xref: /linux/arch/x86/kernel/apic/apic.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
1 /*
2  *	Local APIC handling, local APIC timers
3  *
4  *	(c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
5  *
6  *	Fixes
7  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs;
8  *					thanks to Eric Gilmore
9  *					and Rolf G. Tews
10  *					for testing these extensively.
11  *	Maciej W. Rozycki	:	Various updates and fixes.
12  *	Mikael Pettersson	:	Power Management for UP-APIC.
13  *	Pavel Machek and
14  *	Mikael Pettersson	:	PM converted to driver model.
15  */
16 
17 #include <linux/perf_event.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/mc146818rtc.h>
20 #include <linux/acpi_pmtmr.h>
21 #include <linux/clockchips.h>
22 #include <linux/interrupt.h>
23 #include <linux/bootmem.h>
24 #include <linux/ftrace.h>
25 #include <linux/ioport.h>
26 #include <linux/module.h>
27 #include <linux/syscore_ops.h>
28 #include <linux/delay.h>
29 #include <linux/timex.h>
30 #include <linux/i8253.h>
31 #include <linux/dmar.h>
32 #include <linux/init.h>
33 #include <linux/cpu.h>
34 #include <linux/dmi.h>
35 #include <linux/smp.h>
36 #include <linux/mm.h>
37 
38 #include <asm/irq_remapping.h>
39 #include <asm/perf_event.h>
40 #include <asm/x86_init.h>
41 #include <asm/pgalloc.h>
42 #include <linux/atomic.h>
43 #include <asm/mpspec.h>
44 #include <asm/i8259.h>
45 #include <asm/proto.h>
46 #include <asm/apic.h>
47 #include <asm/io_apic.h>
48 #include <asm/desc.h>
49 #include <asm/hpet.h>
50 #include <asm/idle.h>
51 #include <asm/mtrr.h>
52 #include <asm/time.h>
53 #include <asm/smp.h>
54 #include <asm/mce.h>
55 #include <asm/tsc.h>
56 #include <asm/hypervisor.h>
57 
58 unsigned int num_processors;
59 
60 unsigned disabled_cpus __cpuinitdata;
61 
62 /* Processor that is doing the boot up */
63 unsigned int boot_cpu_physical_apicid = -1U;
64 
65 /*
66  * The highest APIC ID seen during enumeration.
67  */
68 unsigned int max_physical_apicid;
69 
70 /*
71  * Bitmask of physically existing CPUs:
72  */
73 physid_mask_t phys_cpu_present_map;
74 
75 /*
76  * Map cpu index to physical APIC ID
77  */
78 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
79 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
80 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
81 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
82 
83 #ifdef CONFIG_X86_32
84 
85 /*
86  * On x86_32, the mapping between cpu and logical apicid may vary
87  * depending on apic in use.  The following early percpu variable is
88  * used for the mapping.  This is where the behaviors of x86_64 and 32
89  * actually diverge.  Let's keep it ugly for now.
90  */
91 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
92 
93 /* Local APIC was disabled by the BIOS and enabled by the kernel */
94 static int enabled_via_apicbase;
95 
96 /*
97  * Handle interrupt mode configuration register (IMCR).
98  * This register controls whether the interrupt signals
99  * that reach the BSP come from the master PIC or from the
100  * local APIC. Before entering Symmetric I/O Mode, either
101  * the BIOS or the operating system must switch out of
102  * PIC Mode by changing the IMCR.
103  */
104 static inline void imcr_pic_to_apic(void)
105 {
106 	/* select IMCR register */
107 	outb(0x70, 0x22);
108 	/* NMI and 8259 INTR go through APIC */
109 	outb(0x01, 0x23);
110 }
111 
112 static inline void imcr_apic_to_pic(void)
113 {
114 	/* select IMCR register */
115 	outb(0x70, 0x22);
116 	/* NMI and 8259 INTR go directly to BSP */
117 	outb(0x00, 0x23);
118 }
119 #endif
120 
121 /*
122  * Knob to control our willingness to enable the local APIC.
123  *
124  * +1=force-enable
125  */
126 static int force_enable_local_apic __initdata;
127 /*
128  * APIC command line parameters
129  */
130 static int __init parse_lapic(char *arg)
131 {
132 	if (config_enabled(CONFIG_X86_32) && !arg)
133 		force_enable_local_apic = 1;
134 	else if (!strncmp(arg, "notscdeadline", 13))
135 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
136 	return 0;
137 }
138 early_param("lapic", parse_lapic);
139 
140 #ifdef CONFIG_X86_64
141 static int apic_calibrate_pmtmr __initdata;
142 static __init int setup_apicpmtimer(char *s)
143 {
144 	apic_calibrate_pmtmr = 1;
145 	notsc_setup(NULL);
146 	return 0;
147 }
148 __setup("apicpmtimer", setup_apicpmtimer);
149 #endif
150 
151 int x2apic_mode;
152 #ifdef CONFIG_X86_X2APIC
153 /* x2apic enabled before OS handover */
154 int x2apic_preenabled;
155 static int x2apic_disabled;
156 static int nox2apic;
157 static __init int setup_nox2apic(char *str)
158 {
159 	if (x2apic_enabled()) {
160 		int apicid = native_apic_msr_read(APIC_ID);
161 
162 		if (apicid >= 255) {
163 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
164 				   apicid);
165 			return 0;
166 		}
167 
168 		pr_warning("x2apic already enabled. will disable it\n");
169 	} else
170 		setup_clear_cpu_cap(X86_FEATURE_X2APIC);
171 
172 	nox2apic = 1;
173 
174 	return 0;
175 }
176 early_param("nox2apic", setup_nox2apic);
177 #endif
178 
179 unsigned long mp_lapic_addr;
180 int disable_apic;
181 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
182 static int disable_apic_timer __initdata;
183 /* Local APIC timer works in C2 */
184 int local_apic_timer_c2_ok;
185 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
186 
187 int first_system_vector = 0xfe;
188 
189 /*
190  * Debug level, exported for io_apic.c
191  */
192 unsigned int apic_verbosity;
193 
194 int pic_mode;
195 
196 /* Have we found an MP table */
197 int smp_found_config;
198 
199 static struct resource lapic_resource = {
200 	.name = "Local APIC",
201 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
202 };
203 
204 unsigned int lapic_timer_frequency = 0;
205 
206 static void apic_pm_activate(void);
207 
208 static unsigned long apic_phys;
209 
210 /*
211  * Get the LAPIC version
212  */
213 static inline int lapic_get_version(void)
