sched: adjust scheduler cpu power for stolen time

[linux-2.6/libata-dev.git] / drivers / watchdog / cpwd.c

/* cpwd.c - driver implementation for hardware watchdog
 * timers found on Sun Microsystems CP1400 and CP1500 boards.
 *
 * This device supports both the generic Linux watchdog
 * interface and Solaris-compatible ioctls as best it is
 * able.
 *
 * NOTE:        CP1400 systems appear to have a defective intr_mask
 *                      register on the PLD, preventing the disabling of
 *                      timer interrupts.  We use a timer to periodically
 *                      reset 'stopped' watchdogs on affected platforms.
 *
 * Copyright (c) 2000 Eric Brower (ebrower@usa.net)
 * Copyright (C) 2008 David S. Miller <davem@davemloft.net>
 */

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/errno.h>
#include <linux/major.h>
#include <linux/init.h>
#include <linux/miscdevice.h>
#include <linux/interrupt.h>
#include <linux/ioport.h>
#include <linux/timer.h>
#include <linux/slab.h>
#include <linux/mutex.h>
#include <linux/io.h>
#include <linux/of.h>
#include <linux/of_device.h>
#include <linux/uaccess.h>

#include <asm/irq.h>
#include <asm/watchdog.h>

#define DRIVER_NAME     "cpwd"
#define PFX             DRIVER_NAME ": "

#define WD_OBPNAME      "watchdog"
#define WD_BADMODEL     "SUNW,501-5336"
#define WD_BTIMEOUT     (jiffies + (HZ * 1000))
#define WD_BLIMIT       0xFFFF

#define WD0_MINOR       212
#define WD1_MINOR       213
#define WD2_MINOR       214

/* Internal driver definitions.  */
#define WD0_ID                  0
#define WD1_ID                  1
#define WD2_ID                  2
#define WD_NUMDEVS              3

#define WD_INTR_OFF             0
#define WD_INTR_ON              1

#define WD_STAT_INIT    0x01    /* Watchdog timer is initialized        */
#define WD_STAT_BSTOP   0x02    /* Watchdog timer is brokenstopped      */
#define WD_STAT_SVCD    0x04    /* Watchdog interrupt occurred          */

/* Register value definitions
 */
#define WD0_INTR_MASK   0x01    /* Watchdog device interrupt masks      */
#define WD1_INTR_MASK   0x02
#define WD2_INTR_MASK   0x04

#define WD_S_RUNNING    0x01    /* Watchdog device status running       */
#define WD_S_EXPIRED    0x02    /* Watchdog device status expired       */

struct cpwd {
        void __iomem    *regs;
        spinlock_t      lock;

        unsigned int    irq;

        unsigned long   timeout;
        bool            enabled;
        bool            reboot;
        bool            broken;
        bool            initialized;

        struct {
                struct miscdevice       misc;
                void __iomem            *regs;
                u8                      intr_mask;
                u8                      runstatus;
                u16                     timeout;
        } devs[WD_NUMDEVS];
};

static DEFINE_MUTEX(cpwd_mutex);
static struct cpwd *cpwd_device;

/* Sun uses Altera PLD EPF8820ATC144-4
 * providing three hardware watchdogs:
 *
 * 1) RIC - sends an interrupt when triggered
 * 2) XIR - asserts XIR_B_RESET when triggered, resets CPU
 * 3) POR - asserts POR_B_RESET when triggered, resets CPU, backplane, board
 *
 *** Timer register block definition (struct wd_timer_regblk)
 *
 * dcntr and limit registers (halfword access):
 * -------------------
 * | 15 | ...| 1 | 0 |
 * -------------------
 * |-  counter val  -|
 * -------------------
 * dcntr -      Current 16-bit downcounter value.
 *                      When downcounter reaches '0' watchdog expires.
 *                      Reading this register resets downcounter with
 *                      'limit' value.
 * limit -      16-bit countdown value in 1/10th second increments.
 *                      Writing this register begins countdown with input value.
 *                      Reading from this register does not affect counter.
 * NOTES:       After watchdog reset, dcntr and limit contain '1'
 *
 * status register (byte access):
 * ---------------------------
 * | 7 | ... | 2 |  1  |  0  |
 * --------------+------------
 * |-   UNUSED  -| EXP | RUN |
 * ---------------------------
 * status-      Bit 0 - Watchdog is running
 *                      Bit 1 - Watchdog has expired
 *
 *** PLD register block definition (struct wd_pld_regblk)
 *
 * intr_mask register (byte access):
 * ---------------------------------
 * | 7 | ... | 3 |  2  |  1  |  0  |
 * +-------------+------------------
 * |-   UNUSED  -| WD3 | WD2 | WD1 |
 * ---------------------------------
 * WD3 -  1 == Interrupt disabled for watchdog 3
 * WD2 -  1 == Interrupt disabled for watchdog 2
 * WD1 -  1 == Interrupt disabled for watchdog 1
 *
 * pld_status register (byte access):
 * UNKNOWN, MAGICAL MYSTERY REGISTER
 *
 */
#define WD_TIMER_REGSZ  16
#define WD0_OFF         0
#define WD1_OFF         (WD_TIMER_REGSZ * 1)
#define WD2_OFF         (WD_TIMER_REGSZ * 2)
#define PLD_OFF         (WD_TIMER_REGSZ * 3)

