sh: CMT clockevent platform driver
Magnus Damm [Thu, 22 Jan 2009 09:55:59 +0000 (09:55 +0000)]
SuperH CMT clockevent driver.

Both 16-bit and 32-bit CMT versions are supported, but only 32-bit
is tested. This driver contains support for both clockevents and
clocksources, but no unregistration is supported at this point.

Works fine as clock source and/or event in periodic or oneshot mode.
Tested on sh7722 and sh7723, but should work with any cpu/architecture.

This version is lacking clocksource and early platform driver support
for now - this to minimize the amount of dependencies.

Signed-off-by: Magnus Damm <damm@igel.co.jp>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>

arch/sh/Kconfig
drivers/clocksource/Makefile
drivers/clocksource/sh_cmt.c [new file with mode: 0644]
include/linux/sh_cmt.h [new file with mode: 0644]

index ebabe51..5407e12 100644 (file)
@@ -397,6 +397,14 @@ source "arch/sh/boards/Kconfig"
 
 menu "Timer and clock configuration"
 
+config SH_TIMER_CMT
+       def_bool n
+       prompt "CMT support"
+       select GENERIC_TIME
+       select GENERIC_CLOCKEVENTS
+       help
+         This enables build of the CMT system timer driver.
+
 config SH_TMU
        def_bool y
        prompt "TMU timer support"
index 1525882..1efb287 100644 (file)
@@ -2,3 +2,4 @@ obj-$(CONFIG_ATMEL_TCB_CLKSRC)  += tcb_clksrc.o
 obj-$(CONFIG_X86_CYCLONE_TIMER)        += cyclone.o
 obj-$(CONFIG_X86_PM_TIMER)     += acpi_pm.o
 obj-$(CONFIG_SCx200HR_TIMER)   += scx200_hrt.o
+obj-$(CONFIG_SH_TIMER_CMT)     += sh_cmt.o
diff --git a/drivers/clocksource/sh_cmt.c b/drivers/clocksource/sh_cmt.c
new file mode 100644 (file)
index 0000000..7783b42
--- /dev/null
@@ -0,0 +1,615 @@
+/*
+ * SuperH Timer Support - CMT
+ *
+ *  Copyright (C) 2008 Magnus Damm
+ *
+ * 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
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/interrupt.h>
+#include <linux/ioport.h>
+#include <linux/io.h>
+#include <linux/clk.h>
+#include <linux/irq.h>
+#include <linux/err.h>
+#include <linux/clocksource.h>
+#include <linux/clockchips.h>
+#include <linux/sh_cmt.h>
+
+struct sh_cmt_priv {
+       void __iomem *mapbase;
+       struct clk *clk;
+       unsigned long width; /* 16 or 32 bit version of hardware block */
+       unsigned long overflow_bit;
+       unsigned long clear_bits;
+       struct irqaction irqaction;
+       struct platform_device *pdev;
+
+       unsigned long flags;
+       unsigned long match_value;
+       unsigned long next_match_value;
+       unsigned long max_match_value;
+       unsigned long rate;
+       spinlock_t lock;
+       struct clock_event_device ced;
+       unsigned long total_cycles;
+};
+
+static DEFINE_SPINLOCK(sh_cmt_lock);
+
+#define CMSTR -1 /* shared register */
+#define CMCSR 0 /* channel register */
+#define CMCNT 1 /* channel register */
+#define CMCOR 2 /* channel register */
+
+static inline unsigned long sh_cmt_read(struct sh_cmt_priv *p, int reg_nr)
+{
+       struct sh_cmt_config *cfg = p->pdev->dev.platform_data;
+       void __iomem *base = p->mapbase;
+       unsigned long offs;
+
+       if (reg_nr == CMSTR) {
+               offs = 0;
+               base -= cfg->channel_offset;
+       } else
+               offs = reg_nr;
+
+       if (p->width == 16)
+               offs <<= 1;
+       else {
+               offs <<= 2;
+               if ((reg_nr == CMCNT) || (reg_nr == CMCOR))
+                       return ioread32(base + offs);
+       }
+
+       return ioread16(base + offs);
+}
+
+static inline void sh_cmt_write(struct sh_cmt_priv *p, int reg_nr,