214 {
215 	return GET_APIC_VERSION(apic_read(APIC_LVR));
216 }
217 
218 /*
219  * Check, if the APIC is integrated or a separate chip
220  */
221 static inline int lapic_is_integrated(void)
222 {
223 #ifdef CONFIG_X86_64
224 	return 1;
225 #else
226 	return APIC_INTEGRATED(lapic_get_version());
227 #endif
228 }
229 
230 /*
231  * Check, whether this is a modern or a first generation APIC
232  */
233 static int modern_apic(void)
234 {
235 	/* AMD systems use old APIC versions, so check the CPU */
236 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
237 	    boot_cpu_data.x86 >= 0xf)
238 		return 1;
239 	return lapic_get_version() >= 0x14;
240 }
241 
242 /*
243  * right after this call apic become NOOP driven
244  * so apic->write/read doesn't do anything
245  */
246 static void __init apic_disable(void)
247 {
248 	pr_info("APIC: switched to apic NOOP\n");
249 	apic = &apic_noop;
250 }
251 
252 void native_apic_wait_icr_idle(void)
253 {
254 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
255 		cpu_relax();
256 }
257 
258 u32 native_safe_apic_wait_icr_idle(void)
259 {
260 	u32 send_status;
261 	int timeout;
262 
263 	timeout = 0;
264 	do {
265 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
266 		if (!send_status)
267 			break;
268 		inc_irq_stat(icr_read_retry_count);
269 		udelay(100);
270 	} while (timeout++ < 1000);
271 
272 	return send_status;
273 }
274 
275 void native_apic_icr_write(u32 low, u32 id)
276 {
277 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
278 	apic_write(APIC_ICR, low);
279 }
280 
281 u64 native_apic_icr_read(void)
282 {
283 	u32 icr1, icr2;
284 
285 	icr2 = apic_read(APIC_ICR2);
286 	icr1 = apic_read(APIC_ICR);
287 
288 	return icr1 | ((u64)icr2 << 32);
289 }
290 
291 #ifdef CONFIG_X86_32
292 /**
293  * get_physical_broadcast - Get number of physical broadcast IDs
294  */
295 int get_physical_broadcast(void)
296 {
297 	return modern_apic() ? 0xff : 0xf;
298 }
299 #endif
300 
301 /**
302  * lapic_get_maxlvt - get the maximum number of local vector table entries
303  */
304 int lapic_get_maxlvt(void)
305 {
306 	unsigned int v;
307 
308 	v = apic_read(APIC_LVR);
309 	/*
310 	 * - we always have APIC integrated on 64bit mode
311 	 * - 82489DXs do not report # of LVT entries
312 	 */
313 	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
314 }
315 
316 /*
317  * Local APIC timer
318  */
319 
320 /* Clock divisor */
321 #define APIC_DIVISOR 16
322 #define TSC_DIVISOR  32
323 
324 /*
325  * This function sets up the local APIC timer, with a timeout of
326  * 'clocks' APIC bus clock. During calibration we actually call
327  * this function twice on the boot CPU, once with a bogus timeout
328  * value, second time for real. The other (noncalibrating) CPUs
329  * call this function only once, with the real, calibrated value.
330  *
331  * We do reads before writes even if unnecessary, to get around the
332  * P5 APIC double write bug.
333  */
334 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
335 {
336 	unsigned int lvtt_value, tmp_value;
337 
338 	lvtt_value = LOCAL_TIMER_VECTOR;
339 	if (!oneshot)
340 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
341 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
342 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
343 
344 	if (!lapic_is_integrated())
345 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
346 
347 	if (!irqen)
348 		lvtt_value |= APIC_LVT_MASKED;
349 
350 	apic_write(APIC_LVTT, lvtt_value);
351 
352 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
353 		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
354 		return;
355 	}
356 
357 	/*
358 	 * Divide PICLK by 16
359 	 */
360 	tmp_value = apic_read(APIC_TDCR);
361 	apic_write(APIC_TDCR,
362 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
363 		APIC_TDR_DIV_16);
364 
365 	if (!oneshot)
366 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
367 }
368 
369 /*
370  * Setup extended LVT, AMD specific
371  *
372  * Software should use the LVT offsets the BIOS provides.  The offsets
373  * are determined by the subsystems using it like those for MCE
374  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
375  * are supported. Beginning with family 10h at least 4 offsets are
376  * available.
377  *
378  * Since the offsets must be consistent for all cores, we keep track
379  * of the LVT offsets in software and reserve the offset for the same
380  * vector also to be used on other cores. An offset is freed by
381  * setting the entry to APIC_EILVT_MASKED.
382  *
383  * If the BIOS is right, there should be no conflicts. Otherwise a
384  * "[Firmware Bug]: ..." error message is generated. However, if
385  * software does not properly determines the offsets, it is not
386  * necessarily a BIOS bug.
387  */
388 
389 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
390 
391 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
392 {
393 	return (old & APIC_EILVT_MASKED)
394 		|| (new == APIC_EILVT_MASKED)
395 		|| ((new & ~APIC_EILVT_MASKED) == old);
396 }
397 
398 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
399 {
400 	unsigned int rsvd, vector;
401 
402 	if (offset >= APIC_EILVT_NR_MAX)
403 		return ~0;
404 
405 	rsvd = atomic_read(&eilvt_offsets[offset]);
406 	do {
407 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
408 		if (vector && !eilvt_entry_is_changeable(vector, new))
409 			/* may not change if vectors are different */
410 			return rsvd;
411 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
412 	} while (rsvd != new);
413 
414 	rsvd &= ~APIC_EILVT_MASKED;
415 	if (rsvd && rsvd != vector)
416 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
417 			offset, rsvd);
418 
419 	return new;
420 }
421 
422 /*
423  * If mask=1, the LVT entry does not generate interrupts while mask=0
424  * enables the vector. See also the BKDGs. Must be called with
425  * preemption disabled.
