On 01/28/2016 04:49 PM, Janosch Frank wrote:
> KVM statistics for VMs (no. of exits, halts and other special
> instructions) are currently only available in a summarized manner for
> all VMs. They are exported to userland through files in the kvm
> debugfs directory and used for performance monitoring, as well as VM
> problem detection with helper tools like kvm_stat. If a VM has
> problems and therefore creates a large number of exits, one can not
> easily find out which one it is, as there is no VM specific data.
>
> This patch adds a kvm debugfs subdirectory for each VM, which is named
> after its pid and file descriptor. They contain the same kind of files
> that are already in the kvm debugfs directory, but the data that is
> exported through them is now VM specific.
>
> CC: Dan Carpenter <dan.carpenter@xxxxxxxxxx> [includes fixes by Dan Carpenter]
> Signed-off-by: Janosch Frank <frankja@xxxxxxxxxxxxxxxxxx>
FWIW, the newly created subfolders,
require QEMU commit 6590045e5dd2fb0b1d7cdc047ae0c52fd4bb5276
scripts/kvm/kvm_stat: Replaced os.listdir with os.walk
Otherwise you might get errors like
Traceback (most recent call last):
File "scripts/kvm/kvm_stat", line 640, in <module>
curses.wrapper(tui, stats)
File "/usr/lib64/python2.7/curses/wrapper.py", line 43, in wrapper
return func(stdscr, *args, **kwds)
File "scripts/kvm/kvm_stat", line 547, in tui
refresh(sleeptime)
File "scripts/kvm/kvm_stat", line 523, in refresh
s = stats.get()
File "scripts/kvm/kvm_stat", line 483, in get
new = d.read()
File "scripts/kvm/kvm_stat", line 37, in read
return dict([(key, val(key)) for key in self._fields])
File "scripts/kvm/kvm_stat", line 36, in val
return int(file(self.base + '/' + key).read())
IOError: [Errno 21] Is a directory: '/sys/kernel/debug/kvm/13123-31'
when using the "-d" option.
Paolo, I still think that this a valuable addon for debugging. Any guidance?
> ---
> include/linux/kvm_host.h | 7 ++
> virt/kvm/kvm_main.c | 187 ++++++++++++++++++++++++++++++++++++++++++++---
> 2 files changed, 184 insertions(+), 10 deletions(-)
>
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index f707f74..3f237e1 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -406,6 +406,8 @@ struct kvm {
> #endif
> long tlbs_dirty;
> struct list_head devices;
> + struct dentry *debugfs_dentry;
> + struct kvm_stat_data **debugfs_stat_data;
> };
>
> #define kvm_err(fmt, ...) \
> @@ -982,6 +984,11 @@ enum kvm_stat_kind {
> KVM_STAT_VCPU,
> };
>
> +struct kvm_stat_data {
> + int offset;
> + struct kvm *kvm;
> +};
> +
> struct kvm_stats_debugfs_item {
> const char *name;
> int offset;
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 314c777..d530e60 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -63,6 +63,9 @@
> #define CREATE_TRACE_POINTS
> #include <trace/events/kvm.h>
>
> +/* Worst case buffer size needed for holding an integer. */
> +#define ITOA_MAX_LEN 12
> +
> MODULE_AUTHOR("Qumranet");
> MODULE_LICENSE("GPL");
>
> @@ -100,6 +103,9 @@ static __read_mostly struct preempt_ops kvm_preempt_ops;
> struct dentry *kvm_debugfs_dir;
> EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
>
> +static int kvm_debugfs_num_entries;
> +static const struct file_operations *stat_fops_per_vm[];
> +
> static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
> unsigned long arg);
> #ifdef CONFIG_KVM_COMPAT
> @@ -529,6 +535,58 @@ static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
> kvfree(slots);
> }
>
> +static void kvm_destroy_vm_debugfs(struct kvm *kvm)
> +{
> + int i;
> +
> + if (!kvm->debugfs_dentry)
> + return;
> +
> + debugfs_remove_recursive(kvm->debugfs_dentry);
> +
> + for (i = 0; i < kvm_debugfs_num_entries; i++)
> + kfree(kvm->debugfs_stat_data[i]);
> + kfree(kvm->debugfs_stat_data);
> +}
> +
> +static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
> +{
> + char dir_name[ITOA_MAX_LEN * 2];
> + struct kvm_stat_data *stat_data;
> + struct kvm_stats_debugfs_item *p;
> +
> + if (!debugfs_initialized())
> + return 0;
> +
> + snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
> + kvm->debugfs_dentry = debugfs_create_dir(dir_name,
> + kvm_debugfs_dir);
> + if (!kvm->debugfs_dentry)
> + return -ENOMEM;
> +
> + kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
> + sizeof(*kvm->debugfs_stat_data),
> + GFP_KERNEL);
> + if (!kvm->debugfs_stat_data)
> + return -ENOMEM;
> +
> + for (p = debugfs_entries; p->name; p++) {
> + stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
> + if (!stat_data)
> + return -ENOMEM;
> +
> + stat_data->kvm = kvm;
> + stat_data->offset = p->offset;
> + kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
> + if (!debugfs_create_file(p->name, 0444,
> + kvm->debugfs_dentry,
> + stat_data,
> + stat_fops_per_vm[p->kind]))
> + return -ENOMEM;
> + }
> + return 0;
> +}
> +
> static struct kvm *kvm_create_vm(unsigned long type)
> {
> int r, i;
> @@ -636,6 +694,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
> int i;
> struct mm_struct *mm = kvm->mm;
>
> + kvm_destroy_vm_debugfs(kvm);
> kvm_arch_sync_events(kvm);
> spin_lock(&kvm_lock);
> list_del(&kvm->vm_list);
> @@ -2962,8 +3021,15 @@ static int kvm_dev_ioctl_create_vm(unsigned long type)
> }
> #endif
> r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
> - if (r < 0)
> + if (r < 0) {
> kvm_put_kvm(kvm);
> + return r;
> + }
> +
> + if (kvm_create_vm_debugfs(kvm, r) < 0) {
> + kvm_put_kvm(kvm);
> + return -ENOMEM;
> + }
>
> return r;
> }
> @@ -3388,15 +3454,114 @@ static struct notifier_block kvm_cpu_notifier = {
> .notifier_call = kvm_cpu_hotplug,
> };
>
> +static int kvm_debugfs_open(struct inode *inode, struct file *file,
> + int (*get)(void *, u64 *), int (*set)(void *, u64),
> + const char *fmt)
> +{
> + struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
> + inode->i_private;
> +
> + /* The debugfs files are a reference to the kvm struct which
> + * is still valid when kvm_destroy_vm is called.
