smpboot.c

来自「是关于linux2.5.1的完全源码」· C语言 代码 · 共 1,256 行 · 第 1/3 页

/*
 *	x86 SMP booting functions
 *
 *	(c) 1995 Alan Cox, Building #3 <[e-mail]>
 *	(c) 1998, 1999, 2000 Ingo Molnar <[e-mail]>
 *
 *	Much of the core SMP work is based on previous work by Thomas Radke, to
 *	whom a great many thanks are extended.
 *
 *	Thanks to Intel for making available several different Pentium,
 *	Pentium Pro and Pentium-II/Xeon MP machines.
 *	Original development of Linux SMP code supported by Caldera.
 *
 *	This code is released under the GNU General Public License version 2 or
 *	later.
 *
 *	Fixes
 *		Felix Koop	:	NR_CPUS used properly
 *		Jose Renau	:	Handle single CPU case.
 *		Alan Cox	:	By repeated request 8) - Total BogoMIP report.
 *		Greg Wright	:	Fix for kernel stacks panic.
 *		Erich Boleyn	:	MP v1.4 and additional changes.
 *	Matthias Sattler	:	Changes for 2.1 kernel map.
 *	Michel Lespinasse	:	Changes for 2.1 kernel map.
 *	Michael Chastain	:	Change trampoline.S to gnu as.
 *		Alan Cox	:	Dumb bug: 'B' step PPro's are fine
 *		Ingo Molnar	:	Added APIC timers, based on code
 *					from Jose Renau
 *		Ingo Molnar	:	various cleanups and rewrites
 *		Tigran Aivazian	:	fixed "0.00 in /proc/uptime on SMP" bug.
 *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs
 *		Martin J. Bligh	: 	Added support for multi-quad systems
 *		Dave Jones	:	Report invalid combinations of Athlon CPUs.
 */

#include <linux/config.h>
#include <linux/init.h>
#include <linux/kernel.h>

#include <linux/mm.h>
#include <linux/kernel_stat.h>
#include <linux/smp_lock.h>
#include <linux/irq.h>
#include <linux/bootmem.h>

#include <linux/delay.h>
#include <linux/mc146818rtc.h>
#include <asm/mtrr.h>
#include <asm/pgalloc.h>
#include <asm/tlbflush.h>
#include <asm/smpboot.h>

/* Set if we find a B stepping CPU			*/
static int smp_b_stepping;

/* Setup configured maximum number of CPUs to activate */
static int max_cpus = -1;

/* Total count of live CPUs */
int smp_num_cpus = 1;

/* Number of siblings per CPU package */
int smp_num_siblings = 1;
int __initdata phys_proc_id[NR_CPUS]; /* Package ID of each logical CPU */

/* Bitmask of currently online CPUs */
unsigned long cpu_online_map;

static volatile unsigned long cpu_callin_map;
static volatile unsigned long cpu_callout_map;

/* Per CPU bogomips and other parameters */
struct cpuinfo_x86 cpu_data[NR_CPUS] __cacheline_aligned;

/* Set when the idlers are all forked */
int smp_threads_ready;

/*
 * Setup routine for controlling SMP activation
 *
 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 * activation entirely (the MPS table probe still happens, though).
 *
 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 * greater than 0, limits the maximum number of CPUs activated in
 * SMP mode to <NUM>.
 */

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);

/*
 * Trampoline 80x86 program as an array.
 */

extern unsigned char trampoline_data [];
extern unsigned char trampoline_end  [];
static unsigned char *trampoline_base;

/*
 * Currently trivial. Write the real->protected mode
 * bootstrap into the page concerned. The caller
 * has made sure it's suitably aligned.
 */

static unsigned long __init setup_trampoline(void)
{
	memcpy(trampoline_base, trampoline_data, trampoline_end - trampoline_data);
	return virt_to_phys(trampoline_base);
}

/*
 * We are called very early to get the low memory for the
 * SMP bootup trampoline page.
 */
void __init smp_alloc_memory(void)
{
	trampoline_base = (void *) alloc_bootmem_low_pages(PAGE_SIZE);
	/*
	 * Has to be in very low memory so we can execute
	 * real-mode AP code.
	 */
	if (__pa(trampoline_base) >= 0x9F000)
		BUG();
}