426  */
427 
428 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
429 {
430 	unsigned long reg = APIC_EILVTn(offset);
431 	unsigned int new, old, reserved;
432 
433 	new = (mask << 16) | (msg_type << 8) | vector;
434 	old = apic_read(reg);
435 	reserved = reserve_eilvt_offset(offset, new);
436 
437 	if (reserved != new) {
438 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
439 		       "vector 0x%x, but the register is already in use for "
440 		       "vector 0x%x on another cpu\n",
441 		       smp_processor_id(), reg, offset, new, reserved);
442 		return -EINVAL;
443 	}
444 
445 	if (!eilvt_entry_is_changeable(old, new)) {
446 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
447 		       "vector 0x%x, but the register is already in use for "
448 		       "vector 0x%x on this cpu\n",
449 		       smp_processor_id(), reg, offset, new, old);
450 		return -EBUSY;
451 	}
452 
453 	apic_write(reg, new);
454 
455 	return 0;
456 }
457 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
458 
459 /*
460  * Program the next event, relative to now
461  */
462 static int lapic_next_event(unsigned long delta,
463 			    struct clock_event_device *evt)
464 {
465 	apic_write(APIC_TMICT, delta);
466 	return 0;
467 }
468 
469 static int lapic_next_deadline(unsigned long delta,
470 			       struct clock_event_device *evt)
471 {
472 	u64 tsc;
473 
474 	rdtscll(tsc);
475 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
476 	return 0;
477 }
478 
479 /*
480  * Setup the lapic timer in periodic or oneshot mode
481  */
482 static void lapic_timer_setup(enum clock_event_mode mode,
483 			      struct clock_event_device *evt)
484 {
485 	unsigned long flags;
486 	unsigned int v;
487 
488 	/* Lapic used as dummy for broadcast ? */
489 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
490 		return;
491 
492 	local_irq_save(flags);
493 
494 	switch (mode) {
495 	case CLOCK_EVT_MODE_PERIODIC:
496 	case CLOCK_EVT_MODE_ONESHOT:
497 		__setup_APIC_LVTT(lapic_timer_frequency,
498 				  mode != CLOCK_EVT_MODE_PERIODIC, 1);
499 		break;
500 	case CLOCK_EVT_MODE_UNUSED:
501 	case CLOCK_EVT_MODE_SHUTDOWN:
502 		v = apic_read(APIC_LVTT);
503 		v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
504 		apic_write(APIC_LVTT, v);
505 		apic_write(APIC_TMICT, 0);
506 		break;
507 	case CLOCK_EVT_MODE_RESUME:
508 		/* Nothing to do here */
509 		break;
510 	}
511 
512 	local_irq_restore(flags);
513 }
514 
515 /*
516  * Local APIC timer broadcast function
517  */
518 static void lapic_timer_broadcast(const struct cpumask *mask)
519 {
520 #ifdef CONFIG_SMP
521 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
522 #endif
523 }
524 
525 
526 /*
527  * The local apic timer can be used for any function which is CPU local.
528  */
529 static struct clock_event_device lapic_clockevent = {
530 	.name		= "lapic",
531 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
532 			| CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
533 	.shift		= 32,
534 	.set_mode	= lapic_timer_setup,
535 	.set_next_event	= lapic_next_event,
536 	.broadcast	= lapic_timer_broadcast,
537 	.rating		= 100,
538 	.irq		= -1,
539 };
540 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
541 
542 /*
543  * Setup the local APIC timer for this CPU. Copy the initialized values
544  * of the boot CPU and register the clock event in the framework.
545  */
546 static void __cpuinit setup_APIC_timer(void)
547 {
548 	struct clock_event_device *levt = &__get_cpu_var(lapic_events);
549 
550 	if (this_cpu_has(X86_FEATURE_ARAT)) {
551 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
552 		/* Make LAPIC timer preferrable over percpu HPET */
553 		lapic_clockevent.rating = 150;
554 	}
555 
556 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
557 	levt->cpumask = cpumask_of(smp_processor_id());
558 
559 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
560 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
561 				    CLOCK_EVT_FEAT_DUMMY);
562 		levt->set_next_event = lapic_next_deadline;
563 		clockevents_config_and_register(levt,
564 						(tsc_khz / TSC_DIVISOR) * 1000,
565 						0xF, ~0UL);
566 	} else
567 		clockevents_register_device(levt);
568 }
569 
570 /*
571  * In this functions we calibrate APIC bus clocks to the external timer.
572  *
573  * We want to do the calibration only once since we want to have local timer
574  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
575  * frequency.
576  *
577  * This was previously done by reading the PIT/HPET and waiting for a wrap
578  * around to find out, that a tick has elapsed. I have a box, where the PIT
579  * readout is broken, so it never gets out of the wait loop again. This was
580  * also reported by others.
581  *
582  * Monitoring the jiffies value is inaccurate and the clockevents
583  * infrastructure allows us to do a simple substitution of the interrupt
584  * handler.
585  *
586  * The calibration routine also uses the pm_timer when possible, as the PIT
587  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
588  * back to normal later in the boot process).
589  */
590 
591 #define LAPIC_CAL_LOOPS		(HZ/10)
592 
593 static __initdata int lapic_cal_loops = -1;
594 static __initdata long lapic_cal_t1, lapic_cal_t2;
595 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
596 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
597 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
598 
599 /*
600  * Temporary interrupt handler.
