smp.c

来自「是关于linux2.5.1的完全源码」· C语言 代码 · 共 1,263 行 · 第 1/2 页

/*
 *	linux/arch/alpha/kernel/smp.c
 *
 *      2001-07-09 Phil Ezolt ([e-mail])
 *            Renamed modified smp_call_function to smp_call_function_on_cpu()
 *            Created an function that conforms to the old calling convention
 *            of smp_call_function().
 *
 *            This is helpful for DCPI.
 *
 */

#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/kernel_stat.h>
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/threads.h>
#include <linux/smp.h>
#include <linux/smp_lock.h>
#include <linux/interrupt.h>
#include <linux/init.h>
#include <linux/delay.h>
#include <linux/spinlock.h>
#include <linux/irq.h>
#include <linux/cache.h>

#include <asm/hwrpb.h>
#include <asm/ptrace.h>
#include <asm/atomic.h>

#include <asm/io.h>
#include <asm/irq.h>
#include <asm/bitops.h>
#include <asm/pgtable.h>
#include <asm/pgalloc.h>
#include <asm/hardirq.h>
#include <asm/softirq.h>
#include <asm/mmu_context.h>

#define __KERNEL_SYSCALLS__
#include <asm/unistd.h>

#include "proto.h"
#include "irq_impl.h"


#define DEBUG_SMP 0
#if DEBUG_SMP
#define DBGS(args)	printk args
#else
#define DBGS(args)
#endif

/* A collection of per-processor data.  */
struct cpuinfo_alpha cpu_data[NR_CPUS];

/* A collection of single bit ipi messages.  */
static struct {
	unsigned long bits ____cacheline_aligned;
} ipi_data[NR_CPUS] __cacheline_aligned;

enum ipi_message_type {
	IPI_RESCHEDULE,
	IPI_MIGRATION,
	IPI_CALL_FUNC,
	IPI_CPU_STOP,
};

spinlock_t kernel_flag __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;

/* Set to a secondary's cpuid when it comes online.  */
static int smp_secondary_alive __initdata = 0;

/* Which cpus ids came online.  */
unsigned long cpu_present_mask;

/* cpus reported in the hwrpb */
static unsigned long hwrpb_cpu_present_mask __initdata = 0;

static int max_cpus = -1;	/* Command-line limitation.  */
int smp_num_probed;		/* Internal processor count */
int smp_num_cpus = 1;		/* Number that came online.  */
int smp_threads_ready;		/* True once the per process idle is forked. */
cycles_t cacheflush_time;
unsigned long cache_decay_ticks;

int __cpu_number_map[NR_CPUS];
int __cpu_logical_map[NR_CPUS];

extern void calibrate_delay(void);
extern asmlinkage void entInt(void);


static int __init nosmp(char *str)
{
	max_cpus = 0;
	return 1;
}

__setup("nosmp", nosmp);

static int __init maxcpus(char *str)
{
	get_option(&str, &max_cpus);
	return 1;
}

__setup("maxcpus", maxcpus);


/*
 * Called by both boot and secondaries to move global data into
 *  per-processor storage.
 */
static inline void __init
smp_store_cpu_info(int cpuid)
{
	cpu_data[cpuid].loops_per_jiffy = loops_per_jiffy;
	cpu_data[cpuid].last_asn = ASN_FIRST_VERSION;
	cpu_data[cpuid].need_new_asn = 0;
	cpu_data[cpuid].asn_lock = 0;
	local_irq_count(cpuid) = 0;
	local_bh_count(cpuid) = 0;
}

/*
 * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
 */
static inline void __init
smp_setup_percpu_timer(int cpuid)
{
	cpu_data[cpuid].prof_counter = 1;
	cpu_data[cpuid].prof_multiplier = 1;
}

static void __init
wait_boot_cpu_to_stop(int cpuid)
{
	long stop = jiffies + 10*HZ;

	while (time_before(jiffies, stop)) {
	        if (!smp_secondary_alive)
			return;
		barrier();
	}

	printk("wait_boot_cpu_to_stop: FAILED on CPU %d, hanging now\n", cpuid);
	for (;;)
		barrier();
}

