIGN_FPSTATE	(1 << 14)
 #define KVM_ARM64_ON_UNSUPPORTED_CPU	(1 << 15) /* Physical CPU not in supported_cpus */
+#define KVM_ARM64_WFIT			(1 << 16) /* WFIT instruction trapped */
 
 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
 				 KVM_GUESTDBG_USE_SW_BP | \
arch/arm64/kvm/arch_timer.c +20 -2
index c92a68190f6a..4e39ace073af 100644
@@ -239,6 +239,20 @@ static bool kvm_timer_irq_can_fire(struct arch_timer_context *timer_ctx)
 		  (ARCH_TIMER_CTRL_IT_MASK | ARCH_TIMER_CTRL_ENABLE)) == ARCH_TIMER_CTRL_ENABLE);
 }
 
+static bool vcpu_has_wfit_active(struct kvm_vcpu *vcpu)
+{
+	return (cpus_have_final_cap(ARM64_HAS_WFXT) &&
+		(vcpu->arch.flags & KVM_ARM64_WFIT));
+}
+
+static u64 wfit_delay_ns(struct kvm_vcpu *vcpu)
+{
+	struct arch_timer_context *ctx = vcpu_vtimer(vcpu);
+	u64 val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu));
+
+	return kvm_counter_compute_delta(ctx, val);
+}
+
 /*
  * Returns the earliest expiration time in ns among guest timers.
  * Note that it will return 0 if none of timers can fire.
@@ -256,6 +270,9 @@ static u64 kvm_timer_earliest_exp(struct kvm_vcpu *vcpu)
 			min_delta = min(min_delta, kvm_timer_compute_delta(ctx));
 	}
 
+	if (vcpu_has_wfit_active(vcpu))
+		min_delta = min(min_delta, wfit_delay_ns(vcpu));
+
 	/* If none of timers can fire, then return 0 */
 	if (min_delta == ULLONG_MAX)
 		return 0;
@@ -355,7 +372,7 @@ static bool kvm_timer_should_fire(struct arch_timer_context *timer_ctx)
 
 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 {
-	return 0;
+	return vcpu_has_wfit_active(vcpu) && wfit_delay_ns(vcpu) == 0;
 }
 
 /*
@@ -481,7 +498,8 @@ static void kvm_timer_blocking(struct kvm_vcpu *vcpu)
 	 */
 	if (!kvm_timer_irq_can_fire(map.direct_vtimer) &&
 	    !kvm_timer_irq_can_fire(map.direct_ptimer) &&
-	    !kvm_timer_irq_can_fire(map.emul_ptimer))
+	    !kvm_timer_irq_can_fire(map.emul_ptimer) &&
+	    !vcpu_has_wfit_active(vcpu))
 		return;
 
 	/*
arch/arm64/kvm/arm.c +1 -0
index 2122c699af06..e7cb8a4d2e81 100644
@@ -634,6 +634,7 @@ void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
 	preempt_enable();
 
 	kvm_vcpu_halt(vcpu);
+	vcpu->arch.flags &= ~KVM_ARM64_WFIT;
 	kvm_clear_request(KVM_REQ_UNHALT, vcpu);
 
 	preempt_disable();
arch/arm64/kvm/handle_exit.c +6 -1
index 97fe14aab1a3..4260f2cd1971 100644
@@ -85,16 +85,21 @@ static int handle_no_fpsimd(struct kvm_vcpu *vcpu)
  * WFI: Simply call kvm_vcpu_halt(), which will halt execution of
  * world-switches and schedule other host processes until there is an
  * incoming IRQ or FIQ to the VM.
+ * WFIT: Same as WFI, with a timed wakeup implemented as a background timer
  */
 static int kvm_handle_wfx(struct kvm_vcpu *vcpu)
 {
-	if (kvm_vcpu_get_esr(vcpu) & ESR_ELx_WFx_ISS_WFE) {
+	u64 esr = kvm_vcpu_get_esr(vcpu);
+
+	if (esr & ESR_ELx_WFx_ISS_WFE) {
 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true);
 		vcpu->stat.wfe_exit_stat++;
 		kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu));
 	} else {
 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false);
 		vcpu->stat.wfi_exit_stat++;
+		if ((esr & (ESR_ELx_WFx_ISS_RV | ESR_ELx_WFx_ISS_WFxT)) == (ESR_ELx_WFx_ISS_RV | ESR_ELx_WFx_ISS_WFxT))
+			vcpu->arch.flags |= KVM_ARM64_WFIT;
 		kvm_vcpu_wfi(vcpu);
 	}
 
