[linux-2.6-block.git] / arch / sparc / kernel / leon_smp.c

// SPDX-License-Identifier: GPL-2.0
/* leon_smp.c: Sparc-Leon SMP support.
 *
 * based on sun4m_smp.c
 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
 * Copyright (C) 2009 Daniel Hellstrom (daniel@gaisler.com) Aeroflex Gaisler AB
 * Copyright (C) 2009 Konrad Eisele (konrad@gaisler.com) Aeroflex Gaisler AB
 */

#include <asm/head.h>

#include <linux/kernel.h>
#include <linux/sched/mm.h>
#include <linux/threads.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/kernel_stat.h>
#include <linux/of.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/mm.h>
#include <linux/swap.h>
#include <linux/profile.h>
#include <linux/pm.h>
#include <linux/delay.h>
#include <linux/gfp.h>
#include <linux/cpu.h>
#include <linux/clockchips.h>

#include <asm/cacheflush.h>
#include <asm/tlbflush.h>

#include <asm/ptrace.h>
#include <linux/atomic.h>
#include <asm/irq_regs.h>
#include <asm/traps.h>

#include <asm/delay.h>
#include <asm/irq.h>
#include <asm/page.h>
#include <asm/oplib.h>
#include <asm/cpudata.h>
#include <asm/asi.h>
#include <asm/leon.h>
#include <asm/leon_amba.h>
#include <asm/timer.h>

#include "kernel.h"

#include "irq.h"

extern ctxd_t *srmmu_ctx_table_phys;
static int smp_processors_ready;
extern volatile unsigned long cpu_callin_map[NR_CPUS];
extern cpumask_t smp_commenced_mask;
void leon_configure_cache_smp(void);
static void leon_ipi_init(void);

/* IRQ number of LEON IPIs */
int leon_ipi_irq = LEON3_IRQ_IPI_DEFAULT;

static inline unsigned long do_swap(volatile unsigned long *ptr,
				    unsigned long val)
{
	__asm__ __volatile__("swapa [%2] %3, %0\n\t" : "=&r"(val)
			     : "0"(val), "r"(ptr), "i"(ASI_LEON_DCACHE_MISS)
			     : "memory");
	return val;
}

void leon_cpu_pre_starting(void *arg)
{
	leon_configure_cache_smp();
}

void leon_cpu_pre_online(void *arg)
{
	int cpuid = hard_smp_processor_id();

	/* Allow master to continue. The master will then give us the
	 * go-ahead by setting the smp_commenced_mask and will wait without
	 * timeouts until our setup is completed fully (signified by
	 * our bit being set in the cpu_online_mask).
	 */
	do_swap(&cpu_callin_map[cpuid], 1);

	local_ops->cache_all();
	local_ops->tlb_all();

	/* Fix idle thread fields. */
	__asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(&current_set[cpuid])
			     : "memory" /* paranoid */);

	/* Attach to the address space of init_task. */
	mmgrab(&init_mm);
	current->active_mm = &init_mm;

	while (!cpumask_test_cpu(cpuid, &smp_commenced_mask))
		mb();
}

/*
 *	Cycle through the processors asking the PROM to start each one.
 */

extern struct linux_prom_registers smp_penguin_ctable;

void leon_configure_cache_smp(void)
{
	unsigned long cfg = sparc_leon3_get_dcachecfg();
	int me = smp_processor_id();

	if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) {
		printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n",
		     (unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg),
		     (unsigned int)cfg, (unsigned int)me);
		sparc_leon3_disable_cache();
	} else {
		if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) {
			sparc_leon3_enable_snooping();
		} else {
			printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n",
			     me);
			sparc_leon3_disable_cache();
		}
	}

	local_ops->cache_all();
	local_ops->tlb_all();
}

static void leon_smp_setbroadcast(unsigned int mask)
{
	int broadcast =
	    ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
	      LEON3_IRQMPSTATUS_BROADCAST) & 1);
	if (!broadcast) {
		prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n",
		     leon_smp_nrcpus());
		if (leon_smp_nrcpus() > 1) {
			BUG();
		} else {
			prom_printf("continue anyway\n");
			return;
		}
	}
	LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask);
}

int leon_smp_nrcpus(void)
{
	int nrcpu =
	    ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
	      LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1;
	return nrcpu;
}

void __init leon_boot_cpus(void)
{
	int nrcpu = leon_smp_nrcpus();
	int me = smp_processor_id();

