Skip to contentSkip to conversation

KVM: arm64: nv: Reduce FP/SVE overhead on exception/exception return

PATCH Applied Marc Zyngier 3 participants 8 messages May 20 – May 21

Subsystems (3)
lore
Share & tools
lore

Marc Zyngier

author maintainer
#0 lore raw

v2 cover KVM: arm64: nv: Reduce FP/SVE overhead on exception/exception return

This is the second version of this short series optimising away a lot of unnecessary FPSIMD/SVE context switch with NV.

* From v1 [1]:

- New commit message on patch #2 (Mark)

- Additional comments and WARN_ON_ONCE() (Mark)

If nobody screams, I'll stick that into -next.

Thanks,

	M.

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

Marc Zyngier (2):
KVM: arm64: nv: Track L2 to L1 exception emulation
KVM: arm64: nv: Don't save/restore FP register during a nested ERET or
exception

 arch/arm64/include/asm/kvm_host.h |  3 ++-
 arch/arm64/kvm/emulate-nested.c   |  4 ++++
 arch/arm64/kvm/fpsimd.c           | 23 +++++++++++++++++++++++
 3 files changed, 29 insertions(+), 1 deletion(-)
Signature · 1 line
2.47.3
author maintainer
#1 lore raw

PATCH v2 1/2 KVM: arm64: nv: Track L2 to L1 exception emulation

While we currently track that we are emulating a nested ERET from L1 to L2, we currently don't track the reverse direction (an exception going from L2 to L1).

Add a new vcpu state flag for this purpose, which will see some use shortly.

---
 arch/arm64/include/asm/kvm_host.h | 3 ++-
 arch/arm64/kvm/emulate-nested.c   | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
arch/arm64/include/asm/kvm_host.h +2 -1
index 65eead8362e0b..c79747d5f4dd1 100644
@@ -1112,7 +1112,8 @@ struct kvm_vcpu_arch {
 #define IN_NESTED_ERET		__vcpu_single_flag(sflags, BIT(7))
 /* SError pending for nested guest */
 #define NESTED_SERROR_PENDING	__vcpu_single_flag(sflags, BIT(8))
-
+/* KVM is currently emulating an L2 to L1 exception */
+#define IN_NESTED_EXCEPTION	__vcpu_single_flag(sflags, BIT(9))
 
 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) +	\
arch/arm64/kvm/emulate-nested.c +4 -0
index dba7ced74ca5e..15c691a6266d5 100644
@@ -2862,6 +2862,8 @@ static int kvm_inject_nested(struct kvm_vcpu *vcpu, u64 esr_el2,
 
 	preempt_disable();
 
+	vcpu_set_flag(vcpu, IN_NESTED_EXCEPTION);
+
 	/*
 	 * We may have an exception or PC update in the EL0/EL1 context.
 	 * Commit it before entering EL2.
@@ -2884,6 +2886,8 @@ static int kvm_inject_nested(struct kvm_vcpu *vcpu, u64 esr_el2,
 	__kvm_adjust_pc(vcpu);
 
 	kvm_arch_vcpu_load(vcpu, smp_processor_id());
+	vcpu_clear_flag(vcpu, IN_NESTED_EXCEPTION);
+
 	preempt_enable();
 
 	if (kvm_vcpu_has_pmu(vcpu))
Signature · 1 line
2.47.3
author maintainer
#2 lore raw

PATCH v2 2/2 KVM: arm64: nv: Don't save/restore FP register during a nested ERET or exception

In reply to Marc Zyngier

When switching between L1 and L2, we save the old state using kvm_arch_vcpu_put(), mutate the state in memory, then load the new state using kvm_arch_vcpu_load(). Any live FPSIMD/SVE state is saved and unbound, such that it can be lazily restored on a subsequent trap.

The FPSIMD/SVE state is shared by exception levels, and only a handful of related control registers need to be changed when transitioning between L1 and L2. The save/restore of the common state is needless overhead, especially as trapping becomes exponentially more expensive with nesting.

Avoid this overhead by leaving the common FPSIMD/SVE state live on the CPU, and only switching the state that is distinct for L1 and L2:

- the trap controls: the effective values are recomputed on each entry into the guest to take the EL into account and merge the L0 and L1 configuration if in a nested context, or directly use the L0 configuration in non-nested context (see __activate_traps()).

- the VL settings: the effective values are are also recomputed on each entry into the guest (see fpsimd_lazy_switch_to_guest()).

Since we appear to cover all bases, use the vcpu flags indicating the handling of a nested ERET or exception delivery to avoid the whole FP save/restore shenanigans. SME will have to be similarly dealt with when it eventually gets supported.

For an EL1 L3 guest where L1 and L2 have this optimisation, this results in at least a 10% wall clock reduction when running an I/O heavy workload, generating a high rate of nested exceptions.

