ALSA: seq: Fix regression by incorrect ioctl_mutex usages

This is the revised backport of the upstream commit
b3defb79

We had another backport (e.g. 623e5c8a in 4.4.115), but it applies
the new mutex also to the code paths that are invoked via faked
kernel-to-kernel ioctls.  As reported recently, this leads to a
deadlock at suspend (or other scenarios triggering the kernel
sequencer client).

This patch addresses the issue by taking the mutex only in the code
paths invoked by user-space, just like the original fix patch does.
Reported-and-tested-by: Andres Bertens <abertensu@yahoo.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

sound/core/seq/seq_clientmgr.c

@@ -2196,7 +2196,6 @@ static int snd_seq_do_ioctl(struct snd_seq_client *client, unsigned int cmd,
 			    void __user *arg)
 {
 	struct seq_ioctl_table *p;
-	int ret;
 
 	switch (cmd) {
 	case SNDRV_SEQ_IOCTL_PVERSION:
@@ -2210,12 +2209,8 @@ static int snd_seq_do_ioctl(struct snd_seq_client *client, unsigned int cmd,
 	if (! arg)
 		return -EFAULT;
 	for (p = ioctl_tables; p->cmd; p++) {
-		if (p->cmd == cmd) {
-			mutex_lock(&client->ioctl_mutex);
-			ret = p->func(client, arg);
-			mutex_unlock(&client->ioctl_mutex);
-			return ret;
-		}
+		if (p->cmd == cmd)
+			return p->func(client, arg);
 	}
 	pr_debug("ALSA: seq unknown ioctl() 0x%x (type='%c', number=0x%02x)\n",
 		   cmd, _IOC_TYPE(cmd), _IOC_NR(cmd));
@@ -2226,11 +2221,15 @@ static int snd_seq_do_ioctl(struct snd_seq_client *client, unsigned int cmd,
 static long snd_seq_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
 	struct snd_seq_client *client = file->private_data;
+	long ret;
 
 	if (snd_BUG_ON(!client))
 		return -ENXIO;
 		
-	return snd_seq_do_ioctl(client, cmd, (void __user *) arg);
+	mutex_lock(&client->ioctl_mutex);
+	ret = snd_seq_do_ioctl(client, cmd, (void __user *) arg);
+	mutex_unlock(&client->ioctl_mutex);
+	return ret;
 }
 
 #ifdef CONFIG_COMPAT