	/* Setup IPI */
	leon_ipi_init();

	printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me,
	       (unsigned int)nrcpu, (unsigned int)NR_CPUS,
	       (unsigned int)&(leon3_irqctrl_regs->mpstatus));

	leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me);
	leon_enable_irq_cpu(LEON3_IRQ_TICKER, me);
	leon_enable_irq_cpu(leon_ipi_irq, me);

	leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER);

	leon_configure_cache_smp();
	local_ops->cache_all();

}

int leon_boot_one_cpu(int i, struct task_struct *idle)
{
	int timeout;

	current_set[i] = task_thread_info(idle);

	/* See trampoline.S:leon_smp_cpu_startup for details...
	 * Initialize the contexts table
	 * Since the call to prom_startcpu() trashes the structure,
	 * we need to re-initialize it for each cpu
	 */
	smp_penguin_ctable.which_io = 0;
	smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys;
	smp_penguin_ctable.reg_size = 0;

	/* whirrr, whirrr, whirrrrrrrrr... */
	printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i,
	       (unsigned int)&leon3_irqctrl_regs->mpstatus);
	local_ops->cache_all();

	/* Make sure all IRQs are of from the start for this new CPU */
	LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0);

	/* Wake one CPU */
	LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i);

	/* wheee... it's going... */
	for (timeout = 0; timeout < 10000; timeout++) {
		if (cpu_callin_map[i])
			break;
		udelay(200);
	}
	printk(KERN_INFO "Started CPU %d\n", (unsigned int)i);

	if (!(cpu_callin_map[i])) {
		printk(KERN_ERR "Processor %d is stuck.\n", i);
		return -ENODEV;
	} else {
		leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i);
		leon_enable_irq_cpu(LEON3_IRQ_TICKER, i);
		leon_enable_irq_cpu(leon_ipi_irq, i);
	}

	local_ops->cache_all();
	return 0;
}

void __init leon_smp_done(void)
{

	int i, first;
	int *prev;

	/* setup cpu list for irq rotation */
	first = 0;
	prev = &first;
	for (i = 0; i < NR_CPUS; i++) {
		if (cpu_online(i)) {
			*prev = i;
			prev = &cpu_data(i).next;
		}
	}
	*prev = first;
	local_ops->cache_all();

	/* Free unneeded trap tables */
	if (!cpu_present(1)) {
		free_reserved_page(virt_to_page(&trapbase_cpu1[0]));
	}
	if (!cpu_present(2)) {
		free_reserved_page(virt_to_page(&trapbase_cpu2[0]));
	}
	if (!cpu_present(3)) {
		free_reserved_page(virt_to_page(&trapbase_cpu3[0]));
	}
	/* Ok, they are spinning and ready to go. */
	smp_processors_ready = 1;

}

struct leon_ipi_work {
	int single;
	int msk;
	int resched;
};

static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work);

/* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ
 * is used for all three types of IPIs.
 */
static void __init leon_ipi_init(void)
{
	int cpu, len;
	struct leon_ipi_work *work;
	struct property *pp;
	struct device_node *rootnp;
	struct tt_entry *trap_table;
	unsigned long flags;

	/* Find IPI IRQ or stick with default value */
	rootnp = of_find_node_by_path("/ambapp0");
	if (rootnp) {
		pp = of_find_property(rootnp, "ipi_num", &len);
		if (pp && (*(int *)pp->value))
			leon_ipi_irq = *(int *)pp->value;
	}
	printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq);

	/* Adjust so that we jump directly to smpleon_ipi */
	local_irq_save(flags);
	trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)];
	trap_table->inst_three += smpleon_ipi - real_irq_entry;
	local_ops->cache_all();
	local_irq_restore(flags);

	for_each_possible_cpu(cpu) {
		work = &per_cpu(leon_ipi_work, cpu);
		work->single = work->msk = work->resched = 0;
	}
}

static void leon_send_ipi(int cpu, int level)
{
	unsigned long mask;
	mask = leon_get_irqmask(level);
	LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask);
}

static void leon_ipi_single(int cpu)
{
	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);

	/* Mark work */
	work->single = 1;

	/* Generate IRQ on the CPU */
	leon_send_ipi(cpu, leon_ipi_irq);
}

static void leon_ipi_mask_one(int cpu)
{
	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);

	/* Mark work */
	work->msk = 1;

	/* Generate IRQ on the CPU */
	leon_send_ipi(cpu, leon_ipi_irq);
}

static void leon_ipi_resched(int cpu)
{
	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);

	/* Mark work */
	work->resched = 1;

	/* Generate IRQ on the CPU (any IRQ will cause resched) */
	leon_send_ipi(cpu, leon_ipi_irq);
}

void leonsmp_ipi_interrupt(void)
{
	struct leon_ipi_work *work = this_cpu_ptr(&leon_ipi_work);

	if (work->single) {
		work->single = 0;
		smp_call_function_single_interrupt();
	}
	if (work->msk) {
		work->msk = 0;
		smp_call_function_interrupt();
	}
	if (work->resched) {
		work->resched = 0;
		smp_resched_interrupt();
	}
}

static struct smp_funcall {
	void *func;
	unsigned long arg1;
	unsigned long arg2;
	unsigned long arg3;
	unsigned long arg4;
	unsigned long arg5;
	unsigned long processors_in[NR_CPUS];	/* Set when ipi entered. */
	unsigned long processors_out[NR_CPUS];	/* Set when ipi exited. */
} ccall_info __attribute__((aligned(8)));

static DEFINE_SPINLOCK(cross_call_lock);

/* Cross calls must be serialized, at least currently. */
static void leon_cross_call(void *func, cpumask_t mask, unsigned long arg1,
			    unsigned long arg2, unsigned long arg3,
			    unsigned long arg4)
{
	if (smp_processors_ready) {
		register int high = NR_CPUS - 1;
		unsigned long flags;

		spin_lock_irqsave(&cross_call_lock, flags);

		{
			/* If you make changes here, make sure gcc generates proper code... */
			register void *f asm("i0") = func;
			register unsigned long a1 asm("i1") = arg1;
			register unsigned long a2 asm("i2") = arg2;
			register unsigned long a3 asm("i3") = arg3;
			register unsigned long a4 asm("i4") = arg4;
			register unsigned long a5 asm("i5") = 0;

			__asm__ __volatile__("std %0, [%6]\n\t"
					     "std %2, [%6 + 8]\n\t"
					     "std %4, [%6 + 16]\n\t" : :
					     "r"(f), "r"(a1), "r"(a2), "r"(a3),
					     "r"(a4), "r"(a5),
					     "r"(&ccall_info.func));
		}

		/* Init receive/complete mapping, plus fire the IPI's off. */
		{
			register int i;

			cpumask_clear_cpu(smp_processor_id(), &mask);
			cpumask_and(&mask, cpu_online_mask, &mask);
			for (i = 0; i <= high; i++) {
				if (cpumask_test_cpu(i, &mask)) {
					ccall_info.processors_in[i] = 0;
					ccall_info.processors_out[i] = 0;
					leon_send_ipi(i, LEON3_IRQ_CROSS_CALL);

				}
			}
		}

		{
			register int i;

			i = 0;
			do {
				if (!cpumask_test_cpu(i, &mask))
					continue;

				while (!ccall_info.processors_in[i])
					barrier();
			} while (++i <= high);

			i = 0;
			do {
				if (!cpumask_test_cpu(i, &mask))
					continue;

				while (!ccall_info.processors_out[i])
					barrier();
			} while (++i <= high);
		}

		spin_unlock_irqrestore(&cross_call_lock, flags);
	}
}

/* Running cross calls. */
void leon_cross_call_irq(void)
{
	void (*func)(unsigned long, unsigned long, unsigned long, unsigned long,
		     unsigned long) = ccall_info.func;
	int i = smp_processor_id();

	ccall_info.processors_in[i] = 1;
	func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3, ccall_info.arg4,
	     ccall_info.arg5);
	ccall_info.processors_out[i] = 1;
}

static const struct sparc32_ipi_ops leon_ipi_ops = {
	.cross_call = leon_cross_call,
	.resched    = leon_ipi_resched,
	.single     = leon_ipi_single,
	.mask_one   = leon_ipi_mask_one,
};

void __init leon_init_smp(void)
{
	/* Patch ipi15 trap table */
	t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m);

	sparc32_ipi_ops = &leon_ipi_ops;
}
Commit	Line	Data
b2441318	1	// SPDX-License-Identifier: GPL-2.0
8401707f KE	2	/* leon_smp.c: Sparc-Leon SMP support.
	3	*
	4	* based on sun4m_smp.c
	5	* Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
	6	* Copyright (C) 2009 Daniel Hellstrom (daniel@gaisler.com) Aeroflex Gaisler AB
	7	* Copyright (C) 2009 Konrad Eisele (konrad@gaisler.com) Aeroflex Gaisler AB
	8	*/
	9
	10	#include <asm/head.h>
	11
	12	#include <linux/kernel.h>
68e21be2	13	#include <linux/sched/mm.h>
8401707f KE	14	#include <linux/threads.h>
8401707f KE	15	#include <linux/smp.h>
8401707f KE	16	#include <linux/interrupt.h>
8401707f KE	17	#include <linux/kernel_stat.h>
1ca0c808	18	#include <linux/of.h>
8401707f KE	19	#include <linux/init.h>
	20	#include <linux/spinlock.h>
	21	#include <linux/mm.h>
	22	#include <linux/swap.h>
	23	#include <linux/profile.h>
	24	#include <linux/pm.h>
	25	#include <linux/delay.h>
5a0e3ad6	26	#include <linux/gfp.h>
62f08283 TK	27	#include <linux/cpu.h>
62f08283 TK	28	#include <linux/clockchips.h>
8401707f KE	29
	30	#include <asm/cacheflush.h>
	31	#include <asm/tlbflush.h>
	32
	33	#include <asm/ptrace.h>
60063497	34	#include <linux/atomic.h>
8401707f	35	#include <asm/irq_regs.h>
1ca0c808	36	#include <asm/traps.h>
8401707f KE	37
	38	#include <asm/delay.h>
	39	#include <asm/irq.h>
	40	#include <asm/page.h>
8401707f KE	41	#include <asm/oplib.h>
	42	#include <asm/cpudata.h>
	43	#include <asm/asi.h>
	44	#include <asm/leon.h>
	45	#include <asm/leon_amba.h>
62f08283	46	#include <asm/timer.h>
8401707f	47
a2a211cb SR	48	#include "kernel.h"
a2a211cb SR	49
8401707f KE	50	#include "irq.h"
	51
	52	extern ctxd_t *srmmu_ctx_table_phys;
	53	static int smp_processors_ready;
	54	extern volatile unsigned long cpu_callin_map[NR_CPUS];
8401707f	55	extern cpumask_t smp_commenced_mask;
2066aadd	56	void leon_configure_cache_smp(void);
1ca0c808 DH	57	static void leon_ipi_init(void);
	58
	59	/* IRQ number of LEON IPIs */
	60	int leon_ipi_irq = LEON3_IRQ_IPI_DEFAULT;
8401707f KE	61
	62	static inline unsigned long do_swap(volatile unsigned long *ptr,
	63	unsigned long val)
	64	{
502279a7 DH	65	__asm__ __volatile__("swapa [%2] %3, %0\n\t" : "=&r"(val)
502279a7 DH	66	: "0"(val), "r"(ptr), "i"(ASI_LEON_DCACHE_MISS)
8401707f KE	67	: "memory");
	68	return val;
	69	}
	70
2066aadd	71	void leon_cpu_pre_starting(void *arg)
8401707f	72	{
8401707f	73	leon_configure_cache_smp();
f9fd3488	74	}
8401707f	75
2066aadd	76	void leon_cpu_pre_online(void *arg)
f9fd3488 SR	77	{
f9fd3488 SR	78	int cpuid = hard_smp_processor_id();
8401707f	79
f9fd3488 SR	80	/* Allow master to continue. The master will then give us the
	81	* go-ahead by setting the smp_commenced_mask and will wait without
	82	* timeouts until our setup is completed fully (signified by
	83	* our bit being set in the cpu_online_mask).
8401707f KE	84	*/
	85	do_swap(&cpu_callin_map[cpuid], 1);
	86
5d83d666 DM	87	local_ops->cache_all();
5d83d666 DM	88	local_ops->tlb_all();
8401707f	89
8401707f KE	90	/* Fix idle thread fields. */
	91	__asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(&current_set[cpuid])
	92	: "memory" /* paranoid */);
	93
	94	/* Attach to the address space of init_task. */
f1f10076	95	mmgrab(&init_mm);
8401707f KE	96	current->active_mm = &init_mm;
8401707f KE	97
fb1fece5	98	while (!cpumask_test_cpu(cpuid, &smp_commenced_mask))
8401707f	99	mb();
8401707f KE	100	}
	101
	102	/*
	103	* Cycle through the processors asking the PROM to start each one.
	104	*/
	105
	106	extern struct linux_prom_registers smp_penguin_ctable;
	107
2066aadd	108	void leon_configure_cache_smp(void)
8401707f KE	109	{
	110	unsigned long cfg = sparc_leon3_get_dcachecfg();
	111	int me = smp_processor_id();
	112
	113	if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) {
	114	printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n",
	115	(unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg),
	116	(unsigned int)cfg, (unsigned int)me);
	117	sparc_leon3_disable_cache();
	118	} else {
	119	if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) {
	120	sparc_leon3_enable_snooping();
	121	} else {
	122	printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n",
	123	me);
	124	sparc_leon3_disable_cache();
	125	}
	126	}
	127
5d83d666 DM	128	local_ops->cache_all();
5d83d666 DM	129	local_ops->tlb_all();
8401707f KE	130	}
8401707f KE	131
1bbc9060	132	static void leon_smp_setbroadcast(unsigned int mask)
8401707f KE	133	{
	134	int broadcast =
	135	((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
	136	LEON3_IRQMPSTATUS_BROADCAST) & 1);
	137	if (!broadcast) {
	138	prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n",
	139	leon_smp_nrcpus());
	140	if (leon_smp_nrcpus() > 1) {
	141	BUG();
	142	} else {
	143	prom_printf("continue anyway\n");
	144	return;
	145	}
	146	}
	147	LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask);
	148	}
	149
8401707f KE	150	int leon_smp_nrcpus(void)
	151	{
	152	int nrcpu =
	153	((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
	154	LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1;
	155	return nrcpu;
	156	}
	157
	158	void __init leon_boot_cpus(void)
	159	{
	160	int nrcpu = leon_smp_nrcpus();
	161	int me = smp_processor_id();
	162
1ca0c808 DH	163	/* Setup IPI */
	164	leon_ipi_init();
	165
0da2b300	166	printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me,
8401707f KE	167	(unsigned int)nrcpu, (unsigned int)NR_CPUS,
	168	(unsigned int)&(leon3_irqctrl_regs->mpstatus));
	169
	170	leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me);
	171	leon_enable_irq_cpu(LEON3_IRQ_TICKER, me);
1ca0c808	172	leon_enable_irq_cpu(leon_ipi_irq, me);
8401707f KE	173
	174	leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER);
	175
	176	leon_configure_cache_smp();
5d83d666	177	local_ops->cache_all();
8401707f KE	178
	179	}
	180
2066aadd	181	int leon_boot_one_cpu(int i, struct task_struct *idle)
8401707f	182	{
8401707f KE	183	int timeout;
8401707f KE	184
f0a2bc7e	185	current_set[i] = task_thread_info(idle);
8401707f KE	186
	187	/* See trampoline.S:leon_smp_cpu_startup for details...
	188	* Initialize the contexts table
	189	* Since the call to prom_startcpu() trashes the structure,
	190	* we need to re-initialize it for each cpu
	191	*/
	192	smp_penguin_ctable.which_io = 0;
	193	smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys;
	194	smp_penguin_ctable.reg_size = 0;
	195
	196	/* whirrr, whirrr, whirrrrrrrrr... */
	197	printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i,
	198	(unsigned int)&leon3_irqctrl_regs->mpstatus);
5d83d666	199	local_ops->cache_all();
8401707f	200
970def65 DH	201	/* Make sure all IRQs are of from the start for this new CPU */
	202	LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0);
	203
	204	/* Wake one CPU */
8401707f KE	205	LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i);
	206
	207	/* wheee... it's going... */
	208	for (timeout = 0; timeout < 10000; timeout++) {
	209	if (cpu_callin_map[i])
	210	break;
	211	udelay(200);
	212	}
0da2b300	213	printk(KERN_INFO "Started CPU %d\n", (unsigned int)i);
8401707f KE	214
	215	if (!(cpu_callin_map[i])) {
	216	printk(KERN_ERR "Processor %d is stuck.\n", i);
	217	return -ENODEV;
	218	} else {
	219	leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i);
	220	leon_enable_irq_cpu(LEON3_IRQ_TICKER, i);
1ca0c808	221	leon_enable_irq_cpu(leon_ipi_irq, i);
8401707f KE	222	}
8401707f KE	223
5d83d666	224	local_ops->cache_all();
8401707f KE	225	return 0;
	226	}
	227
	228	void __init leon_smp_done(void)
	229	{
	230
	231	int i, first;
	232	int *prev;
	233
	234	/* setup cpu list for irq rotation */
	235	first = 0;
	236	prev = &first;
	237	for (i = 0; i < NR_CPUS; i++) {
	238	if (cpu_online(i)) {
	239	*prev = i;
	240	prev = &cpu_data(i).next;
	241	}
	242	}
	243	*prev = first;
5d83d666	244	local_ops->cache_all();
8401707f KE	245
8401707f KE	246	/* Free unneeded trap tables */
fb1fece5	247	if (!cpu_present(1)) {
7bfba2ca	248	free_reserved_page(virt_to_page(&trapbase_cpu1[0]));
8401707f	249	}
fb1fece5	250	if (!cpu_present(2)) {
7bfba2ca	251	free_reserved_page(virt_to_page(&trapbase_cpu2[0]));
8401707f	252	}
fb1fece5	253	if (!cpu_present(3)) {
7bfba2ca	254	free_reserved_page(virt_to_page(&trapbase_cpu3[0]));
8401707f KE	255	}
	256	/* Ok, they are spinning and ready to go. */
	257	smp_processors_ready = 1;
	258
	259	}
	260
1ca0c808 DH	261	struct leon_ipi_work {
	262	int single;
	263	int msk;
	264	int resched;
	265	};
	266
	267	static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work);
	268
	269	/* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ
	270	* is used for all three types of IPIs.
	271	*/
	272	static void __init leon_ipi_init(void)
	273	{
	274	int cpu, len;
	275	struct leon_ipi_work *work;
	276	struct property *pp;
	277	struct device_node *rootnp;
	278	struct tt_entry *trap_table;
	279	unsigned long flags;
	280
	281	/* Find IPI IRQ or stick with default value */
	282	rootnp = of_find_node_by_path("/ambapp0");
	283	if (rootnp) {
	284	pp = of_find_property(rootnp, "ipi_num", &len);
	285	if (pp && ((int )pp->value))
	286	leon_ipi_irq = (int )pp->value;
	287	}
	288	printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq);
	289
	290	/* Adjust so that we jump directly to smpleon_ipi */
	291	local_irq_save(flags);
	292	trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)];
	293	trap_table->inst_three += smpleon_ipi - real_irq_entry;
5d83d666	294	local_ops->cache_all();
1ca0c808 DH	295	local_irq_restore(flags);
	296
	297	for_each_possible_cpu(cpu) {
	298	work = &per_cpu(leon_ipi_work, cpu);
	299	work->single = work->msk = work->resched = 0;
	300	}
	301	}
	302
4ba22b16 SR	303	static void leon_send_ipi(int cpu, int level)
	304	{
	305	unsigned long mask;
	306	mask = leon_get_irqmask(level);
	307	LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask);
	308	}
	309
1ca0c808 DH	310	static void leon_ipi_single(int cpu)
	311	{
	312	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
	313
	314	/* Mark work */
	315	work->single = 1;
	316
	317	/* Generate IRQ on the CPU */
4ba22b16	318	leon_send_ipi(cpu, leon_ipi_irq);
1ca0c808 DH	319	}
	320
	321	static void leon_ipi_mask_one(int cpu)
	322	{
	323	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
	324
	325	/* Mark work */
	326	work->msk = 1;
	327
	328	/* Generate IRQ on the CPU */
4ba22b16	329	leon_send_ipi(cpu, leon_ipi_irq);
1ca0c808 DH	330	}
	331
	332	static void leon_ipi_resched(int cpu)
	333	{
	334	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
	335
	336	/* Mark work */
	337	work->resched = 1;
	338
	339	/* Generate IRQ on the CPU (any IRQ will cause resched) */
4ba22b16	340	leon_send_ipi(cpu, leon_ipi_irq);
1ca0c808 DH	341	}
	342
	343	void leonsmp_ipi_interrupt(void)
	344	{
494fc421	345	struct leon_ipi_work *work = this_cpu_ptr(&leon_ipi_work);
1ca0c808 DH	346
	347	if (work->single) {
	348	work->single = 0;
	349	smp_call_function_single_interrupt();
	350	}
	351	if (work->msk) {
	352	work->msk = 0;
	353	smp_call_function_interrupt();
	354	}
	355	if (work->resched) {
	356	work->resched = 0;
	357	smp_resched_interrupt();
	358	}
	359	}
	360
8401707f	361	static struct smp_funcall {
17006e86	362	void *func;
8401707f KE	363	unsigned long arg1;
	364	unsigned long arg2;
	365	unsigned long arg3;
	366	unsigned long arg4;
	367	unsigned long arg5;
	368	unsigned long processors_in[NR_CPUS]; /* Set when ipi entered. */
	369	unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */
92f268ee	370	} ccall_info __attribute__((aligned(8)));
8401707f KE	371
	372	static DEFINE_SPINLOCK(cross_call_lock);
	373
	374	/* Cross calls must be serialized, at least currently. */
17006e86	375	static void leon_cross_call(void *func, cpumask_t mask, unsigned long arg1,
8401707f KE	376	unsigned long arg2, unsigned long arg3,
	377	unsigned long arg4)
	378	{
	379	if (smp_processors_ready) {
	380	register int high = NR_CPUS - 1;
	381	unsigned long flags;
	382
	383	spin_lock_irqsave(&cross_call_lock, flags);
	384
	385	{
	386	/* If you make changes here, make sure gcc generates proper code... */
17006e86	387	register void *f asm("i0") = func;
8401707f KE	388	register unsigned long a1 asm("i1") = arg1;
	389	register unsigned long a2 asm("i2") = arg2;
	390	register unsigned long a3 asm("i3") = arg3;
	391	register unsigned long a4 asm("i4") = arg4;
	392	register unsigned long a5 asm("i5") = 0;
	393
	394	__asm__ __volatile__("std %0, [%6]\n\t"
	395	"std %2, [%6 + 8]\n\t"
	396	"std %4, [%6 + 16]\n\t" : :
	397	"r"(f), "r"(a1), "r"(a2), "r"(a3),
	398	"r"(a4), "r"(a5),
	399	"r"(&ccall_info.func));
	400	}
	401
	402	/* Init receive/complete mapping, plus fire the IPI's off. */
	403	{
	404	register int i;
	405
fb1fece5 KM	406	cpumask_clear_cpu(smp_processor_id(), &mask);
fb1fece5 KM	407	cpumask_and(&mask, cpu_online_mask, &mask);
8401707f	408	for (i = 0; i <= high; i++) {
fb1fece5	409	if (cpumask_test_cpu(i, &mask)) {
8401707f KE	410	ccall_info.processors_in[i] = 0;
8401707f KE	411	ccall_info.processors_out[i] = 0;
4ba22b16	412	leon_send_ipi(i, LEON3_IRQ_CROSS_CALL);
8401707f KE	413
	414	}
	415	}
	416	}
	417
	418	{
	419	register int i;
	420
	421	i = 0;
	422	do {
fb1fece5	423	if (!cpumask_test_cpu(i, &mask))
8401707f KE	424	continue;
	425
	426	while (!ccall_info.processors_in[i])
	427	barrier();
	428	} while (++i <= high);
	429
	430	i = 0;
	431	do {
fb1fece5	432	if (!cpumask_test_cpu(i, &mask))
8401707f KE	433	continue;
	434
	435	while (!ccall_info.processors_out[i])
	436	barrier();
	437	} while (++i <= high);
	438	}
	439
	440	spin_unlock_irqrestore(&cross_call_lock, flags);
	441	}
	442	}
	443
	444	/* Running cross calls. */
	445	void leon_cross_call_irq(void)
	446	{
17006e86 BVA	447	void (*func)(unsigned long, unsigned long, unsigned long, unsigned long,
17006e86 BVA	448	unsigned long) = ccall_info.func;
8401707f KE	449	int i = smp_processor_id();
	450
	451	ccall_info.processors_in[i] = 1;
17006e86 BVA	452	func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3, ccall_info.arg4,
17006e86 BVA	453	ccall_info.arg5);
8401707f KE	454	ccall_info.processors_out[i] = 1;
	455	}
	456
4ba22b16 SR	457	static const struct sparc32_ipi_ops leon_ipi_ops = {
	458	.cross_call = leon_cross_call,
	459	.resched = leon_ipi_resched,
	460	.single = leon_ipi_single,
	461	.mask_one = leon_ipi_mask_one,
	462	};
	463
8401707f KE	464	void __init leon_init_smp(void)
	465	{
	466	/* Patch ipi15 trap table */
	467	t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m);
	468
4ba22b16	469	sparc32_ipi_ops = &leon_ipi_ops;
8401707f	470	}