Update to 3.4-final.
[linux-flexiantxendom0-3.2.10.git] / drivers / char / virtio_console.c
index 6207e37..cdf2f54 100644 (file)
@@ -1,6 +1,7 @@
 /*
  * Copyright (C) 2006, 2007, 2009 Rusty Russell, IBM Corporation
- * Copyright (C) 2009, 2010 Red Hat, Inc.
+ * Copyright (C) 2009, 2010, 2011 Red Hat, Inc.
+ * Copyright (C) 2009, 2010, 2011 Amit Shah <amit.shah@redhat.com>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
  */
 #include <linux/cdev.h>
 #include <linux/debugfs.h>
+#include <linux/completion.h>
 #include <linux/device.h>
 #include <linux/err.h>
+#include <linux/freezer.h>
 #include <linux/fs.h>
 #include <linux/init.h>
 #include <linux/list.h>
@@ -31,7 +34,8 @@
 #include <linux/virtio_console.h>
 #include <linux/wait.h>
 #include <linux/workqueue.h>
-#include "hvc_console.h"
+#include <linux/module.h>
+#include "../tty/hvc/hvc_console.h"
 
 /*
  * This is a global struct for storing common data for all the devices
@@ -48,6 +52,9 @@ struct ports_driver_data {
        /* Used for exporting per-port information to debugfs */
        struct dentry *debugfs_dir;
 
+       /* List of all the devices we're handling */
+       struct list_head portdevs;
+
        /* Number of devices this driver is handling */
        unsigned int index;
 
@@ -69,6 +76,7 @@ struct ports_driver_data {
 static struct ports_driver_data pdrvdata;
 
 DEFINE_SPINLOCK(pdrvdata_lock);
+DECLARE_COMPLETION(early_console_added);
 
 /* This struct holds information that's relevant only for console ports */
 struct console {
@@ -78,6 +86,9 @@ struct console {
        /* The hvc device associated with this console port */
        struct hvc_struct *hvc;
 
+       /* The size of the console */
+       struct winsize ws;
+
        /*
         * This number identifies the number that we used to register
         * with hvc in hvc_instantiate() and hvc_alloc(); this is the
@@ -105,6 +116,9 @@ struct port_buffer {
  * ports for that device (vdev->priv).
  */
 struct ports_device {
+       /* Next portdev in the list, head is in the pdrvdata struct */
+       struct list_head list;
+
        /*
         * Workqueue handlers where we process deferred work after
         * notification
@@ -141,6 +155,10 @@ struct ports_device {
        int chr_major;
 };
 
+struct port_stats {
+       unsigned long bytes_sent, bytes_received, bytes_discarded;
+};
+
 /* This struct holds the per-port data */
 struct port {
        /* Next port in the list, head is in the ports_device */
@@ -159,6 +177,9 @@ struct port {
         */
        spinlock_t inbuf_lock;
 
+       /* Protect the operations on the out_vq. */
+       spinlock_t outvq_lock;
+
        /* The IO vqs for this port */
        struct virtqueue *in_vq, *out_vq;
 
@@ -166,24 +187,39 @@ struct port {
        struct dentry *debugfs_file;
 
        /*
+        * Keep count of the bytes sent, received and discarded for
+        * this port for accounting and debugging purposes.  These
+        * counts are not reset across port open / close events.
+        */
+       struct port_stats stats;
+
+       /*
         * The entries in this struct will be valid if this port is
         * hooked up to an hvc console
         */
        struct console cons;
 
        /* Each port associates with a separate char device */
-       struct cdev cdev;
+       struct cdev *cdev;
        struct device *dev;
 
+       /* Reference-counting to handle port hot-unplugs and file operations */
+       struct kref kref;
+
        /* A waitqueue for poll() or blocking read operations */
        wait_queue_head_t waitqueue;
 
        /* The 'name' of the port that we expose via sysfs properties */
        char *name;
 
+       /* We can notify apps of host connect / disconnect events via SIGIO */
+       struct fasync_struct *async_queue;
+
        /* The 'id' to identify the port with the Host */
        u32 id;
 
+       bool outvq_full;
+
        /* Is the host device open */
        bool host_connected;
 
@@ -213,6 +249,41 @@ out:
        return port;
 }
 
+static struct port *find_port_by_devt_in_portdev(struct ports_device *portdev,
+                                                dev_t dev)
+{
+       struct port *port;
+       unsigned long flags;
+
+       spin_lock_irqsave(&portdev->ports_lock, flags);
+       list_for_each_entry(port, &portdev->ports, list)
+               if (port->cdev->dev == dev)
+                       goto out;
+       port = NULL;
+out:
+       spin_unlock_irqrestore(&portdev->ports_lock, flags);
+
+       return port;
+}
+
+static struct port *find_port_by_devt(dev_t dev)
+{
+       struct ports_device *portdev;
+       struct port *port;
+       unsigned long flags;
+
+       spin_lock_irqsave(&pdrvdata_lock, flags);
+       list_for_each_entry(portdev, &pdrvdata.portdevs, list) {
+               port = find_port_by_devt_in_portdev(portdev, dev);
+               if (port)
+                       goto out;
+       }
+       port = NULL;
+out:
+       spin_unlock_irqrestore(&pdrvdata_lock, flags);
+       return port;
+}
+
 static struct port *find_port_by_id(struct ports_device *portdev, u32 id)
 {
        struct port *port;
@@ -291,17 +362,19 @@ fail:
 }
 
 /* Callers should take appropriate locks */
-static void *get_inbuf(struct port *port)
+static struct port_buffer *get_inbuf(struct port *port)
 {
        struct port_buffer *buf;
-       struct virtqueue *vq;
        unsigned int len;
 
