Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / watchdog / watchdog_dev.c
diff --git a/kernel/drivers/watchdog/watchdog_dev.c b/kernel/drivers/watchdog/watchdog_dev.c
new file mode 100644 (file)
index 0000000..6aaefba
--- /dev/null
@@ -0,0 +1,604 @@
+/*
+ *     watchdog_dev.c
+ *
+ *     (c) Copyright 2008-2011 Alan Cox <alan@lxorguk.ukuu.org.uk>,
+ *                                             All Rights Reserved.
+ *
+ *     (c) Copyright 2008-2011 Wim Van Sebroeck <wim@iguana.be>.
+ *
+ *
+ *     This source code is part of the generic code that can be used
+ *     by all the watchdog timer drivers.
+ *
+ *     This part of the generic code takes care of the following
+ *     misc device: /dev/watchdog.
+ *
+ *     Based on source code of the following authors:
+ *       Matt Domsch <Matt_Domsch@dell.com>,
+ *       Rob Radez <rob@osinvestor.com>,
+ *       Rusty Lynch <rusty@linux.co.intel.com>
+ *       Satyam Sharma <satyam@infradead.org>
+ *       Randy Dunlap <randy.dunlap@oracle.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 the Free Software Foundation; either version
+ *     2 of the License, or (at your option) any later version.
+ *
+ *     Neither Alan Cox, CymruNet Ltd., Wim Van Sebroeck nor Iguana vzw.
+ *     admit liability nor provide warranty for any of this software.
+ *     This material is provided "AS-IS" and at no charge.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/module.h>      /* For module stuff/... */
+#include <linux/types.h>       /* For standard types (like size_t) */
+#include <linux/errno.h>       /* For the -ENODEV/... values */
+#include <linux/kernel.h>      /* For printk/panic/... */
+#include <linux/fs.h>          /* For file operations */
+#include <linux/watchdog.h>    /* For watchdog specific items */
+#include <linux/miscdevice.h>  /* For handling misc devices */
+#include <linux/init.h>                /* For __init/__exit/... */
+#include <linux/uaccess.h>     /* For copy_to_user/put_user/... */
+
+#include "watchdog_core.h"
+
+/* the dev_t structure to store the dynamically allocated watchdog devices */
+static dev_t watchdog_devt;
+/* the watchdog device behind /dev/watchdog */
+static struct watchdog_device *old_wdd;
+
+/*
+ *     watchdog_ping: ping the watchdog.
+ *     @wddev: the watchdog device to ping
+ *
+ *     If the watchdog has no own ping operation then it needs to be
+ *     restarted via the start operation. This wrapper function does
+ *     exactly that.
+ *     We only ping when the watchdog device is running.
+ */
+
+static int watchdog_ping(struct watchdog_device *wddev)
+{
+       int err = 0;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_ping;
+       }
+
+       if (!watchdog_active(wddev))
+               goto out_ping;
+
+       if (wddev->ops->ping)
+               err = wddev->ops->ping(wddev);  /* ping the watchdog */
+       else
+               err = wddev->ops->start(wddev); /* restart watchdog */
+
+out_ping:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_start: wrapper to start the watchdog.
+ *     @wddev: the watchdog device to start
+ *
+ *     Start the watchdog if it is not active and mark it active.
+ *     This function returns zero on success or a negative errno code for
+ *     failure.
+ */
+
+static int watchdog_start(struct watchdog_device *wddev)
+{
+       int err = 0;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_start;
+       }
+
+       if (watchdog_active(wddev))
+               goto out_start;
+
+       err = wddev->ops->start(wddev);
+       if (err == 0)
+               set_bit(WDOG_ACTIVE, &wddev->status);
+
+out_start:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_stop: wrapper to stop the watchdog.
+ *     @wddev: the watchdog device to stop
+ *
+ *     Stop the watchdog if it is still active and unmark it active.
+ *     This function returns zero on success or a negative errno code for
+ *     failure.
+ *     If the 'nowayout' feature was set, the watchdog cannot be stopped.
+ */
+
+static int watchdog_stop(struct watchdog_device *wddev)
+{
+       int err = 0;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_stop;
+       }
+
+       if (!watchdog_active(wddev))
+               goto out_stop;
+
+       if (test_bit(WDOG_NO_WAY_OUT, &wddev->status)) {
+               dev_info(wddev->dev, "nowayout prevents watchdog being stopped!\n");
+               err = -EBUSY;
+               goto out_stop;
+       }
+
+       err = wddev->ops->stop(wddev);
+       if (err == 0)
+               clear_bit(WDOG_ACTIVE, &wddev->status);
+
+out_stop:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_get_status: wrapper to get the watchdog status
+ *     @wddev: the watchdog device to get the status from
+ *     @status: the status of the watchdog device
+ *
+ *     Get the watchdog's status flags.
