KVM: ppc: Implement in-kernel exit timing statistics

Existing KVM statistics are either just counters (kvm_stat) reported for
KVM generally or trace based aproaches like kvm_trace.
For KVM on powerpc we had the need to track the timings of the different exit
types. While this could be achieved parsing data created with a kvm_trace
extension this adds too much overhead (at least on embedded PowerPC) slowing
down the workloads we wanted to measure.

Therefore this patch adds a in-kernel exit timing statistic to the powerpc kvm
code. These statistic is available per vm&vcpu under the kvm debugfs directory.
As this statistic is low, but still some overhead it can be enabled via a
.config entry and should be off by default.

Since this patch touched all powerpc kvm_stat code anyway this code is now
merged and simplified together with the exit timing statistic code (still
working with exit timing disabled in .config).

Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

1 files changed, 262 insertions, 0 deletions

diff --git a/arch/powerpc/kvm/timing.c b/arch/powerpc/kvm/timing.c
new file mode 100644
index 000000000000..f42d2728a6a5
--- /dev/null
+++ b/arch/powerpc/kvm/timing.c
@@ -0,0 +1,262 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
+ *
+ * Copyright IBM Corp. 2007
+ *
+ * Authors: Hollis Blanchard <hollisb@us.ibm.com>
+ *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
+ */
+
+#include <linux/kvm_host.h>
+#include <linux/fs.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
+#include <linux/uaccess.h>
+
+#include "timing.h"
+#include <asm/time.h>
+#include <asm-generic/div64.h>
+
+void kvmppc_init_timing_stats(struct kvm_vcpu *vcpu)
+{
+        int i;
+
+        /* pause guest execution to avoid concurrent updates */
+        local_irq_disable();
+        mutex_lock(&vcpu->mutex);
+
+        vcpu->arch.last_exit_type = 0xDEAD;
+        for (i = 0; i < __NUMBER_OF_KVM_EXIT_TYPES; i++) {
+                vcpu->arch.timing_count_type[i] = 0;
+                vcpu->arch.timing_max_duration[i] = 0;
+                vcpu->arch.timing_min_duration[i] = 0xFFFFFFFF;
+                vcpu->arch.timing_sum_duration[i] = 0;
+                vcpu->arch.timing_sum_quad_duration[i] = 0;
+        }
+        vcpu->arch.timing_last_exit = 0;
+        vcpu->arch.timing_exit.tv64 = 0;
+        vcpu->arch.timing_last_enter.tv64 = 0;
+
+        mutex_unlock(&vcpu->mutex);
+        local_irq_enable();
+}
+
+static void add_exit_timing(struct kvm_vcpu *vcpu,
+                                        u64 duration, int type)
+{
+        u64 old;
+
+        do_div(duration, tb_ticks_per_usec);
+        if (unlikely(duration > 0xFFFFFFFF)) {
+                printk(KERN_ERR"%s - duration too big -> overflow"
+                        " duration %lld type %d exit #%d\n",
+                        __func__, duration, type,
+                        vcpu->arch.timing_count_type[type]);
+                return;
+        }
+
+        vcpu->arch.timing_count_type[type]++;
+
+        /* sum */
+        old = vcpu->arch.timing_sum_duration[type];
+        vcpu->arch.timing_sum_duration[type] += duration;
+        if (unlikely(old > vcpu->arch.timing_sum_duration[type])) {
+                printk(KERN_ERR"%s - wrap adding sum of durations"
+                        " old %lld new %lld type %d exit # of type %d\n",
+                        __func__, old, vcpu->arch.timing_sum_duration[type],
+                        type, vcpu->arch.timing_count_type[type]);
+        }
+
+        /* square sum */
+        old = vcpu->arch.timing_sum_quad_duration[type];
+        vcpu->arch.timing_sum_quad_duration[type] += (duration*duration);
+        if (unlikely(old > vcpu->arch.timing_sum_quad_duration[type])) {
+                printk(KERN_ERR"%s - wrap adding sum of squared durations"
+                        " old %lld new %lld type %d exit # of type %d\n",
+                        __func__, old,
+                        vcpu->arch.timing_sum_quad_duration[type],
+                        type, vcpu->arch.timing_count_type[type]);
+        }
+
+        /* set min/max */
+        if (unlikely(duration < vcpu->arch.timing_min_duration[type]))
+                vcpu->arch.timing_min_duration[type] = duration;
+        if (unlikely(duration > vcpu->arch.timing_max_duration[type]))
+                vcpu->arch.timing_max_duration[type] = duration;
+}
+
+void kvmppc_update_timing_stats(struct kvm_vcpu *vcpu)
+{
+        u64 exit = vcpu->arch.timing_last_exit;
+        u64 enter = vcpu->arch.timing_last_enter.tv64;
+
+        /* save exit time, used next exit when the reenter time is known */
+        vcpu->arch.timing_last_exit = vcpu->arch.timing_exit.tv64;
+
+        if (unlikely(vcpu->arch.last_exit_type == 0xDEAD || exit == 0))
+                return; /* skip incomplete cycle (e.g. after reset) */
+
+        /* update statistics for average and standard deviation */
+        add_exit_timing(vcpu, (enter - exit), vcpu->arch.last_exit_type);
+        /* enter -> timing_last_exit is time spent in guest - log this too */
+        add_exit_timing(vcpu, (vcpu->arch.timing_last_exit - enter),
+                        TIMEINGUEST);
+}
+
+static const char *kvm_exit_names[__NUMBER_OF_KVM_EXIT_TYPES] = {
+        [MMIO_EXITS] =                  "MMIO",