arch/arm64/tools/cpucaps +1 -0
index 3ed418f70e3b..01f7d253dec4 100644
@@ -38,6 +38,7 @@ HAS_STAGE2_FWB
 HAS_SYSREG_GIC_CPUIF
 HAS_TLB_RANGE
 HAS_VIRT_HOST_EXTN
+HAS_WFXT
 HW_DBM
 KVM_PROTECTED_MODE
 MISMATCHED_CACHE_TYPE
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#8 lore raw

PATCH v2 6/10 KVM: arm64: Offer early resume for non-blocking WFxT instructions

In reply to Marc Zyngier

For WFxT instructions used with very small delays, it is not unlikely that the deadline is already expired by the time we reach the WFx handling code.

Check for this condition as soon as possible, and return to the guest immediately if we can.

---
 arch/arm64/kvm/handle_exit.c | 30 ++++++++++++++++++++++++++----
 1 file changed, 26 insertions(+), 4 deletions(-)
arch/arm64/kvm/handle_exit.c +26 -4
index 4260f2cd1971..7726b01dc09a 100644
@@ -80,12 +80,14 @@ static int handle_no_fpsimd(struct kvm_vcpu *vcpu)
  *
  * @vcpu:	the vcpu pointer
  *
- * WFE: Yield the CPU and come back to this vcpu when the scheduler
+ * WFE[T]: Yield the CPU and come back to this vcpu when the scheduler
  * decides to.
  * WFI: Simply call kvm_vcpu_halt(), which will halt execution of
  * world-switches and schedule other host processes until there is an
  * incoming IRQ or FIQ to the VM.
  * WFIT: Same as WFI, with a timed wakeup implemented as a background timer
+ *
+ * WF{I,E}T can immediately return if the deadline has already expired.
  */
 static int kvm_handle_wfx(struct kvm_vcpu *vcpu)
 {
@@ -94,15 +96,35 @@ static int kvm_handle_wfx(struct kvm_vcpu *vcpu)
 	if (esr & ESR_ELx_WFx_ISS_WFE) {
 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true);
 		vcpu->stat.wfe_exit_stat++;
-		kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu));
 	} else {
 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false);
 		vcpu->stat.wfi_exit_stat++;
-		if ((esr & (ESR_ELx_WFx_ISS_RV | ESR_ELx_WFx_ISS_WFxT)) == (ESR_ELx_WFx_ISS_RV | ESR_ELx_WFx_ISS_WFxT))
+	}
+
+	if (esr & ESR_ELx_WFx_ISS_WFxT) {
+		if (esr & ESR_ELx_WFx_ISS_RV) {
+			u64 val, now;
+
+			now = kvm_arm_timer_get_reg(vcpu, KVM_REG_ARM_TIMER_CNT);
+			val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu));
+
+			if (now >= val)
+				goto out;
+		} else {
+			/* Treat WFxT as WFx if RN is invalid */
+			esr &= ~ESR_ELx_WFx_ISS_WFxT;
+		}
+	}
+
+	if (esr & ESR_ELx_WFx_ISS_WFE) {
+		kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu));
+	} else {
+		if (esr & ESR_ELx_WFx_ISS_WFxT)
 			vcpu->arch.flags |= KVM_ARM64_WFIT;
+
 		kvm_vcpu_wfi(vcpu);
 	}
-
+out:
 	kvm_incr_pc(vcpu);
 
 	return 1;
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#9 lore raw

PATCH v2 7/10 KVM: arm64: Expose the WFXT feature to guests

In reply to Marc Zyngier

Plumb in the capability, and expose WFxT to guests when available.