-       vq = port->in_vq;
-       buf = virtqueue_get_buf(vq, &len);
+       if (port->inbuf)
+               return port->inbuf;
+
+       buf = virtqueue_get_buf(port->in_vq, &len);
        if (buf) {
                buf->len = len;
                buf->offset = 0;
+               port->stats.bytes_received += len;
        }
        return buf;
 }
@@ -319,7 +392,7 @@ static int add_inbuf(struct virtqueue *vq, struct port_buffer *buf)
 
        sg_init_one(sg, buf->buf, buf->size);
 
-       ret = virtqueue_add_buf(vq, sg, 0, 1, buf);
+       ret = virtqueue_add_buf(vq, sg, 0, 1, buf, GFP_ATOMIC);
        virtqueue_kick(vq);
        return ret;
 }
@@ -328,28 +401,27 @@ static int add_inbuf(struct virtqueue *vq, struct port_buffer *buf)
 static void discard_port_data(struct port *port)
 {
        struct port_buffer *buf;
-       struct virtqueue *vq;
-       unsigned int len;
-       int ret;
+       unsigned int err;
 
-       vq = port->in_vq;
-       if (port->inbuf)
-               buf = port->inbuf;
-       else
-               buf = virtqueue_get_buf(vq, &len);
+       if (!port->portdev) {
+               /* Device has been unplugged.  vqs are already gone. */
+               return;
+       }
+       buf = get_inbuf(port);
 
-       ret = 0;
+       err = 0;
        while (buf) {
-               if (add_inbuf(vq, buf) < 0) {
-                       ret++;
+               port->stats.bytes_discarded += buf->len - buf->offset;
+               if (add_inbuf(port->in_vq, buf) < 0) {
+                       err++;
                        free_buf(buf);
                }
-               buf = virtqueue_get_buf(vq, &len);
+               port->inbuf = NULL;
+               buf = get_inbuf(port);
        }
-       port->inbuf = NULL;
-       if (ret)
+       if (err)
                dev_warn(port->dev, "Errors adding %d buffers back to vq\n",
-                        ret);
+                        err);
 }
 
 static bool port_has_data(struct port *port)
@@ -357,18 +429,12 @@ static bool port_has_data(struct port *port)
        unsigned long flags;
        bool ret;
 
+       ret = false;
        spin_lock_irqsave(&port->inbuf_lock, flags);
-       if (port->inbuf) {
-               ret = true;
-               goto out;
-       }
        port->inbuf = get_inbuf(port);
-       if (port->inbuf) {
+       if (port->inbuf)
                ret = true;
-               goto out;
-       }
-       ret = false;
-out:
+
        spin_unlock_irqrestore(&port->inbuf_lock, flags);
        return ret;
 }
@@ -391,7 +457,7 @@ static ssize_t __send_control_msg(struct ports_device *portdev, u32 port_id,
        vq = portdev->c_ovq;
 
        sg_init_one(sg, &cpkt, sizeof(cpkt));
-       if (virtqueue_add_buf(vq, sg, 1, 0, &cpkt) >= 0) {
+       if (virtqueue_add_buf(vq, sg, 1, 0, &cpkt, GFP_ATOMIC) >= 0) {
                virtqueue_kick(vq);
                while (!virtqueue_get_buf(vq, &len))
                        cpu_relax();
@@ -402,34 +468,79 @@ static ssize_t __send_control_msg(struct ports_device *portdev, u32 port_id,
 static ssize_t send_control_msg(struct port *port, unsigned int event,
                                unsigned int value)
 {
-       return __send_control_msg(port->portdev, port->id, event, value);
+       /* Did the port get unplugged before userspace closed it? */
+       if (port->portdev)
+               return __send_control_msg(port->portdev, port->id, event, value);
+       return 0;
 }
 
-static ssize_t send_buf(struct port *port, void *in_buf, size_t in_count)
+/* Callers must take the port->outvq_lock */
+static void reclaim_consumed_buffers(struct port *port)
+{
+       void *buf;
+       unsigned int len;
+
+       if (!port->portdev) {
+               /* Device has been unplugged.  vqs are already gone. */
+               return;
+       }
+       while ((buf = virtqueue_get_buf(port->out_vq, &len))) {
+               kfree(buf);
+               port->outvq_full = false;
+       }
+}
+
+static ssize_t send_buf(struct port *port, void *in_buf, size_t in_count,
+                       bool nonblock)
 {
        struct scatterlist sg[1];
        struct virtqueue *out_vq;
        ssize_t ret;
+       unsigned long flags;
        unsigned int len;
 
        out_vq = port->out_vq;
 
+       spin_lock_irqsave(&port->outvq_lock, flags);
+
+       reclaim_consumed_buffers(port);
+
        sg_init_one(sg, in_buf, in_count);
-       ret = virtqueue_add_buf(out_vq, sg, 1, 0, in_buf);
+       ret = virtqueue_add_buf(out_vq, sg, 1, 0, in_buf, GFP_ATOMIC);
 
        /* Tell Host to go! */
        virtqueue_kick(out_vq);
 
        if (ret < 0) {
                in_count = 0;
-               goto fail;
+               goto done;
        }
 
-       /* Wait till the host acknowledges it pushed out the data we sent. */
+       if (ret == 0)
+               port->outvq_full = true;
+
+       if (nonblock)
+               goto done;
+
+       /*
+        * Wait till the host acknowledges it pushed out the data we
+        * sent.  This is done for data from the hvc_console; the tty
+        * operations are performed with spinlocks held so we can't
+        * sleep here.  An alternative would be to copy the data to a
+        * buffer and relax the spinning requirement.  The downside is
+        * we need to kmalloc a GFP_ATOMIC buffer each time the
+        * console driver writes something out.
+        */
        while (!virtqueue_get_buf(out_vq, &len))
                cpu_relax();
-fail:
-       /* We're expected to return the amount of data we wrote */
+done:
+       spin_unlock_irqrestore(&port->outvq_lock, flags);
+
+       port->stats.bytes_sent += in_count;
+       /*
+        * We're expected to return the amount of data we wrote -- all
+        * of it
+        */
        return in_count;
 }
 