+                               unsigned long value)
+{
+       struct sh_cmt_config *cfg = p->pdev->dev.platform_data;
+       void __iomem *base = p->mapbase;
+       unsigned long offs;
+
+       if (reg_nr == CMSTR) {
+               offs = 0;
+               base -= cfg->channel_offset;
+       } else
+               offs = reg_nr;
+
+       if (p->width == 16)
+               offs <<= 1;
+       else {
+               offs <<= 2;
+               if ((reg_nr == CMCNT) || (reg_nr == CMCOR)) {
+                       iowrite32(value, base + offs);
+                       return;
+               }
+       }
+
+       iowrite16(value, base + offs);
+}
+
+static unsigned long sh_cmt_get_counter(struct sh_cmt_priv *p,
+                                       int *has_wrapped)
+{
+       unsigned long v1, v2, v3;
+
+       /* Make sure the timer value is stable. Stolen from acpi_pm.c */
+       do {
+               v1 = sh_cmt_read(p, CMCNT);
+               v2 = sh_cmt_read(p, CMCNT);
+               v3 = sh_cmt_read(p, CMCNT);
+       } while (unlikely((v1 > v2 && v1 < v3) || (v2 > v3 && v2 < v1)
+                         || (v3 > v1 && v3 < v2)));
+
+       *has_wrapped = sh_cmt_read(p, CMCSR) & p->overflow_bit;
+       return v2;
+}
+
+
+static void sh_cmt_start_stop_ch(struct sh_cmt_priv *p, int start)
+{
+       struct sh_cmt_config *cfg = p->pdev->dev.platform_data;
+       unsigned long flags, value;
+
+       /* start stop register shared by multiple timer channels */
+       spin_lock_irqsave(&sh_cmt_lock, flags);
+       value = sh_cmt_read(p, CMSTR);
+
+       if (start)
+               value |= 1 << cfg->timer_bit;
+       else
+               value &= ~(1 << cfg->timer_bit);
+
+       sh_cmt_write(p, CMSTR, value);
+       spin_unlock_irqrestore(&sh_cmt_lock, flags);
+}
+
+static int sh_cmt_enable(struct sh_cmt_priv *p, unsigned long *rate)
+{
+       struct sh_cmt_config *cfg = p->pdev->dev.platform_data;
+       int ret;
+
+       /* enable clock */
+       ret = clk_enable(p->clk);
+       if (ret) {
+               pr_err("sh_cmt: cannot enable clock \"%s\"\n", cfg->clk);
+               return ret;
+       }
+       *rate = clk_get_rate(p->clk) / 8;
+
+       /* make sure channel is disabled */
+       sh_cmt_start_stop_ch(p, 0);
+
+       /* configure channel, periodic mode and maximum timeout */
+       if (p->width == 16)
+               sh_cmt_write(p, CMCSR, 0);
+       else
+               sh_cmt_write(p, CMCSR, 0x01a4);
+
+       sh_cmt_write(p, CMCOR, 0xffffffff);
+       sh_cmt_write(p, CMCNT, 0);
+
+       /* enable channel */
+       sh_cmt_start_stop_ch(p, 1);
+       return 0;
+}
+
+static void sh_cmt_disable(struct sh_cmt_priv *p)
+{
+       /* disable channel */
+       sh_cmt_start_stop_ch(p, 0);
+
+       /* stop clock */
+       clk_disable(p->clk);
+}
+
+/* private flags */
+#define FLAG_CLOCKEVENT (1 << 0)
+#define FLAG_CLOCKSOURCE (1 << 1)
+#define FLAG_REPROGRAM (1 << 2)
+#define FLAG_SKIPEVENT (1 << 3)
+#define FLAG_IRQCONTEXT (1 << 4)
+
+static void sh_cmt_clock_event_program_verify(struct sh_cmt_priv *p,
+                                             int absolute)
+{
+       unsigned long new_match;
+       unsigned long value = p->next_match_value;
+       unsigned long delay = 0;
+       unsigned long now = 0;
+       int has_wrapped;
+
+       now = sh_cmt_get_counter(p, &has_wrapped);
+       p->flags |= FLAG_REPROGRAM; /* force reprogram */
+
+       if (has_wrapped) {
+               /* we're competing with the interrupt handler.