+ */
+
+static int watchdog_get_status(struct watchdog_device *wddev,
+                                                       unsigned int *status)
+{
+       int err = 0;
+
+       *status = 0;
+       if (!wddev->ops->status)
+               return -EOPNOTSUPP;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_status;
+       }
+
+       *status = wddev->ops->status(wddev);
+
+out_status:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_set_timeout: set the watchdog timer timeout
+ *     @wddev: the watchdog device to set the timeout for
+ *     @timeout: timeout to set in seconds
+ */
+
+static int watchdog_set_timeout(struct watchdog_device *wddev,
+                                                       unsigned int timeout)
+{
+       int err;
+
+       if ((wddev->ops->set_timeout == NULL) ||
+           !(wddev->info->options & WDIOF_SETTIMEOUT))
+               return -EOPNOTSUPP;
+
+       if (watchdog_timeout_invalid(wddev, timeout))
+               return -EINVAL;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_timeout;
+       }
+
+       err = wddev->ops->set_timeout(wddev, timeout);
+
+out_timeout:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_get_timeleft: wrapper to get the time left before a reboot
+ *     @wddev: the watchdog device to get the remaining time from
+ *     @timeleft: the time that's left
+ *
+ *     Get the time before a watchdog will reboot (if not pinged).
+ */
+
+static int watchdog_get_timeleft(struct watchdog_device *wddev,
+                                                       unsigned int *timeleft)
+{
+       int err = 0;
+
+       *timeleft = 0;
+       if (!wddev->ops->get_timeleft)
+               return -EOPNOTSUPP;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_timeleft;
+       }
+
+       *timeleft = wddev->ops->get_timeleft(wddev);
+
+out_timeleft:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_ioctl_op: call the watchdog drivers ioctl op if defined
+ *     @wddev: the watchdog device to do the ioctl on
+ *     @cmd: watchdog command
+ *     @arg: argument pointer
+ */
+
+static int watchdog_ioctl_op(struct watchdog_device *wddev, unsigned int cmd,
+                                                       unsigned long arg)
+{
+       int err;
+
+       if (!wddev->ops->ioctl)
+               return -ENOIOCTLCMD;
+
+       mutex_lock(&wddev->lock);
+
+       if (test_bit(WDOG_UNREGISTERED, &wddev->status)) {
+               err = -ENODEV;
+               goto out_ioctl;
+       }
+
+       err = wddev->ops->ioctl(wddev, cmd, arg);
+
+out_ioctl:
+       mutex_unlock(&wddev->lock);
+       return err;
+}
+
+/*
+ *     watchdog_write: writes to the watchdog.
+ *     @file: file from VFS
+ *     @data: user address of data
+ *     @len: length of data
+ *     @ppos: pointer to the file offset
+ *
+ *     A write to a watchdog device is defined as a keepalive ping.
+ *     Writing the magic 'V' sequence allows the next close to turn
+ *     off the watchdog (if 'nowayout' is not set).
+ */
+
+static ssize_t watchdog_write(struct file *file, const char __user *data,
+                                               size_t len, loff_t *ppos)
+{
+       struct watchdog_device *wdd = file->private_data;
+       size_t i;
+       char c;
+
+       if (len == 0)
+               return 0;
+
+       /*
+        * Note: just in case someone wrote the magic character
+        * five months ago...
+        */
+       clear_bit(WDOG_ALLOW_RELEASE, &wdd->status);
+
+       /* scan to see whether or not we got the magic character */
+       for (i = 0; i != len; i++) {
+               if (get_user(c, data + i))
+                       return -EFAULT;
+               if (c == 'V')
+                       set_bit(WDOG_ALLOW_RELEASE, &wdd->status);
+       }
+
+       /* someone wrote to us, so we send the watchdog a keepalive ping */
+       watchdog_ping(wdd);
+
+       return len;
+}
+
+/*
+ *     watchdog_ioctl: handle the different ioctl's for the watchdog device.
+ *     @file: file handle to the device
+ *     @cmd: watchdog command
+ *     @arg: argument pointer
+ *
+ *     The watchdog API defines a common set of functions for all watchdogs
+ *     according to their available features.
