Drop kvm_arch_sched_in() and instead add and use kvm_vcpu.scheduled_out
to communicate to kvm_arch_vcpu_load() that the vCPU is being scheduling
back in.
While fiddling with an idea for optimizing state management on AMD CPUs,
I wanted to skip re-saving certain host state when a vCPU is scheduled back
in, as the state (theoretically) shouldn't change for the task while it's
scheduled out. Actually doing that was annoying and unnecessarily brittle
due to having a separate API for the kvm_sched_in() case (the state save
needed to be in kvm_arch_vcpu_load() for the common path).
The other motivation for this is to avoid yet another arch hook, and more
arbitrary ordering, if there's a future need to hook kvm_sched_out() (we've
come close on the x86 side several times). E.g. kvm_arch_vcpu_put() can
simply check kvm_vcpu.scheduled_out if it needs to something specific for
the vCPU being scheduled out.
v2:
- Add scheduled_out flag instead of passing a bool to kvm_arch_vcpu_load().
[Oliver]
- Tack on patches to clean up x86's setting of l1tf_flush_l1d in
kvm_arch_sched_load() (the code looked slightly less weird when the flag
was being set by kvm_arch_sched_in()).
v1: https://lore.kernel.org/all/[email protected]
Sean Christopherson (6):
KVM: Add a flag to track if a loaded vCPU is scheduled out
KVM: VMX: Move PLE grow/shrink helpers above vmx_vcpu_load()
KVM: x86: Fold kvm_arch_sched_in() into kvm_arch_vcpu_load()
KVM: Delete the now unused kvm_arch_sched_in()
KVM: x86: Unconditionally set l1tf_flush_l1d during vCPU load
KVM: x86: Drop now-superflous setting of l1tf_flush_l1d in vcpu_run()
arch/arm64/include/asm/kvm_host.h | 1 -
arch/loongarch/include/asm/kvm_host.h | 1 -
arch/mips/include/asm/kvm_host.h | 1 -
arch/powerpc/include/asm/kvm_host.h | 1 -
arch/riscv/include/asm/kvm_host.h | 1 -
arch/s390/include/asm/kvm_host.h | 1 -
arch/x86/include/asm/kvm-x86-ops.h | 1 -
arch/x86/include/asm/kvm_host.h | 2 -
arch/x86/kvm/pmu.c | 6 +-
arch/x86/kvm/svm/svm.c | 11 +---
arch/x86/kvm/vmx/main.c | 2 -
arch/x86/kvm/vmx/vmx.c | 80 +++++++++++++--------------
arch/x86/kvm/vmx/x86_ops.h | 1 -
arch/x86/kvm/x86.c | 22 +++-----
include/linux/kvm_host.h | 3 +-
virt/kvm/kvm_main.c | 5 +-
16 files changed, 59 insertions(+), 80 deletions(-)
base-commit: 4aad0b1893a141f114ba40ed509066f3c9bc24b0
--
2.45.0.215.g3402c0e53f-goog
Now that KVM unconditionally sets l1tf_flush_l1d in kvm_arch_vcpu_load(),
drop the redundant store from vcpu_run(). The flag is cleared only when
VM-Enter is imminent, deep below vcpu_run(), i.e. barring a KVM bug, it's
impossible for l1tf_flush_l1d to be cleared between loading the vCPU and
calling vcpu_run().
Signed-off-by: Sean Christopherson <[email protected]>
---
arch/x86/kvm/vmx/vmx.c | 7 ++++---
arch/x86/kvm/x86.c | 1 -
2 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index da2f95385a12..552b6a9887a5 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -6672,9 +6672,10 @@ static noinstr void vmx_l1d_flush(struct kvm_vcpu *vcpu)
bool flush_l1d;
/*
- * Clear the per-vcpu flush bit, it gets set again
- * either from vcpu_run() or from one of the unsafe
- * VMEXIT handlers.
+ * Clear the per-vcpu flush bit, it gets set again if the vCPU
+ * is reloaded, i.e. if the vCPU is scheduled out or if KVM
+ * exits to userspace, or if KVM reaches one of the unsafe
+ * VMEXIT handlers, e.g. if KVM calls into the emulator.
*/
flush_l1d = vcpu->arch.l1tf_flush_l1d;
vcpu->arch.l1tf_flush_l1d = false;
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 60fea297f91f..86ae7392cc59 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -11264,7 +11264,6 @@ static int vcpu_run(struct kvm_vcpu *vcpu)
int r;
vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
- vcpu->arch.l1tf_flush_l1d = true;
for (;;) {
/*
--
2.45.0.215.g3402c0e53f-goog
On 22/05/2024 1:40 pm, Sean Christopherson wrote:
> Drop kvm_arch_sched_in() and instead add and use kvm_vcpu.scheduled_out
> to communicate to kvm_arch_vcpu_load() that the vCPU is being scheduling
> back in.
>
For this series,
Acked-by: Kai Huang <[email protected]>
On Tue, 21 May 2024 18:40:07 -0700, Sean Christopherson wrote:
> Drop kvm_arch_sched_in() and instead add and use kvm_vcpu.scheduled_out
> to communicate to kvm_arch_vcpu_load() that the vCPU is being scheduling
> back in.
>
> While fiddling with an idea for optimizing state management on AMD CPUs,
> I wanted to skip re-saving certain host state when a vCPU is scheduled back
> in, as the state (theoretically) shouldn't change for the task while it's
> scheduled out. Actually doing that was annoying and unnecessarily brittle
> due to having a separate API for the kvm_sched_in() case (the state save
> needed to be in kvm_arch_vcpu_load() for the common path).
>
> [...]
Applied to kvm-x86 generic, thanks!
[1/6] KVM: Add a flag to track if a loaded vCPU is scheduled out
https://github.com/kvm-x86/linux/commit/d1ae567fb8b5
[2/6] KVM: VMX: Move PLE grow/shrink helpers above vmx_vcpu_load()
https://github.com/kvm-x86/linux/commit/5d9c07febb86
[3/6] KVM: x86: Fold kvm_arch_sched_in() into kvm_arch_vcpu_load()
https://github.com/kvm-x86/linux/commit/8fbb696a8f53
[4/6] KVM: Delete the now unused kvm_arch_sched_in()
https://github.com/kvm-x86/linux/commit/2a27c4314007
[5/6] KVM: x86: Unconditionally set l1tf_flush_l1d during vCPU load
https://github.com/kvm-x86/linux/commit/ef2e18ef3750
[6/6] KVM: x86: Drop now-superflous setting of l1tf_flush_l1d in vcpu_run()
https://github.com/kvm-x86/linux/commit/3dee3b187499
--
https://github.com/kvm-x86/linux/tree/next