+                *  -> let the interrupt handler reprogram the timer.
+                *  -> interrupt number two handles the event.
+                */
+               p->flags |= FLAG_SKIPEVENT;
+               return;
+       }
+
+       if (absolute)
+               now = 0;
+
+       do {
+               /* reprogram the timer hardware,
+                * but don't save the new match value yet.
+                */
+               new_match = now + value + delay;
+               if (new_match > p->max_match_value)
+                       new_match = p->max_match_value;
+
+               sh_cmt_write(p, CMCOR, new_match);
+
+               now = sh_cmt_get_counter(p, &has_wrapped);
+               if (has_wrapped && (new_match > p->match_value)) {
+                       /* we are changing to a greater match value,
+                        * so this wrap must be caused by the counter
+                        * matching the old value.
+                        * -> first interrupt reprograms the timer.
+                        * -> interrupt number two handles the event.
+                        */
+                       p->flags |= FLAG_SKIPEVENT;
+                       break;
+               }
+
+               if (has_wrapped) {
+                       /* we are changing to a smaller match value,
+                        * so the wrap must be caused by the counter
+                        * matching the new value.
+                        * -> save programmed match value.
+                        * -> let isr handle the event.
+                        */
+                       p->match_value = new_match;
+                       break;
+               }
+
+               /* be safe: verify hardware settings */
+               if (now < new_match) {
+                       /* timer value is below match value, all good.
+                        * this makes sure we won't miss any match events.
+                        * -> save programmed match value.
+                        * -> let isr handle the event.
+                        */
+                       p->match_value = new_match;
+                       break;
+               }
+
+               /* the counter has reached a value greater
+                * than our new match value. and since the
+                * has_wrapped flag isn't set we must have
+                * programmed a too close event.
+                * -> increase delay and retry.
+                */
+               if (delay)
+                       delay <<= 1;
+               else
+                       delay = 1;
+
+               if (!delay)
+                       pr_warning("sh_cmt: too long delay\n");
+
+       } while (delay);
+}
+
+static void sh_cmt_set_next(struct sh_cmt_priv *p, unsigned long delta)
+{
+       unsigned long flags;
+
+       if (delta > p->max_match_value)
+               pr_warning("sh_cmt: delta out of range\n");
+
+       spin_lock_irqsave(&p->lock, flags);
+       p->next_match_value = delta;
+       sh_cmt_clock_event_program_verify(p, 0);
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static irqreturn_t sh_cmt_interrupt(int irq, void *dev_id)
+{
+       struct sh_cmt_priv *p = dev_id;
+
+       /* clear flags */
+       sh_cmt_write(p, CMCSR, sh_cmt_read(p, CMCSR) & p->clear_bits);
+
+       /* update clock source counter to begin with if enabled
+        * the wrap flag should be cleared by the timer specific
+        * isr before we end up here.
