linux/arch/parisc/kernel/time.c
Helge Deller 54b6680090 parisc: Add native high-resolution sched_clock() implementation
Add a native implementation for the sched_clock() function which utilizes the
processor-internal cycle counter (Control Register 16) as high-resolution time
source.

With this patch we now get much more fine-grained resolutions in various
in-kernel time measurements (e.g. when viewing the function tracing logs), and
probably a more accurate scheduling on SMP systems.

There are a few specific implementation details in this patch:

1. On a 32bit kernel we emulate the higher 32bits of the required 64-bit
resolution of sched_clock() by increasing a per-cpu counter at every
wrap-around of the 32bit cycle counter.

2. In a SMP system, the cycle counters of the various CPUs are not syncronized
(similiar to the TSC in a x86_64 system). To cope with this we define
HAVE_UNSTABLE_SCHED_CLOCK and let the upper layers do the adjustment work.

3. Since we need HAVE_UNSTABLE_SCHED_CLOCK, we need to provide a cmpxchg64()
function even on a 32-bit kernel.

4. A 64-bit SMP kernel which is started on a UP system will mark the
sched_clock() implementation as "stable", which means that we don't expect any
jumps in the returned counter. This is true because we then run only on one
CPU.

Signed-off-by: Helge Deller <deller@gmx.de>
2016-05-22 21:39:25 +02:00

322 lines
8.9 KiB
C

/*
* linux/arch/parisc/kernel/time.c
*
* Copyright (C) 1991, 1992, 1995 Linus Torvalds
* Modifications for ARM (C) 1994, 1995, 1996,1997 Russell King
* Copyright (C) 1999 SuSE GmbH, (Philipp Rumpf, prumpf@tux.org)
*
* 1994-07-02 Alan Modra
* fixed set_rtc_mmss, fixed time.year for >= 2000, new mktime
* 1998-12-20 Updated NTP code according to technical memorandum Jan '96
* "A Kernel Model for Precision Timekeeping" by Dave Mills
*/
#include <linux/errno.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/param.h>
#include <linux/string.h>
#include <linux/mm.h>
#include <linux/interrupt.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/smp.h>
#include <linux/profile.h>
#include <linux/clocksource.h>
#include <linux/platform_device.h>
#include <linux/ftrace.h>
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/irq.h>
#include <asm/page.h>
#include <asm/param.h>
#include <asm/pdc.h>
#include <asm/led.h>
#include <linux/timex.h>
static unsigned long clocktick __read_mostly; /* timer cycles per tick */
#ifndef CONFIG_64BIT
/*
* The processor-internal cycle counter (Control Register 16) is used as time
* source for the sched_clock() function. This register is 64bit wide on a
* 64-bit kernel and 32bit on a 32-bit kernel. Since sched_clock() always
* requires a 64bit counter we emulate on the 32-bit kernel the higher 32bits
* with a per-cpu variable which we increase every time the counter
* wraps-around (which happens every ~4 secounds).
*/
static DEFINE_PER_CPU(unsigned long, cr16_high_32_bits);
#endif
/*
* We keep time on PA-RISC Linux by using the Interval Timer which is
* a pair of registers; one is read-only and one is write-only; both
* accessed through CR16. The read-only register is 32 or 64 bits wide,
* and increments by 1 every CPU clock tick. The architecture only
* guarantees us a rate between 0.5 and 2, but all implementations use a
* rate of 1. The write-only register is 32-bits wide. When the lowest