#define WD_DCNTR        0x00
#define WD_LIMIT        0x04
#define WD_STATUS       0x08

#define PLD_IMASK       (PLD_OFF + 0x00)
#define PLD_STATUS      (PLD_OFF + 0x04)

static struct timer_list cpwd_timer;

static int wd0_timeout;
static int wd1_timeout;
static int wd2_timeout;

module_param(wd0_timeout, int, 0);
MODULE_PARM_DESC(wd0_timeout, "Default watchdog0 timeout in 1/10secs");
module_param(wd1_timeout, int, 0);
MODULE_PARM_DESC(wd1_timeout, "Default watchdog1 timeout in 1/10secs");
module_param(wd2_timeout, int, 0);
MODULE_PARM_DESC(wd2_timeout, "Default watchdog2 timeout in 1/10secs");

MODULE_AUTHOR("Eric Brower <ebrower@usa.net>");
MODULE_DESCRIPTION("Hardware watchdog driver for Sun Microsystems CP1400/1500");
MODULE_LICENSE("GPL");
MODULE_SUPPORTED_DEVICE("watchdog");

static void cpwd_writew(u16 val, void __iomem *addr)
{
        writew(cpu_to_le16(val), addr);
}
static u16 cpwd_readw(void __iomem *addr)
{
        u16 val = readw(addr);

        return le16_to_cpu(val);
}

static void cpwd_writeb(u8 val, void __iomem *addr)
{
        writeb(val, addr);
}

static u8 cpwd_readb(void __iomem *addr)
{
        return readb(addr);
}

/* Enable or disable watchdog interrupts
 * Because of the CP1400 defect this should only be
 * called during initialzation or by wd_[start|stop]timer()
 *
 * index        - sub-device index, or -1 for 'all'
 * enable       - non-zero to enable interrupts, zero to disable
 */
static void cpwd_toggleintr(struct cpwd *p, int index, int enable)
{
        unsigned char curregs = cpwd_readb(p->regs + PLD_IMASK);
        unsigned char setregs =
                (index == -1) ?
                (WD0_INTR_MASK | WD1_INTR_MASK | WD2_INTR_MASK) :
                (p->devs[index].intr_mask);

        if (enable == WD_INTR_ON)
                curregs &= ~setregs;
        else
                curregs |= setregs;

        cpwd_writeb(curregs, p->regs + PLD_IMASK);
}

/* Restarts timer with maximum limit value and
 * does not unset 'brokenstop' value.
 */
static void cpwd_resetbrokentimer(struct cpwd *p, int index)
{
        cpwd_toggleintr(p, index, WD_INTR_ON);
        cpwd_writew(WD_BLIMIT, p->devs[index].regs + WD_LIMIT);
}

/* Timer method called to reset stopped watchdogs--
 * because of the PLD bug on CP1400, we cannot mask
 * interrupts within the PLD so me must continually
 * reset the timers ad infinitum.
 */
static void cpwd_brokentimer(unsigned long data)
{
        struct cpwd *p = (struct cpwd *) data;
        int id, tripped = 0;

        /* kill a running timer instance, in case we
         * were called directly instead of by kernel timer
         */
        if (timer_pending(&cpwd_timer))
                del_timer(&cpwd_timer);

        for (id = 0; id < WD_NUMDEVS; id++) {
                if (p->devs[id].runstatus & WD_STAT_BSTOP) {
                        ++tripped;
                        cpwd_resetbrokentimer(p, id);
                }
        }

        if (tripped) {
                /* there is at least one timer brokenstopped-- reschedule */
                cpwd_timer.expires = WD_BTIMEOUT;
                add_timer(&cpwd_timer);
        }
}

