[PATCH] debugfs directory for each VM

[PATCH] debugfs directory for each VM

[Ryota OZAKI]

Hi all,

This patch allows a VM to have own directory on debugfs,
that contains statics only for the VM. Each directory
is identified by the pid of the VM (ie qemu).

I tried this patch under several host kernel versions,
.22, .23, and .24, and confirmed things work out well.

Thanks,
ozaki-r

Signed-off-by: Ryota Ozaki <ozaki.ryota@gmail.com>
--
 arch/x86/kvm/x86.c       |    5 ++-
 include/linux/kvm_host.h |    8 ++++
 virt/kvm/kvm_main.c      |   81 ++++++++++++++++++++++++++++++++++++++++------
 3 files changed, 82 insertions(+), 12 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 5339ab1..7cab8a6 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -53,8 +53,8 @@ static u64 __read_mostly efer_reserved_bits =
0xfffffffffffffafeULL;
 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
 #endif

-#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
-#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
+#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM, NULL
+#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, NULL

 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
 				    struct kvm_cpuid_entry2 __user *entries);
@@ -93,6 +93,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ NULL }
 };

+size_t debugfs_entries_size =
sizeof(debugfs_entries)/sizeof(debugfs_entries[0]);

 unsigned long segment_base(u16 selector)
 {
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 958e003..037bab1 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -43,6 +43,11 @@
 struct kvm_vcpu;
 extern struct kmem_cache *kvm_vcpu_cache;

+struct kvm_debugfs {
+	struct dentry *dentry;
+	struct kvm_stats_debugfs_item *debugfs_entries;
+};
+
 struct kvm_guest_debug {
 	int enabled;
 	unsigned long bp[4];
@@ -127,6 +132,7 @@ struct kvm {
 	struct kvm_io_bus pio_bus;
 	struct kvm_vm_stat stat;
 	struct kvm_arch arch;
+	struct kvm_debugfs debugfs;
 };

 /* The guest did something we don't support. */
@@ -304,7 +310,9 @@ struct kvm_stats_debugfs_item {
 	int offset;
 	enum kvm_stat_kind kind;
 	struct dentry *dentry;
+	struct kvm *kvm;
 };
 extern struct kvm_stats_debugfs_item debugfs_entries[];
+extern size_t debugfs_entries_size;

 #endif
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 30bf832..357f827 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -60,6 +60,8 @@ EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
 static __read_mostly struct preempt_ops kvm_preempt_ops;

 static struct dentry *debugfs_dir;
+static void kvm_debugfs_init(struct kvm *kvm);
+static void kvm_debugfs_destroy(struct kvm *kvm);

 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 			   unsigned long arg);
@@ -192,6 +194,7 @@ static struct kvm *kvm_create_vm(void)
 	kvm_io_bus_init(&kvm->pio_bus);
 	mutex_init(&kvm->lock);
 	kvm_io_bus_init(&kvm->mmio_bus);
+	kvm_debugfs_init(kvm);
 	init_rwsem(&kvm->slots_lock);
 	spin_lock(&kvm_lock);
 	list_add(&kvm->vm_list, &vm_list);
@@ -236,6 +239,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
 	spin_lock(&kvm_lock);
 	list_del(&kvm->vm_list);
 	spin_unlock(&kvm_lock);
+	kvm_debugfs_destroy(kvm);
 	kvm_io_bus_destroy(&kvm->pio_bus);
 	kvm_io_bus_destroy(&kvm->mmio_bus);
 	kvm_arch_destroy_vm(kvm);
@@ -1248,36 +1252,48 @@ static struct notifier_block kvm_cpu_notifier = {
 	.priority = 20, /* must be > scheduler priority */
 };

-static int vm_stat_get(void *_offset, u64 *val)
+static int vm_stat_get(void *_item, u64 *val)
 {
-	unsigned offset = (long)_offset;
-	struct kvm *kvm;
+	unsigned offset = (long)((struct kvm_stats_debugfs_item *)_item)->offset;
+	struct kvm *kvm = ((struct kvm_stats_debugfs_item *)_item)->kvm;

