smpboot.c

xen虚拟机源代码安装包

/*
 * SMP boot-related support
 *
 * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
 *	David Mosberger-Tang <[e-mail]>
 * Copyright (C) 2001, 2004-2005 Intel Corp
 * 	Rohit Seth <[e-mail]>
 * 	Suresh Siddha <[e-mail]>
 * 	Gordon Jin <[e-mail]>
 *	Ashok Raj  <[e-mail]>
 *
 * 01/05/16 Rohit Seth <[e-mail]>	Moved SMP booting functions from smp.c to here.
 * 01/04/27 David Mosberger <[e-mail]>	Added ITC synching code.
 * 02/07/31 David Mosberger <[e-mail]>	Switch over to hotplug-CPU boot-sequence.
 *						smp_boot_cpus()/smp_commence() is replaced by
 *						smp_prepare_cpus()/__cpu_up()/smp_cpus_done().
 * 04/06/21 Ashok Raj		<[e-mail]> Added CPU Hotplug Support
 * 04/12/26 Jin Gordon <[e-mail]>
 * 04/12/26 Rohit Seth <[e-mail]>
 *						Add multi-threading and multi-core detection
 * 05/01/30 Suresh Siddha <[e-mail]>
 *						Setup cpu_sibling_map and cpu_core_map
 */
#include <linux/config.h>

#include <linux/module.h>
#include <linux/acpi.h>
#include <linux/bootmem.h>
#include <linux/cpu.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/irq.h>
#include <linux/kernel.h>
#include <linux/kernel_stat.h>
#include <linux/mm.h>
#include <linux/notifier.h>	/* hg add me */
#include <linux/smp.h>
#include <linux/smp_lock.h>
#include <linux/spinlock.h>
#include <linux/efi.h>
#include <linux/percpu.h>
#include <linux/bitops.h>

#include <asm/atomic.h>
#include <asm/cache.h>
#include <asm/current.h>
#include <asm/delay.h>
#include <asm/ia32.h>
#include <asm/io.h>
#include <asm/irq.h>
#include <asm/machvec.h>
#include <asm/mca.h>
#include <asm/page.h>
#include <asm/pgalloc.h>
#include <asm/pgtable.h>
#include <asm/processor.h>
#include <asm/ptrace.h>
#include <asm/sal.h>
#include <asm/system.h>
#include <asm/tlbflush.h>
#include <asm/unistd.h>

#ifdef XEN
#include <xen/domain.h>
#include <asm/hw_irq.h>
#include <asm/vmx.h>
#ifndef CONFIG_SMP
cpumask_t cpu_online_map = CPU_MASK_CPU0;
EXPORT_SYMBOL(cpu_online_map);
#endif
#endif

#ifdef CONFIG_SMP /* ifdef XEN */

#define SMP_DEBUG 0

#if SMP_DEBUG
#define Dprintk(x...)  printk(x)
#else
#define Dprintk(x...)
#endif

#ifdef CONFIG_HOTPLUG_CPU
/*
 * Store all idle threads, this can be reused instead of creating
 * a new thread. Also avoids complicated thread destroy functionality
 * for idle threads.
 */
struct task_struct *idle_thread_array[NR_CPUS];

/*
 * Global array allocated for NR_CPUS at boot time
 */
struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];

/*
 * start_ap in head.S uses this to store current booting cpu
 * info.
 */
struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];

#define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);

#define get_idle_for_cpu(x)		(idle_thread_array[(x)])
#define set_idle_for_cpu(x,p)	(idle_thread_array[(x)] = (p))

#else

#define get_idle_for_cpu(x)		(NULL)
#define set_idle_for_cpu(x,p)
#define set_brendez_area(x)
#endif


/*
 * ITC synchronization related stuff:
 */
#define MASTER	0
#define SLAVE	(SMP_CACHE_BYTES/8)

#define NUM_ROUNDS	64	/* magic value */
#define NUM_ITERS	5	/* likewise */

static DEFINE_SPINLOCK(itc_sync_lock);
static volatile unsigned long go[SLAVE + 1];

#define DEBUG_ITC_SYNC	0

extern void __devinit calibrate_delay (void);
extern void start_ap (void);
extern unsigned long ia64_iobase;

task_t *task_for_booting_cpu;