/*
 * Where secondaries begin a life of C.
 */
void __init
smp_callin(void)
{
	int cpuid = hard_smp_processor_id();

	/* Turn on machine checks.  */
	wrmces(7);

	/* Set trap vectors.  */
	trap_init();

	/* Set interrupt vector.  */
	wrent(entInt, 0);

	/* Get our local ticker going. */
	smp_setup_percpu_timer(cpuid);

	/* All kernel threads share the same mm context.  */
	atomic_inc(&init_mm.mm_count);
	current->active_mm = &init_mm;

	/* Must have completely accurate bogos.  */
	__sti();

	/* Wait boot CPU to stop with irq enabled before running
	   calibrate_delay. */
	wait_boot_cpu_to_stop(cpuid);
	mb();
	calibrate_delay();

	smp_store_cpu_info(cpuid);
	/* Allow master to continue only after we written loops_per_jiffy.  */
	wmb();
	smp_secondary_alive = 1;

	/* Wait for the go code.  */
	while (!smp_threads_ready)
		barrier();

	DBGS(("smp_callin: commencing CPU %d current %p active_mm %p\n",
	      cpuid, current, current->active_mm));

	/* Do nothing.  */
	cpu_idle();
}


/*
 * Rough estimation for SMP scheduling, this is the number of cycles it
 * takes for a fully memory-limited process to flush the SMP-local cache.
 *
 * We are not told how much cache there is, so we have to guess.
 */
static void __init
smp_tune_scheduling (int cpuid)
{
	struct percpu_struct *cpu;
	unsigned long on_chip_cache;	/* kB */
	unsigned long freq;		/* Hz */
	unsigned long bandwidth = 350;	/* MB/s */

	cpu = (struct percpu_struct*)((char*)hwrpb + hwrpb->processor_offset
				      + cpuid * hwrpb->processor_size);
	switch (cpu->type)
	{
	case EV45_CPU:
		on_chip_cache = 16 + 16;
		break;

	case EV5_CPU:
	case EV56_CPU:
		on_chip_cache = 8 + 8 + 96;
		break;

	case PCA56_CPU:
		on_chip_cache = 16 + 8;
		break;

	case EV6_CPU:
	case EV67_CPU:
	default:
		on_chip_cache = 64 + 64;
		break;
	}

	freq = hwrpb->cycle_freq ? : est_cycle_freq;

	cacheflush_time = (freq / 1000000) * (on_chip_cache << 10) / bandwidth;
	cache_decay_ticks = cacheflush_time / (freq / 1000) * HZ / 1000;

	printk("per-CPU timeslice cutoff: %ld.%02ld usecs.\n",
	       cacheflush_time/(freq/1000000),
	       (cacheflush_time*100/(freq/1000000)) % 100);
	printk("task migration cache decay timeout: %ld msecs.\n",
	       (cache_decay_ticks + 1) * 1000 / HZ);
}

/* Wait until hwrpb->txrdy is clear for cpu.  Return -1 on timeout.  */
static int __init
wait_for_txrdy (unsigned long cpumask)
{
	unsigned long timeout;

	if (!(hwrpb->txrdy & cpumask))
		return 0;

	timeout = jiffies + 10*HZ;
	while (time_before(jiffies, timeout)) {
		if (!(hwrpb->txrdy & cpumask))
			return 0;
		udelay(10);
		barrier();
	}

	return -1;
}

/*
 * Send a message to a secondary's console.  "START" is one such
 * interesting message.  ;-)
 */
static void __init
send_secondary_console_msg(char *str, int cpuid)
{
	struct percpu_struct *cpu;
	register char *cp1, *cp2;
	unsigned long cpumask;
	size_t len;

	cpu = (struct percpu_struct *)
		((char*)hwrpb
		 + hwrpb->processor_offset
		 + cpuid * hwrpb->processor_size);

	cpumask = (1UL << cpuid);
	if (wait_for_txrdy(cpumask))
		goto timeout;

	cp2 = str;
	len = strlen(cp2);
	*(unsigned int *)&cpu->ipc_buffer[0] = len;
	cp1 = (char *) &cpu->ipc_buffer[1];
	memcpy(cp1, cp2, len);