---
 arch/arm64/kernel/cpufeature.c | 12 ++++++++++++
 arch/arm64/kvm/sys_regs.c      |  2 ++
 2 files changed, 14 insertions(+)
arch/arm64/kernel/cpufeature.c +12 -0
index d72c4b4d389c..db6d9ab685e5 100644
@@ -237,6 +237,7 @@ static const struct arm64_ftr_bits ftr_id_aa64isar2[] = {
 	ARM64_FTR_BITS(FTR_VISIBLE_IF_IS_ENABLED(CONFIG_ARM64_PTR_AUTH),
 		       FTR_STRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_GPA3_SHIFT, 4, 0),
 	ARM64_FTR_BITS(FTR_VISIBLE, FTR_NONSTRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_RPRES_SHIFT, 4, 0),
+	ARM64_FTR_BITS(FTR_HIDDEN, FTR_NONSTRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_WFXT_SHIFT, 4, 0),
 	ARM64_FTR_END,
 };
 
@@ -2442,6 +2443,17 @@ static const struct arm64_cpu_capabilities arm64_features[] = {
 		.matches = has_cpuid_feature,
 		.min_field_value = 1,
 	},
+	{
+		.desc = "WFx with timeout",
+		.capability = ARM64_HAS_WFXT,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.sys_reg = SYS_ID_AA64ISAR2_EL1,
+		.sign = FTR_UNSIGNED,
+		.field_pos = ID_AA64ISAR2_WFXT_SHIFT,
+		.field_width = 4,
+		.matches = has_cpuid_feature,
+		.min_field_value = ID_AA64ISAR2_WFXT_SUPPORTED,
+	},
 	{},
 };
 
arch/arm64/kvm/sys_regs.c +2 -0
index 7b45c040cc27..cc9a77546cc0 100644
@@ -1144,6 +1144,8 @@ static u64 read_id_reg(const struct kvm_vcpu *vcpu,
 		if (!vcpu_has_ptrauth(vcpu))
 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR2_APA3) |
 				 ARM64_FEATURE_MASK(ID_AA64ISAR2_GPA3));
+		if (!cpus_have_final_cap(ARM64_HAS_WFXT))
+			val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_WFXT);
 		break;
 	case SYS_ID_AA64DFR0_EL1:
 		/* Limit debug to ARMv8.0 */
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#11 lore raw

PATCH v2 8/10 arm64: Add HWCAP advertising FEAT_WFXT

In reply to Marc Zyngier

In order to allow userspace to enjoy WFET, add a new HWCAP that advertises it when available.

---
 Documentation/arm64/cpu-feature-registers.rst | 2 ++
 Documentation/arm64/elf_hwcaps.rst            | 4 ++++
 arch/arm64/include/asm/hwcap.h                | 1 +
 arch/arm64/include/uapi/asm/hwcap.h           | 1 +
 arch/arm64/kernel/cpufeature.c                | 3 ++-
 arch/arm64/kernel/cpuinfo.c                   | 1 +
 6 files changed, 11 insertions(+), 1 deletion(-)
Documentation/arm64/cpu-feature-registers.rst +2 -0
index 749ae970c319..04ba83e1965f 100644
@@ -290,6 +290,8 @@ infrastructure:
      +------------------------------+---------+---------+
      | RPRES                        | [7-4]   |    y    |
      +------------------------------+---------+---------+
+     | WFXT                         | [3-0]   |    y    |
+     +------------------------------+---------+---------+
 
 
 Appendix I: Example
Documentation/arm64/elf_hwcaps.rst +4 -0
index a8f30963e550..af3ab524e826 100644
@@ -264,6 +264,10 @@ HWCAP2_MTE3
     Functionality implied by ID_AA64PFR1_EL1.MTE == 0b0011, as described
     by Documentation/arm64/memory-tagging-extension.rst.
 