/* Reset countdown timer with 'limit' value and continue countdown.
 * This will not start a stopped timer.
 */
static void cpwd_pingtimer(struct cpwd *p, int index)
{
        if (cpwd_readb(p->devs[index].regs + WD_STATUS) & WD_S_RUNNING)
                cpwd_readw(p->devs[index].regs + WD_DCNTR);
}

/* Stop a running watchdog timer-- the timer actually keeps
 * running, but the interrupt is masked so that no action is
 * taken upon expiration.
 */
static void cpwd_stoptimer(struct cpwd *p, int index)
{
        if (cpwd_readb(p->devs[index].regs + WD_STATUS) & WD_S_RUNNING) {
                cpwd_toggleintr(p, index, WD_INTR_OFF);

                if (p->broken) {
                        p->devs[index].runstatus |= WD_STAT_BSTOP;
                        cpwd_brokentimer((unsigned long) p);
                }
        }
}

/* Start a watchdog timer with the specified limit value
 * If the watchdog is running, it will be restarted with
 * the provided limit value.
 *
 * This function will enable interrupts on the specified
 * watchdog.
 */
static void cpwd_starttimer(struct cpwd *p, int index)
{
        if (p->broken)
                p->devs[index].runstatus &= ~WD_STAT_BSTOP;

        p->devs[index].runstatus &= ~WD_STAT_SVCD;

        cpwd_writew(p->devs[index].timeout, p->devs[index].regs + WD_LIMIT);
        cpwd_toggleintr(p, index, WD_INTR_ON);
}

static int cpwd_getstatus(struct cpwd *p, int index)
{
        unsigned char stat = cpwd_readb(p->devs[index].regs + WD_STATUS);
        unsigned char intr = cpwd_readb(p->devs[index].regs + PLD_IMASK);
        unsigned char ret  = WD_STOPPED;

        /* determine STOPPED */
        if (!stat)
                return ret;

        /* determine EXPIRED vs FREERUN vs RUNNING */
        else if (WD_S_EXPIRED & stat) {
                ret = WD_EXPIRED;
        } else if (WD_S_RUNNING & stat) {
                if (intr & p->devs[index].intr_mask) {
                        ret = WD_FREERUN;
                } else {
                        /* Fudge WD_EXPIRED status for defective CP1400--
                         * IF timer is running
                         *      AND brokenstop is set
                         *      AND an interrupt has been serviced
                         * we are WD_EXPIRED.
                         *
                         * IF timer is running
                         *      AND brokenstop is set
                         *      AND no interrupt has been serviced
                         * we are WD_FREERUN.
                         */
                        if (p->broken &&
                            (p->devs[index].runstatus & WD_STAT_BSTOP)) {
                                if (p->devs[index].runstatus & WD_STAT_SVCD) {
                                        ret = WD_EXPIRED;
                                } else {
                                        /* we could as well pretend
                                         * we are expired */
                                        ret = WD_FREERUN;
                                }
                        } else {
                                ret = WD_RUNNING;
                        }
                }
        }

        /* determine SERVICED */
        if (p->devs[index].runstatus & WD_STAT_SVCD)
                ret |= WD_SERVICED;

        return ret;
}

static irqreturn_t cpwd_interrupt(int irq, void *dev_id)
{
        struct cpwd *p = dev_id;

        /* Only WD0 will interrupt-- others are NMI and we won't
         * see them here....
         */
        spin_lock_irq(&p->lock);

        cpwd_stoptimer(p, WD0_ID);
        p->devs[WD0_ID].runstatus |=  WD_STAT_SVCD;

        spin_unlock_irq(&p->lock);

        return IRQ_HANDLED;
}

static int cpwd_open(struct inode *inode, struct file *f)
{
        struct cpwd *p = cpwd_device;

        mutex_lock(&cpwd_mutex);
        switch (iminor(inode)) {
        case WD0_MINOR:
        case WD1_MINOR:
        case WD2_MINOR:
                break;

        default:
                mutex_unlock(&cpwd_mutex);
                return -ENODEV;
        }