	/* atomic test and set */
	wmb();
	set_bit(cpuid, &hwrpb->rxrdy);

	if (wait_for_txrdy(cpumask))
		goto timeout;
	return;

 timeout:
	printk("Processor %x not ready\n", cpuid);
}

/*
 * A secondary console wants to send a message.  Receive it.
 */
static void
recv_secondary_console_msg(void)
{
	int mycpu, i, cnt;
	unsigned long txrdy = hwrpb->txrdy;
	char *cp1, *cp2, buf[80];
	struct percpu_struct *cpu;

	DBGS(("recv_secondary_console_msg: TXRDY 0x%lx.\n", txrdy));

	mycpu = hard_smp_processor_id();

	for (i = 0; i < NR_CPUS; i++) {
		if (!(txrdy & (1UL << i)))
			continue;

		DBGS(("recv_secondary_console_msg: "
		      "TXRDY contains CPU %d.\n", i));

		cpu = (struct percpu_struct *)
		  ((char*)hwrpb
		   + hwrpb->processor_offset
		   + i * hwrpb->processor_size);

 		DBGS(("recv_secondary_console_msg: on %d from %d"
		      " HALT_REASON 0x%lx FLAGS 0x%lx\n",
		      mycpu, i, cpu->halt_reason, cpu->flags));

		cnt = cpu->ipc_buffer[0] >> 32;
		if (cnt <= 0 || cnt >= 80)
			strcpy(buf, "<<< BOGUS MSG >>>");
		else {
			cp1 = (char *) &cpu->ipc_buffer[11];
			cp2 = buf;
			strcpy(cp2, cp1);
			
			while ((cp2 = strchr(cp2, '\r')) != 0) {
				*cp2 = ' ';
				if (cp2[1] == '\n')
					cp2[1] = ' ';
			}
		}

		DBGS((KERN_INFO "recv_secondary_console_msg: on %d "
		      "message is '%s'\n", mycpu, buf));
	}

	hwrpb->txrdy = 0;
}

/*
 * Convince the console to have a secondary cpu begin execution.
 */
static int __init
secondary_cpu_start(int cpuid, struct task_struct *idle)
{
	struct percpu_struct *cpu;
	struct pcb_struct *hwpcb, *ipcb;
	long timeout;
	  
	cpu = (struct percpu_struct *)
		((char*)hwrpb
		 + hwrpb->processor_offset
		 + cpuid * hwrpb->processor_size);
	hwpcb = (struct pcb_struct *) cpu->hwpcb;
	ipcb = &idle->thread_info->pcb;

	/* Initialize the CPU's HWPCB to something just good enough for
	   us to get started.  Immediately after starting, we'll swpctx
	   to the target idle task's pcb.  Reuse the stack in the mean
	   time.  Precalculate the target PCBB.  */
	hwpcb->ksp = (unsigned long)ipcb + sizeof(union thread_union) - 16;
	hwpcb->usp = 0;
	hwpcb->ptbr = ipcb->ptbr;
	hwpcb->pcc = 0;
	hwpcb->asn = 0;
	hwpcb->unique = virt_to_phys(ipcb);
	hwpcb->flags = ipcb->flags;
	hwpcb->res1 = hwpcb->res2 = 0;

