[PATCH v3 3/4] KVM: arm64: Pass pmu events to hyp via vcpu
Fuad Tabba
tabba at google.com
Mon May 9 23:43:45 PDT 2022
Hi Oliver,
On Mon, May 9, 2022 at 9:21 PM Oliver Upton <oupton at google.com> wrote:
>
> On Mon, May 09, 2022 at 12:03:29PM +0000, Fuad Tabba wrote:
> > Instead of the host accessing hyp data directly, pass the pmu
> > events of the current cpu to hyp via the vcpu.
> >
> > This adds 64 bits (in two fields) to the vcpu that need to be
> > synced before every vcpu run in nvhe and protected modes.
> > However, it isolates the hypervisor from the host, which allows
> > us to use pmu in protected mode in a subsequent patch.
> >
> > No visible side effects in behavior intended.
> >
> > Signed-off-by: Fuad Tabba <tabba at google.com>
> > ---
> > arch/arm64/include/asm/kvm_host.h | 8 ++------
> > arch/arm64/kvm/arm.c | 15 +++++++++++++++
> > arch/arm64/kvm/hyp/nvhe/switch.c | 20 ++++++--------------
> > arch/arm64/kvm/pmu.c | 12 ++++--------
> > include/kvm/arm_pmu.h | 6 ++++++
> > 5 files changed, 33 insertions(+), 28 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> > index dfd360404dd8..90476e713643 100644
> > --- a/arch/arm64/include/asm/kvm_host.h
> > +++ b/arch/arm64/include/asm/kvm_host.h
> > @@ -273,14 +273,8 @@ struct kvm_cpu_context {
> > struct kvm_vcpu *__hyp_running_vcpu;
> > };
> >
> > -struct kvm_pmu_events {
> > - u32 events_host;
> > - u32 events_guest;
> > -};
> > -
>
> Looks like you're moving this to arm_pmu.h as well. Probably a better
> home for it, but unclear why it is done in this patch.
Like you said, I thought it was a better home, and it's not needed
here anymore. I could maybe move it to the repacking patch and make it
as a general "cleanup" patch, if you think that would be clearer.
>
> > struct kvm_host_data {
> > struct kvm_cpu_context host_ctxt;
> > - struct kvm_pmu_events pmu_events;
> > };
> >
>
> Are we going to need this struct any more since it now has a single
> member?
I thought about removing it, but it would cause a bit of code churn.
That said, I could remove it in a new patch that I have as the last
one, and leave it to the maintainer to decide whether to take it.
>
> > struct kvm_host_psci_config {
> > @@ -763,6 +757,7 @@ void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
> > struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
> >
> > DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
> > +DECLARE_PER_CPU(struct kvm_pmu_events, kvm_pmu_events);
>
> Why do you need this declaration? I don't see the percpu data being
> accessed outside of pmu.c.
You're right. At one previous iteration I did need it, which is why
it's here, but not anymore. Will remove it.
> > static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
> > {
> > @@ -821,6 +816,7 @@ void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
> > void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
> > void kvm_clr_pmu_events(u32 clr);
> >
> > +struct kvm_pmu_events *kvm_get_pmu_events(void);
> > void kvm_vcpu_pmu_restore_guest(struct kvm_vcpu *vcpu);
> > void kvm_vcpu_pmu_restore_host(struct kvm_vcpu *vcpu);
> > #else
> > diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> > index 2adb5832a756..86bcdb2a23a8 100644
> > --- a/arch/arm64/kvm/arm.c
> > +++ b/arch/arm64/kvm/arm.c
> > @@ -817,6 +817,19 @@ static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
> > return ret;
> > }
> >
> > +/*
> > + * Updates the vcpu's view of the pmu events for this cpu.
> > + * Must be called before every vcpu run after disabling interrupts, to ensure
> > + * that an interrupt cannot fire and update the structure.
> > + */
> > +static void kvm_pmu_update_vcpu_events(struct kvm_vcpu *vcpu)
> > +{
> > + if (has_vhe() || !kvm_vcpu_has_pmu(vcpu))
> > + return;
> > +
> > + vcpu->arch.pmu.events = *kvm_get_pmu_events();
> > +}
> > +
> > /**
> > * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
> > * @vcpu: The VCPU pointer
> > @@ -882,6 +895,8 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
> >
> > kvm_vgic_flush_hwstate(vcpu);
> >
> > + kvm_pmu_update_vcpu_events(vcpu);
> > +
> > /*
> > * Ensure we set mode to IN_GUEST_MODE after we disable
> > * interrupts and before the final VCPU requests check.