+HWCAP2_WFXT
+
+    Functionality implied by ID_AA64ISAR2_EL1.WFXT == 0b0010.
+
 4. Unused AT_HWCAP bits
 -----------------------
 
arch/arm64/include/asm/hwcap.h +1 -0
index 8db5ec0089db..909337b50e1f 100644
@@ -109,6 +109,7 @@
 #define KERNEL_HWCAP_AFP		__khwcap2_feature(AFP)
 #define KERNEL_HWCAP_RPRES		__khwcap2_feature(RPRES)
 #define KERNEL_HWCAP_MTE3		__khwcap2_feature(MTE3)
+#define KERNEL_HWCAP_WFXT		__khwcap2_feature(WFXT)
 
 /*
  * This yields a mask that user programs can use to figure out what
arch/arm64/include/uapi/asm/hwcap.h +1 -0
index 99cb5d383048..9dddde1d046c 100644
@@ -79,5 +79,6 @@
 #define HWCAP2_AFP		(1 << 20)
 #define HWCAP2_RPRES		(1 << 21)
 #define HWCAP2_MTE3		(1 << 22)
+#define HWCAP2_WFXT		(1 << 23)
 
 #endif /* _UAPI__ASM_HWCAP_H */
arch/arm64/kernel/cpufeature.c +2 -1
index db6d9ab685e5..945190ebadd5 100644
@@ -237,7 +237,7 @@ static const struct arm64_ftr_bits ftr_id_aa64isar2[] = {
 	ARM64_FTR_BITS(FTR_VISIBLE_IF_IS_ENABLED(CONFIG_ARM64_PTR_AUTH),
 		       FTR_STRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_GPA3_SHIFT, 4, 0),
 	ARM64_FTR_BITS(FTR_VISIBLE, FTR_NONSTRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_RPRES_SHIFT, 4, 0),
-	ARM64_FTR_BITS(FTR_HIDDEN, FTR_NONSTRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_WFXT_SHIFT, 4, 0),
+	ARM64_FTR_BITS(FTR_VISIBLE, FTR_NONSTRICT, FTR_LOWER_SAFE, ID_AA64ISAR2_WFXT_SHIFT, 4, 0),
 	ARM64_FTR_END,
 };
 
@@ -2587,6 +2587,7 @@ static const struct arm64_cpu_capabilities arm64_elf_hwcaps[] = {
 	HWCAP_CAP(SYS_ID_AA64MMFR0_EL1, ID_AA64MMFR0_ECV_SHIFT, 4, FTR_UNSIGNED, 1, CAP_HWCAP, KERNEL_HWCAP_ECV),
 	HWCAP_CAP(SYS_ID_AA64MMFR1_EL1, ID_AA64MMFR1_AFP_SHIFT, 4, FTR_UNSIGNED, 1, CAP_HWCAP, KERNEL_HWCAP_AFP),
 	HWCAP_CAP(SYS_ID_AA64ISAR2_EL1, ID_AA64ISAR2_RPRES_SHIFT, 4, FTR_UNSIGNED, 1, CAP_HWCAP, KERNEL_HWCAP_RPRES),
+	HWCAP_CAP(SYS_ID_AA64ISAR2_EL1, ID_AA64ISAR2_WFXT_SHIFT, 4, FTR_UNSIGNED, ID_AA64ISAR2_WFXT_SUPPORTED, CAP_HWCAP, KERNEL_HWCAP_WFXT),
 	{},
 };
 
arch/arm64/kernel/cpuinfo.c +1 -0
index 330b92ea863a..781eab314794 100644
@@ -98,6 +98,7 @@ static const char *const hwcap_str[] = {
 	[KERNEL_HWCAP_AFP]		= "afp",
 	[KERNEL_HWCAP_RPRES]		= "rpres",
 	[KERNEL_HWCAP_MTE3]		= "mte3",
+	[KERNEL_HWCAP_WFXT]		= "wfxt",
 };
 
 #ifdef CONFIG_COMPAT
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
maintainer Acked-by on v2 8/10
#12 lore raw

Re: [PATCH v2 08/10] arm64: Add HWCAP advertising FEAT_WFXT

Quoted 4 lines · Marc Zyngier

In order to allow userspace to enjoy WFET, add a new HWCAP that advertises it when available.

I assume this series will go in via the kvm tree? Could we have the HWCAP changes at the beginning, I'm sure they'll conflict with other series (I plan to queue the SME patches).