601  */
602 static void __init lapic_cal_handler(struct clock_event_device *dev)
603 {
604 	unsigned long long tsc = 0;
605 	long tapic = apic_read(APIC_TMCCT);
606 	unsigned long pm = acpi_pm_read_early();
607 
608 	if (cpu_has_tsc)
609 		rdtscll(tsc);
610 
611 	switch (lapic_cal_loops++) {
612 	case 0:
613 		lapic_cal_t1 = tapic;
614 		lapic_cal_tsc1 = tsc;
615 		lapic_cal_pm1 = pm;
616 		lapic_cal_j1 = jiffies;
617 		break;
618 
619 	case LAPIC_CAL_LOOPS:
620 		lapic_cal_t2 = tapic;
621 		lapic_cal_tsc2 = tsc;
622 		if (pm < lapic_cal_pm1)
623 			pm += ACPI_PM_OVRRUN;
624 		lapic_cal_pm2 = pm;
625 		lapic_cal_j2 = jiffies;
626 		break;
627 	}
628 }
629 
630 static int __init
631 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
632 {
633 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
634 	const long pm_thresh = pm_100ms / 100;
635 	unsigned long mult;
636 	u64 res;
637 
638 #ifndef CONFIG_X86_PM_TIMER
639 	return -1;
640 #endif
641 
642 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
643 
644 	/* Check, if the PM timer is available */
645 	if (!deltapm)
646 		return -1;
647 
648 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
649 
650 	if (deltapm > (pm_100ms - pm_thresh) &&
651 	    deltapm < (pm_100ms + pm_thresh)) {
652 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
653 		return 0;
654 	}
655 
656 	res = (((u64)deltapm) *  mult) >> 22;
657 	do_div(res, 1000000);
658 	pr_warning("APIC calibration not consistent "
659 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
660 
661 	/* Correct the lapic counter value */
662 	res = (((u64)(*delta)) * pm_100ms);
663 	do_div(res, deltapm);
664 	pr_info("APIC delta adjusted to PM-Timer: "
665 		"%lu (%ld)\n", (unsigned long)res, *delta);
666 	*delta = (long)res;
667 
668 	/* Correct the tsc counter value */
669 	if (cpu_has_tsc) {
670 		res = (((u64)(*deltatsc)) * pm_100ms);
671 		do_div(res, deltapm);
672 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
673 					  "PM-Timer: %lu (%ld)\n",
674 					(unsigned long)res, *deltatsc);
675 		*deltatsc = (long)res;
676 	}
677 
678 	return 0;
679 }
680 
681 static int __init calibrate_APIC_clock(void)
682 {
683 	struct clock_event_device *levt = &__get_cpu_var(lapic_events);
684 	void (*real_handler)(struct clock_event_device *dev);
685 	unsigned long deltaj;
686 	long delta, deltatsc;
687 	int pm_referenced = 0;
688 
689 	/**
690 	 * check if lapic timer has already been calibrated by platform
691 	 * specific routine, such as tsc calibration code. if so, we just fill
692 	 * in the clockevent structure and return.
693 	 */
694 
695 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
696 		return 0;
697 	} else if (lapic_timer_frequency) {
698 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
699 				lapic_timer_frequency);
700 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
701 					TICK_NSEC, lapic_clockevent.shift);
702 		lapic_clockevent.max_delta_ns =
703 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
704 		lapic_clockevent.min_delta_ns =
705 			clockevent_delta2ns(0xF, &lapic_clockevent);
706 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
707 		return 0;
708 	}
709 
710 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
711 		    "calibrating APIC timer ...\n");
712 
713 	local_irq_disable();
714 
715 	/* Replace the global interrupt handler */
716 	real_handler = global_clock_event->event_handler;
717 	global_clock_event->event_handler = lapic_cal_handler;
718 
719 	/*
720 	 * Setup the APIC counter to maximum. There is no way the lapic
721 	 * can underflow in the 100ms detection time frame
722 	 */
723 	__setup_APIC_LVTT(0xffffffff, 0, 0);
724 
725 	/* Let the interrupts run */
726 	local_irq_enable();
727 
728 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
729 		cpu_relax();
730 
731 	local_irq_disable();
732 
733 	/* Restore the real event handler */
734 	global_clock_event->event_handler = real_handler;
735 
736 	/* Build delta t1-t2 as apic timer counts down */
737 	delta = lapic_cal_t1 - lapic_cal_t2;
738 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
739 
740 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
741 
742 	/* we trust the PM based calibration if possible */
743 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
744 					&delta, &deltatsc);
745 
746 	/* Calculate the scaled math multiplication factor */
747 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
748 				       lapic_clockevent.shift);
749 	lapic_clockevent.max_delta_ns =
750 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
751 	lapic_clockevent.min_delta_ns =
752 		clockevent_delta2ns(0xF, &lapic_clockevent);
753 
754 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
755 
756 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
757 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
758 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
759 		    lapic_timer_frequency);
760 
761 	if (cpu_has_tsc) {
762 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
763 			    "%ld.%04ld MHz.\n",
764 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
765 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
766 	}
767 
768 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
769 		    "%u.%04u MHz.\n",
770 		    lapic_timer_frequency / (1000000 / HZ),
771 		    lapic_timer_frequency % (1000000 / HZ));
772 
773 	/*
774 	 * Do a sanity check on the APIC calibration result
775 	 */
776 	if (lapic_timer_frequency < (1000000 / HZ)) {
777 		local_irq_enable();
778 		pr_warning("APIC frequency too slow, disabling apic timer\n");
779 		return -1;
780 	}
781 
782 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
783 
784 	/*
785 	 * PM timer calibration failed or not turned on
786 	 * so lets try APIC timer based calibration
787 	 */
788 	if (!pm_referenced) {
789 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
790 
791 		/*
792 		 * Setup the apic timer manually
793 		 */
794 		levt->event_handler = lapic_cal_handler;
795 		lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
796 		lapic_cal_loops = -1;
797 
798 		/* Let the interrupts run */
799 		local_irq_enable();
800 
801 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
802 			cpu_relax();
803 
804 		/* Stop the lapic timer */
805 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
806 
807 		/* Jiffies delta */
808 		deltaj = lapic_cal_j2 - lapic_cal_j1;
809 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
810 
811 		/* Check, if the jiffies result is consistent */
812 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
813 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
814 		else
815 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
816 	} else
817 		local_irq_enable();
818 
819 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
820 		pr_warning("APIC timer disabled due to verification failure\n");
821 			return -1;
822 	}
823 
824 	return 0;
825 }
826 
827 /*
828  * Setup the boot APIC
829  *
830  * Calibrate and verify the result.
831  */
832 void __init setup_boot_APIC_clock(void)
833 {
834 	/*
835 	 * The local apic timer can be disabled via the kernel
836 	 * commandline or from the CPU detection code. Register the lapic
837 	 * timer as a dummy clock event source on SMP systems, so the
838 	 * broadcast mechanism is used. On UP systems simply ignore it.
839 	 */
840 	if (disable_apic_timer) {
841 		pr_info("Disabling APIC timer\n");
842 		/* No broadcast on UP ! */
843 		if (num_possible_cpus() > 1) {
844 			lapic_clockevent.mult = 1;
845 			setup_APIC_timer();
846 		}
847 		return;
848 	}
849 
850 	if (calibrate_APIC_clock()) {
851 		/* No broadcast on UP ! */
852 		if (num_possible_cpus() > 1)
853 			setup_APIC_timer();
854 		return;
855 	}
856 
857 	/*
858 	 * If nmi_watchdog is set to IO_APIC, we need the
859 	 * PIT/HPET going.  Otherwise register lapic as a dummy
860 	 * device.