> > diff --git a/arch/arm64/kvm/hyp/nvhe/switch.c b/arch/arm64/kvm/hyp/nvhe/switch.c
> > index 0716163313d6..c61120ec8d1a 100644
> > --- a/arch/arm64/kvm/hyp/nvhe/switch.c
> > +++ b/arch/arm64/kvm/hyp/nvhe/switch.c
> > @@ -153,13 +153,9 @@ static void __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
> > /*
> > * Disable host events, enable guest events
> > */
> > -static bool __pmu_switch_to_guest(struct kvm_cpu_context *host_ctxt)
> > +static bool __pmu_switch_to_guest(struct kvm_vcpu *vcpu)
> > {
> > - struct kvm_host_data *host;
> > - struct kvm_pmu_events *pmu;
> > -
> > - host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
> > - pmu = &host->pmu_events;
> > + struct kvm_pmu_events *pmu = &vcpu->arch.pmu.events;
> >
> > if (pmu->events_host)
> > write_sysreg(pmu->events_host, pmcntenclr_el0);
> > @@ -173,13 +169,9 @@ static bool __pmu_switch_to_guest(struct kvm_cpu_context *host_ctxt)
> > /*
> > * Disable guest events, enable host events
> > */
> > -static void __pmu_switch_to_host(struct kvm_cpu_context *host_ctxt)
> > +static void __pmu_switch_to_host(struct kvm_vcpu *vcpu)
> > {
> > - struct kvm_host_data *host;
> > - struct kvm_pmu_events *pmu;
> > -
> > - host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
> > - pmu = &host->pmu_events;
> > + struct kvm_pmu_events *pmu = &vcpu->arch.pmu.events;
> >
> > if (pmu->events_guest)
> > write_sysreg(pmu->events_guest, pmcntenclr_el0);
> > @@ -304,7 +296,7 @@ int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
> > host_ctxt->__hyp_running_vcpu = vcpu;
> > guest_ctxt = &vcpu->arch.ctxt;
> >
> > - pmu_switch_needed = __pmu_switch_to_guest(host_ctxt);
> > + pmu_switch_needed = __pmu_switch_to_guest(vcpu);
> >
> > __sysreg_save_state_nvhe(host_ctxt);
> > /*
> > @@ -366,7 +358,7 @@ int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
> > __debug_restore_host_buffers_nvhe(vcpu);
> >
> > if (pmu_switch_needed)
> > - __pmu_switch_to_host(host_ctxt);
> > + __pmu_switch_to_host(vcpu);
> >
> > /* Returning to host will clear PSR.I, remask PMR if needed */
> > if (system_uses_irq_prio_masking())
> > diff --git a/arch/arm64/kvm/pmu.c b/arch/arm64/kvm/pmu.c
> > index 4bd38ff34221..c19bf6e4969e 100644
> > --- a/arch/arm64/kvm/pmu.c
> > +++ b/arch/arm64/kvm/pmu.c
> > @@ -5,7 +5,8 @@
> > */
> > #include <linux/kvm_host.h>
> > #include <linux/perf_event.h>
> > -#include <asm/kvm_hyp.h>
> > +
> > +DEFINE_PER_CPU(struct kvm_pmu_events, kvm_pmu_events);
> >
> > /*
> > * Given the perf event attributes and system type, determine
> > @@ -25,14 +26,9 @@ static bool kvm_pmu_switch_needed(struct perf_event_attr *attr)
> > return (attr->exclude_host != attr->exclude_guest);
> > }
> >
> > -static struct kvm_pmu_events *kvm_get_pmu_events(void)
> > +struct kvm_pmu_events *kvm_get_pmu_events(void)
>
> Why not make this function visible in patch 1? It seems benign even
> though there are no other users at that moment outside of the
> compilation unit.
Because I wanted every patch to be self-standing, but no preference
really. I can change that to reduce code churn.
Cheers,
/fuad
>
> --
> Thanks,
> Oliver
More information about the linux-arm-kernel
mailing list