@@ -479,9 +590,36 @@ static ssize_t fill_readbuf(struct port *port, char *out_buf, size_t out_count,
 }
 
 /* The condition that must be true for polling to end */
-static bool wait_is_over(struct port *port)
+static bool will_read_block(struct port *port)
 {
-       return port_has_data(port) || !port->host_connected;
+       if (!port->guest_connected) {
+               /* Port got hot-unplugged. Let's exit. */
+               return false;
+       }
+       return !port_has_data(port) && port->host_connected;
+}
+
+static bool will_write_block(struct port *port)
+{
+       bool ret;
+
+       if (!port->guest_connected) {
+               /* Port got hot-unplugged. Let's exit. */
+               return false;
+       }
+       if (!port->host_connected)
+               return true;
+
+       spin_lock_irq(&port->outvq_lock);
+       /*
+        * Check if the Host has consumed any buffers since we last
+        * sent data (this is only applicable for nonblocking ports).
+        */
+       reclaim_consumed_buffers(port);
+       ret = port->outvq_full;
+       spin_unlock_irq(&port->outvq_lock);
+
+       return ret;
 }
 
 static ssize_t port_fops_read(struct file *filp, char __user *ubuf,
@@ -503,11 +641,14 @@ static ssize_t port_fops_read(struct file *filp, char __user *ubuf,
                if (filp->f_flags & O_NONBLOCK)
                        return -EAGAIN;
 
-               ret = wait_event_interruptible(port->waitqueue,
-                                              wait_is_over(port));
+               ret = wait_event_freezable(port->waitqueue,
+                                          !will_read_block(port));
                if (ret < 0)
                        return ret;
        }
+       /* Port got hot-unplugged. */
+       if (!port->guest_connected)
+               return -ENODEV;
        /*
         * We could've received a disconnection message while we were
         * waiting for more data.
@@ -530,9 +671,29 @@ static ssize_t port_fops_write(struct file *filp, const char __user *ubuf,
        struct port *port;
        char *buf;
        ssize_t ret;
+       bool nonblock;
+
+       /* Userspace could be out to fool us */
+       if (!count)
+               return 0;
 
        port = filp->private_data;
 
+       nonblock = filp->f_flags & O_NONBLOCK;
+
+       if (will_write_block(port)) {
+               if (nonblock)
+                       return -EAGAIN;
+
+               ret = wait_event_freezable(port->waitqueue,
+                                          !will_write_block(port));
+               if (ret < 0)
+                       return ret;
+       }
+       /* Port got hot-unplugged. */
+       if (!port->guest_connected)
+               return -ENODEV;
+
        count = min((size_t)(32 * 1024), count);
 
        buf = kmalloc(count, GFP_KERNEL);
@@ -545,9 +706,22 @@ static ssize_t port_fops_write(struct file *filp, const char __user *ubuf,
                goto free_buf;
        }
 
-       ret = send_buf(port, buf, count);
+       /*
+        * We now ask send_buf() to not spin for generic ports -- we
+        * can re-use the same code path that non-blocking file
+        * descriptors take for blocking file descriptors since the
+        * wait is already done and we're certain the write will go
+        * through to the host.
+        */
+       nonblock = true;
+       ret = send_buf(port, buf, count, nonblock);
+
+       if (nonblock && ret > 0)
+               goto out;
+
 free_buf:
        kfree(buf);
+out:
        return ret;
 }
 
@@ -559,10 +733,14 @@ static unsigned int port_fops_poll(struct file *filp, poll_table *wait)
        port = filp->private_data;
        poll_wait(filp, &port->waitqueue, wait);
 
+       if (!port->guest_connected) {
+               /* Port got unplugged */
+               return POLLHUP;
+       }
        ret = 0;
-       if (port->inbuf)
+       if (!will_read_block(port))
                ret |= POLLIN | POLLRDNORM;
-       if (port->host_connected)
+       if (!will_write_block(port))
                ret |= POLLOUT;
        if (!port->host_connected)
                ret |= POLLHUP;
@@ -570,6 +748,8 @@ static unsigned int port_fops_poll(struct file *filp, poll_table *wait)
        return ret;
 }
 
+static void remove_port(struct kref *kref);
+
 static int port_fops_release(struct inode *inode, struct file *filp)
 {
        struct port *port;
@@ -586,6 +766,20 @@ static int port_fops_release(struct inode *inode, struct file *filp)
 
        spin_unlock_irq(&port->inbuf_lock);
 
+       spin_lock_irq(&port->outvq_lock);
+       reclaim_consumed_buffers(port);
+       spin_unlock_irq(&port->outvq_lock);
+
+       /*
+        * Locks aren't necessary here as a port can't be opened after
+        * unplug, and if a port isn't unplugged, a kref would already
+        * exist for the port.  Plus, taking ports_lock here would
+        * create a dependency on other locks taken by functions
+        * inside remove_port if we're the last holder of the port,
+        * creating many problems.
+        */
+       kref_put(&port->kref, remove_port);
+
        return 0;
 }
 
@@ -593,31 +787,62 @@ static int port_fops_open(struct inode *inode, struct file *filp)
 {
        struct cdev *cdev = inode->i_cdev;
        struct port *port;
+       int ret;
 
-       port = container_of(cdev, struct port, cdev);
+       port = find_port_by_devt(cdev->dev);
        filp->private_data = port;
 