861 	 */
862 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
863 
864 	/* Setup the lapic or request the broadcast */
865 	setup_APIC_timer();
866 }
867 
868 void __cpuinit setup_secondary_APIC_clock(void)
869 {
870 	setup_APIC_timer();
871 }
872 
873 /*
874  * The guts of the apic timer interrupt
875  */
876 static void local_apic_timer_interrupt(void)
877 {
878 	int cpu = smp_processor_id();
879 	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
880 
881 	/*
882 	 * Normally we should not be here till LAPIC has been initialized but
883 	 * in some cases like kdump, its possible that there is a pending LAPIC
884 	 * timer interrupt from previous kernel's context and is delivered in
885 	 * new kernel the moment interrupts are enabled.
886 	 *
887 	 * Interrupts are enabled early and LAPIC is setup much later, hence
888 	 * its possible that when we get here evt->event_handler is NULL.
889 	 * Check for event_handler being NULL and discard the interrupt as
890 	 * spurious.
891 	 */
892 	if (!evt->event_handler) {
893 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
894 		/* Switch it off */
895 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
896 		return;
897 	}
898 
899 	/*
900 	 * the NMI deadlock-detector uses this.
901 	 */
902 	inc_irq_stat(apic_timer_irqs);
903 
904 	evt->event_handler(evt);
905 }
906 
907 /*
908  * Local APIC timer interrupt. This is the most natural way for doing
909  * local interrupts, but local timer interrupts can be emulated by
910  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
911  *
912  * [ if a single-CPU system runs an SMP kernel then we call the local
913  *   interrupt as well. Thus we cannot inline the local irq ... ]
914  */
915 void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
916 {
917 	struct pt_regs *old_regs = set_irq_regs(regs);
918 
919 	/*
920 	 * NOTE! We'd better ACK the irq immediately,
921 	 * because timer handling can be slow.
922 	 */
923 	ack_APIC_irq();
924 	/*
925 	 * update_process_times() expects us to have done irq_enter().
926 	 * Besides, if we don't timer interrupts ignore the global
927 	 * interrupt lock, which is the WrongThing (tm) to do.
928 	 */
929 	irq_enter();
930 	exit_idle();
931 	local_apic_timer_interrupt();
932 	irq_exit();
933 
934 	set_irq_regs(old_regs);
935 }
936 
937 int setup_profiling_timer(unsigned int multiplier)
938 {
939 	return -EINVAL;
940 }
941 
942 /*
943  * Local APIC start and shutdown
944  */
945 
946 /**
947  * clear_local_APIC - shutdown the local APIC
948  *
949  * This is called, when a CPU is disabled and before rebooting, so the state of
950  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
951  * leftovers during boot.
952  */
953 void clear_local_APIC(void)
954 {
955 	int maxlvt;
956 	u32 v;
957 
958 	/* APIC hasn't been mapped yet */
959 	if (!x2apic_mode && !apic_phys)
960 		return;
961 
962 	maxlvt = lapic_get_maxlvt();
963 	/*
964 	 * Masking an LVT entry can trigger a local APIC error
965 	 * if the vector is zero. Mask LVTERR first to prevent this.
966 	 */
967 	if (maxlvt >= 3) {
968 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
969 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
970 	}
971 	/*
972 	 * Careful: we have to set masks only first to deassert
973 	 * any level-triggered sources.
974 	 */
975 	v = apic_read(APIC_LVTT);
976 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
977 	v = apic_read(APIC_LVT0);
978 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
979 	v = apic_read(APIC_LVT1);
980 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
981 	if (maxlvt >= 4) {
982 		v = apic_read(APIC_LVTPC);
983 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
984 	}
985 
986 	/* lets not touch this if we didn't frob it */
987 #ifdef CONFIG_X86_THERMAL_VECTOR
988 	if (maxlvt >= 5) {
989 		v = apic_read(APIC_LVTTHMR);
990 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
991 	}
992 #endif
993 #ifdef CONFIG_X86_MCE_INTEL
994 	if (maxlvt >= 6) {
995 		v = apic_read(APIC_LVTCMCI);
996 		if (!(v & APIC_LVT_MASKED))
997 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
998 	}
999 #endif
1000 
1001 	/*
1002 	 * Clean APIC state for other OSs:
1003 	 */
1004 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1005 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1006 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1007 	if (maxlvt >= 3)
1008 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1009 	if (maxlvt >= 4)
1010 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1011 
1012 	/* Integrated APIC (!82489DX) ? */
1013 	if (lapic_is_integrated()) {
1014 		if (maxlvt > 3)
1015 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1016 			apic_write(APIC_ESR, 0);
1017 		apic_read(APIC_ESR);
1018 	}
1019 }
1020 
1021 /**
1022  * disable_local_APIC - clear and disable the local APIC
1023  */
1024 void disable_local_APIC(void)
1025 {
1026 	unsigned int value;
1027 
1028 	/* APIC hasn't been mapped yet */
1029 	if (!x2apic_mode && !apic_phys)
1030 		return;
1031 
1032 	clear_local_APIC();
1033 
1034 	/*
1035 	 * Disable APIC (implies clearing of registers
1036 	 * for 82489DX!).
1037 	 */
1038 	value = apic_read(APIC_SPIV);
1039 	value &= ~APIC_SPIV_APIC_ENABLED;
1040 	apic_write(APIC_SPIV, value);
1041 
1042 #ifdef CONFIG_X86_32
1043 	/*
1044 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1045 	 * restore the disabled state.
1046 	 */
1047 	if (enabled_via_apicbase) {
1048 		unsigned int l, h;
1049 
1050 		rdmsr(MSR_IA32_APICBASE, l, h);
1051 		l &= ~MSR_IA32_APICBASE_ENABLE;
1052 		wrmsr(MSR_IA32_APICBASE, l, h);
1053 	}
1054 #endif
1055 }
1056 
1057 /*
1058  * If Linux enabled the LAPIC against the BIOS default disable it down before
1059  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1060  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1061  * for the case where Linux didn't enable the LAPIC.
1062  */
1063 void lapic_shutdown(void)
1064 {
1065 	unsigned long flags;
1066 
1067 	if (!cpu_has_apic && !apic_from_smp_config())
1068 		return;
1069 
1070 	local_irq_save(flags);
1071 
1072 #ifdef CONFIG_X86_32
1073 	if (!enabled_via_apicbase)
1074 		clear_local_APIC();
1075 	else
1076 #endif
1077 		disable_local_APIC();
1078 
1079 
1080 	local_irq_restore(flags);
1081 }
1082 
1083 /*
1084  * This is to verify that we're looking at a real local APIC.
