// SPDX-License-Identifier: GPL-2.0-only
/*
 * Time related functions for Hexagon architecture
 *
 * Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
 */

#include <linux/init.h>
#include <linux/clockchips.h>
#include <linux/clocksource.h>
#include <linux/interrupt.h>
#include <linux/err.h>
#include <linux/platform_device.h>
#include <linux/ioport.h>
#include <linux/of.h>
#include <linux/of_address.h>
#include <linux/of_irq.h>
#include <linux/module.h>

#include <asm/timer-regs.h>
#include <asm/hexagon_vm.h>

/*
 * For the clocksource we need:
 *      pcycle frequency (600MHz)
 * For the loops_per_jiffy we need:
 *      thread/cpu frequency (100MHz)
 * And for the timer, we need:
 *      sleep clock rate
 */

cycles_t        pcycle_freq_mhz;
cycles_t        thread_freq_mhz;
cycles_t        sleep_clk_freq;

static struct resource rtos_timer_resources[] = {
        {
                .start  = RTOS_TIMER_REGS_ADDR,
                .end    = RTOS_TIMER_REGS_ADDR+PAGE_SIZE-1,
                .flags  = IORESOURCE_MEM,
        },
};

static struct platform_device rtos_timer_device = {
        .name           = "rtos_timer",
        .id             = -1,
        .num_resources  = ARRAY_SIZE(rtos_timer_resources),
        .resource       = rtos_timer_resources,
};

/*  A lot of this stuff should move into a platform specific section.  */
struct adsp_hw_timer_struct {
        u32 match;   /*  Match value  */
        u32 count;
        u32 enable;  /*  [1] - CLR_ON_MATCH_EN, [0] - EN  */
        u32 clear;   /*  one-shot register that clears the count  */
};

/*  Look for "TCX0" for related constants.  */
static __iomem struct adsp_hw_timer_struct *rtos_timer;

static u64 timer_get_cycles(struct clocksource *cs)
{
        return (u64) __vmgettime();
}

static struct clocksource hexagon_clocksource = {
        .name           = "pcycles",
        .rating         = 250,
        .read           = timer_get_cycles,
        .mask           = CLOCKSOURCE_MASK(64),
        .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
};

static int set_next_event(unsigned long delta, struct clock_event_device *evt)
{
        /*  Assuming the timer will be disabled when we enter here.  */

        iowrite32(1, &rtos_timer->clear);
        iowrite32(0, &rtos_timer->clear);

        iowrite32(delta, &rtos_timer->match);
        iowrite32(1 << TIMER_ENABLE, &rtos_timer->enable);
        return 0;
}

#ifdef CONFIG_SMP
/*  Broadcast mechanism  */
static void broadcast(const struct cpumask *mask)
{
        send_ipi(mask, IPI_TIMER);
}
#endif

/* XXX Implement set_state_shutdown() */
static struct clock_event_device hexagon_clockevent_dev = {
        .name           = "clockevent",
        .features       = CLOCK_EVT_FEAT_ONESHOT,
        .rating         = 400,
        .irq            = RTOS_TIMER_INT,
        .set_next_event = set_next_event,
#ifdef CONFIG_SMP
        .broadcast      = broadcast,
#endif
};

#ifdef CONFIG_SMP
static DEFINE_PER_CPU(struct clock_event_device, clock_events);

void setup_percpu_clockdev(void)
{
        int cpu = smp_processor_id();
        struct clock_event_device *ce_dev = &hexagon_clockevent_dev;
        struct clock_event_device *dummy_clock_dev =
                &per_cpu(clock_events, cpu);

        memcpy(dummy_clock_dev, ce_dev, sizeof(*dummy_clock_dev));
        INIT_LIST_HEAD(&dummy_clock_dev->list);

        dummy_clock_dev->features = CLOCK_EVT_FEAT_DUMMY;
        dummy_clock_dev->cpumask = cpumask_of(cpu);

        clockevents_register_device(dummy_clock_dev);
}

/*  Called from smp.c for each CPU's timer ipi call  */
void ipi_timer(void)
{
        int cpu = smp_processor_id();
        struct clock_event_device *ce_dev = &per_cpu(clock_events, cpu);

        ce_dev->event_handler(ce_dev);
}
#endif /* CONFIG_SMP */

static irqreturn_t timer_interrupt(int irq, void *devid)
{
        struct clock_event_device *ce_dev = &hexagon_clockevent_dev;

        iowrite32(0, &rtos_timer->enable);
        ce_dev->event_handler(ce_dev);

        return IRQ_HANDLED;
}

/*  This should also be pulled from devtree  */
static struct irqaction rtos_timer_intdesc = {
        .handler = timer_interrupt,
        .flags = IRQF_TIMER | IRQF_TRIGGER_RISING,
        .name = "rtos_timer"
};

/*
 * time_init_deferred - called by start_kernel to set up timer/clock source
 *
 * Install the IRQ handler for the clock, setup timers.
 * This is done late, as that way, we can use ioremap().
 *
 * This runs just before the delay loop is calibrated, and
 * is used for delay calibration.
 */
void __init time_init_deferred(void)
{
        struct resource *resource = NULL;
        struct clock_event_device *ce_dev = &hexagon_clockevent_dev;

        ce_dev->cpumask = cpu_all_mask;

        if (!resource)
                resource = rtos_timer_device.resource;

        /*  ioremap here means this has to run later, after paging init  */
        rtos_timer = ioremap(resource->start, resource_size(resource));

        if (!rtos_timer) {
                release_mem_region(resource->start, resource_size(resource));
        }
        clocksource_register_khz(&hexagon_clocksource, pcycle_freq_mhz * 1000);

        /*  Note: the sim generic RTOS clock is apparently really 18750Hz  */

        /*
         * Last arg is some guaranteed seconds for which the conversion will
         * work without overflow.
         */
        clockevents_calc_mult_shift(ce_dev, sleep_clk_freq, 4);

        ce_dev->max_delta_ns = clockevent_delta2ns(0x7fffffff, ce_dev);
        ce_dev->max_delta_ticks = 0x7fffffff;
        ce_dev->min_delta_ns = clockevent_delta2ns(0xf, ce_dev);
        ce_dev->min_delta_ticks = 0xf;

#ifdef CONFIG_SMP
        setup_percpu_clockdev();
#endif

        clockevents_register_device(ce_dev);
        setup_irq(ce_dev->irq, &rtos_timer_intdesc);
}

void __init time_init(void)
{
        late_time_init = time_init_deferred;
}

void __delay(unsigned long cycles)
{
        unsigned long long start = __vmgettime();

        while ((__vmgettime() - start) < cycles)
                cpu_relax();
}
EXPORT_SYMBOL(__delay);

/*
 * This could become parametric or perhaps even computed at run-time,
 * but for now we take the observed simulator jitter.
 */
static long long fudgefactor = 350;  /* Maybe lower if kernel optimized. */

void __udelay(unsigned long usecs)
{
        unsigned long long start = __vmgettime();
        unsigned long long finish = (pcycle_freq_mhz * usecs) - fudgefactor;

        while ((__vmgettime() - start) < finish)
                cpu_relax(); /*  not sure how this improves readability  */
}
EXPORT_SYMBOL(__udelay);