Configure the hypervisor tracing clock before starting tracing. For
tracing purpose, the boot clock is interesting as it doesn't stop on
suspend. However, it is corrected on a regular basis, which implies we
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>
Cc: Lakshmi Sowjanya D <lakshmi.sowjanya.d@intel.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 d549d7d491c3..96490f8c3ff2 100644
--- a/arch/arm64/include/asm/kvm_asm.h
+++ b/arch/arm64/include/asm/kvm_asm.h
@@ -79,6 +79,7 @@ enum __kvm_host_smccc_func {
__KVM_HOST_SMCCC_FUNC___pkvm_init_vm,
__KVM_HOST_SMCCC_FUNC___pkvm_init_vcpu,
__KVM_HOST_SMCCC_FUNC___pkvm_teardown_vm,
+ __KVM_HOST_SMCCC_FUNC___pkvm_update_clock_tracing,
__KVM_HOST_SMCCC_FUNC___pkvm_load_tracing,
__KVM_HOST_SMCCC_FUNC___pkvm_teardown_tracing,
__KVM_HOST_SMCCC_FUNC___pkvm_enable_tracing,
diff --git a/arch/arm64/kvm/hyp/include/nvhe/trace.h b/arch/arm64/kvm/hyp/include/nvhe/trace.h
index a7c0c73af0e5..df17683a3b12 100644
--- a/arch/arm64/kvm/hyp/include/nvhe/trace.h
+++ b/arch/arm64/kvm/hyp/include/nvhe/trace.h
@@ -16,6 +16,7 @@ struct hyp_buffer_page {
void *tracing_reserve_entry(unsigned long length);
void tracing_commit_entry(void);
+void __pkvm_update_clock_tracing(u32 mult, u32 shift, u64 epoch_ns, u64 epoch_cyc);
int __pkvm_load_tracing(unsigned long desc_va, size_t desc_size);
void __pkvm_teardown_tracing(void);
int __pkvm_enable_tracing(bool enable);
@@ -24,6 +25,8 @@ int __pkvm_swap_reader_tracing(unsigned int cpu);
static inline void *tracing_reserve_entry(unsigned long length) { return NULL; }
static inline void tracing_commit_entry(void) { }
+static inline
+void __pkvm_update_clock_tracing(u32 mult, u32 shift, u64 epoch_ns, u64 epoch_cyc) { }
static inline int __pkvm_load_tracing(unsigned long desc_va, size_t desc_size) { return -ENODEV; }
static inline void __pkvm_teardown_tracing(void) { }
static inline int __pkvm_enable_tracing(bool enable) { return -ENODEV; }
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 1fb3391e122a..7f5c3e888960 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -374,6 +374,18 @@ static void handle___pkvm_teardown_vm(struct kvm_cpu_context *host_ctxt)
cpu_reg(host_ctxt, 1) = __pkvm_teardown_vm(handle);
}
+static void handle___pkvm_update_clock_tracing(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);
+
+ __pkvm_update_clock_tracing(mult, shift, epoch_ns, epoch_cyc);
+
+ cpu_reg(host_ctxt, 1) = 0;
+}
+
static void handle___pkvm_load_tracing(struct kvm_cpu_context *host_ctxt)
{
DECLARE_REG(unsigned long, desc_hva, host_ctxt, 1);
@@ -435,6 +447,7 @@ static const hcall_t host_hcall[] = {
HANDLE_FUNC(__pkvm_init_vm),
HANDLE_FUNC(__pkvm_init_vcpu),
HANDLE_FUNC(__pkvm_teardown_vm),
+ HANDLE_FUNC(__pkvm_update_clock_tracing),
HANDLE_FUNC(__pkvm_load_tracing),
HANDLE_FUNC(__pkvm_teardown_tracing),
HANDLE_FUNC(__pkvm_enable_tracing),
diff --git a/arch/arm64/kvm/hyp/nvhe/trace.c b/arch/arm64/kvm/hyp/nvhe/trace.c
index debb3ee7dd3a..022fe2e24f82 100644
--- a/arch/arm64/kvm/hyp/nvhe/trace.c
+++ b/arch/arm64/kvm/hyp/nvhe/trace.c
@@ -468,6 +468,21 @@ static void rb_teardown_bpage_backing(void)
hyp_buffer_pages_backing.size = 0;
}
+void __pkvm_update_clock_tracing(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 hyp_rb_per_cpu *cpu_buffer = per_cpu_ptr(&trace_rb, cpu);
+
+ while (atomic_read(&cpu_buffer->status) == HYP_RB_WRITING);
+ }
+
+ /* ...we can now override the old one and swap. */
+ trace_clock_update(mult, shift, epoch_ns, epoch_cyc);
+}
+
int __pkvm_swap_reader_tracing(unsigned int cpu)
{
struct hyp_rb_per_cpu *cpu_buffer;
diff --git a/arch/arm64/kvm/hyp_trace.c b/arch/arm64/kvm/hyp_trace.c
index b9d1f96d0678..1720daeda8ae 100644
--- a/arch/arm64/kvm/hyp_trace.c
+++ b/arch/arm64/kvm/hyp_trace.c
@@ -16,10 +16,25 @@
#define RB_POLL_MS 100
+/* Same 10min used by clocksource when width is more than 32-bits */
+#define CLOCK_MAX_CONVERSION_S 600
+#define CLOCK_INIT_MS 100
+#define CLOCK_POLL_MS 500
+
#define TRACEFS_DIR "hypervisor"
#define TRACEFS_MODE_WRITE 0640
#define TRACEFS_MODE_READ 0440
+struct hyp_trace_clock {
+ u64 cycles;
+ u64 max_delta;
+ u64 boot;
+ u32 mult;
+ u32 shift;
+ struct delayed_work work;
+ struct completion ready;
+};
+
static struct hyp_trace_buffer {
struct hyp_trace_desc *desc;
struct ring_buffer_writer writer;
@@ -28,6 +43,7 @@ static struct hyp_trace_buffer {
bool tracing_on;
int nr_readers;
struct mutex lock;
+ struct hyp_trace_clock clock;
} hyp_trace_buffer = {
.lock = __MUTEX_INITIALIZER(hyp_trace_buffer.lock),
};
@@ -74,6 +90,107 @@ bpage_backing_free(struct hyp_buffer_pages_backing *bpage_backing)
free_pages_exact((void *)bpage_backing->start, bpage_backing->size);
}
+static void __hyp_clock_work(struct work_struct *work)
+{
+ struct delayed_work *dwork = to_delayed_work(work);
+ struct hyp_trace_buffer *hyp_buffer;
+ struct hyp_trace_clock *hyp_clock;
+ struct system_time_snapshot snap;
+ u64 rate, delta_cycles;
+ u64 boot, delta_boot;
+ u64 err = 0;
+
+ hyp_clock = container_of(dwork, struct hyp_trace_clock, work);
+ hyp_buffer = container_of(hyp_clock, struct hyp_trace_buffer, clock);
+
+ ktime_get_snapshot(&snap);
+ boot = ktime_to_ns(snap.boot);
+
+ delta_boot = boot - hyp_clock->boot;
+ delta_cycles = snap.cycles - hyp_clock->cycles;
+
+ /* Compare hyp clock with the kernel boot clock */
+ if (hyp_clock->mult) {
+ u64 cur = delta_cycles;
+
+ cur *= hyp_clock->mult;
+ cur >>= hyp_clock->shift;
+ cur += hyp_clock->boot;
+
+ err = abs_diff(cur, boot);
+
+ /* No deviation, only update epoch if necessary */
+ if (!err) {
+ if (delta_cycles >= hyp_clock->max_delta)
+ goto update_hyp;
+
+ goto resched;
+ }
+
+ /* Warn if the error is above tracing precision (1us) */
+ if (hyp_buffer->tracing_on && err > NSEC_PER_USEC)
+ pr_warn_ratelimited("hyp trace clock off by %lluus\n",
+ err / NSEC_PER_USEC);
+ }
+
+ if (delta_boot > U32_MAX) {
+ do_div(delta_boot, NSEC_PER_SEC);
+ rate = delta_cycles;
+ } else {
+ rate = delta_cycles * NSEC_PER_SEC;
+ }
+
+ do_div(rate, delta_boot);
+
+ clocks_calc_mult_shift(&hyp_clock->mult, &hyp_clock->shift,
+ rate, NSEC_PER_SEC, CLOCK_MAX_CONVERSION_S);
+
+update_hyp:
+ hyp_clock->max_delta = (U64_MAX / hyp_clock->mult) >> 1;
+ hyp_clock->cycles = snap.cycles;
+ hyp_clock->boot = boot;
+ kvm_call_hyp_nvhe(__pkvm_update_clock_tracing, hyp_clock->mult,
+ hyp_clock->shift, hyp_clock->boot, hyp_clock->cycles);
+ complete(&hyp_clock->ready);
+
+ pr_debug("hyp trace clock update mult=%u shift=%u max_delta=%llu err=%llu\n",
+ hyp_clock->mult, hyp_clock->shift, hyp_clock->max_delta, err);
+
+resched:
+ schedule_delayed_work(&hyp_clock->work,
+ msecs_to_jiffies(CLOCK_POLL_MS));
+}
+
+static void hyp_clock_start(struct hyp_trace_buffer *hyp_buffer)
+{
+ struct hyp_trace_clock *hyp_clock = &hyp_buffer->clock;
+ struct system_time_snapshot snap;
+
+ ktime_get_snapshot(&snap);
+
+ hyp_clock->boot = ktime_to_ns(snap.boot);
+ hyp_clock->cycles = snap.cycles;
+ hyp_clock->mult = 0;
+
+ init_completion(&hyp_clock->ready);
+ INIT_DELAYED_WORK(&hyp_clock->work, __hyp_clock_work);
+ schedule_delayed_work(&hyp_clock->work, msecs_to_jiffies(CLOCK_INIT_MS));
+}
+
+static void hyp_clock_stop(struct hyp_trace_buffer *hyp_buffer)
+{
+ struct hyp_trace_clock *hyp_clock = &hyp_buffer->clock;
+
+ cancel_delayed_work_sync(&hyp_clock->work);
+}
+
+static void hyp_clock_wait(struct hyp_trace_buffer *hyp_buffer)
+{
+ struct hyp_trace_clock *hyp_clock = &hyp_buffer->clock;
+
+ wait_for_completion(&hyp_clock->ready);
+}
+
static int __get_reader_page(int cpu)
{
return kvm_call_hyp_nvhe(__pkvm_swap_reader_tracing, cpu);
@@ -294,10 +411,14 @@ static int hyp_trace_start(void)
if (hyp_buffer->tracing_on)
goto out;
+ hyp_clock_start(hyp_buffer);
+
ret = hyp_trace_buffer_load(hyp_buffer, hyp_trace_buffer_size);
if (ret)
goto out;
+ hyp_clock_wait(hyp_buffer);
+
ret = kvm_call_hyp_nvhe(__pkvm_enable_tracing, true);
if (ret) {
hyp_trace_buffer_teardown(hyp_buffer);
@@ -307,6 +428,9 @@ static int hyp_trace_start(void)
hyp_buffer->tracing_on = true;
out:
+ if (!hyp_buffer->tracing_on)
+ hyp_clock_stop(hyp_buffer);
+
mutex_unlock(&hyp_buffer->lock);
return ret;
@@ -324,6 +448,7 @@ static void hyp_trace_stop(void)
ret = kvm_call_hyp_nvhe(__pkvm_enable_tracing, false);
if (!ret) {
+ hyp_clock_stop(hyp_buffer);
ring_buffer_poll_writer(hyp_buffer->trace_buffer,
RING_BUFFER_ALL_CPUS);
hyp_buffer->tracing_on = false;
@@ -615,6 +740,14 @@ static const struct file_operations hyp_trace_pipe_fops = {
.llseek = no_llseek,
};
+static int hyp_trace_clock_show(struct seq_file *m, void *v)
+{
+ seq_puts(m, "[boot]\n");
+
+ return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(hyp_trace_clock);
+
int hyp_trace_init_tracefs(void)
{
struct dentry *root, *per_cpu_root;
@@ -639,6 +772,9 @@ int hyp_trace_init_tracefs(void)
tracefs_create_file("trace_pipe", TRACEFS_MODE_WRITE, root,
(void *)RING_BUFFER_ALL_CPUS, &hyp_trace_pipe_fops);
+ tracefs_create_file("trace_clock", TRACEFS_MODE_READ, root, NULL,
+ &hyp_trace_clock_fops);
+
per_cpu_root = tracefs_create_dir("per_cpu", root);
if (!per_cpu_root) {
pr_err("Failed to create tracefs folder "TRACEFS_DIR"/per_cpu/\n");
--
2.46.0.598.g6f2099f65c-googOn Wed, Sep 11, 2024 at 2:31 AM Vincent Donnefort <vdonnefort@google.com> wrote:
>
> Configure the hypervisor tracing clock before starting tracing. For
> tracing purpose, the boot clock is interesting as it doesn't stop on
> suspend. However, it is corrected on a regular basis, which implies we
> 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>
> Cc: Lakshmi Sowjanya D <lakshmi.sowjanya.d@intel.com>
> Signed-off-by: Vincent Donnefort <vdonnefort@google.com>
>
...
> +static void __hyp_clock_work(struct work_struct *work)
> +{
> + struct delayed_work *dwork = to_delayed_work(work);
> + struct hyp_trace_buffer *hyp_buffer;
> + struct hyp_trace_clock *hyp_clock;
> + struct system_time_snapshot snap;
> + u64 rate, delta_cycles;
> + u64 boot, delta_boot;
> + u64 err = 0;
> +
> + hyp_clock = container_of(dwork, struct hyp_trace_clock, work);
> + hyp_buffer = container_of(hyp_clock, struct hyp_trace_buffer, clock);
> +
> + ktime_get_snapshot(&snap);
> + boot = ktime_to_ns(snap.boot);
> +
> + delta_boot = boot - hyp_clock->boot;
> + delta_cycles = snap.cycles - hyp_clock->cycles;
> +
> + /* Compare hyp clock with the kernel boot clock */
> + if (hyp_clock->mult) {
> + u64 cur = delta_cycles;
> +
> + cur *= hyp_clock->mult;
Mult overflow protection (I see you already have a max_delta value) is
probably needed here.
> + cur >>= hyp_clock->shift;
> + cur += hyp_clock->boot;
> +
> + err = abs_diff(cur, boot);
> +
> + /* No deviation, only update epoch if necessary */
> + if (!err) {
> + if (delta_cycles >= hyp_clock->max_delta)
> + goto update_hyp;
> +
> + goto resched;
> + }
> +
> + /* Warn if the error is above tracing precision (1us) */
> + if (hyp_buffer->tracing_on && err > NSEC_PER_USEC)
> + pr_warn_ratelimited("hyp trace clock off by %lluus\n",
> + err / NSEC_PER_USEC);
I'm curious in practice, does this come up often? If so, does it
converge down nicely? Have you done much disruption testing using
adjtimex?
> + }
> +
> + if (delta_boot > U32_MAX) {
> + do_div(delta_boot, NSEC_PER_SEC);
> + rate = delta_cycles;
> + } else {
> + rate = delta_cycles * NSEC_PER_SEC;
> + }
> +
> + do_div(rate, delta_boot);
> +
> + clocks_calc_mult_shift(&hyp_clock->mult, &hyp_clock->shift,
> + rate, NSEC_PER_SEC, CLOCK_MAX_CONVERSION_S);
> +
> +update_hyp:
> + hyp_clock->max_delta = (U64_MAX / hyp_clock->mult) >> 1;
> + hyp_clock->cycles = snap.cycles;
> + hyp_clock->boot = boot;
> + kvm_call_hyp_nvhe(__pkvm_update_clock_tracing, hyp_clock->mult,
> + hyp_clock->shift, hyp_clock->boot, hyp_clock->cycles);
> + complete(&hyp_clock->ready);
I'm very forgetful, so maybe it's unnecessary, but for future-you or
just other's like me, it might be worth adding some extra comments to
clarify the assumptions in these calculations.
thanks
-john
On Fri, Sep 13, 2024 at 04:21:05PM -0700, 'John Stultz' via kernel-team wrote:
> On Wed, Sep 11, 2024 at 2:31 AM Vincent Donnefort <vdonnefort@google.com> wrote:
> >
> > Configure the hypervisor tracing clock before starting tracing. For
> > tracing purpose, the boot clock is interesting as it doesn't stop on
> > suspend. However, it is corrected on a regular basis, which implies we
> > 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>
> > Cc: Lakshmi Sowjanya D <lakshmi.sowjanya.d@intel.com>
> > Signed-off-by: Vincent Donnefort <vdonnefort@google.com>
> >
> ...
> > +static void __hyp_clock_work(struct work_struct *work)
> > +{
> > + struct delayed_work *dwork = to_delayed_work(work);
> > + struct hyp_trace_buffer *hyp_buffer;
> > + struct hyp_trace_clock *hyp_clock;
> > + struct system_time_snapshot snap;
> > + u64 rate, delta_cycles;
> > + u64 boot, delta_boot;
> > + u64 err = 0;
> > +
> > + hyp_clock = container_of(dwork, struct hyp_trace_clock, work);
> > + hyp_buffer = container_of(hyp_clock, struct hyp_trace_buffer, clock);
> > +
> > + ktime_get_snapshot(&snap);
> > + boot = ktime_to_ns(snap.boot);
> > +
> > + delta_boot = boot - hyp_clock->boot;
> > + delta_cycles = snap.cycles - hyp_clock->cycles;
> > +
> > + /* Compare hyp clock with the kernel boot clock */
> > + if (hyp_clock->mult) {
> > + u64 cur = delta_cycles;
> > +
> > + cur *= hyp_clock->mult;
>
> Mult overflow protection (I see you already have a max_delta value) is
> probably needed here.
That should never happen really with the max_delta. But I could add a WARN_ON
and fallback to a 128-bits compute instead here too?
>
> > + cur >>= hyp_clock->shift;
> > + cur += hyp_clock->boot;
> > +
> > + err = abs_diff(cur, boot);
> > +
> > + /* No deviation, only update epoch if necessary */
> > + if (!err) {
> > + if (delta_cycles >= hyp_clock->max_delta)
> > + goto update_hyp;
> > +
> > + goto resched;
> > + }
> > +
> > + /* Warn if the error is above tracing precision (1us) */
> > + if (hyp_buffer->tracing_on && err > NSEC_PER_USEC)
> > + pr_warn_ratelimited("hyp trace clock off by %lluus\n",
> > + err / NSEC_PER_USEC);
>
> I'm curious in practice, does this come up often? If so, does it
> converge down nicely? Have you done much disruption testing using
> adjtimex?
So far, I haven't seen any error above ~100 ns on the machine I have tested
with, but that's a good point, I'll check how it looks when the boot clock is
less stable.
>
> > + }
> > +
> > + if (delta_boot > U32_MAX) {
> > + do_div(delta_boot, NSEC_PER_SEC);
> > + rate = delta_cycles;
> > + } else {
> > + rate = delta_cycles * NSEC_PER_SEC;
> > + }
> > +
> > + do_div(rate, delta_boot);
> > +
> > + clocks_calc_mult_shift(&hyp_clock->mult, &hyp_clock->shift,
> > + rate, NSEC_PER_SEC, CLOCK_MAX_CONVERSION_S);
> > +
> > +update_hyp:
> > + hyp_clock->max_delta = (U64_MAX / hyp_clock->mult) >> 1;
> > + hyp_clock->cycles = snap.cycles;
> > + hyp_clock->boot = boot;
> > + kvm_call_hyp_nvhe(__pkvm_update_clock_tracing, hyp_clock->mult,
> > + hyp_clock->shift, hyp_clock->boot, hyp_clock->cycles);
> > + complete(&hyp_clock->ready);
>
> I'm very forgetful, so maybe it's unnecessary, but for future-you or
> just other's like me, it might be worth adding some extra comments to
> clarify the assumptions in these calculations.
Ack.
>
>
> thanks
> -john
Thanks for your time!
--
Vincent
© 2016 - 2026 Red Hat, Inc.