        /* Register IRQ on first open of device */
        if (!p->initialized) {
                if (request_irq(p->irq, &cpwd_interrupt,
                                IRQF_SHARED, DRIVER_NAME, p)) {
                        printk(KERN_ERR PFX "Cannot register IRQ %d\n",
                                p->irq);
                        mutex_unlock(&cpwd_mutex);
                        return -EBUSY;
                }
                p->initialized = true;
        }

        mutex_unlock(&cpwd_mutex);

        return nonseekable_open(inode, f);
}

static int cpwd_release(struct inode *inode, struct file *file)
{
        return 0;
}

static long cpwd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
        static const struct watchdog_info info = {
                .options                = WDIOF_SETTIMEOUT,
                .firmware_version       = 1,
                .identity               = DRIVER_NAME,
        };
        void __user *argp = (void __user *)arg;
        struct inode *inode = file->f_path.dentry->d_inode;
        int index = iminor(inode) - WD0_MINOR;
        struct cpwd *p = cpwd_device;
        int setopt = 0;

        switch (cmd) {
        /* Generic Linux IOCTLs */
        case WDIOC_GETSUPPORT:
                if (copy_to_user(argp, &info, sizeof(struct watchdog_info)))
                        return -EFAULT;
                break;

        case WDIOC_GETSTATUS:
        case WDIOC_GETBOOTSTATUS:
                if (put_user(0, (int __user *)argp))
                        return -EFAULT;
                break;

        case WDIOC_KEEPALIVE:
                cpwd_pingtimer(p, index);
                break;

        case WDIOC_SETOPTIONS:
                if (copy_from_user(&setopt, argp, sizeof(unsigned int)))
                        return -EFAULT;

                if (setopt & WDIOS_DISABLECARD) {
                        if (p->enabled)
                                return -EINVAL;
                        cpwd_stoptimer(p, index);
                } else if (setopt & WDIOS_ENABLECARD) {
                        cpwd_starttimer(p, index);
                } else {
                        return -EINVAL;
                }
                break;

        /* Solaris-compatible IOCTLs */
        case WIOCGSTAT:
                setopt = cpwd_getstatus(p, index);
                if (copy_to_user(argp, &setopt, sizeof(unsigned int)))
                        return -EFAULT;
                break;

        case WIOCSTART:
                cpwd_starttimer(p, index);
                break;

        case WIOCSTOP:
                if (p->enabled)
                        return -EINVAL;

                cpwd_stoptimer(p, index);
                break;

        default:
                return -EINVAL;
        }

        return 0;
}

static long cpwd_compat_ioctl(struct file *file, unsigned int cmd,
                              unsigned long arg)
{
        int rval = -ENOIOCTLCMD;

        switch (cmd) {
        /* solaris ioctls are specific to this driver */
        case WIOCSTART:
        case WIOCSTOP:
        case WIOCGSTAT:
                mutex_lock(&cpwd_mutex);
                rval = cpwd_ioctl(file, cmd, arg);
                mutex_unlock(&cpwd_mutex);
                break;

        /* everything else is handled by the generic compat layer */
        default:
                break;
        }

        return rval;
}

static ssize_t cpwd_write(struct file *file, const char __user *buf,
                          size_t count, loff_t *ppos)
{
        struct inode *inode = file->f_path.dentry->d_inode;
        struct cpwd *p = cpwd_device;
        int index = iminor(inode);

        if (count) {
                cpwd_pingtimer(p, index);
                return 1;
        }

        return 0;
}

static ssize_t cpwd_read(struct file *file, char __user *buffer,
                         size_t count, loff_t *ppos)
{
        return -EINVAL;
}

static const struct file_operations cpwd_fops = {
        .owner =                THIS_MODULE,
        .unlocked_ioctl =       cpwd_ioctl,
        .compat_ioctl =         cpwd_compat_ioctl,
        .open =                 cpwd_open,
        .write =                cpwd_write,
        .read =                 cpwd_read,
        .release =              cpwd_release,
        .llseek =               no_llseek,
};

static int __devinit cpwd_probe(struct platform_device *op)
{
        struct device_node *options;
        const char *str_prop;
        const void *prop_val;
        int i, err = -EINVAL;
        struct cpwd *p;

        if (cpwd_device)
                return -EINVAL;

        p = kzalloc(sizeof(*p), GFP_KERNEL);
        err = -ENOMEM;
        if (!p) {
                printk(KERN_ERR PFX "Unable to allocate struct cpwd.\n");
                goto out;
        }

        p->irq = op->archdata.irqs[0];

        spin_lock_init(&p->lock);

        p->regs = of_ioremap(&op->resource[0], 0,
                             4 * WD_TIMER_REGSZ, DRIVER_NAME);
        if (!p->regs) {
                printk(KERN_ERR PFX "Unable to map registers.\n");
                goto out_free;
        }