+        */
+       if (p->flags & FLAG_CLOCKSOURCE)
+               p->total_cycles += p->match_value;
+
+       if (!(p->flags & FLAG_REPROGRAM))
+               p->next_match_value = p->max_match_value;
+
+       p->flags |= FLAG_IRQCONTEXT;
+
+       if (p->flags & FLAG_CLOCKEVENT) {
+               if (!(p->flags & FLAG_SKIPEVENT)) {
+                       if (p->ced.mode == CLOCK_EVT_MODE_ONESHOT) {
+                               p->next_match_value = p->max_match_value;
+                               p->flags |= FLAG_REPROGRAM;
+                       }
+
+                       p->ced.event_handler(&p->ced);
+               }
+       }
+
+       p->flags &= ~FLAG_SKIPEVENT;
+
+       if (p->flags & FLAG_REPROGRAM) {
+               p->flags &= ~FLAG_REPROGRAM;
+               sh_cmt_clock_event_program_verify(p, 1);
+
+               if (p->flags & FLAG_CLOCKEVENT)
+                       if ((p->ced.mode == CLOCK_EVT_MODE_SHUTDOWN)
+                           || (p->match_value == p->next_match_value))
+                               p->flags &= ~FLAG_REPROGRAM;
+       }
+
+       p->flags &= ~FLAG_IRQCONTEXT;
+
+       return IRQ_HANDLED;
+}
+
+static int sh_cmt_start(struct sh_cmt_priv *p, unsigned long flag)
+{
+       int ret = 0;
+       unsigned long flags;
+
+       spin_lock_irqsave(&p->lock, flags);
+
+       if (!(p->flags & (FLAG_CLOCKEVENT | FLAG_CLOCKSOURCE)))
+               ret = sh_cmt_enable(p, &p->rate);
+
+       if (ret)
+               goto out;
+       p->flags |= flag;
+
+       /* setup timeout if no clockevent */
+       if ((flag == FLAG_CLOCKSOURCE) && (!(p->flags & FLAG_CLOCKEVENT)))
+               sh_cmt_set_next(p, p->max_match_value);
+ out:
+       spin_unlock_irqrestore(&p->lock, flags);
+
+       return ret;
+}
+
+static void sh_cmt_stop(struct sh_cmt_priv *p, unsigned long flag)
+{
+       unsigned long flags;
+       unsigned long f;
+
+       spin_lock_irqsave(&p->lock, flags);
+
+       f = p->flags & (FLAG_CLOCKEVENT | FLAG_CLOCKSOURCE);
+       p->flags &= ~flag;
+
+       if (f && !(p->flags & (FLAG_CLOCKEVENT | FLAG_CLOCKSOURCE)))
+               sh_cmt_disable(p);
+
+       /* adjust the timeout to maximum if only clocksource left */
+       if ((flag == FLAG_CLOCKEVENT) && (p->flags & FLAG_CLOCKSOURCE))
+               sh_cmt_set_next(p, p->max_match_value);
+
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static struct sh_cmt_priv *ced_to_sh_cmt(struct clock_event_device *ced)
+{
+       return container_of(ced, struct sh_cmt_priv, ced);
+}
+
+static void sh_cmt_clock_event_start(struct sh_cmt_priv *p, int periodic)
+{
+       struct clock_event_device *ced = &p->ced;
+
+       sh_cmt_start(p, FLAG_CLOCKEVENT);
+
+       /* TODO: calculate good shift from rate and counter bit width */
+
+       ced->shift = 32;
+       ced->mult = div_sc(p->rate, NSEC_PER_SEC, ced->shift);
+       ced->max_delta_ns = clockevent_delta2ns(p->max_match_value, ced);
+       ced->min_delta_ns = clockevent_delta2ns(0x1f, ced);
+
+       if (periodic)
+               sh_cmt_set_next(p, (p->rate + HZ/2) / HZ);
+       else
+               sh_cmt_set_next(p, p->max_match_value);
+}
+
+static void sh_cmt_clock_event_mode(enum clock_event_mode mode,
+                                   struct clock_event_device *ced)
+{
