RISC-V: KVM: Add skeleton support for perf

This patch only adds barebone structure of perf implementation. Most
of the function returns zero at this point and will be implemented
fully in the future.

Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Signed-off-by: Atish Patra <atishp@rivosinc.com>
Signed-off-by: Anup Patel <anup@brainfault.org>

5 files changed, 227 insertions, 0 deletions

diff --git a/arch/riscv/include/asm/kvm_host.h b/arch/riscv/include/asm/kvm_host.h
index 93f43a3e7886..b90be9a2cc24 100644
--- a/arch/riscv/include/asm/kvm_host.h
+++ b/arch/riscv/include/asm/kvm_host.h
@@ -18,6 +18,7 @@
 #include <asm/kvm_vcpu_insn.h>
 #include <asm/kvm_vcpu_sbi.h>
 #include <asm/kvm_vcpu_timer.h>
+#include <asm/kvm_vcpu_pmu.h>
 
 #define KVM_MAX_VCPUS			1024
 
@@ -228,6 +229,9 @@ struct kvm_vcpu_arch {
 
 	/* Don't run the VCPU (blocked) */
 	bool pause;
+
+	/* Performance monitoring context */
+	struct kvm_pmu pmu_context;
 };
 
 static inline void kvm_arch_hardware_unsetup(void) {}
diff --git a/arch/riscv/include/asm/kvm_vcpu_pmu.h b/arch/riscv/include/asm/kvm_vcpu_pmu.h
new file mode 100644
index 000000000000..c4cd5cd84ef2
--- /dev/null
+++ b/arch/riscv/include/asm/kvm_vcpu_pmu.h
@@ -0,0 +1,74 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2023 Rivos Inc
+ *
+ * Authors:
+ *     Atish Patra <atishp@rivosinc.com>
+ */
+
+#ifndef __KVM_VCPU_RISCV_PMU_H
+#define __KVM_VCPU_RISCV_PMU_H
+
+#include <linux/perf/riscv_pmu.h>
+#include <asm/sbi.h>
+
+#ifdef CONFIG_RISCV_PMU_SBI
+#define RISCV_KVM_MAX_FW_CTRS	32
+#define RISCV_KVM_MAX_HW_CTRS	32
+#define RISCV_KVM_MAX_COUNTERS	(RISCV_KVM_MAX_HW_CTRS + RISCV_KVM_MAX_FW_CTRS)
+static_assert(RISCV_KVM_MAX_COUNTERS <= 64);
+
+/* Per virtual pmu counter data */
+struct kvm_pmc {
+	u8 idx;
+	struct perf_event *perf_event;
+	u64 counter_val;
+	union sbi_pmu_ctr_info cinfo;
+	/* Event monitoring status */
+	bool started;
+};
+
+/* PMU data structure per vcpu */
+struct kvm_pmu {
+	struct kvm_pmc pmc[RISCV_KVM_MAX_COUNTERS];
+	/* Number of the virtual firmware counters available */
+	int num_fw_ctrs;
+	/* Number of the virtual hardware counters available */
+	int num_hw_ctrs;
+	/* A flag to indicate that pmu initialization is done */
+	bool init_done;
+	/* Bit map of all the virtual counter used */
+	DECLARE_BITMAP(pmc_in_use, RISCV_KVM_MAX_COUNTERS);
+};
+
+#define vcpu_to_pmu(vcpu) (&(vcpu)->arch.pmu_context)
+#define pmu_to_vcpu(pmu)  (container_of((pmu), struct kvm_vcpu, arch.pmu_context))
+
+int kvm_riscv_vcpu_pmu_num_ctrs(struct kvm_vcpu *vcpu, struct kvm_vcpu_sbi_return *retdata);
+int kvm_riscv_vcpu_pmu_ctr_info(struct kvm_vcpu *vcpu, unsigned long cidx,
+				struct kvm_vcpu_sbi_return *retdata);
+int kvm_riscv_vcpu_pmu_ctr_start(struct kvm_vcpu *vcpu, unsigned long ctr_base,
+				 unsigned long ctr_mask, unsigned long flags, u64 ival,
+				 struct kvm_vcpu_sbi_return *retdata);
+int kvm_riscv_vcpu_pmu_ctr_stop(struct kvm_vcpu *vcpu, unsigned long ctr_base,
+				unsigned long ctr_mask, unsigned long flags,
+				struct kvm_vcpu_sbi_return *retdata);
+int kvm_riscv_vcpu_pmu_ctr_cfg_match(struct kvm_vcpu *vcpu, unsigned long ctr_base,
+				     unsigned long ctr_mask, unsigned long flags,
+				     unsigned long eidx, u64 evtdata,
+				     struct kvm_vcpu_sbi_return *retdata);
+int kvm_riscv_vcpu_pmu_ctr_read(struct kvm_vcpu *vcpu, unsigned long cidx,
+				struct kvm_vcpu_sbi_return *retdata);
+void kvm_riscv_vcpu_pmu_init(struct kvm_vcpu *vcpu);
+void kvm_riscv_vcpu_pmu_deinit(struct kvm_vcpu *vcpu);
+void kvm_riscv_vcpu_pmu_reset(struct kvm_vcpu *vcpu);
+
+#else
+struct kvm_pmu {
+};
+
+static inline void kvm_riscv_vcpu_pmu_init(struct kvm_vcpu *vcpu) {}
+static inline void kvm_riscv_vcpu_pmu_deinit(struct kvm_vcpu *vcpu) {}
+static inline void kvm_riscv_vcpu_pmu_reset(struct kvm_vcpu *vcpu) {}
+#endif /* CONFIG_RISCV_PMU_SBI */
+#endif /* !__KVM_VCPU_RISCV_PMU_H */
diff --git a/arch/riscv/kvm/Makefile b/arch/riscv/kvm/Makefile
index 019df9208bdd..5de1053a5ce4 100644
--- a/arch/riscv/kvm/Makefile
+++ b/arch/riscv/kvm/Makefile
@@ -25,3 +25,4 @@ kvm-y += vcpu_sbi_base.o
 kvm-y += vcpu_sbi_replace.o
 kvm-y += vcpu_sbi_hsm.o
 kvm-y += vcpu_timer.o
+kvm-$(CONFIG_RISCV_PMU_SBI) += vcpu_pmu.o
diff --git a/arch/riscv/kvm/vcpu.c b/arch/riscv/kvm/vcpu.c
index 7c08567097f0..7d010b0be54e 100644
--- a/arch/riscv/kvm/vcpu.c
+++ b/arch/riscv/kvm/vcpu.c
@@ -138,6 +138,8 @@ static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu)
 	WRITE_ONCE(vcpu->arch.irqs_pending, 0);
 	WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
 
