sched_clock.c 4.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/*
 * sched_clock.c: support for extending counters to full 64-bit ns counter
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */
#include <linux/clocksource.h>
#include <linux/init.h>
#include <linux/jiffies.h>
#include <linux/kernel.h>
12
#include <linux/moduleparam.h>
13
#include <linux/sched.h>
14
#include <linux/syscore_ops.h>
15 16 17 18
#include <linux/timer.h>

#include <asm/sched_clock.h>

19 20 21 22 23 24
struct clock_data {
	u64 epoch_ns;
	u32 epoch_cyc;
	u32 epoch_cyc_copy;
	u32 mult;
	u32 shift;
25 26
	bool suspended;
	bool needs_suspend;
27 28
};

29 30
static void sched_clock_poll(unsigned long wrap_ticks);
static DEFINE_TIMER(sched_clock_timer, sched_clock_poll, 0, 0);
31 32 33
static int irqtime = -1;

core_param(irqtime, irqtime, int, 0400);
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

static struct clock_data cd = {
	.mult	= NSEC_PER_SEC / HZ,
};

static u32 __read_mostly sched_clock_mask = 0xffffffff;

static u32 notrace jiffy_sched_clock_read(void)
{
	return (u32)(jiffies - INITIAL_JIFFIES);
}

static u32 __read_mostly (*read_sched_clock)(void) = jiffy_sched_clock_read;

static inline u64 cyc_to_ns(u64 cyc, u32 mult, u32 shift)
{
	return (cyc * mult) >> shift;
}

static unsigned long long cyc_to_sched_clock(u32 cyc, u32 mask)
{
	u64 epoch_ns;
	u32 epoch_cyc;

58 59 60
	if (cd.suspended)
		return cd.epoch_ns;

61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102
	/*
	 * Load the epoch_cyc and epoch_ns atomically.  We do this by
	 * ensuring that we always write epoch_cyc, epoch_ns and
	 * epoch_cyc_copy in strict order, and read them in strict order.
	 * If epoch_cyc and epoch_cyc_copy are not equal, then we're in
	 * the middle of an update, and we should repeat the load.
	 */
	do {
		epoch_cyc = cd.epoch_cyc;
		smp_rmb();
		epoch_ns = cd.epoch_ns;
		smp_rmb();
	} while (epoch_cyc != cd.epoch_cyc_copy);

	return epoch_ns + cyc_to_ns((cyc - epoch_cyc) & mask, cd.mult, cd.shift);
}

/*
 * Atomically update the sched_clock epoch.
 */
static void notrace update_sched_clock(void)
{
	unsigned long flags;
	u32 cyc;
	u64 ns;

	cyc = read_sched_clock();
	ns = cd.epoch_ns +
		cyc_to_ns((cyc - cd.epoch_cyc) & sched_clock_mask,
			  cd.mult, cd.shift);
	/*
	 * Write epoch_cyc and epoch_ns in a way that the update is
	 * detectable in cyc_to_fixed_sched_clock().
	 */
	raw_local_irq_save(flags);
	cd.epoch_cyc = cyc;
	smp_wmb();
	cd.epoch_ns = ns;
	smp_wmb();
	cd.epoch_cyc_copy = cyc;
	raw_local_irq_restore(flags);
}
103 104 105 106

static void sched_clock_poll(unsigned long wrap_ticks)
{
	mod_timer(&sched_clock_timer, round_jiffies(jiffies + wrap_ticks));
107
	update_sched_clock();
108 109
}

110
void __init setup_sched_clock(u32 (*read)(void), int bits, unsigned long rate)
111 112 113 114 115
{
	unsigned long r, w;
	u64 res, wrap;
	char r_unit;

116 117 118 119 120
	BUG_ON(bits > 32);
	WARN_ON(!irqs_disabled());
	WARN_ON(read_sched_clock != jiffy_sched_clock_read);
	read_sched_clock = read;
	sched_clock_mask = (1 << bits) - 1;
121 122

	/* calculate the mult/shift to convert counter ticks to ns. */
123
	clocks_calc_mult_shift(&cd.mult, &cd.shift, rate, NSEC_PER_SEC, 0);
124 125 126 127 128

	r = rate;
	if (r >= 4000000) {
		r /= 1000000;
		r_unit = 'M';
129
	} else if (r >= 1000) {
130 131
		r /= 1000;
		r_unit = 'k';
132 133
	} else
		r_unit = ' ';
134 135

	/* calculate how many ns until we wrap */
136
	wrap = cyc_to_ns((1ULL << bits) - 1, cd.mult, cd.shift);
137 138 139 140
	do_div(wrap, NSEC_PER_MSEC);
	w = wrap;

	/* calculate the ns resolution of this counter */
141
	res = cyc_to_ns(1ULL, cd.mult, cd.shift);
142
	pr_info("sched_clock: %u bits at %lu%cHz, resolution %lluns, wraps every %lums\n",
143
		bits, r, r_unit, res, w);
144 145 146 147 148 149

	/*
	 * Start the timer to keep sched_clock() properly updated and
	 * sets the initial epoch.
	 */
	sched_clock_timer.data = msecs_to_jiffies(w - (w / 10));
150
	update_sched_clock();
151 152 153 154

	/*
	 * Ensure that sched_clock() starts off at 0ns
	 */
155 156
	cd.epoch_ns = 0;

157 158 159 160
	/* Enable IRQ time accounting if we have a fast enough sched_clock */
	if (irqtime > 0 || (irqtime == -1 && rate >= 1000000))
		enable_sched_clock_irqtime();

161 162 163 164 165 166 167
	pr_debug("Registered %pF as sched_clock source\n", read);
}

unsigned long long notrace sched_clock(void)
{
	u32 cyc = read_sched_clock();
	return cyc_to_sched_clock(cyc, sched_clock_mask);
168
}
169 170 171

void __init sched_clock_postinit(void)
{
172 173 174 175 176 177 178
	/*
	 * If no sched_clock function has been provided at that point,
	 * make it the final one one.
	 */
	if (read_sched_clock == jiffy_sched_clock_read)
		setup_sched_clock(jiffy_sched_clock_read, 32, HZ);

179 180
	sched_clock_poll(sched_clock_timer.data);
}
181 182 183 184

static int sched_clock_suspend(void)
{
	sched_clock_poll(sched_clock_timer.data);
185
	cd.suspended = true;
186 187 188
	return 0;
}

189 190
static void sched_clock_resume(void)
{
191 192 193
	cd.epoch_cyc = read_sched_clock();
	cd.epoch_cyc_copy = cd.epoch_cyc;
	cd.suspended = false;
194 195
}

196 197
static struct syscore_ops sched_clock_ops = {
	.suspend = sched_clock_suspend,
198
	.resume = sched_clock_resume,
199 200 201 202 203 204 205 206
};

static int __init sched_clock_syscore_init(void)
{
	register_syscore_ops(&sched_clock_ops);
	return 0;
}
device_initcall(sched_clock_syscore_init);