/*
 * The bootstrap kernel entry code has set these up. Save them for
 * a given CPU
 */

void __init smp_store_cpu_info(int id)
{
	struct cpuinfo_x86 *c = cpu_data + id;

	*c = boot_cpu_data;
	identify_cpu(c);
	/*
	 * Mask B, Pentium, but not Pentium MMX
	 */
	if (c->x86_vendor == X86_VENDOR_INTEL &&
	    c->x86 == 5 &&
	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
	    c->x86_model <= 3)
		/*
		 * Remember we have B step Pentia with bugs
		 */
		smp_b_stepping = 1;

	/*
	 * Certain Athlons might work (for various values of 'work') in SMP
	 * but they are not certified as MP capable.
	 */
	if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {

		/* Athlon 660/661 is valid. */	
		if ((c->x86_model==6) && ((c->x86_mask==0) || (c->x86_mask==1)))
			goto valid_k7;

		/* Duron 670 is valid */
		if ((c->x86_model==7) && (c->x86_mask==0))
			goto valid_k7;

		/* Athlon 662, Duron 671, and Athlon >model 7 have capability bit */
		if (((c->x86_model==6) && (c->x86_mask>=2)) ||
		    ((c->x86_model==7) && (c->x86_mask>=1)) ||
		     (c->x86_model> 7))
			if (cpu_has_mp)
				goto valid_k7;

		/* If we get here, it's not a certified SMP capable AMD system. */
		printk (KERN_INFO "WARNING: This combination of AMD processors is not suitable for SMP.\n");
		tainted |= TAINT_UNSAFE_SMP;
		
	}
valid_k7:

}

/*
 * Architecture specific routine called by the kernel just before init is
 * fired off. This allows the BP to have everything in order [we hope].
 * At the end of this all the APs will hit the system scheduling and off
 * we go. Each AP will load the system gdt's and jump through the kernel
 * init into idle(). At this point the scheduler will one day take over
 * and give them jobs to do. smp_callin is a standard routine
 * we use to track CPUs as they power up.
 */

static atomic_t smp_commenced = ATOMIC_INIT(0);

void __init smp_commence(void)
{
	/*
	 * Lets the callins below out of their loop.
	 */
	Dprintk("Setting commenced=1, go go go\n");

	wmb();
	atomic_set(&smp_commenced,1);
}

/*
 * TSC synchronization.
 *
 * We first check wether all CPUs have their TSC's synchronized,
 * then we print a warning if not, and always resync.
 */

static atomic_t tsc_start_flag = ATOMIC_INIT(0);
static atomic_t tsc_count_start = ATOMIC_INIT(0);
static atomic_t tsc_count_stop = ATOMIC_INIT(0);
static unsigned long long tsc_values[NR_CPUS];

#define NR_LOOPS 5

extern unsigned long fast_gettimeoffset_quotient;

/*
 * accurate 64-bit/32-bit division, expanded to 32-bit divisions and 64-bit
 * multiplication. Not terribly optimized but we need it at boot time only
 * anyway.
 *
 * result == a / b
 *	== (a1 + a2*(2^32)) / b
 *	== a1/b + a2*(2^32/b)
 *	== a1/b + a2*((2^32-1)/b) + a2/b + (a2*((2^32-1) % b))/b
 *		    ^---- (this multiplication can overflow)
 */

static unsigned long long __init div64 (unsigned long long a, unsigned long b0)
{
	unsigned int a1, a2;
	unsigned long long res;

	a1 = ((unsigned int*)&a)[0];
	a2 = ((unsigned int*)&a)[1];

	res = a1/b0 +
		(unsigned long long)a2 * (unsigned long long)(0xffffffff/b0) +
		a2 / b0 +
		(a2 * (0xffffffff % b0)) / b0;

	return res;
}

static void __init synchronize_tsc_bp (void)
{
	int i;
	unsigned long long t0;
	unsigned long long sum, avg;
	long long delta;
	unsigned long one_usec;
	int buggy = 0;

	printk("checking TSC synchronization across CPUs: ");

	one_usec = ((1<<30)/fast_gettimeoffset_quotient)*(1<<2);

	atomic_set(&tsc_start_flag, 1);
	wmb();