+       /* Prevent against a port getting hot-unplugged at the same time */
+       spin_lock_irq(&port->portdev->ports_lock);
+       kref_get(&port->kref);
+       spin_unlock_irq(&port->portdev->ports_lock);
+
        /*
         * Don't allow opening of console port devices -- that's done
         * via /dev/hvc
         */
-       if (is_console_port(port))
-               return -ENXIO;
+       if (is_console_port(port)) {
+               ret = -ENXIO;
+               goto out;
+       }
 
        /* Allow only one process to open a particular port at a time */
        spin_lock_irq(&port->inbuf_lock);
        if (port->guest_connected) {
                spin_unlock_irq(&port->inbuf_lock);
-               return -EMFILE;
+               ret = -EMFILE;
+               goto out;
        }
 
        port->guest_connected = true;
        spin_unlock_irq(&port->inbuf_lock);
 
+       spin_lock_irq(&port->outvq_lock);
+       /*
+        * There might be a chance that we missed reclaiming a few
+        * buffers in the window of the port getting previously closed
+        * and opening now.
+        */
+       reclaim_consumed_buffers(port);
+       spin_unlock_irq(&port->outvq_lock);
+
+       nonseekable_open(inode, filp);
+
        /* Notify host of port being opened */
        send_control_msg(filp->private_data, VIRTIO_CONSOLE_PORT_OPEN, 1);
 
        return 0;
+out:
+       kref_put(&port->kref, remove_port);
+       return ret;
+}
+
+static int port_fops_fasync(int fd, struct file *filp, int mode)
+{
+       struct port *port;
+
+       port = filp->private_data;
+       return fasync_helper(fd, filp, mode, &port->async_queue);
 }
 
 /*
@@ -633,6 +858,8 @@ static const struct file_operations port_fops = {
        .write = port_fops_write,
        .poll  = port_fops_poll,
        .release = port_fops_release,
+       .fasync = port_fops_fasync,
+       .llseek = no_llseek,
 };
 
 /*
@@ -654,7 +881,7 @@ static int put_chars(u32 vtermno, const char *buf, int count)
        if (!port)
                return -EPIPE;
 
-       return send_buf(port, (void *)buf, count);
+       return send_buf(port, (void *)buf, count, false);
 }
 
 /*
@@ -685,22 +912,14 @@ static int get_chars(u32 vtermno, char *buf, int count)
 static void resize_console(struct port *port)
 {
        struct virtio_device *vdev;
-       struct winsize ws;
 
        /* The port could have been hot-unplugged */
-       if (!port)
+       if (!port || !is_console_port(port))
                return;
 
        vdev = port->portdev->vdev;
-       if (virtio_has_feature(vdev, VIRTIO_CONSOLE_F_SIZE)) {
-               vdev->config->get(vdev,
-                                 offsetof(struct virtio_console_config, cols),
-                                 &ws.ws_col, sizeof(u16));
-               vdev->config->get(vdev,
-                                 offsetof(struct virtio_console_config, rows),
-                                 &ws.ws_row, sizeof(u16));
-               hvc_resize(port->cons.hvc, ws);
-       }
+       if (virtio_has_feature(vdev, VIRTIO_CONSOLE_F_SIZE))
+               hvc_resize(port->cons.hvc, port->cons.ws);
 }
 
 /* We set the configuration at this point, since we now have a tty */
@@ -819,12 +1038,6 @@ static struct attribute_group port_attribute_group = {
        .attrs = port_sysfs_entries,
 };
 
-static int debugfs_open(struct inode *inode, struct file *filp)
-{
-       filp->private_data = inode->i_private;
-       return 0;
-}
-
 static ssize_t debugfs_read(struct file *filp, char __user *ubuf,
                            size_t count, loff_t *offp)
 {
@@ -846,6 +1059,16 @@ static ssize_t debugfs_read(struct file *filp, char __user *ubuf,
        out_offset += snprintf(buf + out_offset, out_count - out_offset,
                               "host_connected: %d\n", port->host_connected);
        out_offset += snprintf(buf + out_offset, out_count - out_offset,
+                              "outvq_full: %d\n", port->outvq_full);
+       out_offset += snprintf(buf + out_offset, out_count - out_offset,
+                              "bytes_sent: %lu\n", port->stats.bytes_sent);
+       out_offset += snprintf(buf + out_offset, out_count - out_offset,
+                              "bytes_received: %lu\n",
+                              port->stats.bytes_received);
+       out_offset += snprintf(buf + out_offset, out_count - out_offset,
+                              "bytes_discarded: %lu\n",
+                              port->stats.bytes_discarded);
+       out_offset += snprintf(buf + out_offset, out_count - out_offset,
                               "is_console: %s\n",
                               is_console_port(port) ? "yes" : "no");
        out_offset += snprintf(buf + out_offset, out_count - out_offset,
@@ -858,10 +1081,19 @@ static ssize_t debugfs_read(struct file *filp, char __user *ubuf,
 
 static const struct file_operations port_debugfs_ops = {
        .owner = THIS_MODULE,
-       .open  = debugfs_open,
+       .open  = simple_open,
        .read  = debugfs_read,
 };
 
+static void set_console_size(struct port *port, u16 rows, u16 cols)
+{
+       if (!port || !is_console_port(port))
+               return;
+
+       port->cons.ws.ws_row = rows;
+       port->cons.ws.ws_col = cols;
+}
+
 static unsigned int fill_queue(struct virtqueue *vq, spinlock_t *lock)
 {
        struct port_buffer *buf;
@@ -888,6 +1120,12 @@ static unsigned int fill_queue(struct virtqueue *vq, spinlock_t *lock)
        return nr_added_bufs;
 }
 
