Re: [PATCH v3 4/6] arm64: KVM: Add interface to set guest value for TRFCR register

[Marc Zyngier]

On Thu, 19 Oct 2023 17:55:02 +0100,
James Clark <james.clark@xxxxxxx> wrote:
> 
> Add an interface for the Coresight driver to use to set the value of the
> TRFCR register for the guest. This register controls the exclude
> settings for trace at different exception levels, and is used to honor
> the exclude_host and exclude_guest parameters from the Perf session.
> This will be used to later write TRFCR_EL1 on nVHE at guest switch. For
> VHE, the host trace is controlled by TRFCR_EL2 and thus we can write to
> the TRFCR_EL1 immediately. Because guest writes to the register are
> trapped, the value will persist and can't be modified.
> 
> The settings must be copied to the vCPU before each run in the same
> way that PMU events are, because the per-cpu struct isn't accessible in
> protected mode.

Then maybe we should look at a better way of sharing global data
between EL1 and EL2 instead of copying stuff ad-nauseam?

> 
> Signed-off-by: James Clark <james.clark@xxxxxxx>
> ---
>  arch/arm64/include/asm/kvm_host.h |  3 +++
>  arch/arm64/kvm/arm.c              |  1 +
>  arch/arm64/kvm/debug.c            | 26 ++++++++++++++++++++++++++
>  3 files changed, 30 insertions(+)
> 
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index 0f0bf8e641bd..e1852102550d 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -1125,6 +1125,8 @@ 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);
>  bool kvm_set_pmuserenr(u64 val);
> +void kvm_etm_set_guest_trfcr(u64 trfcr_guest);
> +void kvm_etm_update_vcpu_events(struct kvm_vcpu *vcpu);
>  #else
>  static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
>  static inline void kvm_clr_pmu_events(u32 clr) {}
> @@ -1132,6 +1134,7 @@ static inline bool kvm_set_pmuserenr(u64 val)
>  {
>  	return false;
>  }
> +static inline void kvm_etm_set_guest_trfcr(u64 trfcr_guest) {}
>  #endif
>  
>  void kvm_vcpu_load_sysregs_vhe(struct kvm_vcpu *vcpu);
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index 0f717b6a9151..e4d846f2f665 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -1015,6 +1015,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
>  		kvm_vgic_flush_hwstate(vcpu);
>  
>  		kvm_pmu_update_vcpu_events(vcpu);
> +		kvm_etm_update_vcpu_events(vcpu);
>  
>  		/*
>  		 * Ensure we set mode to IN_GUEST_MODE after we disable
> diff --git a/arch/arm64/kvm/debug.c b/arch/arm64/kvm/debug.c
> index 20cdd40b3c42..2ab41b954512 100644
> --- a/arch/arm64/kvm/debug.c
> +++ b/arch/arm64/kvm/debug.c
> @@ -23,6 +23,12 @@
>  
>  static DEFINE_PER_CPU(u64, mdcr_el2);
>  
> +/*
> + * Per CPU value for TRFCR that should be applied to any guest vcpu that may
> + * run on that core in the future.
> + */
> +static DEFINE_PER_CPU(u64, guest_trfcr);
> +
>  /**
>   * save/restore_guest_debug_regs
>   *
> @@ -356,3 +362,23 @@ void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu)
>  	vcpu_clear_flag(vcpu, DEBUG_STATE_SAVE_TRBE);
>  	vcpu_clear_flag(vcpu, DEBUG_STATE_SAVE_TRFCR);
>  }
> +
> +void kvm_etm_set_guest_trfcr(u64 trfcr_guest)
> +{
> +	if (has_vhe())
> +		write_sysreg_s(trfcr_guest, SYS_TRFCR_EL12);
> +	else
> +		*this_cpu_ptr(&guest_trfcr) = trfcr_guest;
> +}
> +EXPORT_SYMBOL_GPL(kvm_etm_set_guest_trfcr);

How does the ETM code know what guests it impacts? Don't you have some
per-process context already?

> +
> +/*
> + * Updates the vcpu's view of the etm 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.
> + */
> +void kvm_etm_update_vcpu_events(struct kvm_vcpu *vcpu)
> +{
> +	if (!has_vhe() && vcpu_get_flag(vcpu, DEBUG_STATE_SAVE_TRFCR))
> +		ctxt_sys_reg(&vcpu->arch.ctxt, TRFCR_EL1) = *this_cpu_ptr(&guest_trfcr);
> +}

Why this requirement of updating it at all times? Why can't this be
done in a more lazy way, using the flags to instruct the hypervisor
what and when to load it?

	M.

-- 
Without deviation from the norm, progress is not possible.