+ */
+
+static long watchdog_ioctl(struct file *file, unsigned int cmd,
+                                                       unsigned long arg)
+{
+       struct watchdog_device *wdd = file->private_data;
+       void __user *argp = (void __user *)arg;
+       int __user *p = argp;
+       unsigned int val;
+       int err;
+
+       err = watchdog_ioctl_op(wdd, cmd, arg);
+       if (err != -ENOIOCTLCMD)
+               return err;
+
+       switch (cmd) {
+       case WDIOC_GETSUPPORT:
+               return copy_to_user(argp, wdd->info,
+                       sizeof(struct watchdog_info)) ? -EFAULT : 0;
+       case WDIOC_GETSTATUS:
+               err = watchdog_get_status(wdd, &val);
+               if (err == -ENODEV)
+                       return err;
+               return put_user(val, p);
+       case WDIOC_GETBOOTSTATUS:
+               return put_user(wdd->bootstatus, p);
+       case WDIOC_SETOPTIONS:
+               if (get_user(val, p))
+                       return -EFAULT;
+               if (val & WDIOS_DISABLECARD) {
+                       err = watchdog_stop(wdd);
+                       if (err < 0)
+                               return err;
+               }
+               if (val & WDIOS_ENABLECARD) {
+                       err = watchdog_start(wdd);
+                       if (err < 0)
+                               return err;
+               }
+               return 0;
+       case WDIOC_KEEPALIVE:
+               if (!(wdd->info->options & WDIOF_KEEPALIVEPING))
+                       return -EOPNOTSUPP;
+               watchdog_ping(wdd);
+               return 0;
+       case WDIOC_SETTIMEOUT:
+               if (get_user(val, p))
+                       return -EFAULT;
+               err = watchdog_set_timeout(wdd, val);
+               if (err < 0)
+                       return err;
+               /* If the watchdog is active then we send a keepalive ping
+                * to make sure that the watchdog keep's running (and if
+                * possible that it takes the new timeout) */
+               watchdog_ping(wdd);
+               /* Fall */
+       case WDIOC_GETTIMEOUT:
+               /* timeout == 0 means that we don't know the timeout */
+               if (wdd->timeout == 0)
+                       return -EOPNOTSUPP;
+               return put_user(wdd->timeout, p);
+       case WDIOC_GETTIMELEFT:
+               err = watchdog_get_timeleft(wdd, &val);
+               if (err)
+                       return err;
+               return put_user(val, p);
+       default:
+               return -ENOTTY;
+       }
+}
+
+/*
+ *     watchdog_open: open the /dev/watchdog* devices.
+ *     @inode: inode of device
+ *     @file: file handle to device
+ *
+ *     When the /dev/watchdog* device gets opened, we start the watchdog.
+ *     Watch out: the /dev/watchdog device is single open, so we make sure
+ *     it can only be opened once.
+ */
+
+static int watchdog_open(struct inode *inode, struct file *file)
+{
+       int err = -EBUSY;
+       struct watchdog_device *wdd;
+
+       /* Get the corresponding watchdog device */
+       if (imajor(inode) == MISC_MAJOR)
+               wdd = old_wdd;
+       else
+               wdd = container_of(inode->i_cdev, struct watchdog_device, cdev);
+
+       /* the watchdog is single open! */
+       if (test_and_set_bit(WDOG_DEV_OPEN, &wdd->status))
+               return -EBUSY;
+
+       /*
+        * If the /dev/watchdog device is open, we don't want the module
+        * to be unloaded.
+        */
+       if (!try_module_get(wdd->ops->owner))
+               goto out;
+
+       err = watchdog_start(wdd);
+       if (err < 0)
+               goto out_mod;
+
+       file->private_data = wdd;
+
+       if (wdd->ops->ref)
+               wdd->ops->ref(wdd);
+
+       /* dev/watchdog is a virtual (and thus non-seekable) filesystem */
+       return nonseekable_open(inode, file);
+
+out_mod:
+       module_put(wdd->ops->owner);
+out:
+       clear_bit(WDOG_DEV_OPEN, &wdd->status);
+       return err;
+}
+
+/*
+ *     watchdog_release: release the watchdog device.
+ *     @inode: inode of device
+ *     @file: file handle to device
+ *
+ *     This is the code for when /dev/watchdog gets closed. We will only
+ *     stop the watchdog when we have received the magic char (and nowayout
+ *     was not set), else the watchdog will keep running.
+ */
+
+static int watchdog_release(struct inode *inode, struct file *file)
+{
+       struct watchdog_device *wdd = file->private_data;
+       int err = -EBUSY;
+
+       /*
+        * We only stop the watchdog if we received the magic character
+        * or if WDIOF_MAGICCLOSE is not set. If nowayout was set then
+        * watchdog_stop will fail.