+static void send_sigio_to_port(struct port *port)
+{
+       if (port->async_queue && port->guest_connected)
+               kill_fasync(&port->async_queue, SIGIO, POLL_OUT);
+}
+
 static int add_port(struct ports_device *portdev, u32 id)
 {
        char debugfs_name[16];
@@ -902,6 +1140,7 @@ static int add_port(struct ports_device *portdev, u32 id)
                err = -ENOMEM;
                goto fail;
        }
+       kref_init(&port->kref);
 
        port->portdev = portdev;
        port->id = id;
@@ -909,20 +1148,32 @@ static int add_port(struct ports_device *portdev, u32 id)
        port->name = NULL;
        port->inbuf = NULL;
        port->cons.hvc = NULL;
+       port->async_queue = NULL;
+
+       port->cons.ws.ws_row = port->cons.ws.ws_col = 0;
 
        port->host_connected = port->guest_connected = false;
+       port->stats = (struct port_stats) { 0 };
+
+       port->outvq_full = false;
 
        port->in_vq = portdev->in_vqs[port->id];
        port->out_vq = portdev->out_vqs[port->id];
 
-       cdev_init(&port->cdev, &port_fops);
+       port->cdev = cdev_alloc();
+       if (!port->cdev) {
+               dev_err(&port->portdev->vdev->dev, "Error allocating cdev\n");
+               err = -ENOMEM;
+               goto free_port;
+       }
+       port->cdev->ops = &port_fops;
 
        devt = MKDEV(portdev->chr_major, id);
-       err = cdev_add(&port->cdev, devt, 1);
+       err = cdev_add(port->cdev, devt, 1);
        if (err < 0) {
                dev_err(&port->portdev->vdev->dev,
                        "Error %d adding cdev for port %u\n", err, id);
-               goto free_port;
+               goto free_cdev;
        }
        port->dev = device_create(pdrvdata.class, &port->portdev->vdev->dev,
                                  devt, port, "vport%up%u",
@@ -936,6 +1187,7 @@ static int add_port(struct ports_device *portdev, u32 id)
        }
 
        spin_lock_init(&port->inbuf_lock);
+       spin_lock_init(&port->outvq_lock);
        init_waitqueue_head(&port->waitqueue);
 
        /* Fill the in_vq with buffers so the host can send us data. */
@@ -986,61 +1238,89 @@ free_inbufs:
 free_device:
        device_destroy(pdrvdata.class, port->dev->devt);
 free_cdev:
-       cdev_del(&port->cdev);
+       cdev_del(port->cdev);
 free_port:
        kfree(port);
 fail:
        /* The host might want to notify management sw about port add failure */
-       send_control_msg(port, VIRTIO_CONSOLE_PORT_READY, 0);
+       __send_control_msg(portdev, id, VIRTIO_CONSOLE_PORT_READY, 0);
        return err;
 }
 
-/* Remove all port-specific data. */
-static int remove_port(struct port *port)
+/* No users remain, remove all port-specific data. */
+static void remove_port(struct kref *kref)
+{
+       struct port *port;
+
+       port = container_of(kref, struct port, kref);
+
+       sysfs_remove_group(&port->dev->kobj, &port_attribute_group);
+       device_destroy(pdrvdata.class, port->dev->devt);
+       cdev_del(port->cdev);
+
+       kfree(port->name);
+
+       debugfs_remove(port->debugfs_file);
+
+       kfree(port);
+}
+
+static void remove_port_data(struct port *port)
 {
        struct port_buffer *buf;
 
+       /* Remove unused data this port might have received. */
+       discard_port_data(port);
+
+       reclaim_consumed_buffers(port);
+
+       /* Remove buffers we queued up for the Host to send us data in. */
+       while ((buf = virtqueue_detach_unused_buf(port->in_vq)))
+               free_buf(buf);
+}
+
+/*
+ * Port got unplugged.  Remove port from portdev's list and drop the
+ * kref reference.  If no userspace has this port opened, it will
+ * result in immediate removal the port.
+ */
+static void unplug_port(struct port *port)
+{
        spin_lock_irq(&port->portdev->ports_lock);
        list_del(&port->list);
        spin_unlock_irq(&port->portdev->ports_lock);
 
+       if (port->guest_connected) {
+               port->guest_connected = false;
+               port->host_connected = false;
+               wake_up_interruptible(&port->waitqueue);
+
+               /* Let the app know the port is going down. */
+               send_sigio_to_port(port);
+       }
+
        if (is_console_port(port)) {
                spin_lock_irq(&pdrvdata_lock);
                list_del(&port->cons.list);
                spin_unlock_irq(&pdrvdata_lock);
-#if 0
-               /*
-                * hvc_remove() not called as removing one hvc port
-                * results in other hvc ports getting frozen.
-                *
-                * Once this is resolved in hvc, this functionality
-                * will be enabled.  Till that is done, the -EPIPE
-                * return from get_chars() above will help
-                * hvc_console.c to clean up on ports we remove here.
-                */
                hvc_remove(port->cons.hvc);
-#endif
        }
-       if (port->guest_connected)
-               send_control_msg(port, VIRTIO_CONSOLE_PORT_OPEN, 0);
-
-       sysfs_remove_group(&port->dev->kobj, &port_attribute_group);
-       device_destroy(pdrvdata.class, port->dev->devt);
-       cdev_del(&port->cdev);
-
-       /* Remove unused data this port might have received. */
-       discard_port_data(port);
-
-       /* Remove buffers we queued up for the Host to send us data in. */
-       while ((buf = virtqueue_detach_unused_buf(port->in_vq)))
-               free_buf(buf);
 
-       kfree(port->name);
+       remove_port_data(port);
 