> + * To avoid the race between open and the removal of the debugfs
> + * directory we test against the users count.
> + */
> + if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
> + return -ENOENT;
> +
> + if (simple_attr_open(inode, file, get, set, fmt)) {
> + kvm_put_kvm(stat_data->kvm);
> + return -ENOMEM;
> + }
> +
> + return 0;
> +}
> +
> +static int kvm_debugfs_release(struct inode *inode, struct file *file)
> +{
> + struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
> + inode->i_private;
> +
> + simple_attr_release(inode, file);
> + kvm_put_kvm(stat_data->kvm);
> +
> + return 0;
> +}
> +
> +static int vm_stat_get_per_vm(void *data, u64 *val)
> +{
> + struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
> +
> + *val = *(u32 *)((void *)stat_data->kvm + stat_data->offset);
> +
> + return 0;
> +}
> +
> +static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
> +{
> + __simple_attr_check_format("%llu\n", 0ull);
> + return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
> + NULL, "%llu\n");
> +}
> +
> +static const struct file_operations vm_stat_get_per_vm_fops = {
> + .owner = THIS_MODULE,
> + .open = vm_stat_get_per_vm_open,
> + .release = kvm_debugfs_release,
> + .read = simple_attr_read,
> + .write = simple_attr_write,
> + .llseek = generic_file_llseek,
> +};
> +
> +static int vcpu_stat_get_per_vm(void *data, u64 *val)
> +{
> + int i;
> + struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
> + struct kvm_vcpu *vcpu;
> +
> + *val = 0;
> +
> + kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
> + *val += *(u32 *)((void *)vcpu + stat_data->offset);
> +
> + return 0;
> +}
> +
> +static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
> +{
> + __simple_attr_check_format("%llu\n", 0ull);
> + return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
> + NULL, "%llu\n");
> +}
> +
> +static const struct file_operations vcpu_stat_get_per_vm_fops = {
> + .owner = THIS_MODULE,
> + .open = vcpu_stat_get_per_vm_open,
> + .release = kvm_debugfs_release,
> + .read = simple_attr_read,
> + .write = simple_attr_write,
> + .llseek = generic_file_llseek,
> +};
> +
> +static const struct file_operations *stat_fops_per_vm[] = {
> + [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
> + [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
> +};
> +
> static int vm_stat_get(void *_offset, u64 *val)
> {
> unsigned offset = (long)_offset;
> struct kvm *kvm;
> + struct kvm_stat_data stat_tmp = {.offset = offset};
> + u64 tmp_val;
>
> *val = 0;
> spin_lock(&kvm_lock);
> - list_for_each_entry(kvm, &vm_list, vm_list)
> - *val += *(u32 *)((void *)kvm + offset);
> + list_for_each_entry(kvm, &vm_list, vm_list) {
> + stat_tmp.kvm = kvm;
> + vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
> + *val += tmp_val;
> + }
> spin_unlock(&kvm_lock);
> return 0;
> }
> @@ -3407,15 +3572,16 @@ static int vcpu_stat_get(void *_offset, u64 *val)
> {
> unsigned offset = (long)_offset;
> struct kvm *kvm;
> - struct kvm_vcpu *vcpu;
> - int i;
> + struct kvm_stat_data stat_tmp = {.offset = offset};
> + u64 tmp_val;
>
> *val = 0;
> spin_lock(&kvm_lock);
> - list_for_each_entry(kvm, &vm_list, vm_list)
> - kvm_for_each_vcpu(i, vcpu, kvm)
> - *val += *(u32 *)((void *)vcpu + offset);
> -
> + list_for_each_entry(kvm, &vm_list, vm_list) {
> + stat_tmp.kvm = kvm;
> + vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
> + *val += tmp_val;
> + }
> spin_unlock(&kvm_lock);
> return 0;
> }
> @@ -3436,7 +3602,8 @@ static int kvm_init_debug(void)
> if (kvm_debugfs_dir == NULL)
> goto out;
>
> - for (p = debugfs_entries; p->name; ++p) {
> + kvm_debugfs_num_entries = 0;
> + for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
> if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
> (void *)(long)p->offset,
> stat_fops[p->kind]))
>
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