* 32 bits of the read-only register compare equal to the write-only
* register, it raises a maskable external interrupt. Each processor has
* an Interval Timer of its own and they are not synchronised.
*
* We want to generate an interrupt every 1/HZ seconds. So we program
* CR16 to interrupt every @clocktick cycles. The it_value in cpu_data
* is programmed with the intended time of the next tick. We can be
* held off for an arbitrarily long period of time by interrupts being
* disabled, so we may miss one or more ticks.
*/
irqreturn_t __irq_entry timer_interrupt(int irq, void *dev_id)
{
unsigned long now, now2;
unsigned long next_tick;
unsigned long cycles_elapsed, ticks_elapsed = 1;
unsigned long cycles_remainder;
unsigned int cpu = smp_processor_id();
struct cpuinfo_parisc *cpuinfo = &per_cpu(cpu_data, cpu);
/* gcc can optimize for "read-only" case with a local clocktick */
unsigned long cpt = clocktick;
profile_tick(CPU_PROFILING);
/* Initialize next_tick to the expected tick time. */
next_tick = cpuinfo->it_value;
/* Get current cycle counter (Control Register 16). */
now = mfctl(16);
cycles_elapsed = now - next_tick;
if ((cycles_elapsed >> 6) < cpt) {
/* use "cheap" math (add/subtract) instead
* of the more expensive div/mul method
*/
cycles_remainder = cycles_elapsed;
while (cycles_remainder > cpt) {
cycles_remainder -= cpt;
ticks_elapsed++;
}
} else {
/* TODO: Reduce this to one fdiv op */
cycles_remainder = cycles_elapsed % cpt;
ticks_elapsed += cycles_elapsed / cpt;
}
/* convert from "division remainder" to "remainder of clock tick" */
cycles_remainder = cpt - cycles_remainder;
/* Determine when (in CR16 cycles) next IT interrupt will fire.
* We want IT to fire modulo clocktick even if we miss/skip some.
* But those interrupts don't in fact get delivered that regularly.
*/
next_tick = now + cycles_remainder;
cpuinfo->it_value = next_tick;
/* Program the IT when to deliver the next interrupt.
* Only bottom 32-bits of next_tick are writable in CR16!
*/
mtctl(next_tick, 16);
#if !defined(CONFIG_64BIT)
/* check for overflow on a 32bit kernel (every ~4 seconds). */
if (unlikely(next_tick < now))
this_cpu_inc(cr16_high_32_bits);
#endif
/* Skip one clocktick on purpose if we missed next_tick.
* The new CR16 must be "later" than current CR16 otherwise
* itimer would not fire until CR16 wrapped - e.g 4 seconds
* later on a 1Ghz processor. We'll account for the missed
* tick on the next timer interrupt.
*
* "next_tick - now" will always give the difference regardless
* if one or the other wrapped. If "now" is "bigger" we'll end up
* with a very large unsigned number.
*/
now2 = mfctl(16);
if (next_tick - now2 > cpt)
mtctl(next_tick+cpt, 16);
#if 1
/*
* GGG: DEBUG code for how many cycles programming CR16 used.
*/
if (unlikely(now2 - now > 0x3000)) /* 12K cycles */
printk (KERN_CRIT "timer_interrupt(CPU %d): SLOW! 0x%lx cycles!"
" cyc %lX rem %lX "
" next/now %lX/%lX\n",
cpu, now2 - now, cycles_elapsed, cycles_remainder,
next_tick, now );
#endif
/* Can we differentiate between "early CR16" (aka Scenario 1) and
* "long delay" (aka Scenario 3)? I don't think so.
*
* Timer_interrupt will be delivered at least a few hundred cycles
* after the IT fires. But it's arbitrary how much time passes
* before we call it "late". I've picked one second.
*
* It's important NO printk's are between reading CR16 and