-       debugfs_remove(port->debugfs_file);
+       /*
+        * We should just assume the device itself has gone off --
+        * else a close on an open port later will try to send out a
+        * control message.
+        */
+       port->portdev = NULL;
 
-       kfree(port);
-       return 0;
+       /*
+        * Locks around here are not necessary - a port can't be
+        * opened after we removed the port struct from ports_list
+        * above.
+        */
+       kref_put(&port->kref, remove_port);
 }
 
 /* Any private messages that the Host and Guest want to share */
@@ -1079,7 +1359,7 @@ static void handle_control_message(struct ports_device *portdev,
                add_port(portdev, cpkt->id);
                break;
        case VIRTIO_CONSOLE_PORT_REMOVE:
-               remove_port(port);
+               unplug_port(port);
                break;
        case VIRTIO_CONSOLE_CONSOLE_PORT:
                if (!cpkt->value)
@@ -1088,23 +1368,56 @@ static void handle_control_message(struct ports_device *portdev,
                        break;
 
                init_port_console(port);
+               complete(&early_console_added);
                /*
                 * Could remove the port here in case init fails - but
                 * have to notify the host first.
                 */
                break;
-       case VIRTIO_CONSOLE_RESIZE:
+       case VIRTIO_CONSOLE_RESIZE: {
+               struct {
+                       __u16 rows;
+                       __u16 cols;
+               } size;
+
                if (!is_console_port(port))
                        break;
+
+               memcpy(&size, buf->buf + buf->offset + sizeof(*cpkt),
+                      sizeof(size));
+               set_console_size(port, size.rows, size.cols);
+
                port->cons.hvc->irq_requested = 1;
                resize_console(port);
                break;
+       }
        case VIRTIO_CONSOLE_PORT_OPEN:
                port->host_connected = cpkt->value;
                wake_up_interruptible(&port->waitqueue);
+               /*
+                * If the host port got closed and the host had any
+                * unconsumed buffers, we'll be able to reclaim them
+                * now.
+                */
+               spin_lock_irq(&port->outvq_lock);
+               reclaim_consumed_buffers(port);
+               spin_unlock_irq(&port->outvq_lock);
+
+               /*
+                * If the guest is connected, it'll be interested in
+                * knowing the host connection state changed.
+                */
+               send_sigio_to_port(port);
                break;
        case VIRTIO_CONSOLE_PORT_NAME:
                /*
+                * If we woke up after hibernation, we can get this
+                * again.  Skip it in that case.
+                */
+               if (port->name)
+                       break;
+
+               /*
                 * Skip the size of the header and the cpkt to get the size
                 * of the name that was sent
                 */
@@ -1171,6 +1484,17 @@ static void control_work_handler(struct work_struct *work)
        spin_unlock(&portdev->cvq_lock);
 }
 
+static void out_intr(struct virtqueue *vq)
+{
+       struct port *port;
+
+       port = find_port_by_vq(vq->vdev->priv, vq);
+       if (!port)
+               return;
+
+       wake_up_interruptible(&port->waitqueue);
+}
+
 static void in_intr(struct virtqueue *vq)
 {
        struct port *port;
@@ -1181,8 +1505,7 @@ static void in_intr(struct virtqueue *vq)
                return;
 
        spin_lock_irqsave(&port->inbuf_lock, flags);
-       if (!port->inbuf)
-               port->inbuf = get_inbuf(port);
+       port->inbuf = get_inbuf(port);
 
        /*
         * Don't queue up data when port is closed.  This condition
@@ -1198,6 +1521,9 @@ static void in_intr(struct virtqueue *vq)
 
        wake_up_interruptible(&port->waitqueue);
 
+       /* Send a SIGIO indicating new data in case the process asked for it */
+       send_sigio_to_port(port);
+
        if (is_console_port(port) && hvc_poll(port->cons.hvc))
                hvc_kick();
 }
@@ -1216,13 +1542,29 @@ static void config_intr(struct virtio_device *vdev)
 
        portdev = vdev->priv;
 
-       /*
-        * We'll use this way of resizing only for legacy support.
-        * For newer userspace (VIRTIO_CONSOLE_F_MULTPORT+), use
-        * control messages to indicate console size changes so that
-        * it can be done per-port
-        */
-       resize_console(find_port_by_id(portdev, 0));
+       if (!use_multiport(portdev)) {
+               struct port *port;
+               u16 rows, cols;
+
+               vdev->config->get(vdev,
+                                 offsetof(struct virtio_console_config, cols),
+                                 &cols, sizeof(u16));
+               vdev->config->get(vdev,
+                                 offsetof(struct virtio_console_config, rows),
+                                 &rows, sizeof(u16));
+
+               port = find_port_by_id(portdev, 0);
+               set_console_size(port, rows, cols);
+
+               /*
+                * We'll use this way of resizing only for legacy
+                * support.  For newer userspace
+                * (VIRTIO_CONSOLE_F_MULTPORT+), use control messages
+                * to indicate console size changes so that it can be
+                * done per-port.
+                */
+               resize_console(port);
+       }
 }
 
 static int init_vqs(struct ports_device *portdev)
@@ -1237,31 +1579,16 @@ static int init_vqs(struct ports_device *portdev)
        nr_queues = use_multiport(portdev) ? (nr_ports + 1) * 2 : 2;
 
        vqs = kmalloc(nr_queues * sizeof(struct virtqueue *), GFP_KERNEL);
-       if (!vqs) {
-               err = -ENOMEM;
-               goto fail;
-       }
        io_callbacks = kmalloc(nr_queues * sizeof(vq_callback_t *), GFP_KERNEL);
-       if (!io_callbacks) {
-               err = -ENOMEM;
-               goto free_vqs;
-       }
        io_names = kmalloc(nr_queues * sizeof(char *), GFP_KERNEL);
-       if (!io_names) {
-               err = -ENOMEM;
-               goto free_callbacks;
-       }
        portdev->in_vqs = kmalloc(nr_ports * sizeof(struct virtqueue *),
                                  GFP_KERNEL);
-       if (!portdev->in_vqs) {
-               err = -ENOMEM;
-               goto free_names;
-       }
        portdev->out_vqs = kmalloc(nr_ports * sizeof(struct virtqueue *),
                                   GFP_KERNEL);
-       if (!portdev->out_vqs) {
+       if (!vqs || !io_callbacks || !io_names || !portdev->in_vqs ||
+           !portdev->out_vqs) {
                err = -ENOMEM;
-               goto free_invqs;
+               goto free;
        }
 