---
 arch/arm64/kvm/fpsimd.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
arch/arm64/kvm/fpsimd.c +23 -0
index 15e17aca1dec0..aca98752a6e42 100644
@@ -28,6 +28,20 @@ void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu)
 	if (!system_supports_fpsimd())
 		return;
 
+	/*
+	 * Avoid needless save/restore of the guest's common
+	 * FPSIMD/SVE/SME regs during transitions between L1/L2.
+	 *
+	 * These transitions only happens in a non-preemptible context
+	 * where the host regs have already been saved and unbound. The
+	 * live registers are either free or owned by the guest.
+	 */
+	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
+	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
+		WARN_ON_ONCE(host_owns_fp_regs());
+		return;
+	}
+
 	/*
 	 * Ensure that any host FPSIMD/SVE/SME state is saved and unbound such
 	 * that the host kernel is responsible for restoring this state upon
@@ -102,6 +116,15 @@ void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu)
 {
 	unsigned long flags;
 
+	/*
+	 * See comment in kvm_arch_vcpu_load_fp().
+	 */
+	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
+	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
+		WARN_ON_ONCE(host_owns_fp_regs());
+		return;
+	}
+
 	local_irq_save(flags);
 
 	if (guest_owns_fp_regs()) {
Signature · 1 line
2.47.3
reviewer Reviewed-by on v2 2/2
#3 lore raw

Re: [PATCH v2 2/2] KVM: arm64: nv: Don't save/restore FP register during a nested ERET or exception

Hi Marc,

Quoted 5 lines · Marc Zyngier

When switching between L1 and L2, we save the old state using kvm_arch_vcpu_put(), mutate the state in memory, then load the new state using kvm_arch_vcpu_load(). Any live FPSIMD/SVE state is saved and unbound, such that it can be lazily restored on a subsequent trap.

Quoted 8 lines

The FPSIMD/SVE state is shared by exception levels, and only a handful of related control registers need to be changed when transitioning between L1 and L2. The save/restore of the common state is needless overhead, especially as trapping becomes exponentially more expensive with nesting.

Avoid this overhead by leaving the common FPSIMD/SVE state live on the CPU, and only switching the state that is distinct for L1 and L2:

To make sure I understand this part:

L1 sets up L2's FP state live on the CPU
L1 erets
eret traps to L0/host
preemption disabled
kvm_arch_vcpu_put()
kvm_arch_vcpu_put_fp() <-- actually saves the state of the live registers
.. set elr etc ..
kvm_arch_vcpu_load()
kvm_arch_vcpu_load_fp() <-- doesn't actually restore state, but ensures
the CPTR trap will be set
.. returns to L2 (traps on first use of FP and state will be restored)

So this patch is (effectively) removing the put_fp()/load_fp(), because the FP state is common/shared between L1 and L2, so whatever L1 put into that state before the eret, L2 was going to see.

If my understanding is correct:

Thanks,
Joey

Quoted 67 lines

- the trap controls: the effective values are recomputed on each entry into the guest to take the EL into account and merge the L0 and L1 configuration if in a nested context, or directly use the L0 configuration in non-nested context (see __activate_traps()).

- the VL settings: the effective values are are also recomputed on each entry into the guest (see fpsimd_lazy_switch_to_guest()).

Since we appear to cover all bases, use the vcpu flags indicating the handling of a nested ERET or exception delivery to avoid the whole FP save/restore shenanigans. SME will have to be similarly dealt with when it eventually gets supported.

For an EL1 L3 guest where L1 and L2 have this optimisation, this results in at least a 10% wall clock reduction when running an I/O heavy workload, generating a high rate of nested exceptions.

---
 arch/arm64/kvm/fpsimd.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
arch/arm64/kvm/fpsimd.c +23 -0
index 15e17aca1dec0..aca98752a6e42 100644
@@ -28,6 +28,20 @@ void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu)
 	if (!system_supports_fpsimd())
 		return;
 
+	/*
+	 * Avoid needless save/restore of the guest's common
+	 * FPSIMD/SVE/SME regs during transitions between L1/L2.
+	 *
+	 * These transitions only happens in a non-preemptible context
+	 * where the host regs have already been saved and unbound. The
+	 * live registers are either free or owned by the guest.
+	 */
+	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
+	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
+		WARN_ON_ONCE(host_owns_fp_regs());
+		return;
+	}
+
 	/*
 	 * Ensure that any host FPSIMD/SVE/SME state is saved and unbound such
 	 * that the host kernel is responsible for restoring this state upon
@@ -102,6 +116,15 @@ void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu)
 {
 	unsigned long flags;
 