 	*val = 0;
 	spin_lock(&kvm_lock);
-	list_for_each_entry(kvm, &vm_list, vm_list)
-		*val += *(u32 *)((void *)kvm + offset);
+	if (kvm) {
+		*val = *(u32 *)((void *)kvm + offset);
+	} else {
+		list_for_each_entry(kvm, &vm_list, vm_list)
+			*val += *(u32 *)((void *)kvm + offset);
+	}
 	spin_unlock(&kvm_lock);
 	return 0;
 }

 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");

-static int vcpu_stat_get(void *_offset, u64 *val)
+static int vcpu_stat_get(void *_item, u64 *val)
 {
-	unsigned offset = (long)_offset;
-	struct kvm *kvm;
+	unsigned offset = (long)((struct kvm_stats_debugfs_item *)_item)->offset;
+	struct kvm *kvm = ((struct kvm_stats_debugfs_item *)_item)->kvm;
 	struct kvm_vcpu *vcpu;
 	int i;

 	*val = 0;
 	spin_lock(&kvm_lock);
-	list_for_each_entry(kvm, &vm_list, vm_list)
+	if (kvm) {
 		for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 			vcpu = kvm->vcpus[i];
 			if (vcpu)
 				*val += *(u32 *)((void *)vcpu + offset);
 		}
+	} else {
+		list_for_each_entry(kvm, &vm_list, vm_list)
+			for (i = 0; i < KVM_MAX_VCPUS; ++i) {
+				vcpu = kvm->vcpus[i];
+				if (vcpu)
+					*val += *(u32 *)((void *)vcpu + offset);
+			}
+	}
 	spin_unlock(&kvm_lock);
 	return 0;
 }
@@ -1296,7 +1312,7 @@ static void kvm_init_debug(void)
 	debugfs_dir = debugfs_create_dir("kvm", NULL);
 	for (p = debugfs_entries; p->name; ++p)
 		p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
-						(void *)(long)p->offset,
+						(void *)p,
 						stat_fops[p->kind]);
 }

@@ -1309,6 +1325,51 @@ static void kvm_exit_debug(void)
 	debugfs_remove(debugfs_dir);
 }

+#define PROC_NUMBUF 13 /* from fs/proc/base.c */
+
+static void kvm_debugfs_init(struct kvm *kvm)
+{
+	struct kvm_stats_debugfs_item *p;
+	struct kvm_debugfs *kvm_debugfs = &kvm->debugfs;
+	struct dentry *dir;
+	char pid[PROC_NUMBUF];
+
+	sprintf(pid, "%d", pid_nr(task_pid(current)));
+	dir = debugfs_create_dir(pid, debugfs_dir);
+	if (dir == NULL) {
+		return;
+	}
+
+	p = kmalloc(sizeof(struct
kvm_stats_debugfs_item)*debugfs_entries_size, GFP_KERNEL);
+	if (p == NULL) {
+		debugfs_remove(dir);
+		return;
+	}
+	memcpy(p, debugfs_entries, sizeof(struct
kvm_stats_debugfs_item)*debugfs_entries_size);
+	
+	kvm_debugfs->dentry = dir;
+	kvm_debugfs->debugfs_entries = p;
+	for (; p->name; ++p) {
+		p->kvm = kvm;
+		p->dentry = debugfs_create_file(p->name, 0444, dir,
+						(void *)p,
+						stat_fops[p->kind]);
+	}
+	return;
+}
+
+static void kvm_debugfs_destroy(struct kvm *kvm)
+{
+	struct kvm_stats_debugfs_item *p;
+	struct kvm_debugfs *debugfs = &kvm->debugfs;
+
+	p = debugfs->debugfs_entries;
+	for (; p->name; ++p)
+		debugfs_remove(p->dentry);
+	debugfs_remove(debugfs->dentry);
+	kfree(debugfs->debugfs_entries);
+}
+
 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
 {
 	hardware_disable(NULL);
--

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: [PATCH] debugfs directory for each VM

[Avi Kivity]

Ryota OZAKI wrote:
> Hi all,
>
> This patch allows a VM to have own directory on debugfs,
> that contains statics only for the VM. Each directory
> is identified by the pid of the VM (ie qemu).
>
> I tried this patch under several host kernel versions,
> .22, .23, and .24, and confirmed things work out well.
>   

Nice patch; it's certainly useful to improve per-vm trace capabilities.

