kvm: Support for querying fd-based stats

Add support for querying fd-based KVM stats - as introduced by Linux kernel
commit:

cb082bfab59a ("KVM: stats: Add fd-based API to read binary stats data")

This allows the user to analyze the behavior of the VM without access
to debugfs.

Signed-off-by: Mark Kanda <mark.kanda@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

2 files changed, 393 insertions, 1 deletions

diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
index a4c4863f53..7cc9e33bab 100644
--- a/accel/kvm/kvm-all.c
+++ b/accel/kvm/kvm-all.c
@@ -47,6 +47,7 @@
 #include "kvm-cpus.h"
 
 #include "hw/boards.h"
+#include "monitor/stats.h"
 
 /* This check must be after config-host.h is included */
 #ifdef CONFIG_EVENTFD
@@ -2310,6 +2311,9 @@ bool kvm_dirty_ring_enabled(void)
     return kvm_state->kvm_dirty_ring_size ? true : false;
 }
 
+static void query_stats_cb(StatsResultList **result, StatsTarget target, Error **errp);
+static void query_stats_schemas_cb(StatsSchemaList **result, Error **errp);
+
 static int kvm_init(MachineState *ms)
 {
     MachineClass *mc = MACHINE_GET_CLASS(ms);
@@ -2638,6 +2642,10 @@ static int kvm_init(MachineState *ms)
         }
     }
 
+    if (kvm_check_extension(kvm_state, KVM_CAP_BINARY_STATS_FD)) {
+        add_stats_callbacks(query_stats_cb, query_stats_schemas_cb);
+    }
+
     return 0;
 