+        [DCR_EXITS] =                   "DCR",
+        [SIGNAL_EXITS] =                "SIGNAL",
+        [ITLB_REAL_MISS_EXITS] =        "ITLBREAL",
+        [ITLB_VIRT_MISS_EXITS] =        "ITLBVIRT",
+        [DTLB_REAL_MISS_EXITS] =        "DTLBREAL",
+        [DTLB_VIRT_MISS_EXITS] =        "DTLBVIRT",
+        [SYSCALL_EXITS] =               "SYSCALL",
+        [ISI_EXITS] =                   "ISI",
+        [DSI_EXITS] =                   "DSI",
+        [EMULATED_INST_EXITS] =         "EMULINST",
+        [EMULATED_MTMSRWE_EXITS] =      "EMUL_WAIT",
+        [EMULATED_WRTEE_EXITS] =        "EMUL_WRTEE",
+        [EMULATED_MTSPR_EXITS] =        "EMUL_MTSPR",
+        [EMULATED_MFSPR_EXITS] =        "EMUL_MFSPR",
+        [EMULATED_MTMSR_EXITS] =        "EMUL_MTMSR",
+        [EMULATED_MFMSR_EXITS] =        "EMUL_MFMSR",
+        [EMULATED_TLBSX_EXITS] =        "EMUL_TLBSX",
+        [EMULATED_TLBWE_EXITS] =        "EMUL_TLBWE",
+        [EMULATED_RFI_EXITS] =          "EMUL_RFI",
+        [DEC_EXITS] =                   "DEC",
+        [EXT_INTR_EXITS] =              "EXTINT",
+        [HALT_WAKEUP] =                 "HALT",
+        [USR_PR_INST] =                 "USR_PR_INST",
+        [FP_UNAVAIL] =                  "FP_UNAVAIL",
+        [DEBUG_EXITS] =                 "DEBUG",
+        [TIMEINGUEST] =                 "TIMEINGUEST"
+};
+
+static int kvmppc_exit_timing_show(struct seq_file *m, void *private)
+{
+        struct kvm_vcpu *vcpu = m->private;
+        int i;
+        u64 min, max;
+
+        for (i = 0; i < __NUMBER_OF_KVM_EXIT_TYPES; i++) {
+                if (vcpu->arch.timing_min_duration[i] == 0xFFFFFFFF)
+                        min = 0;
+                else
+                        min = vcpu->arch.timing_min_duration[i];
+                if (vcpu->arch.timing_max_duration[i] == 0)
+                        max = 0;
+                else
+                        max = vcpu->arch.timing_max_duration[i];
+
+                seq_printf(m, "%12s: count %10d min %10lld "
+                        "max %10lld sum %20lld sum_quad %20lld\n",
+                        kvm_exit_names[i], vcpu->arch.timing_count_type[i],
+                        vcpu->arch.timing_min_duration[i],
+                        vcpu->arch.timing_max_duration[i],
+                        vcpu->arch.timing_sum_duration[i],
+                        vcpu->arch.timing_sum_quad_duration[i]);
+        }
+        return 0;
+}
+
+static ssize_t kvmppc_exit_timing_write(struct file *file,
+                                       const char __user *user_buf,
+                                       size_t count, loff_t *ppos)
+{
+        size_t len;
+        int err;
+        const char __user *p;
+        char c;
+
+        len = 0;
+        p = user_buf;
+        while (len < count) {
+                if (get_user(c, p++))
+                        err = -EFAULT;
+                if (c == 0 || c == '\n')
+                        break;
+                len++;
+        }
+
+        if (len > 1) {
+                err = -EINVAL;
+                goto done;
+        }
+
+        if (copy_from_user(&c, user_buf, sizeof(c))) {
+                err = -EFAULT;
+                goto done;
+        }
+
+        if (c == 'c') {
+                struct seq_file *seqf = (struct seq_file *)file->private_data;
+                struct kvm_vcpu *vcpu = seqf->private;
+                /* write does not affect out buffers previsously generated with
+                 * show. Seq file is locked here to prevent races of init with
+                 * a show call */
+                mutex_lock(&seqf->lock);
+                kvmppc_init_timing_stats(vcpu);
+                mutex_unlock(&seqf->lock);
+                err = count;
+        } else {
+                err = -EINVAL;
+                goto done;
+        }
+
+done:
+        return err;
+}
+
+static int kvmppc_exit_timing_open(struct inode *inode, struct file *file)
+{
+        return single_open(file, kvmppc_exit_timing_show, inode->i_private);
+}
+
+static struct file_operations kvmppc_exit_timing_fops = {
+        .owner   = THIS_MODULE,
+        .open    = kvmppc_exit_timing_open,
+        .read    = seq_read,
+        .write   = kvmppc_exit_timing_write,
+        .llseek  = seq_lseek,
+        .release = single_release,
+};
+
+void kvmppc_create_vcpu_debugfs(struct kvm_vcpu *vcpu, unsigned int id)
+{
+        static char dbg_fname[50];
+        struct dentry *debugfs_file;
+
+        snprintf(dbg_fname, sizeof(dbg_fname), "vm%u_vcpu%03u_timing",
+                 current->pid, id);
+        debugfs_file = debugfs_create_file(dbg_fname, 0666,
+                                        kvm_debugfs_dir, vcpu,
+                                        &kvmppc_exit_timing_fops);
+
+        if (!debugfs_file) {
+                printk(KERN_ERR"%s: error creating debugfs file %s\n",
+                        __func__, dbg_fname);
+                return;
+        }
+
+        vcpu->arch.debugfs_exit_timing = debugfs_file;
+}
+
+void kvmppc_remove_vcpu_debugfs(struct kvm_vcpu *vcpu)
+{
+        if (vcpu->arch.debugfs_exit_timing) {
+                debugfs_remove(vcpu->arch.debugfs_exit_timing);
+                vcpu->arch.debugfs_exit_timing = NULL;
+        }
+}