* setting up the next value. May introduce huge variance.
*/
if (unlikely(ticks_elapsed > HZ)) {
/* Scenario 3: very long delay? bad in any case */
printk (KERN_CRIT "timer_interrupt(CPU %d): delayed!"
" cycles %lX rem %lX "
" next/now %lX/%lX\n",
cpu,
cycles_elapsed, cycles_remainder,
next_tick, now );
}
/* Done mucking with unreliable delivery of interrupts.
* Go do system house keeping.
*/
if (!--cpuinfo->prof_counter) {
cpuinfo->prof_counter = cpuinfo->prof_multiplier;
update_process_times(user_mode(get_irq_regs()));
}
if (cpu == 0)
xtime_update(ticks_elapsed);
return IRQ_HANDLED;
}
unsigned long profile_pc(struct pt_regs *regs)
{
unsigned long pc = instruction_pointer(regs);
if (regs->gr[0] & PSW_N)
pc -= 4;
#ifdef CONFIG_SMP
if (in_lock_functions(pc))
pc = regs->gr[2];
#endif
return pc;
}
EXPORT_SYMBOL(profile_pc);
/* clock source code */
static cycle_t read_cr16(struct clocksource *cs)
{
return get_cycles();
}
static struct clocksource clocksource_cr16 = {
.name = "cr16",
.rating = 300,
.read = read_cr16,
.mask = CLOCKSOURCE_MASK(BITS_PER_LONG),
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
};
int update_cr16_clocksource(void)
{
/* since the cr16 cycle counters are not synchronized across CPUs,
we'll check if we should switch to a safe clocksource: */
if (clocksource_cr16.rating != 0 && num_online_cpus() > 1) {
clocksource_change_rating(&clocksource_cr16, 0);
return 1;
}
return 0;
}
void __init start_cpu_itimer(void)
{
unsigned int cpu = smp_processor_id();
unsigned long next_tick = mfctl(16) + clocktick;
#if defined(CONFIG_HAVE_UNSTABLE_SCHED_CLOCK) && defined(CONFIG_64BIT)
/* With multiple 64bit CPUs online, the cr16's are not syncronized. */
if (cpu != 0)
clear_sched_clock_stable();
#endif
mtctl(next_tick, 16); /* kick off Interval Timer (CR16) */
per_cpu(cpu_data, cpu).it_value = next_tick;
}
static int __init rtc_init(void)
{
struct platform_device *pdev;
pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
return PTR_ERR_OR_ZERO(pdev);
}
device_initcall(rtc_init);
void read_persistent_clock(struct timespec *ts)
{
static struct pdc_tod tod_data;
if (pdc_tod_read(&tod_data) == 0) {
ts->tv_sec = tod_data.tod_sec;
ts->tv_nsec = tod_data.tod_usec * 1000;
} else {
printk(KERN_ERR "Error reading tod clock\n");
ts->tv_sec = 0;
ts->tv_nsec = 0;
}
}
/*
* sched_clock() framework
*/
static u32 cyc2ns_mul __read_mostly;
static u32 cyc2ns_shift __read_mostly;
u64 sched_clock(void)
{
u64 now;
/* Get current cycle counter (Control Register 16). */
#ifdef CONFIG_64BIT
now = mfctl(16);
#else
now = mfctl(16) + (((u64) this_cpu_read(cr16_high_32_bits)) << 32);
#endif
/* return the value in ns (cycles_2_ns) */
return mul_u64_u32_shr(now, cyc2ns_mul, cyc2ns_shift);
}
/*
* timer interrupt and sched_clock() initialization
*/
void __init time_init(void)
{
unsigned long current_cr16_khz;
current_cr16_khz = PAGE0->mem_10msec/10; /* kHz */
clocktick = (100 * PAGE0->mem_10msec) / HZ;
/* calculate mult/shift values for cr16 */
clocks_calc_mult_shift(&cyc2ns_mul, &cyc2ns_shift, current_cr16_khz,
NSEC_PER_MSEC, 0);
#if defined(CONFIG_HAVE_UNSTABLE_SCHED_CLOCK) && defined(CONFIG_64BIT)
/* At bootup only one 64bit CPU is online and cr16 is "stable" */
set_sched_clock_stable();
#endif
start_cpu_itimer(); /* get CPU 0 started */
/* register at clocksource framework */
clocksource_register_khz(&clocksource_cr16, current_cr16_khz);
}