 err:
@@ -3697,3 +3705,387 @@ static void kvm_type_init(void)
 }
 
 type_init(kvm_type_init);
+
+typedef struct StatsArgs {
+    union StatsResultsType {
+        StatsResultList **stats;
+        StatsSchemaList **schema;
+    } result;
+    Error **errp;
+} StatsArgs;
+
+static StatsList *add_kvmstat_entry(struct kvm_stats_desc *pdesc,
+                                    uint64_t *stats_data,
+                                    StatsList *stats_list,
+                                    Error **errp)
+{
+
+    Stats *stats;
+    uint64List *val_list = NULL;
+
+    /* Only add stats that we understand.  */
+    switch (pdesc->flags & KVM_STATS_TYPE_MASK) {
+    case KVM_STATS_TYPE_CUMULATIVE:
+    case KVM_STATS_TYPE_INSTANT:
+    case KVM_STATS_TYPE_PEAK:
+    case KVM_STATS_TYPE_LINEAR_HIST:
+    case KVM_STATS_TYPE_LOG_HIST:
+        break;
+    default:
+        return stats_list;
+    }
+
+    switch (pdesc->flags & KVM_STATS_UNIT_MASK) {
+    case KVM_STATS_UNIT_NONE:
+    case KVM_STATS_UNIT_BYTES:
+    case KVM_STATS_UNIT_CYCLES:
+    case KVM_STATS_UNIT_SECONDS:
+        break;
+    default:
+        return stats_list;
+    }
+
+    switch (pdesc->flags & KVM_STATS_BASE_MASK) {
+    case KVM_STATS_BASE_POW10:
+    case KVM_STATS_BASE_POW2:
+        break;
+    default:
+        return stats_list;
+    }
+
+    /* Alloc and populate data list */
+    stats = g_new0(Stats, 1);
+    stats->name = g_strdup(pdesc->name);
+    stats->value = g_new0(StatsValue, 1);;
+
+    if (pdesc->size == 1) {
+        stats->value->u.scalar = *stats_data;
+        stats->value->type = QTYPE_QNUM;
+    } else {
+        int i;
+        for (i = 0; i < pdesc->size; i++) {
+            QAPI_LIST_PREPEND(val_list, stats_data[i]);
+        }
+        stats->value->u.list = val_list;
+        stats->value->type = QTYPE_QLIST;
+    }
+
+    QAPI_LIST_PREPEND(stats_list, stats);
+    return stats_list;
+}
+
+static StatsSchemaValueList *add_kvmschema_entry(struct kvm_stats_desc *pdesc,
+                                                 StatsSchemaValueList *list,
+                                                 Error **errp)
+{
+    StatsSchemaValueList *schema_entry = g_new0(StatsSchemaValueList, 1);
+    schema_entry->value = g_new0(StatsSchemaValue, 1);
+
+    switch (pdesc->flags & KVM_STATS_TYPE_MASK) {
+    case KVM_STATS_TYPE_CUMULATIVE:
+        schema_entry->value->type = STATS_TYPE_CUMULATIVE;
+        break;
+    case KVM_STATS_TYPE_INSTANT:
+        schema_entry->value->type = STATS_TYPE_INSTANT;
+        break;
+    case KVM_STATS_TYPE_PEAK:
+        schema_entry->value->type = STATS_TYPE_PEAK;
+        break;
+    case KVM_STATS_TYPE_LINEAR_HIST:
+        schema_entry->value->type = STATS_TYPE_LINEAR_HISTOGRAM;
+        schema_entry->value->bucket_size = pdesc->bucket_size;
+        schema_entry->value->has_bucket_size = true;
+        break;
+    case KVM_STATS_TYPE_LOG_HIST:
+        schema_entry->value->type = STATS_TYPE_LOG2_HISTOGRAM;
+        break;
+    default:
+        goto exit;
+    }
+
+    switch (pdesc->flags & KVM_STATS_UNIT_MASK) {
+    case KVM_STATS_UNIT_NONE:
+        break;
+    case KVM_STATS_UNIT_BYTES:
+        schema_entry->value->has_unit = true;
+        schema_entry->value->unit = STATS_UNIT_BYTES;
+        break;
+    case KVM_STATS_UNIT_CYCLES:
+        schema_entry->value->has_unit = true;
+        schema_entry->value->unit = STATS_UNIT_CYCLES;
+        break;
+    case KVM_STATS_UNIT_SECONDS:
+        schema_entry->value->has_unit = true;
+        schema_entry->value->unit = STATS_UNIT_SECONDS;
+        break;
+    default:
+        goto exit;
+    }
+
+    schema_entry->value->exponent = pdesc->exponent;
+    if (pdesc->exponent) {
+        switch (pdesc->flags & KVM_STATS_BASE_MASK) {
+        case KVM_STATS_BASE_POW10:
+            schema_entry->value->has_base = true;
+            schema_entry->value->base = 10;
+            break;
+        case KVM_STATS_BASE_POW2:
+            schema_entry->value->has_base = true;
+            schema_entry->value->base = 2;
+            break;
+        default:
+            goto exit;
+        }
+    }
+
+    schema_entry->value->name = g_strdup(pdesc->name);