1085  * Check these against your board if the CPUs aren't getting
1086  * started for no apparent reason.
1087  */
1088 int __init verify_local_APIC(void)
1089 {
1090 	unsigned int reg0, reg1;
1091 
1092 	/*
1093 	 * The version register is read-only in a real APIC.
1094 	 */
1095 	reg0 = apic_read(APIC_LVR);
1096 	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1097 	apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1098 	reg1 = apic_read(APIC_LVR);
1099 	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1100 
1101 	/*
1102 	 * The two version reads above should print the same
1103 	 * numbers.  If the second one is different, then we
1104 	 * poke at a non-APIC.
1105 	 */
1106 	if (reg1 != reg0)
1107 		return 0;
1108 
1109 	/*
1110 	 * Check if the version looks reasonably.
1111 	 */
1112 	reg1 = GET_APIC_VERSION(reg0);
1113 	if (reg1 == 0x00 || reg1 == 0xff)
1114 		return 0;
1115 	reg1 = lapic_get_maxlvt();
1116 	if (reg1 < 0x02 || reg1 == 0xff)
1117 		return 0;
1118 
1119 	/*
1120 	 * The ID register is read/write in a real APIC.
1121 	 */
1122 	reg0 = apic_read(APIC_ID);
1123 	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1124 	apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1125 	reg1 = apic_read(APIC_ID);
1126 	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1127 	apic_write(APIC_ID, reg0);
1128 	if (reg1 != (reg0 ^ apic->apic_id_mask))
1129 		return 0;
1130 
1131 	/*
1132 	 * The next two are just to see if we have sane values.
1133 	 * They're only really relevant if we're in Virtual Wire
1134 	 * compatibility mode, but most boxes are anymore.
1135 	 */
1136 	reg0 = apic_read(APIC_LVT0);
1137 	apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1138 	reg1 = apic_read(APIC_LVT1);
1139 	apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1140 
1141 	return 1;
1142 }
1143 
1144 /**
1145  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1146  */
1147 void __init sync_Arb_IDs(void)
1148 {
1149 	/*
1150 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1151 	 * needed on AMD.
1152 	 */
1153 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1154 		return;
1155 
1156 	/*
1157 	 * Wait for idle.
1158 	 */
1159 	apic_wait_icr_idle();
1160 
1161 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1162 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1163 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1164 }
1165 
1166 /*
1167  * An initial setup of the virtual wire mode.
1168  */
1169 void __init init_bsp_APIC(void)
1170 {
1171 	unsigned int value;
1172 
1173 	/*
1174 	 * Don't do the setup now if we have a SMP BIOS as the
1175 	 * through-I/O-APIC virtual wire mode might be active.
1176 	 */
1177 	if (smp_found_config || !cpu_has_apic)
1178 		return;
1179 
1180 	/*
1181 	 * Do not trust the local APIC being empty at bootup.
1182 	 */
1183 	clear_local_APIC();
1184 
1185 	/*
1186 	 * Enable APIC.
1187 	 */
1188 	value = apic_read(APIC_SPIV);
1189 	value &= ~APIC_VECTOR_MASK;
1190 	value |= APIC_SPIV_APIC_ENABLED;
1191 
1192 #ifdef CONFIG_X86_32
1193 	/* This bit is reserved on P4/Xeon and should be cleared */
1194 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1195 	    (boot_cpu_data.x86 == 15))
1196 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1197 	else
1198 #endif
1199 		value |= APIC_SPIV_FOCUS_DISABLED;
1200 	value |= SPURIOUS_APIC_VECTOR;
1201 	apic_write(APIC_SPIV, value);
1202 
1203 	/*
1204 	 * Set up the virtual wire mode.
1205 	 */
1206 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1207 	value = APIC_DM_NMI;
1208 	if (!lapic_is_integrated())		/* 82489DX */
1209 		value |= APIC_LVT_LEVEL_TRIGGER;
1210 	apic_write(APIC_LVT1, value);
1211 }
1212 
1213 static void __cpuinit lapic_setup_esr(void)
1214 {
1215 	unsigned int oldvalue, value, maxlvt;
1216 
1217 	if (!lapic_is_integrated()) {
1218 		pr_info("No ESR for 82489DX.\n");
1219 		return;
1220 	}
1221 
1222 	if (apic->disable_esr) {
1223 		/*
1224 		 * Something untraceable is creating bad interrupts on
1225 		 * secondary quads ... for the moment, just leave the
1226 		 * ESR disabled - we can't do anything useful with the
1227 		 * errors anyway - mbligh
1228 		 */
1229 		pr_info("Leaving ESR disabled.\n");
1230 		return;
1231 	}
1232 
1233 	maxlvt = lapic_get_maxlvt();
1234 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1235 		apic_write(APIC_ESR, 0);
1236 	oldvalue = apic_read(APIC_ESR);
1237 
1238 	/* enables sending errors */
1239 	value = ERROR_APIC_VECTOR;
1240 	apic_write(APIC_LVTERR, value);
1241 
1242 	/*
1243 	 * spec says clear errors after enabling vector.
1244 	 */
1245 	if (maxlvt > 3)
1246 		apic_write(APIC_ESR, 0);
1247 	value = apic_read(APIC_ESR);
1248 	if (value != oldvalue)
1249 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1250 			"vector: 0x%08x  after: 0x%08x\n",
1251 			oldvalue, value);
1252 }
1253 
1254 /**
1255  * setup_local_APIC - setup the local APIC
1256  *
1257  * Used to setup local APIC while initializing BSP or bringin up APs.
1258  * Always called with preemption disabled.
1259  */
1260 void __cpuinit setup_local_APIC(void)
1261 {
1262 	int cpu = smp_processor_id();
1263 	unsigned int value, queued;
1264 	int i, j, acked = 0;
1265 	unsigned long long tsc = 0, ntsc;
1266 	long long max_loops = cpu_khz;
1267 
1268 	if (cpu_has_tsc)
1269 		rdtscll(tsc);
1270 
1271 	if (disable_apic) {
1272 		disable_ioapic_support();
1273 		return;
1274 	}
1275 
1276 #ifdef CONFIG_X86_32
1277 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1278 	if (lapic_is_integrated() && apic->disable_esr) {
1279 		apic_write(APIC_ESR, 0);
1280 		apic_write(APIC_ESR, 0);
1281 		apic_write(APIC_ESR, 0);
1282 		apic_write(APIC_ESR, 0);
1283 	}
1284 #endif
1285 	perf_events_lapic_init();
1286 
1287 	/*
1288 	 * Double-check whether this APIC is really registered.