        /*
@@ -1271,7 +1598,7 @@ static int init_vqs(struct ports_device *portdev)
         */
        j = 0;
        io_callbacks[j] = in_intr;
-       io_callbacks[j + 1] = NULL;
+       io_callbacks[j + 1] = out_intr;
        io_names[j] = "input";
        io_names[j + 1] = "output";
        j += 2;
@@ -1285,7 +1612,7 @@ static int init_vqs(struct ports_device *portdev)
                for (i = 1; i < nr_ports; i++) {
                        j += 2;
                        io_callbacks[j] = in_intr;
-                       io_callbacks[j + 1] = NULL;
+                       io_callbacks[j + 1] = out_intr;
                        io_names[j] = "input";
                        io_names[j + 1] = "output";
                }
@@ -1295,7 +1622,7 @@ static int init_vqs(struct ports_device *portdev)
                                              io_callbacks,
                                              (const char **)io_names);
        if (err)
-               goto free_outvqs;
+               goto free;
 
        j = 0;
        portdev->in_vqs[0] = vqs[0];
@@ -1311,23 +1638,19 @@ static int init_vqs(struct ports_device *portdev)
                        portdev->out_vqs[i] = vqs[j + 1];
                }
        }
-       kfree(io_callbacks);
        kfree(io_names);
+       kfree(io_callbacks);
        kfree(vqs);
 
        return 0;
 
-free_names:
-       kfree(io_names);
-free_callbacks:
-       kfree(io_callbacks);
-free_outvqs:
+free:
        kfree(portdev->out_vqs);
-free_invqs:
        kfree(portdev->in_vqs);
-free_vqs:
+       kfree(io_names);
+       kfree(io_callbacks);
        kfree(vqs);
-fail:
+
        return err;
 }
 
@@ -1335,6 +1658,28 @@ static const struct file_operations portdev_fops = {
        .owner = THIS_MODULE,
 };
 
+static void remove_vqs(struct ports_device *portdev)
+{
+       portdev->vdev->config->del_vqs(portdev->vdev);
+       kfree(portdev->in_vqs);
+       kfree(portdev->out_vqs);
+}
+
+static void remove_controlq_data(struct ports_device *portdev)
+{
+       struct port_buffer *buf;
+       unsigned int len;
+
+       if (!use_multiport(portdev))
+               return;
+
+       while ((buf = virtqueue_get_buf(portdev->c_ivq, &len)))
+               free_buf(buf);
+
+       while ((buf = virtqueue_detach_unused_buf(portdev->c_ivq)))
+               free_buf(buf);
+}
+
 /*
  * Once we're further in boot, we get probed like any other virtio
  * device.
@@ -1348,6 +1693,10 @@ static int __devinit virtcons_probe(struct virtio_device *vdev)
        struct ports_device *portdev;
        int err;
        bool multiport;
+       bool early = early_put_chars != NULL;
+
+       /* Ensure to read early_put_chars now */
+       barrier();
 
        portdev = kmalloc(sizeof(*portdev), GFP_KERNEL);
        if (!portdev) {
@@ -1375,18 +1724,11 @@ static int __devinit virtcons_probe(struct virtio_device *vdev)
 
        multiport = false;
        portdev->config.max_nr_ports = 1;
-       if (virtio_has_feature(vdev, VIRTIO_CONSOLE_F_MULTIPORT)) {
+       if (virtio_config_val(vdev, VIRTIO_CONSOLE_F_MULTIPORT,
+                             offsetof(struct virtio_console_config,
+                                      max_nr_ports),
+                             &portdev->config.max_nr_ports) == 0)
                multiport = true;
-               vdev->features[0] |= 1 << VIRTIO_CONSOLE_F_MULTIPORT;
-
-               vdev->config->get(vdev, offsetof(struct virtio_console_config,
-                                                max_nr_ports),
-                                 &portdev->config.max_nr_ports,
-                                 sizeof(portdev->config.max_nr_ports));
-       }
-
-       /* Let the Host know we support multiple ports.*/
-       vdev->config->finalize_features(vdev);
 
        err = init_vqs(portdev);
        if (err < 0) {
@@ -1418,22 +1760,37 @@ static int __devinit virtcons_probe(struct virtio_device *vdev)
                add_port(portdev, 0);
        }
 