+    schema_entry->next = list;
+    return schema_entry;
+exit:
+    g_free(schema_entry->value);
+    g_free(schema_entry);
+    return list;
+}
+
+/* Cached stats descriptors */
+typedef struct StatsDescriptors {
+    const char *ident; /* cache key, currently the StatsTarget */
+    struct kvm_stats_desc *kvm_stats_desc;
+    struct kvm_stats_header *kvm_stats_header;
+    QTAILQ_ENTRY(StatsDescriptors) next;
+} StatsDescriptors;
+
+static QTAILQ_HEAD(, StatsDescriptors) stats_descriptors =
+    QTAILQ_HEAD_INITIALIZER(stats_descriptors);
+
+/*
+ * Return the descriptors for 'target', that either have already been read
+ * or are retrieved from 'stats_fd'.
+ */
+static StatsDescriptors *find_stats_descriptors(StatsTarget target, int stats_fd,
+                                                Error **errp)
+{
+    StatsDescriptors *descriptors;
+    const char *ident;
+    struct kvm_stats_desc *kvm_stats_desc;
+    struct kvm_stats_header *kvm_stats_header;
+    size_t size_desc;
+    ssize_t ret;
+
+    ident = StatsTarget_str(target);
+    QTAILQ_FOREACH(descriptors, &stats_descriptors, next) {
+        if (g_str_equal(descriptors->ident, ident)) {
+            return descriptors;
+        }
+    }
+
+    descriptors = g_new0(StatsDescriptors, 1);
+
+    /* Read stats header */
+    kvm_stats_header = g_malloc(sizeof(*kvm_stats_header));
+    ret = read(stats_fd, kvm_stats_header, sizeof(*kvm_stats_header));
+    if (ret != sizeof(*kvm_stats_header)) {
+        error_setg(errp, "KVM stats: failed to read stats header: "
+                   "expected %zu actual %zu",
+                   sizeof(*kvm_stats_header), ret);
+        return NULL;
+    }
+    size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size;
+
+    /* Read stats descriptors */
+    kvm_stats_desc = g_malloc0_n(kvm_stats_header->num_desc, size_desc);
+    ret = pread(stats_fd, kvm_stats_desc,
+                size_desc * kvm_stats_header->num_desc,
+                kvm_stats_header->desc_offset);
+
+    if (ret != size_desc * kvm_stats_header->num_desc) {
+        error_setg(errp, "KVM stats: failed to read stats descriptors: "
+                   "expected %zu actual %zu",
+                   size_desc * kvm_stats_header->num_desc, ret);
+        g_free(descriptors);
+        g_free(kvm_stats_desc);
+        return NULL;
+    }
+    descriptors->kvm_stats_header = kvm_stats_header;
+    descriptors->kvm_stats_desc = kvm_stats_desc;
+    descriptors->ident = ident;
+    QTAILQ_INSERT_TAIL(&stats_descriptors, descriptors, next);
+    return descriptors;
+}
+
+static void query_stats(StatsResultList **result, StatsTarget target,
+                        int stats_fd, Error **errp)
+{
+    struct kvm_stats_desc *kvm_stats_desc;
+    struct kvm_stats_header *kvm_stats_header;
+    StatsDescriptors *descriptors;
+    g_autofree uint64_t *stats_data = NULL;
+    struct kvm_stats_desc *pdesc;
+    StatsList *stats_list = NULL;
+    size_t size_desc, size_data = 0;
+    ssize_t ret;
+    int i;
+
+    descriptors = find_stats_descriptors(target, stats_fd, errp);
+    if (!descriptors) {
+        return;
+    }
+
+    kvm_stats_header = descriptors->kvm_stats_header;
+    kvm_stats_desc = descriptors->kvm_stats_desc;
+    size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size;
+
+    /* Tally the total data size; read schema data */
+    for (i = 0; i < kvm_stats_header->num_desc; ++i) {
+        pdesc = (void *)kvm_stats_desc + i * size_desc;
+        size_data += pdesc->size * sizeof(*stats_data);
+    }
+
+    stats_data = g_malloc0(size_data);
+    ret = pread(stats_fd, stats_data, size_data, kvm_stats_header->data_offset);
+
+    if (ret != size_data) {
+        error_setg(errp, "KVM stats: failed to read data: "
+                   "expected %zu actual %zu", size_data, ret);
+        return;
+    }
+
+    for (i = 0; i < kvm_stats_header->num_desc; ++i) {
+        uint64_t *stats;
+        pdesc = (void *)kvm_stats_desc + i * size_desc;
+
+        /* Add entry to the list */
+        stats = (void *)stats_data + pdesc->offset;
+        stats_list = add_kvmstat_entry(pdesc, stats, stats_list, errp);
+    }
+
+    if (!stats_list) {
+        return;