+       struct sh_cmt_priv *p = ced_to_sh_cmt(ced);
+
+       /* deal with old setting first */
+       switch (ced->mode) {
+       case CLOCK_EVT_MODE_PERIODIC:
+       case CLOCK_EVT_MODE_ONESHOT:
+               sh_cmt_stop(p, FLAG_CLOCKEVENT);
+               break;
+       default:
+               break;
+       }
+
+       switch (mode) {
+       case CLOCK_EVT_MODE_PERIODIC:
+               pr_info("sh_cmt: %s used for periodic clock events\n",
+                       ced->name);
+               sh_cmt_clock_event_start(p, 1);
+               break;
+       case CLOCK_EVT_MODE_ONESHOT:
+               pr_info("sh_cmt: %s used for oneshot clock events\n",
+                       ced->name);
+               sh_cmt_clock_event_start(p, 0);
+               break;
+       case CLOCK_EVT_MODE_SHUTDOWN:
+       case CLOCK_EVT_MODE_UNUSED:
+               sh_cmt_stop(p, FLAG_CLOCKEVENT);
+               break;
+       default:
+               break;
+       }
+}
+
+static int sh_cmt_clock_event_next(unsigned long delta,
+                                  struct clock_event_device *ced)
+{
+       struct sh_cmt_priv *p = ced_to_sh_cmt(ced);
+
+       BUG_ON(ced->mode != CLOCK_EVT_MODE_ONESHOT);
+       if (likely(p->flags & FLAG_IRQCONTEXT))
+               p->next_match_value = delta;
+       else
+               sh_cmt_set_next(p, delta);
+
+       return 0;
+}
+
+static void sh_cmt_register_clockevent(struct sh_cmt_priv *p,
+                                      char *name, unsigned long rating)
+{
+       struct clock_event_device *ced = &p->ced;
+
+       memset(ced, 0, sizeof(*ced));
+
+       ced->name = name;
+       ced->features = CLOCK_EVT_FEAT_PERIODIC;
+       ced->features |= CLOCK_EVT_FEAT_ONESHOT;
+       ced->rating = rating;
+       ced->cpumask = cpumask_of(0);
+       ced->set_next_event = sh_cmt_clock_event_next;
+       ced->set_mode = sh_cmt_clock_event_mode;
+
+       pr_info("sh_cmt: %s used for clock events\n", ced->name);
+       ced->mult = 1; /* work around misplaced WARN_ON() in clockevents.c */
+       clockevents_register_device(ced);
+}
+
+int sh_cmt_register(struct sh_cmt_priv *p, char *name,
+                   unsigned long clockevent_rating,
+                   unsigned long clocksource_rating)
+{
+       if (p->width == (sizeof(p->max_match_value) * 8))
+               p->max_match_value = ~0;
+       else
+               p->max_match_value = (1 << p->width) - 1;
+
+       p->match_value = p->max_match_value;
+       spin_lock_init(&p->lock);
+
+       if (clockevent_rating)
+               sh_cmt_register_clockevent(p, name, clockevent_rating);
+
+       return 0;
+}
+
+static int sh_cmt_setup(struct sh_cmt_priv *p, struct platform_device *pdev)
+{
+       struct sh_cmt_config *cfg = pdev->dev.platform_data;
+       struct resource *res;
+       int irq, ret;
+       ret = -ENXIO;
+
+       memset(p, 0, sizeof(*p));
+       p->pdev = pdev;
+
+       if (!cfg) {
+               dev_err(&p->pdev->dev, "missing platform data\n");
+               goto err0;
+       }
+
+       platform_set_drvdata(pdev, p);
+
+       res = platform_get_resource(p->pdev, IORESOURCE_MEM, 0);
+       if (!res) {
+               dev_err(&p->pdev->dev, "failed to get I/O memory\n");
+               goto err0;
+       }
+
+       irq = platform_get_irq(p->pdev, 0);
+       if (irq < 0) {