#if 0
	DBGS(("KSP 0x%lx PTBR 0x%lx VPTBR 0x%lx UNIQUE 0x%lx\n",
	      hwpcb->ksp, hwpcb->ptbr, hwrpb->vptb, hwpcb->unique));
#endif
	DBGS(("Starting secondary cpu %d: state 0x%lx pal_flags 0x%lx\n",
	      cpuid, idle->state, ipcb->flags));

	/* Setup HWRPB fields that SRM uses to activate secondary CPU */
	hwrpb->CPU_restart = __smp_callin;
	hwrpb->CPU_restart_data = (unsigned long) __smp_callin;

	/* Recalculate and update the HWRPB checksum */
	hwrpb_update_checksum(hwrpb);

	/*
	 * Send a "start" command to the specified processor.
	 */

	/* SRM III 3.4.1.3 */
	cpu->flags |= 0x22;	/* turn on Context Valid and Restart Capable */
	cpu->flags &= ~1;	/* turn off Bootstrap In Progress */
	wmb();

	send_secondary_console_msg("START\r\n", cpuid);

	/* Wait 10 seconds for an ACK from the console.  */
	timeout = jiffies + 10*HZ;
	while (time_before(jiffies, timeout)) {
		if (cpu->flags & 1)
			goto started;
		udelay(10);
		barrier();
	}
	printk(KERN_ERR "SMP: Processor %d failed to start.\n", cpuid);
	return -1;

 started:
	DBGS(("secondary_cpu_start: SUCCESS for CPU %d!!!\n", cpuid));
	return 0;
}

static int __init
fork_by_hand(void)
{
	/* Don't care about the contents of regs since we'll never
	   reschedule the forked task. */
	struct pt_regs regs;
	return do_fork(CLONE_VM|CLONE_PID, 0, &regs, 0);
}

/*
 * Bring one cpu online.
 */
static int __init
smp_boot_one_cpu(int cpuid, int cpunum)
{
	struct task_struct *idle;
	long timeout;

	/* Cook up an idler for this guy.  Note that the address we
	   give to kernel_thread is irrelevant -- it's going to start
	   where HWRPB.CPU_restart says to start.  But this gets all
	   the other task-y sort of data structures set up like we
	   wish.  We can't use kernel_thread since we must avoid
	   rescheduling the child.  */
	if (fork_by_hand() < 0)
		panic("failed fork for CPU %d", cpuid);

	idle = prev_task(&init_task);
	if (!idle)
		panic("No idle process for CPU %d", cpuid);

	init_idle(idle, cpuid);
	unhash_process(idle);

	__cpu_logical_map[cpunum] = cpuid;
	__cpu_number_map[cpuid] = cpunum;

	DBGS(("smp_boot_one_cpu: CPU %d state 0x%lx flags 0x%lx\n",
	      cpuid, idle->state, idle->flags));

	/* Signal the secondary to wait a moment.  */
	smp_secondary_alive = -1;

	/* Whirrr, whirrr, whirrrrrrrrr... */
	if (secondary_cpu_start(cpuid, idle))
		return -1;

	/* Notify the secondary CPU it can run calibrate_delay.  */
	mb();
	smp_secondary_alive = 0;

	/* We've been acked by the console; wait one second for
	   the task to start up for real.  */
	timeout = jiffies + 1*HZ;
	while (time_before(jiffies, timeout)) {
		if (smp_secondary_alive == 1)
			goto alive;
		udelay(10);
		barrier();
	}

	/* We must invalidate our stuff as we failed to boot the CPU.  */
	__cpu_logical_map[cpunum] = -1;
	__cpu_number_map[cpuid] = -1;

	printk(KERN_ERR "SMP: Processor %d is stuck.\n", cpuid);
	return -1;

 alive:
	/* Another "Red Snapper". */
	return 0;
}

/*
 * Called from setup_arch.  Detect an SMP system and which processors
 * are present.
 */
void __init
setup_smp(void)
{
	struct percpu_struct *cpubase, *cpu;
	int i;

	if (boot_cpuid != 0) {
		printk(KERN_WARNING "SMP: Booting off cpu %d instead of 0?\n",
		       boot_cpuid);
	}

	if (hwrpb->nr_processors > 1) {
		int boot_cpu_palrev;