Signature · 1 line
Catalin
author maintainer
#13 lore raw

Re: [PATCH v2 08/10] arm64: Add HWCAP advertising FEAT_WFXT

Quoted 5 lines · Catalin Marinas
Quoted 3 lines · Marc Zyngier

In order to allow userspace to enjoy WFET, add a new HWCAP that advertises it when available.

Quoted 7 lines
Quoted 1 line

I assume this series will go in via the kvm tree? Could we have the HWCAP changes at the beginning, I'm sure they'll conflict with other series (I plan to queue the SME patches).

No problem, I can rejig the series to fit that plan. I'll push something tomorrow.

Thanks,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
author maintainer
#14 lore raw

PATCH v2 9/10 arm64: Add wfet()/wfit() helpers

In reply to Marc Zyngier

Just like we have helpers for WFI and WFE, add the WFxT versions. Note that the encoding is that reported by objdump, as no currrent toolchain knows about these instructions yet.

---
 arch/arm64/include/asm/barrier.h | 4 ++++
 1 file changed, 4 insertions(+)
arch/arm64/include/asm/barrier.h +4 -0
index 62217be36217..9f3e2c3d2ca0 100644
@@ -16,7 +16,11 @@
 
 #define sev()		asm volatile("sev" : : : "memory")
 #define wfe()		asm volatile("wfe" : : : "memory")
+#define wfet(val)	asm volatile("msr s0_3_c1_c0_0, %0"	\
+				     : : "r" (val) : "memory")
 #define wfi()		asm volatile("wfi" : : : "memory")
+#define wfit(val)	asm volatile("msr s0_3_c1_c0_1, %0"	\
+				     : : "r" (val) : "memory")
 
 #define isb()		asm volatile("isb" : : : "memory")
 #define dmb(opt)	asm volatile("dmb " #opt : : : "memory")
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#16 lore raw

PATCH v2 10/10 arm64: Use WFxT for __delay() when possible

In reply to Marc Zyngier

Marginally optimise __delay() by using a WFIT/WFET sequence. It probably is a win if no interrupt fires during the delay.

---
 arch/arm64/lib/delay.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
arch/arm64/lib/delay.c +11 -1
index 1688af0a4c97..5b7890139bc2 100644
@@ -27,7 +27,17 @@ void __delay(unsigned long cycles)
 {
 	cycles_t start = get_cycles();
 
-	if (arch_timer_evtstrm_available()) {
+	if (cpus_have_const_cap(ARM64_HAS_WFXT)) {
+		u64 end = start + cycles;
+
+		/*
+		 * Start with WFIT. If an interrupt makes us resume
+		 * early, use a WFET loop to complete the delay.
+		 */
+		wfit(end);
+		while ((get_cycles() - start) < cycles)
+			wfet(end);
+	} else 	if (arch_timer_evtstrm_available()) {
 		const cycles_t timer_evt_period =
 			USECS_TO_CYCLES(ARCH_TIMER_EVT_STREAM_PERIOD_US);
 
Signature · 6 lines
2.34.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
maintainer Acked-by on v2 10/10
#17 lore raw

Re: [PATCH v2 10/10] arm64: Use WFxT for __delay() when possible

Quoted 17 lines · Marc Zyngier

Marginally optimise __delay() by using a WFIT/WFET sequence. It probably is a win if no interrupt fires during the delay.

---
 arch/arm64/lib/delay.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
arch/arm64/lib/delay.c +0 -1
index 1688af0a4c97..5b7890139bc2 100644
@@ -27,7 +27,17 @@ void __delay(unsigned long cycles)
 {
 	cycles_t start = get_cycles();
 
-	if (arch_timer_evtstrm_available()) {

arch/arm64/lib/delay.c void __delay(unsigned long cycles) · line 39

Quoted 10 lines
arch/arm64/lib/delay.c +10 -0
+	if (cpus_have_const_cap(ARM64_HAS_WFXT)) {
+		u64 end = start + cycles;
+
+		/*
+		 * Start with WFIT. If an interrupt makes us resume
+		 * early, use a WFET loop to complete the delay.
+		 */
+		wfit(end);
+		while ((get_cycles() - start) < cycles)
+			wfet(end);