+               dev_err(&p->pdev->dev, "failed to get irq\n");
+               goto err0;
+       }
+
+       /* map memory, let mapbase point to our channel */
+       p->mapbase = ioremap_nocache(res->start, resource_size(res));
+       if (p->mapbase == NULL) {
+               pr_err("sh_cmt: failed to remap I/O memory\n");
+               goto err0;
+       }
+
+       /* request irq using setup_irq() (too early for request_irq()) */
+       p->irqaction.name = cfg->name;
+       p->irqaction.handler = sh_cmt_interrupt;
+       p->irqaction.dev_id = p;
+       p->irqaction.flags = IRQF_DISABLED | IRQF_TIMER | IRQF_IRQPOLL;
+       p->irqaction.mask = CPU_MASK_NONE;
+       ret = setup_irq(irq, &p->irqaction);
+       if (ret) {
+               pr_err("sh_cmt: failed to request irq %d\n", irq);
+               goto err1;
+       }
+
+       /* get hold of clock */
+       p->clk = clk_get(&p->pdev->dev, cfg->clk);
+       if (IS_ERR(p->clk)) {
+               pr_err("sh_cmt: cannot get clock \"%s\"\n", cfg->clk);
+               ret = PTR_ERR(p->clk);
+               goto err2;
+       }
+
+       if (resource_size(res) == 6) {
+               p->width = 16;
+               p->overflow_bit = 0x80;
+               p->clear_bits = ~0xc0;
+       } else {
+               p->width = 32;
+               p->overflow_bit = 0x8000;
+               p->clear_bits = ~0xc000;
+       }
+
+       return sh_cmt_register(p, cfg->name,
+                              cfg->clockevent_rating,
+                              cfg->clocksource_rating);
+ err2:
+       free_irq(irq, p);
+ err1:
+       iounmap(p->mapbase);
+ err0:
+       return ret;
+}
+
+static int __devinit sh_cmt_probe(struct platform_device *pdev)
+{
+       struct sh_cmt_priv *p = platform_get_drvdata(pdev);
+       int ret;
+
+       p = kmalloc(sizeof(*p), GFP_KERNEL);
+       if (p == NULL) {
+               dev_err(&pdev->dev, "failed to allocate driver data\n");
+               return -ENOMEM;
+       }
+
+       ret = sh_cmt_setup(p, pdev);
+       if (ret) {
+               kfree(p);
+
+               platform_set_drvdata(pdev, NULL);
+       }
+       return ret;
+}
+
+static int __devexit sh_cmt_remove(struct platform_device *pdev)
+{
+       return -EBUSY; /* cannot unregister clockevent and clocksource */
+}
+
+static struct platform_driver sh_cmt_device_driver = {
+       .probe          = sh_cmt_probe,
+       .remove         = __devexit_p(sh_cmt_remove),
+       .driver         = {
+               .name   = "sh_cmt",
+       }
+};
+
+static int __init sh_cmt_init(void)
+{
+       return platform_driver_register(&sh_cmt_device_driver);
+}
+
+static void __exit sh_cmt_exit(void)
+{
+       platform_driver_unregister(&sh_cmt_device_driver);
+}
+
+module_init(sh_cmt_init);
+module_exit(sh_cmt_exit);
+
+MODULE_AUTHOR("Magnus Damm");
+MODULE_DESCRIPTION("SuperH CMT Timer Driver");
+MODULE_LICENSE("GPL v2");
diff --git a/include/linux/sh_cmt.h b/include/linux/sh_cmt.h
new file mode 100644 (file)
index 0000000..68cacde
--- /dev/null
@@ -0,0 +1,13 @@
+#ifndef __SH_CMT_H__
+#define __SH_CMT_H__
+
+struct sh_cmt_config {
+       char *name;
+       unsigned long channel_offset;
+       int timer_bit;
+       char *clk;
+       unsigned long clockevent_rating;
+       unsigned long clocksource_rating;
+};
+
+#endif /* __SH_CMT_H__ */