+	/*
+	 * See comment in kvm_arch_vcpu_load_fp().
+	 */
+	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
+	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
+		WARN_ON_ONCE(host_owns_fp_regs());
+		return;
+	}
+
 	local_irq_save(flags);
 
 	if (guest_owns_fp_regs()) {
Signature · 1 line
2.47.3
author maintainer
#4 lore raw

Re: [PATCH v2 2/2] KVM: arm64: nv: Don't save/restore FP register during a nested ERET or exception

Quoted 32 lines · Joey Gouly

Hi Marc,

Quoted 5 lines · Marc Zyngier

When switching between L1 and L2, we save the old state using kvm_arch_vcpu_put(), mutate the state in memory, then load the new state using kvm_arch_vcpu_load(). Any live FPSIMD/SVE state is saved and unbound, such that it can be lazily restored on a subsequent trap.

Quoted 8 lines

The FPSIMD/SVE state is shared by exception levels, and only a handful of related control registers need to be changed when transitioning between L1 and L2. The save/restore of the common state is needless overhead, especially as trapping becomes exponentially more expensive with nesting.

Avoid this overhead by leaving the common FPSIMD/SVE state live on the CPU, and only switching the state that is distinct for L1 and L2:

To make sure I understand this part:

L1 sets up L2's FP state live on the CPU
L1 erets
eret traps to L0/host
preemption disabled
kvm_arch_vcpu_put()
kvm_arch_vcpu_put_fp() <-- actually saves the state of the live registers
.. set elr etc ..
kvm_arch_vcpu_load()
kvm_arch_vcpu_load_fp() <-- doesn't actually restore state, but ensures
the CPTR trap will be set
.. returns to L2 (traps on first use of FP and state will be restored)

Quoted 3 lines

So this patch is (effectively) removing the put_fp()/load_fp(), because the FP state is common/shared between L1 and L2, so whatever L1 put into that state before the eret, L2 was going to see.

Yes, you got it right. The other path is on L1 to L2 exception, which also requires L0 mediation and has a similar shape.

The most horrible thing is that because all these traps can happen at a arbitrary depth, each individual trap usually results in the combination of all of the above.

Quoted 2 lines

If my understanding is correct:

Thanks!

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
reviewer Acked-by on v2 2/2
#5 lore raw

Re: [PATCH v2 2/2] KVM: arm64: nv: Don't save/restore FP register during a nested ERET or exception

In reply to Marc Zyngier
Quoted 22 lines · Marc Zyngier

When switching between L1 and L2, we save the old state using kvm_arch_vcpu_put(), mutate the state in memory, then load the new state using kvm_arch_vcpu_load(). Any live FPSIMD/SVE state is saved and unbound, such that it can be lazily restored on a subsequent trap.

The FPSIMD/SVE state is shared by exception levels, and only a handful of related control registers need to be changed when transitioning between L1 and L2. The save/restore of the common state is needless overhead, especially as trapping becomes exponentially more expensive with nesting.

Avoid this overhead by leaving the common FPSIMD/SVE state live on the CPU, and only switching the state that is distinct for L1 and L2:

- the trap controls: the effective values are recomputed on each entry into the guest to take the EL into account and merge the L0 and L1 configuration if in a nested context, or directly use the L0 configuration in non-nested context (see __activate_traps()).

- the VL settings: the effective values are are also recomputed on each entry into the guest (see fpsimd_lazy_switch_to_guest()).

Quoted 8 lines

Since we appear to cover all bases, use the vcpu flags indicating the handling of a nested ERET or exception delivery to avoid the whole FP save/restore shenanigans. SME will have to be similarly dealt with when it eventually gets supported.

For an EL1 L3 guest where L1 and L2 have this optimisation, this results in at least a 10% wall clock reduction when running an I/O heavy workload, generating a high rate of nested exceptions.

There's on additional thing that's important, but I forgot to mention last time: in the window between kvm_arch_vcpu_put() and kvm_arch_vcpu_load(), it's possible to take an interrupt, and for a softirq handler to try to use kernel mode NEON.

Due to that, kvm_arch_vcpu_put() must leave the L1 guest's maximum VL configured in the host's ZCR_ELx, such that the guest's state can be saved.

That value is configured by fpsimd_lazy_switch_to_host(), so we just need to make sure that kvm_arch_vcpu_put() doesn't clobber it. I *think* that's fine today, but maybe that warrants a comment somewhere.

Other than that, this all looks good to me:

Mark.