+        */
+       if (!test_bit(WDOG_ACTIVE, &wdd->status))
+               err = 0;
+       else if (test_and_clear_bit(WDOG_ALLOW_RELEASE, &wdd->status) ||
+                !(wdd->info->options & WDIOF_MAGICCLOSE))
+               err = watchdog_stop(wdd);
+
+       /* If the watchdog was not stopped, send a keepalive ping */
+       if (err < 0) {
+               mutex_lock(&wdd->lock);
+               if (!test_bit(WDOG_UNREGISTERED, &wdd->status))
+                       dev_crit(wdd->dev, "watchdog did not stop!\n");
+               mutex_unlock(&wdd->lock);
+               watchdog_ping(wdd);
+       }
+
+       /* Allow the owner module to be unloaded again */
+       module_put(wdd->ops->owner);
+
+       /* make sure that /dev/watchdog can be re-opened */
+       clear_bit(WDOG_DEV_OPEN, &wdd->status);
+
+       /* Note wdd may be gone after this, do not use after this! */
+       if (wdd->ops->unref)
+               wdd->ops->unref(wdd);
+
+       return 0;
+}
+
+static const struct file_operations watchdog_fops = {
+       .owner          = THIS_MODULE,
+       .write          = watchdog_write,
+       .unlocked_ioctl = watchdog_ioctl,
+       .open           = watchdog_open,
+       .release        = watchdog_release,
+};
+
+static struct miscdevice watchdog_miscdev = {
+       .minor          = WATCHDOG_MINOR,
+       .name           = "watchdog",
+       .fops           = &watchdog_fops,
+};
+
+/*
+ *     watchdog_dev_register: register a watchdog device
+ *     @watchdog: watchdog device
+ *
+ *     Register a watchdog device including handling the legacy
+ *     /dev/watchdog node. /dev/watchdog is actually a miscdevice and
+ *     thus we set it up like that.
+ */
+
+int watchdog_dev_register(struct watchdog_device *watchdog)
+{
+       int err, devno;
+
+       if (watchdog->id == 0) {
+               old_wdd = watchdog;
+               watchdog_miscdev.parent = watchdog->parent;
+               err = misc_register(&watchdog_miscdev);
+               if (err != 0) {
+                       pr_err("%s: cannot register miscdev on minor=%d (err=%d).\n",
+                               watchdog->info->identity, WATCHDOG_MINOR, err);
+                       if (err == -EBUSY)
+                               pr_err("%s: a legacy watchdog module is probably present.\n",
+                                       watchdog->info->identity);
+                       old_wdd = NULL;
+                       return err;
+               }
+       }
+
+       /* Fill in the data structures */
+       devno = MKDEV(MAJOR(watchdog_devt), watchdog->id);
+       cdev_init(&watchdog->cdev, &watchdog_fops);
+       watchdog->cdev.owner = watchdog->ops->owner;
+
+       /* Add the device */
+       err  = cdev_add(&watchdog->cdev, devno, 1);
+       if (err) {
+               pr_err("watchdog%d unable to add device %d:%d\n",
+                       watchdog->id,  MAJOR(watchdog_devt), watchdog->id);
+               if (watchdog->id == 0) {
+                       misc_deregister(&watchdog_miscdev);
+                       old_wdd = NULL;
+               }
+       }
+       return err;
+}
+
+/*
+ *     watchdog_dev_unregister: unregister a watchdog device
+ *     @watchdog: watchdog device
+ *
+ *     Unregister the watchdog and if needed the legacy /dev/watchdog device.
+ */
+
+int watchdog_dev_unregister(struct watchdog_device *watchdog)
+{
+       mutex_lock(&watchdog->lock);
+       set_bit(WDOG_UNREGISTERED, &watchdog->status);
+       mutex_unlock(&watchdog->lock);
+
+       cdev_del(&watchdog->cdev);
+       if (watchdog->id == 0) {
+               misc_deregister(&watchdog_miscdev);
+               old_wdd = NULL;
+       }
+       return 0;
+}
+
+/*
+ *     watchdog_dev_init: init dev part of watchdog core
+ *
+ *     Allocate a range of chardev nodes to use for watchdog devices
+ */
+
+int __init watchdog_dev_init(void)
+{
+       int err = alloc_chrdev_region(&watchdog_devt, 0, MAX_DOGS, "watchdog");
+       if (err < 0)
+               pr_err("watchdog: unable to allocate char dev region\n");
+       return err;
+}
+
+/*
+ *     watchdog_dev_exit: exit dev part of watchdog core
+ *
+ *     Release the range of chardev nodes used for watchdog devices
+ */
+
+void __exit watchdog_dev_exit(void)
+{
+       unregister_chrdev_region(watchdog_devt, MAX_DOGS);
+}