Re: [PATCH v9 25/30] KVM: arm64: Sync boot clock with the nVHE/pKVM hyp

[Marc Zyngier <maz@kernel.org>]

On Tue, 02 Dec 2025 09:36:18 +0000,
Vincent Donnefort <vdonnefort@google.com> wrote:
> 
> Configure the hypervisor tracing clock with the kernel boot clock. For
> tracing purposes, the boot clock is interesting: it doesn't stop on
> suspend. However, it is corrected on a regular basis, which implies the
> need to re-evaluate it every once in a while.
> 
> Cc: John Stultz <jstultz@google.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Stephen Boyd <sboyd@kernel.org>
> Cc: Christopher S. Hall <christopher.s.hall@intel.com>
> Cc: Richard Cochran <richardcochran@gmail.com>
> Signed-off-by: Vincent Donnefort <vdonnefort@google.com>
> 
> diff --git a/arch/arm64/include/asm/kvm_asm.h b/arch/arm64/include/asm/kvm_asm.h
> index f83650a7aad9..375607c67285 100644
> --- a/arch/arm64/include/asm/kvm_asm.h
> +++ b/arch/arm64/include/asm/kvm_asm.h
> @@ -93,6 +93,7 @@ enum __kvm_host_smccc_func {
>  	__KVM_HOST_SMCCC_FUNC___tracing_unload,
>  	__KVM_HOST_SMCCC_FUNC___tracing_enable,
>  	__KVM_HOST_SMCCC_FUNC___tracing_swap_reader,
> +	__KVM_HOST_SMCCC_FUNC___tracing_update_clock,
>  };
>  
>  #define DECLARE_KVM_VHE_SYM(sym)	extern char sym[]
> diff --git a/arch/arm64/kvm/hyp/include/nvhe/trace.h b/arch/arm64/kvm/hyp/include/nvhe/trace.h
> index 7da8788ce527..fd641e1b1c23 100644
> --- a/arch/arm64/kvm/hyp/include/nvhe/trace.h
> +++ b/arch/arm64/kvm/hyp/include/nvhe/trace.h
> @@ -11,6 +11,7 @@ int __tracing_load(unsigned long desc_va, size_t desc_size);
>  void __tracing_unload(void);
>  int __tracing_enable(bool enable);
>  int __tracing_swap_reader(unsigned int cpu);
> +void __tracing_update_clock(u32 mult, u32 shift, u64 epoch_ns, u64 epoch_cyc);
>  #else
>  static inline void *tracing_reserve_entry(unsigned long length) { return NULL; }
>  static inline void tracing_commit_entry(void) { }
> @@ -19,5 +20,6 @@ static inline int __tracing_load(unsigned long desc_va, size_t desc_size) { retu
>  static inline void __tracing_unload(void) { }
>  static inline int __tracing_enable(bool enable) { return -ENODEV; }
>  static inline int __tracing_swap_reader(unsigned int cpu) { return -ENODEV; }
> +static inline void __tracing_update_clock(u32 mult, u32 shift, u64 epoch_ns, u64 epoch_cyc) { }
>  #endif
>  #endif
> diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
> index 8b78b29c2069..45b8f70828de 100644
> --- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
> +++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
> @@ -613,6 +613,18 @@ static void handle___tracing_swap_reader(struct kvm_cpu_context *host_ctxt)
>  	cpu_reg(host_ctxt, 1) = __tracing_swap_reader(cpu);
>  }
>  
> +static void handle___tracing_update_clock(struct kvm_cpu_context *host_ctxt)
> +{
> +	DECLARE_REG(u32, mult, host_ctxt, 1);
> +	DECLARE_REG(u32, shift, host_ctxt, 2);
> +	DECLARE_REG(u64, epoch_ns, host_ctxt, 3);
> +	DECLARE_REG(u64, epoch_cyc, host_ctxt, 4);
> +
> +	__tracing_update_clock(mult, shift, epoch_ns, epoch_cyc);
> +
> +	cpu_reg(host_ctxt, 1) = 0;

What's the purpose of setting X1 to 0? This is a call returning void,
so I don't immediately see the need for this.

> +}
> +
>  typedef void (*hcall_t)(struct kvm_cpu_context *);
>  
>  #define HANDLE_FUNC(x)	[__KVM_HOST_SMCCC_FUNC_##x] = (hcall_t)handle_##x
> @@ -658,6 +670,7 @@ static const hcall_t host_hcall[] = {
>  	HANDLE_FUNC(__tracing_unload),
>  	HANDLE_FUNC(__tracing_enable),
>  	HANDLE_FUNC(__tracing_swap_reader),
> +	HANDLE_FUNC(__tracing_update_clock),
>  };
>  
>  static void handle_host_hcall(struct kvm_cpu_context *host_ctxt)
> diff --git a/arch/arm64/kvm/hyp/nvhe/trace.c b/arch/arm64/kvm/hyp/nvhe/trace.c
> index df9d66fcb3c9..97e9f6c1a52c 100644
> --- a/arch/arm64/kvm/hyp/nvhe/trace.c
> +++ b/arch/arm64/kvm/hyp/nvhe/trace.c
> @@ -271,3 +271,19 @@ int __tracing_swap_reader(unsigned int cpu)
>  
>  	return ret;
>  }
> +
> +void __tracing_update_clock(u32 mult, u32 shift, u64 epoch_ns, u64 epoch_cyc)
> +{
> +	int cpu;
> +
> +	/* After this loop, all CPUs are observing the new bank... */
> +	for (cpu = 0; cpu < hyp_nr_cpus; cpu++) {
> +		struct simple_rb_per_cpu *simple_rb = per_cpu_ptr(trace_buffer.simple_rbs, cpu);
> +
> +		while (READ_ONCE(simple_rb->status) == SIMPLE_RB_WRITING)
> +			;
> +	}
> +
> +	/* ...we can now override the old one and swap. */
> +	trace_clock_update(mult, shift, epoch_ns, epoch_cyc);
> +}
> diff --git a/arch/arm64/kvm/hyp_trace.c b/arch/arm64/kvm/hyp_trace.c
> index 2866effe28ec..1e5fc27f0e9d 100644
> --- a/arch/arm64/kvm/hyp_trace.c
> +++ b/arch/arm64/kvm/hyp_trace.c
> @@ -4,15 +4,133 @@
>   * Author: Vincent Donnefort <vdonnefort@google.com>
>   */
>  
> +#include <linux/cpumask.h>
>  #include <linux/trace_remote.h>
> +#include <linux/tracefs.h>
>  #include <linux/simple_ring_buffer.h>
>  
> +#include <asm/arch_timer.h>
>  #include <asm/kvm_host.h>
>  #include <asm/kvm_hyptrace.h>
>  #include <asm/kvm_mmu.h>
>  
>  #include "hyp_trace.h"
>  
> +/* Same 10min used by clocksource when width is more than 32-bits */
> +#define CLOCK_MAX_CONVERSION_S	600
> +/*
> + * Time to give for the clock init. Long enough to get a good mult/shift
> + * estimation. Short enough to not delay the tracing start too much.
> + */
> +#define CLOCK_INIT_MS		100
> +/*
> + * Time between clock checks. Must be small enough to catch clock deviation when
> + * it is still tiny.
> + */
> +#define CLOCK_UPDATE_MS		500

If these definitions are common, can't we reuse the existing ones?
Specially given that this isn't EL2 code.

Thanks,

	M.

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