+    }
+
+    switch (target) {
+    case STATS_TARGET_VM:
+        add_stats_entry(result, STATS_PROVIDER_KVM, NULL, stats_list);
+        break;
+    case STATS_TARGET_VCPU:
+        add_stats_entry(result, STATS_PROVIDER_KVM,
+                        current_cpu->parent_obj.canonical_path,
+                        stats_list);
+        break;
+    default:
+        break;
+    }
+}
+
+static void query_stats_schema(StatsSchemaList **result, StatsTarget target,
+                               int stats_fd, Error **errp)
+{
+    struct kvm_stats_desc *kvm_stats_desc;
+    struct kvm_stats_header *kvm_stats_header;
+    StatsDescriptors *descriptors;
+    struct kvm_stats_desc *pdesc;
+    StatsSchemaValueList *stats_list = NULL;
+    size_t size_desc;
+    int i;
+
+    descriptors = find_stats_descriptors(target, stats_fd, errp);
+    if (!descriptors) {
+        return;
+    }
+
+    kvm_stats_header = descriptors->kvm_stats_header;
+    kvm_stats_desc = descriptors->kvm_stats_desc;
+    size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size;
+
+    /* Tally the total data size; read schema data */
+    for (i = 0; i < kvm_stats_header->num_desc; ++i) {
+        pdesc = (void *)kvm_stats_desc + i * size_desc;
+        stats_list = add_kvmschema_entry(pdesc, stats_list, errp);
+    }
+
+    add_stats_schema(result, STATS_PROVIDER_KVM, target, stats_list);
+}
+
+static void query_stats_vcpu(CPUState *cpu, run_on_cpu_data data)
+{
+    StatsArgs *kvm_stats_args = (StatsArgs *) data.host_ptr;
+    int stats_fd = kvm_vcpu_ioctl(cpu, KVM_GET_STATS_FD, NULL);
+    Error *local_err = NULL;
+
+    if (stats_fd == -1) {
+        error_setg_errno(&local_err, errno, "KVM stats: ioctl failed");
+        error_propagate(kvm_stats_args->errp, local_err);
+        return;
+    }
+    query_stats(kvm_stats_args->result.stats, STATS_TARGET_VCPU, stats_fd,
+                kvm_stats_args->errp);
+    close(stats_fd);
+}
+
+static void query_stats_schema_vcpu(CPUState *cpu, run_on_cpu_data data)
+{
+    StatsArgs *kvm_stats_args = (StatsArgs *) data.host_ptr;
+    int stats_fd = kvm_vcpu_ioctl(cpu, KVM_GET_STATS_FD, NULL);
+    Error *local_err = NULL;
+
+    if (stats_fd == -1) {
+        error_setg_errno(&local_err, errno, "KVM stats: ioctl failed");
+        error_propagate(kvm_stats_args->errp, local_err);
+        return;
+    }
+    query_stats_schema(kvm_stats_args->result.schema, STATS_TARGET_VCPU, stats_fd,
+                       kvm_stats_args->errp);
+    close(stats_fd);
+}
+
+static void query_stats_cb(StatsResultList **result, StatsTarget target, Error **errp)
+{
+    KVMState *s = kvm_state;
+    CPUState *cpu;
+    int stats_fd;
+
+    switch (target) {
+    case STATS_TARGET_VM:
+    {
+        stats_fd = kvm_vm_ioctl(s, KVM_GET_STATS_FD, NULL);
+        if (stats_fd == -1) {
+            error_setg_errno(errp, errno, "KVM stats: ioctl failed");
+            return;
+        }
+        query_stats(result, target, stats_fd, errp);
+        close(stats_fd);
+        break;
+    }
+    case STATS_TARGET_VCPU:
+    {
+        StatsArgs stats_args;
+        stats_args.result.stats = result;
+        stats_args.errp = errp;
+        CPU_FOREACH(cpu) {
+            run_on_cpu(cpu, query_stats_vcpu, RUN_ON_CPU_HOST_PTR(&stats_args));
+        }
+        break;
+    }
+    default:
+        break;
+    }
+}
+
+void query_stats_schemas_cb(StatsSchemaList **result, Error **errp)
+{
+    StatsArgs stats_args;
+    KVMState *s = kvm_state;
+    int stats_fd;
+
+    stats_fd = kvm_vm_ioctl(s, KVM_GET_STATS_FD, NULL);
+    if (stats_fd == -1) {
+        error_setg_errno(errp, errno, "KVM stats: ioctl failed");
+        return;
+    }
+    query_stats_schema(result, STATS_TARGET_VM, stats_fd, errp);
+    close(stats_fd);
+
+    stats_args.result.schema = result;
+    stats_args.errp = errp;
+    run_on_cpu(first_cpu, query_stats_schema_vcpu, RUN_ON_CPU_HOST_PTR(&stats_args));
+}
diff --git a/qapi/stats.json b/qapi/stats.json
index ada0fbf26f..df7c4d886c 100644
--- a/qapi/stats.json
+++ b/qapi/stats.json
@@ -52,7 +52,7 @@
 # Since: 7.1
 ##
 { 'enum': 'StatsProvider',
-  'data': [ ] }
+  'data': [ 'kvm' ] }
 
 ##
 # @StatsTarget: