[deliverable/linux.git] / arch / sparc / kernel / leon_smp.c

/* 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.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/pgalloc.h>
#include <asm/pgtable.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. */
	atomic_inc(&init_mm.mm_count);
	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();
}

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);
}

unsigned int leon_smp_getbroadcast(void)
{
	unsigned int mask;
	mask = LEON_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpbroadcast));
	return 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));
	}
	if (!cpu_present(2)) {
		free_reserved_page(virt_to_page(&trapbase_cpu2));
	}
	if (!cpu_present(3)) {
		free_reserved_page(virt_to_page(&trapbase_cpu3));
	}
	/* Ok, they are spinning and ready to go. */
	smp_processors_ready = 1;

}

void leon_irq_rotate(int cpu)
{
}

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 = &__get_cpu_var(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 {
	smpfunc_t 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;

static DEFINE_SPINLOCK(cross_call_lock);

/* Cross calls must be serialized, at least currently. */
static void leon_cross_call(smpfunc_t 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 smpfunc_t 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)
{
	int i = smp_processor_id();

	ccall_info.processors_in[i] = 1;
	ccall_info.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
8401707f KE	1	/* leon_smp.c: Sparc-Leon SMP support.
	2	*
	3	* based on sun4m_smp.c
	4	* Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
	5	* Copyright (C) 2009 Daniel Hellstrom (daniel@gaisler.com) Aeroflex Gaisler AB
	6	* Copyright (C) 2009 Konrad Eisele (konrad@gaisler.com) Aeroflex Gaisler AB
	7	*/
	8
	9	#include <asm/head.h>
	10
	11	#include <linux/kernel.h>
	12	#include <linux/sched.h>
	13	#include <linux/threads.h>
	14	#include <linux/smp.h>
8401707f KE	15	#include <linux/interrupt.h>
8401707f KE	16	#include <linux/kernel_stat.h>
1ca0c808	17	#include <linux/of.h>
8401707f KE	18	#include <linux/init.h>
	19	#include <linux/spinlock.h>
	20	#include <linux/mm.h>
	21	#include <linux/swap.h>
	22	#include <linux/profile.h>
	23	#include <linux/pm.h>
	24	#include <linux/delay.h>
5a0e3ad6	25	#include <linux/gfp.h>
62f08283 TK	26	#include <linux/cpu.h>
62f08283 TK	27	#include <linux/clockchips.h>
8401707f KE	28
	29	#include <asm/cacheflush.h>
	30	#include <asm/tlbflush.h>
	31
	32	#include <asm/ptrace.h>
60063497	33	#include <linux/atomic.h>
8401707f	34	#include <asm/irq_regs.h>
1ca0c808	35	#include <asm/traps.h>
8401707f KE	36
	37	#include <asm/delay.h>
	38	#include <asm/irq.h>
	39	#include <asm/page.h>
	40	#include <asm/pgalloc.h>
	41	#include <asm/pgtable.h>
	42	#include <asm/oplib.h>
	43	#include <asm/cpudata.h>
	44	#include <asm/asi.h>
	45	#include <asm/leon.h>
	46	#include <asm/leon_amba.h>
62f08283	47	#include <asm/timer.h>
8401707f	48
a2a211cb SR	49	#include "kernel.h"
a2a211cb SR	50
8401707f KE	51	#include "irq.h"
	52
	53	extern ctxd_t *srmmu_ctx_table_phys;
	54	static int smp_processors_ready;
	55	extern volatile unsigned long cpu_callin_map[NR_CPUS];
8401707f	56	extern cpumask_t smp_commenced_mask;
2066aadd	57	void leon_configure_cache_smp(void);
1ca0c808 DH	58	static void leon_ipi_init(void);
	59
	60	/* IRQ number of LEON IPIs */
	61	int leon_ipi_irq = LEON3_IRQ_IPI_DEFAULT;
8401707f KE	62
	63	static inline unsigned long do_swap(volatile unsigned long *ptr,
	64	unsigned long val)
	65	{
502279a7 DH	66	__asm__ __volatile__("swapa [%2] %3, %0\n\t" : "=&r"(val)
502279a7 DH	67	: "0"(val), "r"(ptr), "i"(ASI_LEON_DCACHE_MISS)
8401707f KE	68	: "memory");
	69	return val;
	70	}
	71
2066aadd	72	void leon_cpu_pre_starting(void *arg)
8401707f	73	{
8401707f	74	leon_configure_cache_smp();
f9fd3488	75	}
8401707f	76
2066aadd	77	void leon_cpu_pre_online(void *arg)
f9fd3488 SR	78	{
f9fd3488 SR	79	int cpuid = hard_smp_processor_id();
8401707f	80
f9fd3488 SR	81	/* Allow master to continue. The master will then give us the
	82	* go-ahead by setting the smp_commenced_mask and will wait without
	83	* timeouts until our setup is completed fully (signified by
	84	* our bit being set in the cpu_online_mask).
8401707f KE	85	*/
	86	do_swap(&cpu_callin_map[cpuid], 1);
	87
5d83d666 DM	88	local_ops->cache_all();
5d83d666 DM	89	local_ops->tlb_all();
8401707f	90
8401707f KE	91	/* Fix idle thread fields. */
	92	__asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(&current_set[cpuid])
	93	: "memory" /* paranoid */);
	94
	95	/* Attach to the address space of init_task. */
	96	atomic_inc(&init_mm.mm_count);
	97	current->active_mm = &init_mm;
	98
fb1fece5	99	while (!cpumask_test_cpu(cpuid, &smp_commenced_mask))
8401707f	100	mb();
8401707f KE	101	}
	102
	103	/*
	104	* Cycle through the processors asking the PROM to start each one.
	105	*/
	106
	107	extern struct linux_prom_registers smp_penguin_ctable;
	108
2066aadd	109	void leon_configure_cache_smp(void)
8401707f KE	110	{
	111	unsigned long cfg = sparc_leon3_get_dcachecfg();
	112	int me = smp_processor_id();
	113
	114	if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) {
	115	printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n",
	116	(unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg),
	117	(unsigned int)cfg, (unsigned int)me);
	118	sparc_leon3_disable_cache();
	119	} else {
	120	if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) {
	121	sparc_leon3_enable_snooping();
	122	} else {
	123	printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n",
	124	me);
	125	sparc_leon3_disable_cache();
	126	}
	127	}
	128
5d83d666 DM	129	local_ops->cache_all();
5d83d666 DM	130	local_ops->tlb_all();
8401707f KE	131	}
	132
	133	void leon_smp_setbroadcast(unsigned int mask)
	134	{
	135	int broadcast =
	136	((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
	137	LEON3_IRQMPSTATUS_BROADCAST) & 1);
	138	if (!broadcast) {
	139	prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n",
	140	leon_smp_nrcpus());
	141	if (leon_smp_nrcpus() > 1) {
	142	BUG();
	143	} else {
	144	prom_printf("continue anyway\n");
	145	return;
	146	}
	147	}
	148	LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask);
	149	}
	150
	151	unsigned int leon_smp_getbroadcast(void)
	152	{
	153	unsigned int mask;
	154	mask = LEON_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpbroadcast));
	155	return mask;
	156	}
	157
	158	int leon_smp_nrcpus(void)
	159	{
	160	int nrcpu =
	161	((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
	162	LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1;
	163	return nrcpu;
	164	}
	165
	166	void __init leon_boot_cpus(void)
	167	{
	168	int nrcpu = leon_smp_nrcpus();
	169	int me = smp_processor_id();
	170
1ca0c808 DH	171	/* Setup IPI */
	172	leon_ipi_init();
	173
0da2b300	174	printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me,
8401707f KE	175	(unsigned int)nrcpu, (unsigned int)NR_CPUS,
	176	(unsigned int)&(leon3_irqctrl_regs->mpstatus));
	177
	178	leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me);
	179	leon_enable_irq_cpu(LEON3_IRQ_TICKER, me);
1ca0c808	180	leon_enable_irq_cpu(leon_ipi_irq, me);
8401707f KE	181
	182	leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER);
	183
	184	leon_configure_cache_smp();
5d83d666	185	local_ops->cache_all();
8401707f KE	186
	187	}
	188
2066aadd	189	int leon_boot_one_cpu(int i, struct task_struct *idle)
8401707f	190	{
8401707f KE	191	int timeout;
8401707f KE	192
f0a2bc7e	193	current_set[i] = task_thread_info(idle);
8401707f KE	194
	195	/* See trampoline.S:leon_smp_cpu_startup for details...
	196	* Initialize the contexts table
	197	* Since the call to prom_startcpu() trashes the structure,
	198	* we need to re-initialize it for each cpu
	199	*/
	200	smp_penguin_ctable.which_io = 0;
	201	smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys;
	202	smp_penguin_ctable.reg_size = 0;
	203
	204	/* whirrr, whirrr, whirrrrrrrrr... */
	205	printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i,
	206	(unsigned int)&leon3_irqctrl_regs->mpstatus);
5d83d666	207	local_ops->cache_all();
8401707f	208
970def65 DH	209	/* Make sure all IRQs are of from the start for this new CPU */
	210	LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0);
	211
	212	/* Wake one CPU */
8401707f KE	213	LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i);
	214
	215	/* wheee... it's going... */
	216	for (timeout = 0; timeout < 10000; timeout++) {
	217	if (cpu_callin_map[i])
	218	break;
	219	udelay(200);
	220	}
0da2b300	221	printk(KERN_INFO "Started CPU %d\n", (unsigned int)i);
8401707f KE	222
	223	if (!(cpu_callin_map[i])) {
	224	printk(KERN_ERR "Processor %d is stuck.\n", i);
	225	return -ENODEV;
	226	} else {
	227	leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i);
	228	leon_enable_irq_cpu(LEON3_IRQ_TICKER, i);
1ca0c808	229	leon_enable_irq_cpu(leon_ipi_irq, i);
8401707f KE	230	}
8401707f KE	231
5d83d666	232	local_ops->cache_all();
8401707f KE	233	return 0;
	234	}
	235
	236	void __init leon_smp_done(void)
	237	{
	238
	239	int i, first;
	240	int *prev;
	241
	242	/* setup cpu list for irq rotation */
	243	first = 0;
	244	prev = &first;
	245	for (i = 0; i < NR_CPUS; i++) {
	246	if (cpu_online(i)) {
	247	*prev = i;
	248	prev = &cpu_data(i).next;
	249	}
	250	}
	251	*prev = first;
5d83d666	252	local_ops->cache_all();
8401707f KE	253
8401707f KE	254	/* Free unneeded trap tables */
fb1fece5	255	if (!cpu_present(1)) {
70affe45	256	free_reserved_page(virt_to_page(&trapbase_cpu1));
8401707f	257	}
fb1fece5	258	if (!cpu_present(2)) {
70affe45	259	free_reserved_page(virt_to_page(&trapbase_cpu2));
8401707f	260	}
fb1fece5	261	if (!cpu_present(3)) {
70affe45	262	free_reserved_page(virt_to_page(&trapbase_cpu3));
8401707f KE	263	}
	264	/* Ok, they are spinning and ready to go. */
	265	smp_processors_ready = 1;
	266
	267	}
	268
	269	void leon_irq_rotate(int cpu)
	270	{
	271	}
	272
1ca0c808 DH	273	struct leon_ipi_work {
	274	int single;
	275	int msk;
	276	int resched;
	277	};
	278
	279	static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work);
	280
	281	/* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ
	282	* is used for all three types of IPIs.
	283	*/
	284	static void __init leon_ipi_init(void)
	285	{
	286	int cpu, len;
	287	struct leon_ipi_work *work;
	288	struct property *pp;
	289	struct device_node *rootnp;
	290	struct tt_entry *trap_table;
	291	unsigned long flags;
	292
	293	/* Find IPI IRQ or stick with default value */
	294	rootnp = of_find_node_by_path("/ambapp0");
	295	if (rootnp) {
	296	pp = of_find_property(rootnp, "ipi_num", &len);
	297	if (pp && ((int )pp->value))
	298	leon_ipi_irq = (int )pp->value;
	299	}
	300	printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq);
	301
	302	/* Adjust so that we jump directly to smpleon_ipi */
	303	local_irq_save(flags);
	304	trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)];
	305	trap_table->inst_three += smpleon_ipi - real_irq_entry;
5d83d666	306	local_ops->cache_all();
1ca0c808 DH	307	local_irq_restore(flags);
	308
	309	for_each_possible_cpu(cpu) {
	310	work = &per_cpu(leon_ipi_work, cpu);
	311	work->single = work->msk = work->resched = 0;
	312	}
	313	}
	314
4ba22b16 SR	315	static void leon_send_ipi(int cpu, int level)
	316	{
	317	unsigned long mask;
	318	mask = leon_get_irqmask(level);
	319	LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask);
	320	}
	321
1ca0c808 DH	322	static void leon_ipi_single(int cpu)
	323	{
	324	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
	325
	326	/* Mark work */
	327	work->single = 1;
	328
	329	/* Generate IRQ on the CPU */
4ba22b16	330	leon_send_ipi(cpu, leon_ipi_irq);
1ca0c808 DH	331	}
	332
	333	static void leon_ipi_mask_one(int cpu)
	334	{
	335	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
	336
	337	/* Mark work */
	338	work->msk = 1;
	339
	340	/* Generate IRQ on the CPU */
4ba22b16	341	leon_send_ipi(cpu, leon_ipi_irq);
1ca0c808 DH	342	}
	343
	344	static void leon_ipi_resched(int cpu)
	345	{
	346	struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
	347
	348	/* Mark work */
	349	work->resched = 1;
	350
	351	/* Generate IRQ on the CPU (any IRQ will cause resched) */
4ba22b16	352	leon_send_ipi(cpu, leon_ipi_irq);
1ca0c808 DH	353	}
	354
	355	void leonsmp_ipi_interrupt(void)
	356	{
	357	struct leon_ipi_work *work = &__get_cpu_var(leon_ipi_work);
	358
	359	if (work->single) {
	360	work->single = 0;
	361	smp_call_function_single_interrupt();
	362	}
	363	if (work->msk) {
	364	work->msk = 0;
	365	smp_call_function_interrupt();
	366	}
	367	if (work->resched) {
	368	work->resched = 0;
	369	smp_resched_interrupt();
	370	}
	371	}
	372
8401707f KE	373	static struct smp_funcall {
	374	smpfunc_t func;
	375	unsigned long arg1;
	376	unsigned long arg2;
	377	unsigned long arg3;
	378	unsigned long arg4;
	379	unsigned long arg5;
	380	unsigned long processors_in[NR_CPUS]; /* Set when ipi entered. */
	381	unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */
	382	} ccall_info;
	383
	384	static DEFINE_SPINLOCK(cross_call_lock);
	385
	386	/* Cross calls must be serialized, at least currently. */
	387	static void leon_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
	388	unsigned long arg2, unsigned long arg3,
	389	unsigned long arg4)
	390	{
	391	if (smp_processors_ready) {
	392	register int high = NR_CPUS - 1;
	393	unsigned long flags;
	394
	395	spin_lock_irqsave(&cross_call_lock, flags);
	396
	397	{
	398	/* If you make changes here, make sure gcc generates proper code... */
	399	register smpfunc_t f asm("i0") = func;
	400	register unsigned long a1 asm("i1") = arg1;
	401	register unsigned long a2 asm("i2") = arg2;
	402	register unsigned long a3 asm("i3") = arg3;
	403	register unsigned long a4 asm("i4") = arg4;
	404	register unsigned long a5 asm("i5") = 0;
	405
	406	__asm__ __volatile__("std %0, [%6]\n\t"
	407	"std %2, [%6 + 8]\n\t"
	408	"std %4, [%6 + 16]\n\t" : :
	409	"r"(f), "r"(a1), "r"(a2), "r"(a3),
	410	"r"(a4), "r"(a5),
	411	"r"(&ccall_info.func));
	412	}
	413
	414	/* Init receive/complete mapping, plus fire the IPI's off. */
	415	{
	416	register int i;
	417
fb1fece5 KM	418	cpumask_clear_cpu(smp_processor_id(), &mask);
fb1fece5 KM	419	cpumask_and(&mask, cpu_online_mask, &mask);
8401707f	420	for (i = 0; i <= high; i++) {
fb1fece5	421	if (cpumask_test_cpu(i, &mask)) {
8401707f KE	422	ccall_info.processors_in[i] = 0;
8401707f KE	423	ccall_info.processors_out[i] = 0;
4ba22b16	424	leon_send_ipi(i, LEON3_IRQ_CROSS_CALL);
8401707f KE	425
	426	}
	427	}
	428	}
	429
	430	{
	431	register int i;
	432
	433	i = 0;
	434	do {
fb1fece5	435	if (!cpumask_test_cpu(i, &mask))
8401707f KE	436	continue;
	437
	438	while (!ccall_info.processors_in[i])
	439	barrier();
	440	} while (++i <= high);
	441
	442	i = 0;
	443	do {
fb1fece5	444	if (!cpumask_test_cpu(i, &mask))
8401707f KE	445	continue;
	446
	447	while (!ccall_info.processors_out[i])
	448	barrier();
	449	} while (++i <= high);
	450	}
	451
	452	spin_unlock_irqrestore(&cross_call_lock, flags);
	453	}
	454	}
	455
	456	/* Running cross calls. */
	457	void leon_cross_call_irq(void)
	458	{
	459	int i = smp_processor_id();
	460
	461	ccall_info.processors_in[i] = 1;
	462	ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
	463	ccall_info.arg4, ccall_info.arg5);
	464	ccall_info.processors_out[i] = 1;
	465	}
	466
4ba22b16 SR	467	static const struct sparc32_ipi_ops leon_ipi_ops = {
	468	.cross_call = leon_cross_call,
	469	.resched = leon_ipi_resched,
	470	.single = leon_ipi_single,
	471	.mask_one = leon_ipi_mask_one,
	472	};
	473
8401707f KE	474	void __init leon_init_smp(void)
	475	{
	476	/* Patch ipi15 trap table */
	477	t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m);
	478
4ba22b16	479	sparc32_ipi_ops = &leon_ipi_ops;
8401707f	480	}