1289 	 * This is meaningless in clustered apic mode, so we skip it.
1290 	 */
1291 	BUG_ON(!apic->apic_id_registered());
1292 
1293 	/*
1294 	 * Intel recommends to set DFR, LDR and TPR before enabling
1295 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1296 	 * document number 292116).  So here it goes...
1297 	 */
1298 	apic->init_apic_ldr();
1299 
1300 #ifdef CONFIG_X86_32
1301 	/*
1302 	 * APIC LDR is initialized.  If logical_apicid mapping was
1303 	 * initialized during get_smp_config(), make sure it matches the
1304 	 * actual value.
1305 	 */
1306 	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1307 	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1308 	/* always use the value from LDR */
1309 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1310 		logical_smp_processor_id();
1311 
1312 	/*
1313 	 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1314 	 * node mapping during NUMA init.  Now that logical apicid is
1315 	 * guaranteed to be known, give it another chance.  This is already
1316 	 * a bit too late - percpu allocation has already happened without
1317 	 * proper NUMA affinity.
1318 	 */
1319 	if (apic->x86_32_numa_cpu_node)
1320 		set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1321 				   apic->x86_32_numa_cpu_node(cpu));
1322 #endif
1323 
1324 	/*
1325 	 * Set Task Priority to 'accept all'. We never change this
1326 	 * later on.
1327 	 */
1328 	value = apic_read(APIC_TASKPRI);
1329 	value &= ~APIC_TPRI_MASK;
1330 	apic_write(APIC_TASKPRI, value);
1331 
1332 	/*
1333 	 * After a crash, we no longer service the interrupts and a pending
1334 	 * interrupt from previous kernel might still have ISR bit set.
1335 	 *
1336 	 * Most probably by now CPU has serviced that pending interrupt and
1337 	 * it might not have done the ack_APIC_irq() because it thought,
1338 	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1339 	 * does not clear the ISR bit and cpu thinks it has already serivced
1340 	 * the interrupt. Hence a vector might get locked. It was noticed
1341 	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1342 	 */
1343 	do {
1344 		queued = 0;
1345 		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1346 			queued |= apic_read(APIC_IRR + i*0x10);
1347 
1348 		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1349 			value = apic_read(APIC_ISR + i*0x10);
1350 			for (j = 31; j >= 0; j--) {
1351 				if (value & (1<<j)) {
1352 					ack_APIC_irq();
1353 					acked++;
1354 				}
1355 			}
1356 		}
1357 		if (acked > 256) {
1358 			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1359 			       acked);
1360 			break;
1361 		}
1362 		if (queued) {
1363 			if (cpu_has_tsc) {
1364 				rdtscll(ntsc);
1365 				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1366 			} else
1367 				max_loops--;
1368 		}
1369 	} while (queued && max_loops > 0);
1370 	WARN_ON(max_loops <= 0);
1371 
1372 	/*
1373 	 * Now that we are all set up, enable the APIC
1374 	 */
1375 	value = apic_read(APIC_SPIV);
1376 	value &= ~APIC_VECTOR_MASK;
1377 	/*
1378 	 * Enable APIC
1379 	 */
1380 	value |= APIC_SPIV_APIC_ENABLED;
1381 
1382 #ifdef CONFIG_X86_32
1383 	/*
1384 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1385 	 * certain networking cards. If high frequency interrupts are
1386 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1387 	 * entry is masked/unmasked at a high rate as well then sooner or
1388 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1389 	 * from the device. If focus CPU is disabled then the hang goes
1390 	 * away, oh well :-(
1391 	 *
1392 	 * [ This bug can be reproduced easily with a level-triggered
1393 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1394 	 *   BX chipset. ]
1395 	 */
1396 	/*
1397 	 * Actually disabling the focus CPU check just makes the hang less
1398 	 * frequent as it makes the interrupt distributon model be more
1399 	 * like LRU than MRU (the short-term load is more even across CPUs).
1400 	 * See also the comment in end_level_ioapic_irq().  --macro
1401 	 */
1402 
1403 	/*
1404 	 * - enable focus processor (bit==0)
1405 	 * - 64bit mode always use processor focus
1406 	 *   so no need to set it
1407 	 */
1408 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1409 #endif
1410 
1411 	/*
1412 	 * Set spurious IRQ vector
1413 	 */
1414 	value |= SPURIOUS_APIC_VECTOR;
1415 	apic_write(APIC_SPIV, value);
1416 
1417 	/*
1418 	 * Set up LVT0, LVT1:
1419 	 *
1420 	 * set up through-local-APIC on the BP's LINT0. This is not
1421 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1422 	 * we delegate interrupts to the 8259A.
1423 	 */
1424 	/*
1425 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1426 	 */
1427 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1428 	if (!cpu && (pic_mode || !value)) {
1429 		value = APIC_DM_EXTINT;
1430 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1431 	} else {
1432 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1433 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1434 	}
1435 	apic_write(APIC_LVT0, value);
1436 
1437 	/*
1438 	 * only the BP should see the LINT1 NMI signal, obviously.
1439 	 */
1440 	if (!cpu)
1441 		value = APIC_DM_NMI;
1442 	else
1443 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1444 	if (!lapic_is_integrated())		/* 82489DX */
1445 		value |= APIC_LVT_LEVEL_TRIGGER;
1446 	apic_write(APIC_LVT1, value);
1447 
1448 #ifdef CONFIG_X86_MCE_INTEL
1449 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1450 	if (!cpu)
1451 		cmci_recheck();
1452 #endif
1453 }
1454 
1455 void __cpuinit end_local_APIC_setup(void)
1456 {
1457 	lapic_setup_esr();
1458 
1459 #ifdef CONFIG_X86_32
1460 	{
1461 		unsigned int value;
1462 		/* Disable the local apic timer */
1463 		value = apic_read(APIC_LVTT);
1464 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1465 		apic_write(APIC_LVTT, value);
1466 	}
1467 #endif
1468 
1469 	apic_pm_activate();
1470 }
1471 
1472 void __init bsp_end_local_APIC_setup(void)
1473 {
1474 	end_local_APIC_setup();
1475 
1476 	/*
1477 	 * Now that local APIC setup is completed for BP, configure the fault
1478 	 * handling for interrupt remapping.