Do you use WFET here as a pending interrupt would cause WFIT to complete immediately?

arch/arm64/lib/delay.c void __delay(unsigned long cycles) · line 40

Quoted 1 line
arch/arm64/lib/delay.c +1 -0
+	} else 	if (arch_timer_evtstrm_available()) {

Nit: two spaces between else and if ;).

author maintainer
#18 lore raw

Re: [PATCH v2 10/10] arm64: Use WFxT for __delay() when possible

Quoted 18 lines · Catalin Marinas
Quoted 16 lines · Marc Zyngier

Marginally optimise __delay() by using a WFIT/WFET sequence. It probably is a win if no interrupt fires during the delay.

---
 arch/arm64/lib/delay.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
arch/arm64/lib/delay.c +0 -0
index 1688af0a4c97..5b7890139bc2 100644
@@ -27,7 +27,17 @@ void __delay(unsigned long cycles)
 {
 	cycles_t start = get_cycles();
 
Quoted 14 lines

arch/arm64/lib/delay.c void __delay(unsigned long cycles) · line 39

Quoted 11 lines
arch/arm64/lib/delay.c +10 -1
-	if (arch_timer_evtstrm_available()) {
+	if (cpus_have_const_cap(ARM64_HAS_WFXT)) {
+		u64 end = start + cycles;
+
+		/*
+		 * Start with WFIT. If an interrupt makes us resume
+		 * early, use a WFET loop to complete the delay.
+		 */
+		wfit(end);
+		while ((get_cycles() - start) < cycles)
+			wfet(end);

Do you use WFET here as a pending interrupt would cause WFIT to complete immediately?

Yes, that's the idea. Even if a pending interrupt is not immediately present, it could come halfway through, shortening the delay, and making WFIT useless until the interrupt is acknowledged.

I would have loved for WFIT to return a status indicating whether the wakeup was for a pending interrupt or for another reason (such as reaching the timeout), but apparently it was too much to ask... Maybe in ARMv11! ;-)

Quoted 4 lines

arch/arm64/lib/delay.c void __delay(unsigned long cycles) · line 40

Quoted 2 lines
arch/arm64/lib/delay.c +1 -0
+	} else 	if (arch_timer_evtstrm_available()) {

Nit: two spaces between else and if ;).

I'll make sure to fix this! ;-)

Quoted 3 lines

Thanks,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
maintainer
#19 lore raw
In reply to Marc Zyngier
Quoted 3 lines · Marc Zyngier

A potential addition to this series would be to remove the event generation from the counters, and rely on the timeout where it matters (spinlocks?). Feedback welcome.

I think we still need to keep the event generation around, at least for hardware bugs we don't know about. I don't think user-space rely on it though, people tend to come up with weird delays like isb ;). But yes, the WFET should be handy when it turns up in hardware.

Signature · 5 lines
Catalin
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#20 lore raw
Quoted 10 lines · Catalin Marinas
Quoted 3 lines · Marc Zyngier

A potential addition to this series would be to remove the event generation from the counters, and rely on the timeout where it matters (spinlocks?). Feedback welcome.

I think we still need to keep the event generation around, at least for hardware bugs we don't know about. I don't think user-space rely on it though, people tend to come up with weird delays like isb ;). But yes, the WFET should be handy when it turns up in hardware.

My hope was that the trick of using the event generation to work around systems failing to broadcast events could become a thing of the past when WFET is present in the HW. After all, they serve the same purpose (generate a local event to un-wedge the CPU).

But the more I look at it, the more I hate the potential solution. One of the issues is that WFxT takes an absolute deadline, rather than a relative one. So you end up with things like:

	ISB
	MRS	x0, CNTVCT_EL0
	ADD	x0, x0, #some_small_value
	WFET	x0

which is really heavy handed for the slow path of an atomic operation. Even if you have ECV and CNTVCTSS_EL0 (which allows you to get rid of the ISB), it is a royal pain.

It would be much better if there was a *relative* version of WFET that would directly take a timeout relative to the current virtual count, but I can sense HW designers calling me names already, so I'll shut up.

	M.
Signature · 5 lines
Without deviation from the norm, progress is not possible.
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm