[RFC,06/11] KVM: arm64: Add clock support in the nVHE hyp

Commit Message

Vincent Donnefort Aug. 5, 2024, 5:32 p.m. UTC

By default, the arm64 host kernel is using the arch timer as a source
for sched_clock. Conveniently, EL2 has access to that same counter,
allowing to generate clock values that are synchronized.

The clock needs nonetheless to be setup with the same slope values as
the kernel. Introducing at the same time trace_clock() which is expected
to be later configured by the hypervisor tracing.

Signed-off-by: Vincent Donnefort <[e-mail]>

Patch

diff --git a/arch/arm64/include/asm/kvm_hyp.h b/arch/arm64/include/asm/kvm_hyp.h
index c838309e4ec4..9105aba0ad00 100644
--- a/arch/arm64/include/asm/kvm_hyp.h
+++ b/arch/arm64/include/asm/kvm_hyp.h
@@ -145,4 +145,10 @@  extern unsigned long kvm_nvhe_sym(__icache_flags);
 extern unsigned int kvm_nvhe_sym(kvm_arm_vmid_bits);
 extern unsigned int kvm_nvhe_sym(kvm_host_sve_max_vl);
 
+struct kvm_nvhe_clock_data {
+	u32 mult;
+	u32 shift;
+	u64 epoch_ns;
+	u64 epoch_cyc;
+};
 #endif /* __ARM64_KVM_HYP_H__ */
diff --git a/arch/arm64/kvm/hyp/include/nvhe/clock.h b/arch/arm64/kvm/hyp/include/nvhe/clock.h
new file mode 100644
index 000000000000..7e5c2d2b1886
--- /dev/null
+++ b/arch/arm64/kvm/hyp/include/nvhe/clock.h
@@ -0,0 +1,15 @@ 
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __ARM64_KVM_HYP_NVHE_CLOCK_H
+#define __ARM64_KVM_HYP_NVHE_CLOCK_H
+#include <linux/types.h>
+
+#include <asm/kvm_hyp.h>
+
+#ifdef CONFIG_TRACING
+void trace_clock_update(struct kvm_nvhe_clock_data *data);
+u64 trace_clock(void);
+#else
+static inline void trace_clock_update(struct kvm_nvhe_clock_data *data) { }
+static inline u64 trace_clock(void) { return 0; }
+#endif
+#endif
diff --git a/arch/arm64/kvm/hyp/nvhe/Makefile b/arch/arm64/kvm/hyp/nvhe/Makefile
index 782b34b004be..fde8ac1abe7b 100644
--- a/arch/arm64/kvm/hyp/nvhe/Makefile
+++ b/arch/arm64/kvm/hyp/nvhe/Makefile
@@ -26,6 +26,7 @@  hyp-obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o host.o
 hyp-obj-y += ../vgic-v3-sr.o ../aarch32.o ../vgic-v2-cpuif-proxy.o ../entry.o \
 	 ../fpsimd.o ../hyp-entry.o ../exception.o ../pgtable.o
 hyp-obj-$(CONFIG_LIST_HARDENED) += list_debug.o
+hyp-obj-$(CONFIG_TRACING) += clock.o
 hyp-obj-y += $(lib-objs)
 
 ##
diff --git a/arch/arm64/kvm/hyp/nvhe/clock.c b/arch/arm64/kvm/hyp/nvhe/clock.c
new file mode 100644
index 000000000000..4ff87e86787c
--- /dev/null
+++ b/arch/arm64/kvm/hyp/nvhe/clock.c
@@ -0,0 +1,42 @@ 
+// SPDX-License-Identifier: GPL-2.0
+#include <nvhe/clock.h>
+
+#include <asm/arch_timer.h>
+#include <asm/div64.h>
+
+static struct kvm_nvhe_clock_data trace_clock_data;
+
+/*
+ * Update without any locks! This is fine because tracing, the sole user of this
+ * clock is ordering the memory and protects from races between read and
+ * updates.
+ */
+void trace_clock_update(struct kvm_nvhe_clock_data *data)
+{
+	trace_clock_data.mult = data->mult;
+	trace_clock_data.shift = data->shift;
+	trace_clock_data.epoch_ns = data->epoch_ns;
+	trace_clock_data.epoch_cyc = data->epoch_cyc;
+}
+
+/*
+ * This clock is relying on host provided slope and epoch values to return
+ * something synchronized with the host. The downside is we can't trust the
+ * output which must not be used for anything else than debugging.
+ */
+u64 trace_clock(void)
+{
+	u64 cyc = __arch_counter_get_cntpct() - trace_clock_data.epoch_cyc;
+	__uint128_t ns;
+
+	/*
+	 * The host kernel can avoid the 64-bits overflow of the multiplication
+	 * by updating the epoch value with a timer (see
+	 * kernel/time/clocksource.c). The hypervisor doesn't have that option,
+	 * so let's do a more costly 128-bits mult here.
+	 */
+	ns = (__uint128_t)cyc * trace_clock_data.mult;
+	ns >>= trace_clock_data.shift;
+
+	return (u64)ns + trace_clock_data.epoch_ns;
+}