        options = of_find_node_by_path("/options");
        err = -ENODEV;
        if (!options) {
                printk(KERN_ERR PFX "Unable to find /options node.\n");
                goto out_iounmap;
        }

        prop_val = of_get_property(options, "watchdog-enable?", NULL);
        p->enabled = (prop_val ? true : false);

        prop_val = of_get_property(options, "watchdog-reboot?", NULL);
        p->reboot = (prop_val ? true : false);

        str_prop = of_get_property(options, "watchdog-timeout", NULL);
        if (str_prop)
                p->timeout = simple_strtoul(str_prop, NULL, 10);

        /* CP1400s seem to have broken PLD implementations-- the
         * interrupt_mask register cannot be written, so no timer
         * interrupts can be masked within the PLD.
         */
        str_prop = of_get_property(op->dev.of_node, "model", NULL);
        p->broken = (str_prop && !strcmp(str_prop, WD_BADMODEL));

        if (!p->enabled)
                cpwd_toggleintr(p, -1, WD_INTR_OFF);

        for (i = 0; i < WD_NUMDEVS; i++) {
                static const char *cpwd_names[] = { "RIC", "XIR", "POR" };
                static int *parms[] = { &wd0_timeout,
                                        &wd1_timeout,
                                        &wd2_timeout };
                struct miscdevice *mp = &p->devs[i].misc;

                mp->minor = WD0_MINOR + i;
                mp->name = cpwd_names[i];
                mp->fops = &cpwd_fops;

                p->devs[i].regs = p->regs + (i * WD_TIMER_REGSZ);
                p->devs[i].intr_mask = (WD0_INTR_MASK << i);
                p->devs[i].runstatus &= ~WD_STAT_BSTOP;
                p->devs[i].runstatus |= WD_STAT_INIT;
                p->devs[i].timeout = p->timeout;
                if (*parms[i])
                        p->devs[i].timeout = *parms[i];

                err = misc_register(&p->devs[i].misc);
                if (err) {
                        printk(KERN_ERR "Could not register misc device for "
                               "dev %d\n", i);
                        goto out_unregister;
                }
        }

        if (p->broken) {
                init_timer(&cpwd_timer);
                cpwd_timer.function     = cpwd_brokentimer;
                cpwd_timer.data         = (unsigned long) p;
                cpwd_timer.expires      = WD_BTIMEOUT;

                printk(KERN_INFO PFX "PLD defect workaround enabled for "
                       "model " WD_BADMODEL ".\n");
        }

        dev_set_drvdata(&op->dev, p);
        cpwd_device = p;
        err = 0;

out:
        return err;

out_unregister:
        for (i--; i >= 0; i--)
                misc_deregister(&p->devs[i].misc);

out_iounmap:
        of_iounmap(&op->resource[0], p->regs, 4 * WD_TIMER_REGSZ);

out_free:
        kfree(p);
        goto out;
}

static int __devexit cpwd_remove(struct platform_device *op)
{
        struct cpwd *p = dev_get_drvdata(&op->dev);
        int i;

        for (i = 0; i < WD_NUMDEVS; i++) {
                misc_deregister(&p->devs[i].misc);

                if (!p->enabled) {
                        cpwd_stoptimer(p, i);
                        if (p->devs[i].runstatus & WD_STAT_BSTOP)
                                cpwd_resetbrokentimer(p, i);
                }
        }

        if (p->broken)
                del_timer_sync(&cpwd_timer);

        if (p->initialized)
                free_irq(p->irq, p);

        of_iounmap(&op->resource[0], p->regs, 4 * WD_TIMER_REGSZ);
        kfree(p);

        cpwd_device = NULL;

        return 0;
}

static const struct of_device_id cpwd_match[] = {
        {
                .name = "watchdog",
        },
        {},
};
MODULE_DEVICE_TABLE(of, cpwd_match);

static struct platform_driver cpwd_driver = {
        .driver = {
                .name = DRIVER_NAME,
                .owner = THIS_MODULE,
                .of_match_table = cpwd_match,
        },
        .probe          = cpwd_probe,
        .remove         = __devexit_p(cpwd_remove),
};

static int __init cpwd_init(void)
{
        return platform_driver_register(&cpwd_driver);
}

static void __exit cpwd_exit(void)
{
        platform_driver_unregister(&cpwd_driver);
}

module_init(cpwd_init);
module_exit(cpwd_exit);