/*
 * State for each CPU
 */
DEFINE_PER_CPU(int, cpu_state);

/* Bitmasks of currently online, and possible CPUs */
cpumask_t cpu_online_map;
EXPORT_SYMBOL(cpu_online_map);
cpumask_t cpu_possible_map;
EXPORT_SYMBOL(cpu_possible_map);

cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
cpumask_t cpu_sibling_map[NR_CPUS] __cacheline_aligned;
int smp_num_siblings = 1;
int smp_num_cpucores = 1;

/* which logical CPU number maps to which CPU (physical APIC ID) */
volatile int ia64_cpu_to_sapicid[NR_CPUS];
EXPORT_SYMBOL(ia64_cpu_to_sapicid);

static volatile cpumask_t cpu_callin_map;

struct smp_boot_data smp_boot_data __initdata;

unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */

char __initdata no_int_routing;

unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */

static int __init
nointroute (char *str)
{
	no_int_routing = 1;
	printk ("no_int_routing on\n");
	return 1;
}

__setup("nointroute", nointroute);

static void fix_b0_for_bsp(void)
{
#ifdef CONFIG_HOTPLUG_CPU
	int cpuid;
	static int fix_bsp_b0 = 1;

	cpuid = smp_processor_id();

	/*
	 * Cache the b0 value on the first AP that comes up
	 */
	if (!(fix_bsp_b0 && cpuid))
		return;

	sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
	printk ("Fixed BSP b0 value from CPU %d\n", cpuid);

	fix_bsp_b0 = 0;
#endif
}

void
sync_master (void *arg)
{
	unsigned long flags, i;

	go[MASTER] = 0;

	local_irq_save(flags);
	{
		for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
			while (!go[MASTER])
				cpu_relax();
			go[MASTER] = 0;
			go[SLAVE] = ia64_get_itc();
		}
	}
	local_irq_restore(flags);
}

/*
 * Return the number of cycles by which our itc differs from the itc on the master
 * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
 * negative that it is behind.
 */
static inline long
#ifdef XEN /* warning cleanup */
get_delta (unsigned long *rt, unsigned long *master)
#else
get_delta (long *rt, long *master)
#endif
{
	unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
	unsigned long tcenter, t0, t1, tm;
	long i;

	for (i = 0; i < NUM_ITERS; ++i) {
		t0 = ia64_get_itc();
		go[MASTER] = 1;
		while (!(tm = go[SLAVE]))
			cpu_relax();
		go[SLAVE] = 0;
		t1 = ia64_get_itc();

		if (t1 - t0 < best_t1 - best_t0)
			best_t0 = t0, best_t1 = t1, best_tm = tm;
	}

	*rt = best_t1 - best_t0;
	*master = best_tm - best_t0;

	/* average best_t0 and best_t1 without overflow: */
	tcenter = (best_t0/2 + best_t1/2);
	if (best_t0 % 2 + best_t1 % 2 == 2)
		++tcenter;
	return tcenter - best_tm;
}

/*
 * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
 * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
 * unaccounted-for errors (such as getting a machine check in the middle of a calibration
 * step).  The basic idea is for the slave to ask the master what itc value it has and to
 * read its own itc before and after the master responds.  Each iteration gives us three
 * timestamps:
 *
 *	slave		master
 *
 *	t0 ---\
 *             ---\
 *		   --->
 *			tm
 *		   /---
 *	       /---
 *	t1 <---
 *
 *
 * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
 * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
 * between the slave and the master is symmetric.  Even if the interconnect were
 * asymmetric, we would still know that the synchronization error is smaller than the
 * roundtrip latency (t0 - t1).
 *
 * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
 * within one or two cycles.  However, we can only *guarantee* that the synchronization is
 * accurate to within a round-trip time, which is typically in the range of several
 * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
 * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
 * than half a micro second or so.
 */
void
ia64_sync_itc (unsigned int master)
{
	long i, delta, adj, adjust_latency = 0, done = 0;
	unsigned long flags, rt, master_time_stamp, bound;
#if DEBUG_ITC_SYNC
	struct {
		long rt;	/* roundtrip time */
		long master;	/* master's timestamp */
		long diff;	/* difference between midpoint and master's timestamp */
		long lat;	/* estimate of itc adjustment latency */
	} t[NUM_ROUNDS];
#endif

	/*
	 * Make sure local timer ticks are disabled while we sync.  If
	 * they were enabled, we'd have to worry about nasty issues
	 * like setting the ITC ahead of (or a long time before) the
	 * next scheduled tick.
	 */
	BUG_ON((ia64_get_itv() & (1 << 16)) == 0);

	go[MASTER] = 1;

	if (smp_call_function_single(master, sync_master, NULL, 1, 0) < 0) {
		printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
		return;
	}

	while (go[MASTER])
		cpu_relax();	/* wait for master to be ready */

	spin_lock_irqsave(&itc_sync_lock, flags);
	{
		for (i = 0; i < NUM_ROUNDS; ++i) {
			delta = get_delta(&rt, &master_time_stamp);
			if (delta == 0) {
				done = 1;	/* let's lock on to this... */
				bound = rt;
			}

			if (!done) {
				if (i > 0) {
					adjust_latency += -delta;
					adj = -delta + adjust_latency/4;
				} else
					adj = -delta;

				ia64_set_itc(ia64_get_itc() + adj);
			}
#if DEBUG_ITC_SYNC
			t[i].rt = rt;
			t[i].master = master_time_stamp;
			t[i].diff = delta;
			t[i].lat = adjust_latency/4;
#endif
		}
	}
	spin_unlock_irqrestore(&itc_sync_lock, flags);

#if DEBUG_ITC_SYNC
	for (i = 0; i < NUM_ROUNDS; ++i)
		printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
		       t[i].rt, t[i].master, t[i].diff, t[i].lat);
#endif

	printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
	       "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
}

/*
 * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
 */
static inline void __devinit
smp_setup_percpu_timer (void)
{
}

static void __devinit
smp_callin (void)
{
	int cpuid, phys_id;
	extern void ia64_init_itm(void);

#ifdef CONFIG_PERFMON
	extern void pfm_init_percpu(void);
#endif

	cpuid = smp_processor_id();
	phys_id = hard_smp_processor_id();

	if (cpu_online(cpuid)) {
		printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
		       phys_id, cpuid);
		BUG();
	}

	fix_b0_for_bsp();

	lock_ipi_calllock();
	cpu_set(cpuid, cpu_online_map);
	unlock_ipi_calllock();
	per_cpu(cpu_state, cpuid) = CPU_ONLINE;

	smp_setup_percpu_timer();

	ia64_mca_cmc_vector_setup();	/* Setup vector on AP */

#ifdef CONFIG_PERFMON
	pfm_init_percpu();
#endif

	local_irq_enable();

	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
		/*
		 * Synchronize the ITC with the BP.  Need to do this after irqs are
		 * enabled because ia64_sync_itc() calls smp_call_function_single(), which
		 * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
		 * local_bh_enable(), which bugs out if irqs are not enabled...
		 */
		Dprintk("Going to syncup ITC with BP.\n");
		ia64_sync_itc(0);
	}

	/*
	 * Get our bogomips.
	 */
	ia64_init_itm();
#ifndef XEN
	calibrate_delay();
#endif
	local_cpu_data->loops_per_jiffy = loops_per_jiffy;

#ifdef CONFIG_IA32_SUPPORT
	ia32_gdt_init();
#endif

	/*
	 * Allow the master to continue.
	 */
	cpu_set(cpuid, cpu_callin_map);
	Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
}


/*
 * Activate a secondary processor.  head.S calls this.
 */
int __devinit
start_secondary (void *unused)
{
	/* Early console may use I/O ports */
	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
	Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
#ifndef XEN
	efi_map_pal_code();
#endif
	cpu_init();
	smp_callin();

#ifdef XEN
	if (vmx_enabled)
		vmx_init_env(0, 0);

	startup_cpu_idle_loop();
#else
	cpu_idle();
#endif
	return 0;
}

struct pt_regs * __devinit idle_regs(struct pt_regs *regs)
{
	return NULL;
}

#ifndef XEN
struct create_idle {
	struct task_struct *idle;
	struct completion done;
	int cpu;
};

void
do_fork_idle(void *_c_idle)
{
	struct create_idle *c_idle = _c_idle;

	c_idle->idle = fork_idle(c_idle->cpu);
	complete(&c_idle->done);
}
#endif

static int __devinit
do_boot_cpu (int sapicid, int cpu)
{