	/*
	 * We loop a few times to get a primed instruction cache,
	 * then the last pass is more or less synchronized and
	 * the BP and APs set their cycle counters to zero all at
	 * once. This reduces the chance of having random offsets
	 * between the processors, and guarantees that the maximum
	 * delay between the cycle counters is never bigger than
	 * the latency of information-passing (cachelines) between
	 * two CPUs.
	 */
	for (i = 0; i < NR_LOOPS; i++) {
		/*
		 * all APs synchronize but they loop on '== num_cpus'
		 */
		while (atomic_read(&tsc_count_start) != smp_num_cpus-1) mb();
		atomic_set(&tsc_count_stop, 0);
		wmb();
		/*
		 * this lets the APs save their current TSC:
		 */
		atomic_inc(&tsc_count_start);

		rdtscll(tsc_values[smp_processor_id()]);
		/*
		 * We clear the TSC in the last loop:
		 */
		if (i == NR_LOOPS-1)
			write_tsc(0, 0);

		/*
		 * Wait for all APs to leave the synchronization point:
		 */
		while (atomic_read(&tsc_count_stop) != smp_num_cpus-1) mb();
		atomic_set(&tsc_count_start, 0);
		wmb();
		atomic_inc(&tsc_count_stop);
	}

	sum = 0;
	for (i = 0; i < smp_num_cpus; i++) {
		t0 = tsc_values[i];
		sum += t0;
	}
	avg = div64(sum, smp_num_cpus);

	sum = 0;
	for (i = 0; i < smp_num_cpus; i++) {
		delta = tsc_values[i] - avg;
		if (delta < 0)
			delta = -delta;
		/*
		 * We report bigger than 2 microseconds clock differences.
		 */
		if (delta > 2*one_usec) {
			long realdelta;
			if (!buggy) {
				buggy = 1;
				printk("\n");
			}
			realdelta = div64(delta, one_usec);
			if (tsc_values[i] < avg)
				realdelta = -realdelta;

			printk("BIOS BUG: CPU#%d improperly initialized, has %ld usecs TSC skew! FIXED.\n", i, realdelta);
		}

		sum += delta;
	}
	if (!buggy)
		printk("passed.\n");
		;
}

static void __init synchronize_tsc_ap (void)
{
	int i;

	/*
	 * smp_num_cpus is not necessarily known at the time
	 * this gets called, so we first wait for the BP to
	 * finish SMP initialization:
	 */
	while (!atomic_read(&tsc_start_flag)) mb();

	for (i = 0; i < NR_LOOPS; i++) {
		atomic_inc(&tsc_count_start);
		while (atomic_read(&tsc_count_start) != smp_num_cpus) mb();

		rdtscll(tsc_values[smp_processor_id()]);
		if (i == NR_LOOPS-1)
			write_tsc(0, 0);

		atomic_inc(&tsc_count_stop);
		while (atomic_read(&tsc_count_stop) != smp_num_cpus) mb();
	}
}
#undef NR_LOOPS

extern void calibrate_delay(void);

static atomic_t init_deasserted;

void __init smp_callin(void)
{
	int cpuid, phys_id;
	unsigned long timeout;

	/*
	 * If waken up by an INIT in an 82489DX configuration
	 * we may get here before an INIT-deassert IPI reaches
	 * our local APIC.  We have to wait for the IPI or we'll
	 * lock up on an APIC access.
	 */
	if (!clustered_apic_mode) 
		while (!atomic_read(&init_deasserted));

	/*
	 * (This works even if the APIC is not enabled.)
	 */
	phys_id = GET_APIC_ID(apic_read(APIC_ID));
	cpuid = smp_processor_id();
	if (test_and_set_bit(cpuid, &cpu_online_map)) {
		printk("huh, phys CPU#%d, CPU#%d already present??\n",
					phys_id, cpuid);
		BUG();
	}
	Dprintk("CPU#%d (phys ID: %d) waiting for CALLOUT\n", cpuid, phys_id);

	/*
	 * STARTUP IPIs are fragile beasts as they might sometimes
	 * trigger some glue motherboard logic. Complete APIC bus
	 * silence for 1 second, this overestimates the time the
	 * boot CPU is spending to send the up to 2 STARTUP IPIs
	 * by a factor of two. This should be enough.
	 */

	/*
	 * Waiting 2s total for startup (udelay is not yet working)
	 */
	timeout = jiffies + 2*HZ;
	while (time_before(jiffies, timeout)) {
		/*
		 * Has the boot CPU finished it's STARTUP sequence?
		 */