+	kvm_riscv_vcpu_pmu_reset(vcpu);
+
 	vcpu->arch.hfence_head = 0;
 	vcpu->arch.hfence_tail = 0;
 	memset(vcpu->arch.hfence_queue, 0, sizeof(vcpu->arch.hfence_queue));
@@ -194,6 +196,9 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
 	/* Setup VCPU timer */
 	kvm_riscv_vcpu_timer_init(vcpu);
 
+	/* setup performance monitoring */
+	kvm_riscv_vcpu_pmu_init(vcpu);
+
 	/* Reset VCPU */
 	kvm_riscv_reset_vcpu(vcpu);
 
@@ -216,6 +221,8 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 	/* Cleanup VCPU timer */
 	kvm_riscv_vcpu_timer_deinit(vcpu);
 
+	kvm_riscv_vcpu_pmu_deinit(vcpu);
+
 	/* Free unused pages pre-allocated for G-stage page table mappings */
 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
 }
diff --git a/arch/riscv/kvm/vcpu_pmu.c b/arch/riscv/kvm/vcpu_pmu.c
new file mode 100644
index 000000000000..dc3c118efb57
--- /dev/null
+++ b/arch/riscv/kvm/vcpu_pmu.c
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2023 Rivos Inc
+ *
+ * Authors:
+ *     Atish Patra <atishp@rivosinc.com>
+ */
+
+#define pr_fmt(fmt)	"riscv-kvm-pmu: " fmt
+#include <linux/errno.h>
+#include <linux/err.h>
+#include <linux/kvm_host.h>
+#include <linux/perf/riscv_pmu.h>
+#include <asm/csr.h>
+#include <asm/kvm_vcpu_sbi.h>
+#include <asm/kvm_vcpu_pmu.h>
+
+#define kvm_pmu_num_counters(pmu) ((pmu)->num_hw_ctrs + (pmu)->num_fw_ctrs)
+
+int kvm_riscv_vcpu_pmu_num_ctrs(struct kvm_vcpu *vcpu,
+				struct kvm_vcpu_sbi_return *retdata)
+{
+	struct kvm_pmu *kvpmu = vcpu_to_pmu(vcpu);
+
+	retdata->out_val = kvm_pmu_num_counters(kvpmu);
+
+	return 0;
+}
+
+int kvm_riscv_vcpu_pmu_ctr_info(struct kvm_vcpu *vcpu, unsigned long cidx,
+				struct kvm_vcpu_sbi_return *retdata)
+{
+	struct kvm_pmu *kvpmu = vcpu_to_pmu(vcpu);
+
+	if (cidx > RISCV_KVM_MAX_COUNTERS || cidx == 1) {
+		retdata->err_val = SBI_ERR_INVALID_PARAM;
+		return 0;
+	}
+
+	retdata->out_val = kvpmu->pmc[cidx].cinfo.value;
+
+	return 0;
+}
+
+int kvm_riscv_vcpu_pmu_ctr_start(struct kvm_vcpu *vcpu, unsigned long ctr_base,
+				 unsigned long ctr_mask, unsigned long flags, u64 ival,
+				 struct kvm_vcpu_sbi_return *retdata)
+{
+	/* TODO */
+	return 0;
+}
+
+int kvm_riscv_vcpu_pmu_ctr_stop(struct kvm_vcpu *vcpu, unsigned long ctr_base,
+				unsigned long ctr_mask, unsigned long flags,
+				struct kvm_vcpu_sbi_return *retdata)
+{
+	/* TODO */
+	return 0;
+}
+