+       spin_lock_irq(&pdrvdata_lock);
+       list_add_tail(&portdev->list, &pdrvdata.portdevs);
+       spin_unlock_irq(&pdrvdata_lock);
+
        __send_control_msg(portdev, VIRTIO_CONSOLE_BAD_ID,
                           VIRTIO_CONSOLE_DEVICE_READY, 1);
+
+       /*
+        * If there was an early virtio console, assume that there are no
+        * other consoles. We need to wait until the hvc_alloc matches the
+        * hvc_instantiate, otherwise tty_open will complain, resulting in
+        * a "Warning: unable to open an initial console" boot failure.
+        * Without multiport this is done in add_port above. With multiport
+        * this might take some host<->guest communication - thus we have to
+        * wait.
+        */
+       if (multiport && early)
+               wait_for_completion(&early_console_added);
+
        return 0;
 
 free_vqs:
-       vdev->config->del_vqs(vdev);
-       kfree(portdev->in_vqs);
-       kfree(portdev->out_vqs);
+       /* The host might want to notify mgmt sw about device add failure */
+       __send_control_msg(portdev, VIRTIO_CONSOLE_BAD_ID,
+                          VIRTIO_CONSOLE_DEVICE_READY, 0);
+       remove_vqs(portdev);
 free_chrdev:
        unregister_chrdev(portdev->chr_major, "virtio-portsdev");
 free:
        kfree(portdev);
 fail:
-       /* The host might want to notify mgmt sw about device add failure */
-       __send_control_msg(portdev, VIRTIO_CONSOLE_BAD_ID,
-                          VIRTIO_CONSOLE_DEVICE_READY, 0);
        return err;
 }
 
@@ -1441,28 +1798,33 @@ static void virtcons_remove(struct virtio_device *vdev)
 {
        struct ports_device *portdev;
        struct port *port, *port2;
-       struct port_buffer *buf;
-       unsigned int len;
 
        portdev = vdev->priv;
 
+       spin_lock_irq(&pdrvdata_lock);
+       list_del(&portdev->list);
+       spin_unlock_irq(&pdrvdata_lock);
+
+       /* Disable interrupts for vqs */
+       vdev->config->reset(vdev);
+       /* Finish up work that's lined up */
        cancel_work_sync(&portdev->control_work);
 
        list_for_each_entry_safe(port, port2, &portdev->ports, list)
-               remove_port(port);
+               unplug_port(port);
 
        unregister_chrdev(portdev->chr_major, "virtio-portsdev");
 
-       while ((buf = virtqueue_get_buf(portdev->c_ivq, &len)))
-               free_buf(buf);
-
-       while ((buf = virtqueue_detach_unused_buf(portdev->c_ivq)))
-               free_buf(buf);
-
-       vdev->config->del_vqs(vdev);
-       kfree(portdev->in_vqs);
-       kfree(portdev->out_vqs);
-
+       /*
+        * When yanking out a device, we immediately lose the
+        * (device-side) queues.  So there's no point in keeping the
+        * guest side around till we drop our final reference.  This
+        * also means that any ports which are in an open state will
+        * have to just stop using the port, as the vqs are going
+        * away.
+        */
+       remove_controlq_data(portdev);
+       remove_vqs(portdev);
        kfree(portdev);
 }
 
@@ -1476,6 +1838,75 @@ static unsigned int features[] = {
        VIRTIO_CONSOLE_F_MULTIPORT,
 };
 
+#ifdef CONFIG_PM
+static int virtcons_freeze(struct virtio_device *vdev)
+{
+       struct ports_device *portdev;
+       struct port *port;
+
+       portdev = vdev->priv;
+
+       vdev->config->reset(vdev);
+
+       virtqueue_disable_cb(portdev->c_ivq);
+       cancel_work_sync(&portdev->control_work);
+       /*
+        * Once more: if control_work_handler() was running, it would
+        * enable the cb as the last step.
+        */
+       virtqueue_disable_cb(portdev->c_ivq);
+       remove_controlq_data(portdev);
+
+       list_for_each_entry(port, &portdev->ports, list) {
+               virtqueue_disable_cb(port->in_vq);
+               virtqueue_disable_cb(port->out_vq);
+               /*
+                * We'll ask the host later if the new invocation has
+                * the port opened or closed.
+                */
+               port->host_connected = false;
+               remove_port_data(port);
+       }
+       remove_vqs(portdev);
+
+       return 0;
+}
+
+static int virtcons_restore(struct virtio_device *vdev)
+{
+       struct ports_device *portdev;
+       struct port *port;
+       int ret;
+
+       portdev = vdev->priv;
+
+       ret = init_vqs(portdev);
+       if (ret)
+               return ret;
+
+       if (use_multiport(portdev))
+               fill_queue(portdev->c_ivq, &portdev->cvq_lock);
+
+       list_for_each_entry(port, &portdev->ports, list) {
+               port->in_vq = portdev->in_vqs[port->id];
+               port->out_vq = portdev->out_vqs[port->id];
+
+               fill_queue(port->in_vq, &port->inbuf_lock);
+
+               /* Get port open/close status on the host */
+               send_control_msg(port, VIRTIO_CONSOLE_PORT_READY, 1);
+
+               /*
+                * If a port was open at the time of suspending, we
+                * have to let the host know that it's still open.
+                */
+               if (port->guest_connected)
+                       send_control_msg(port, VIRTIO_CONSOLE_PORT_OPEN, 1);
+       }
+       return 0;
+}
+#endif
+
 static struct virtio_driver virtio_console = {
        .feature_table = features,
        .feature_table_size = ARRAY_SIZE(features),
@@ -1485,6 +1916,10 @@ static struct virtio_driver virtio_console = {
        .probe =        virtcons_probe,
        .remove =       virtcons_remove,
        .config_changed = config_intr,
+#ifdef CONFIG_PM
+       .freeze =       virtcons_freeze,
+       .restore =      virtcons_restore,
+#endif
 };
 
 static int __init init(void)
@@ -1504,6 +1939,7 @@ static int __init init(void)
                           PTR_ERR(pdrvdata.debugfs_dir));
        }
        INIT_LIST_HEAD(&pdrvdata.consoles);
+       INIT_LIST_HEAD(&pdrvdata.portdevs);
 
        return register_virtio_driver(&virtio_console);
 }