Quoted 49 lines
---
 arch/arm64/kvm/fpsimd.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
arch/arm64/kvm/fpsimd.c +23 -0
index 15e17aca1dec0..aca98752a6e42 100644
@@ -28,6 +28,20 @@ void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu)
 	if (!system_supports_fpsimd())
 		return;
 
+	/*
+	 * Avoid needless save/restore of the guest's common
+	 * FPSIMD/SVE/SME regs during transitions between L1/L2.
+	 *
+	 * These transitions only happens in a non-preemptible context
+	 * where the host regs have already been saved and unbound. The
+	 * live registers are either free or owned by the guest.
+	 */
+	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
+	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
+		WARN_ON_ONCE(host_owns_fp_regs());
+		return;
+	}
+
 	/*
 	 * Ensure that any host FPSIMD/SVE/SME state is saved and unbound such
 	 * that the host kernel is responsible for restoring this state upon
@@ -102,6 +116,15 @@ void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu)
 {
 	unsigned long flags;
 
+	/*
+	 * See comment in kvm_arch_vcpu_load_fp().
+	 */
+	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
+	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
+		WARN_ON_ONCE(host_owns_fp_regs());
+		return;
+	}
+
 	local_irq_save(flags);
 
 	if (guest_owns_fp_regs()) {
Signature · 1 line
2.47.3
author maintainer
#6 lore raw

Re: [PATCH v2 2/2] KVM: arm64: nv: Don't save/restore FP register during a nested ERET or exception

Quoted 38 lines · Mark Rutland
Quoted 22 lines · Marc Zyngier

When switching between L1 and L2, we save the old state using kvm_arch_vcpu_put(), mutate the state in memory, then load the new state using kvm_arch_vcpu_load(). Any live FPSIMD/SVE state is saved and unbound, such that it can be lazily restored on a subsequent trap.

The FPSIMD/SVE state is shared by exception levels, and only a handful of related control registers need to be changed when transitioning between L1 and L2. The save/restore of the common state is needless overhead, especially as trapping becomes exponentially more expensive with nesting.

Avoid this overhead by leaving the common FPSIMD/SVE state live on the CPU, and only switching the state that is distinct for L1 and L2:

- the trap controls: the effective values are recomputed on each entry into the guest to take the EL into account and merge the L0 and L1 configuration if in a nested context, or directly use the L0 configuration in non-nested context (see __activate_traps()).

- the VL settings: the effective values are are also recomputed on each entry into the guest (see fpsimd_lazy_switch_to_guest()).

Quoted 8 lines

Since we appear to cover all bases, use the vcpu flags indicating the handling of a nested ERET or exception delivery to avoid the whole FP save/restore shenanigans. SME will have to be similarly dealt with when it eventually gets supported.

For an EL1 L3 guest where L1 and L2 have this optimisation, this results in at least a 10% wall clock reduction when running an I/O heavy workload, generating a high rate of nested exceptions.

There's on additional thing that's important, but I forgot to mention last time: in the window between kvm_arch_vcpu_put() and kvm_arch_vcpu_load(), it's possible to take an interrupt, and for a softirq handler to try to use kernel mode NEON.

Quoted 7 lines

Due to that, kvm_arch_vcpu_put() must leave the L1 guest's maximum VL configured in the host's ZCR_ELx, such that the guest's state can be saved.

That value is configured by fpsimd_lazy_switch_to_host(), so we just need to make sure that kvm_arch_vcpu_put() doesn't clobber it. I *think* that's fine today, but maybe that warrants a comment somewhere.

I have slapped this onto this patch:

arch/arm64/kvm/fpsimd.c +4 -1
index aca98752a6e42..3f6b1e29cd6b9 100644
@@ -117,7 +117,10 @@ void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu)
 	unsigned long flags;
 
 	/*
-	 * See comment in kvm_arch_vcpu_load_fp().
+	 * See comment in kvm_arch_vcpu_load_fp(). Note that we also rely on
+	 * the guest's max VL to have been set by fpsimd_lazy_switch_to_host()
+	 * so that any intervening kernel-mode SIMD (NEON or otherwise)
+	 * operation sees the full guest state that needs saving.
 	 */
 	if (vcpu_get_flag(vcpu, IN_NESTED_ERET) ||
 	    vcpu_get_flag(vcpu, IN_NESTED_EXCEPTION)) {
Quoted 3 lines

Other than that, this all looks good to me:

Thanks,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
author maintainer Applied
#7 lore raw
In reply to Marc Zyngier
Quoted 8 lines · Marc Zyngier

This is the second version of this short series optimising away a lot of unnecessary FPSIMD/SVE context switch with NV.

* From v1 [1]:

- New commit message on patch #2 (Mark)

[...]

Applied to next, thanks!

[1/2] KVM: arm64: nv: Track L2 to L1 exception emulation

      commit: 27ae400e6e888153ded1ad807a94a94e506dd2df
[2/2] KVM: arm64: nv: Don't save/restore FP register during a nested ERET or exception
      commit: 435c466196148ae116f616e6cda97c33281defc2

Cheers,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.