+int kvm_riscv_vcpu_pmu_ctr_cfg_match(struct kvm_vcpu *vcpu, unsigned long ctr_base,
+				     unsigned long ctr_mask, unsigned long flags,
+				     unsigned long eidx, u64 evtdata,
+				     struct kvm_vcpu_sbi_return *retdata)
+{
+	/* TODO */
+	return 0;
+}
+
+int kvm_riscv_vcpu_pmu_ctr_read(struct kvm_vcpu *vcpu, unsigned long cidx,
+				struct kvm_vcpu_sbi_return *retdata)
+{
+	/* TODO */
+	return 0;
+}
+
+void kvm_riscv_vcpu_pmu_init(struct kvm_vcpu *vcpu)
+{
+	int i = 0, ret, num_hw_ctrs = 0, hpm_width = 0;
+	struct kvm_pmu *kvpmu = vcpu_to_pmu(vcpu);
+	struct kvm_pmc *pmc;
+
+	ret = riscv_pmu_get_hpm_info(&hpm_width, &num_hw_ctrs);
+	if (ret < 0 || !hpm_width || !num_hw_ctrs)
+		return;
+
+	/*
+	 * Increase the number of hardware counters to offset the time counter.
+	 */
+	kvpmu->num_hw_ctrs = num_hw_ctrs + 1;
+	kvpmu->num_fw_ctrs = SBI_PMU_FW_MAX;
+
+	if (kvpmu->num_hw_ctrs > RISCV_KVM_MAX_HW_CTRS) {
+		pr_warn_once("Limiting the hardware counters to 32 as specified by the ISA");
+		kvpmu->num_hw_ctrs = RISCV_KVM_MAX_HW_CTRS;
+	}
+
+	/*
+	 * There is no correlation between the logical hardware counter and virtual counters.
+	 * However, we need to encode a hpmcounter CSR in the counter info field so that
+	 * KVM can trap n emulate the read. This works well in the migration use case as
+	 * KVM doesn't care if the actual hpmcounter is available in the hardware or not.
+	 */
+	for (i = 0; i < kvm_pmu_num_counters(kvpmu); i++) {
+		/* TIME CSR shouldn't be read from perf interface */
+		if (i == 1)
+			continue;
+		pmc = &kvpmu->pmc[i];
+		pmc->idx = i;
+		if (i < kvpmu->num_hw_ctrs) {
+			pmc->cinfo.type = SBI_PMU_CTR_TYPE_HW;
+			if (i < 3)
+				/* CY, IR counters */
+				pmc->cinfo.width = 63;
+			else
+				pmc->cinfo.width = hpm_width;
+			/*
+			 * The CSR number doesn't have any relation with the logical
+			 * hardware counters. The CSR numbers are encoded sequentially
+			 * to avoid maintaining a map between the virtual counter
+			 * and CSR number.
+			 */
+			pmc->cinfo.csr = CSR_CYCLE + i;
+		} else {
+			pmc->cinfo.type = SBI_PMU_CTR_TYPE_FW;
+			pmc->cinfo.width = BITS_PER_LONG - 1;
+		}
+	}
+
+	kvpmu->init_done = true;
+}
+
+void kvm_riscv_vcpu_pmu_deinit(struct kvm_vcpu *vcpu)
+{
+	/* TODO */
+}
+
+void kvm_riscv_vcpu_pmu_reset(struct kvm_vcpu *vcpu)
+{
+	kvm_riscv_vcpu_pmu_deinit(vcpu);
+}