It would be nice to keep the summary, since that allows running kvm_stat 
with no arguments.

On the other hand, the kvmtrace patchset recently posted adds much more 
detailed data; we should probably concentrate on that as a means of 
improving performance tracing.

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: [PATCH] debugfs directory for each VM

[Ryota OZAKI]

Hi Avi,

2008/3/17, Avi Kivity <avi@qumranet.com>:
> Ryota OZAKI wrote:
>  > Hi all,
>  >
>  > This patch allows a VM to have own directory on debugfs,
>  > that contains statics only for the VM. Each directory
>  > is identified by the pid of the VM (ie qemu).
>  >
>  > I tried this patch under several host kernel versions,
>  > .22, .23, and .24, and confirmed things work out well.
>  >
>
>
> Nice patch; it's certainly useful to improve per-vm trace capabilities.
>
>  It would be nice to keep the summary, since that allows running kvm_stat
>  with no arguments.

My patch keeps the summary. But it actually breaks a kvm_stat's assumption
so kvm_stat doesn't work under kvm with my patch. I wrote a patch to fix this
problem. See the following patch. (i'm not familiar with python, sorry.)


>  On the other hand, the kvmtrace patchset recently posted adds much more
>  detailed data; we should probably concentrate on that as a means of
>  improving performance tracing.

The patchset is useful for me, too. I will concentrate that as well.

Thanks,
ozaki-r

>  --
>  error compiling committee.c: too many arguments to function


Signed-off-by: Ryota Ozaki <ozaki.ryota@gmail.com>
--
 kvm_stat |   25 ++++++++++++++++++-------
 1 files changed, 18 insertions(+), 7 deletions(-)

diff --git a/kvm_stat b/kvm_stat
index 07773b0..e507e62 100755
--- a/kvm_stat
+++ b/kvm_stat
@@ -4,11 +4,15 @@ import curses
 import sys, os, time, optparse

 class Stats:
-    def __init__(self):
-        self.base = '/sys/kernel/debug/kvm'
+    def __init__(self, pid = None):
+        if pid:
+            self.base = '/sys/kernel/debug/kvm/' + pid
+        else:
+            self.base = '/sys/kernel/debug/kvm'
         self.values = {}
         for key in os.listdir(self.base):
-            self.values[key] = None
+            if not os.path.isdir(self.base + '/' + key):
+                self.values[key] = None
     def get(self):
         for key, oldval in self.values.iteritems():
             newval = int(file(self.base + '/' + key).read())
@@ -26,7 +30,6 @@ if not os.access('/sys/kernel/debug/kvm', os.F_OK):
     print "and ensure the kvm modules are loaded"
     sys.exit(1)

-stats = Stats()

 label_width = 20
 number_width = 10
@@ -73,14 +76,22 @@ def batch(stats):
         values = s[key]
         print '%-22s%10d%10d' % (key, values[0], values[1])

-options = optparse.OptionParser()
-options.add_option('-1', '--once', '--batch',
+usage = "usage: %prog [options] [<pid>]"
+parser = optparse.OptionParser(usage)
+parser.add_option('-1', '--once', '--batch',
                    action = 'store_true',
                    default = False,
                    dest = 'once',
                    help = 'run in batch mode for one second',
                    )
-(options, args) = options.parse_args(sys.argv)
+(options, args) = parser.parse_args(sys.argv)
+
+if len(args) == 1:
+    stats = Stats()
+elif len(args) == 2:
+    stats = Stats(args[1])
+else:
+    parser.error("too many arguments.")

 if not options.once:
     import curses.wrapper
--

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/