		DBGS(("setup_smp: nr_processors %ld\n",
		      hwrpb->nr_processors));

		cpubase = (struct percpu_struct *)
			((char*)hwrpb + hwrpb->processor_offset);
		boot_cpu_palrev = cpubase->pal_revision;

		for (i = 0; i < hwrpb->nr_processors; i++ ) {
			cpu = (struct percpu_struct *)
				((char *)cpubase + i*hwrpb->processor_size);
			if ((cpu->flags & 0x1cc) == 0x1cc) {
				smp_num_probed++;
				/* Assume here that "whami" == index */
				hwrpb_cpu_present_mask |= (1UL << i);
				cpu->pal_revision = boot_cpu_palrev;
			}

			DBGS(("setup_smp: CPU %d: flags 0x%lx type 0x%lx\n",
			      i, cpu->flags, cpu->type));
			DBGS(("setup_smp: CPU %d: PAL rev 0x%lx\n",
			      i, cpu->pal_revision));
		}
	} else {
		smp_num_probed = 1;
		hwrpb_cpu_present_mask = (1UL << boot_cpuid);
	}
	cpu_present_mask = 1UL << boot_cpuid;

	printk(KERN_INFO "SMP: %d CPUs probed -- cpu_present_mask = %lx\n",
	       smp_num_probed, hwrpb_cpu_present_mask);
}

/*
 * Called by smp_init bring all the secondaries online and hold them.
 */
void __init
smp_boot_cpus(void)
{
	int cpu_count, i;
	unsigned long bogosum;

	/* Take care of some initial bookkeeping.  */
	memset(__cpu_number_map, -1, sizeof(__cpu_number_map));
	memset(__cpu_logical_map, -1, sizeof(__cpu_logical_map));
	memset(ipi_data, 0, sizeof(ipi_data));

	__cpu_number_map[boot_cpuid] = 0;
	__cpu_logical_map[0] = boot_cpuid;
	current_thread_info()->cpu = boot_cpuid;

	smp_store_cpu_info(boot_cpuid);
	smp_tune_scheduling(boot_cpuid);
	smp_setup_percpu_timer(boot_cpuid);

	/* Nothing to do on a UP box, or when told not to.  */
	if (smp_num_probed == 1 || max_cpus == 0) {
		cpu_present_mask = 1UL << boot_cpuid;
		printk(KERN_INFO "SMP mode deactivated.\n");
		return;
	}

	printk(KERN_INFO "SMP starting up secondaries.\n");

	cpu_count = 1;
	for (i = 0; i < NR_CPUS; i++) {
		if (i == boot_cpuid)
			continue;

		if (((hwrpb_cpu_present_mask >> i) & 1) == 0)
			continue;

		if (smp_boot_one_cpu(i, cpu_count))
			continue;

		cpu_present_mask |= 1UL << i;
		cpu_count++;
	}

	if (cpu_count == 1) {
		printk(KERN_ERR "SMP: Only one lonely processor alive.\n");
		return;
	}

	bogosum = 0;
	for (i = 0; i < NR_CPUS; i++) {
		if (cpu_present_mask & (1UL << i))
			bogosum += cpu_data[i].loops_per_jiffy;
	}
	printk(KERN_INFO "SMP: Total of %d processors activated "
	       "(%lu.%02lu BogoMIPS).\n",
	       cpu_count, (bogosum + 2500) / (500000/HZ),
	       ((bogosum + 2500) / (5000/HZ)) % 100);

	smp_num_cpus = cpu_count;
}

/*
 * Called by smp_init to release the blocking online cpus once they 
 * are all started.
 */
void __init
smp_commence(void)
{
	/* smp_init sets smp_threads_ready -- that's enough.  */
	mb();
}