1479 	 */
1480 	if (irq_remapping_enabled)
1481 		irq_remap_enable_fault_handling();
1482 
1483 }
1484 
1485 #ifdef CONFIG_X86_X2APIC
1486 /*
1487  * Need to disable xapic and x2apic at the same time and then enable xapic mode
1488  */
1489 static inline void __disable_x2apic(u64 msr)
1490 {
1491 	wrmsrl(MSR_IA32_APICBASE,
1492 	       msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1493 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1494 }
1495 
1496 static __init void disable_x2apic(void)
1497 {
1498 	u64 msr;
1499 
1500 	if (!cpu_has_x2apic)
1501 		return;
1502 
1503 	rdmsrl(MSR_IA32_APICBASE, msr);
1504 	if (msr & X2APIC_ENABLE) {
1505 		u32 x2apic_id = read_apic_id();
1506 
1507 		if (x2apic_id >= 255)
1508 			panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1509 
1510 		pr_info("Disabling x2apic\n");
1511 		__disable_x2apic(msr);
1512 
1513 		if (nox2apic) {
1514 			clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1515 			setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1516 		}
1517 
1518 		x2apic_disabled = 1;
1519 		x2apic_mode = 0;
1520 
1521 		register_lapic_address(mp_lapic_addr);
1522 	}
1523 }
1524 
1525 void check_x2apic(void)
1526 {
1527 	if (x2apic_enabled()) {
1528 		pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1529 		x2apic_preenabled = x2apic_mode = 1;
1530 	}
1531 }
1532 
1533 void enable_x2apic(void)
1534 {
1535 	u64 msr;
1536 
1537 	rdmsrl(MSR_IA32_APICBASE, msr);
1538 	if (x2apic_disabled) {
1539 		__disable_x2apic(msr);
1540 		return;
1541 	}
1542 
1543 	if (!x2apic_mode)
1544 		return;
1545 
1546 	if (!(msr & X2APIC_ENABLE)) {
1547 		printk_once(KERN_INFO "Enabling x2apic\n");
1548 		wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1549 	}
1550 }
1551 #endif /* CONFIG_X86_X2APIC */
1552 
1553 int __init enable_IR(void)
1554 {
1555 #ifdef CONFIG_IRQ_REMAP
1556 	if (!irq_remapping_supported()) {
1557 		pr_debug("intr-remapping not supported\n");
1558 		return -1;
1559 	}
1560 
1561 	if (!x2apic_preenabled && skip_ioapic_setup) {
1562 		pr_info("Skipped enabling intr-remap because of skipping "
1563 			"io-apic setup\n");
1564 		return -1;
1565 	}
1566 
1567 	return irq_remapping_enable();
1568 #endif
1569 	return -1;
1570 }
1571 
1572 void __init enable_IR_x2apic(void)
1573 {
1574 	unsigned long flags;
1575 	int ret, x2apic_enabled = 0;
1576 	int hardware_init_ret;
1577 
1578 	/* Make sure irq_remap_ops are initialized */
1579 	setup_irq_remapping_ops();
1580 
1581 	hardware_init_ret = irq_remapping_prepare();
1582 	if (hardware_init_ret && !x2apic_supported())
1583 		return;
1584 
1585 	ret = save_ioapic_entries();
1586 	if (ret) {
1587 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1588 		return;
1589 	}
1590 
1591 	local_irq_save(flags);
1592 	legacy_pic->mask_all();
1593 	mask_ioapic_entries();
1594 
1595 	if (x2apic_preenabled && nox2apic)
1596 		disable_x2apic();
1597 
1598 	if (hardware_init_ret)
1599 		ret = -1;
1600 	else
1601 		ret = enable_IR();
1602 
1603 	if (!x2apic_supported())
1604 		goto skip_x2apic;
1605 
1606 	if (ret < 0) {
1607 		/* IR is required if there is APIC ID > 255 even when running
1608 		 * under KVM
1609 		 */
1610 		if (max_physical_apicid > 255 ||
1611 		    !hypervisor_x2apic_available()) {
1612 			if (x2apic_preenabled)
1613 				disable_x2apic();
1614 			goto skip_x2apic;
1615 		}
1616 		/*
1617 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1618 		 * only in physical mode
1619 		 */
1620 		x2apic_force_phys();
1621 	}
1622 
1623 	if (ret == IRQ_REMAP_XAPIC_MODE) {
1624 		pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1625 		goto skip_x2apic;
1626 	}
1627 
1628 	x2apic_enabled = 1;
1629 
1630 	if (x2apic_supported() && !x2apic_mode) {
1631 		x2apic_mode = 1;
1632 		enable_x2apic();
1633 		pr_info("Enabled x2apic\n");
1634 	}
1635 
1636 skip_x2apic:
1637 	if (ret < 0) /* IR enabling failed */
1638 		restore_ioapic_entries();
1639 	legacy_pic->restore_mask();
1640 	local_irq_restore(flags);
1641 }
1642 
1643 #ifdef CONFIG_X86_64
1644 /*
1645  * Detect and enable local APICs on non-SMP boards.
1646  * Original code written by Keir Fraser.
1647  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1648  * not correctly set up (usually the APIC timer won't work etc.)
1649  */
1650 static int __init detect_init_APIC(void)
1651 {
1652 	if (!cpu_has_apic) {
1653 		pr_info("No local APIC present\n");
1654 		return -1;
1655 	}
1656 
1657 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1658 	return 0;
1659 }
1660 #else
1661 
1662 static int __init apic_verify(void)
1663 {
1664 	u32 features, h, l;
1665 
1666 	/*
1667 	 * The APIC feature bit should now be enabled
1668 	 * in `cpuid'
1669 	 */
1670 	features = cpuid_edx(1);
1671 	if (!(features & (1 << X86_FEATURE_APIC))) {
1672 		pr_warning("Could not enable APIC!\n");
1673 		return -1;
1674 	}
1675 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1676 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1677 
1678 	/* The BIOS may have set up the APIC at some other address */
1679 	if (boot_cpu_data.x86 >= 6) {
1680 		rdmsr(MSR_IA32_APICBASE, l, h);
1681 		if (l & MSR_IA32_APICBASE_ENABLE)
1682 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1683 	}
1684 
1685 	pr_info("Found and enabled local APIC!\n");
1686 	return 0;
1687 }
1688 
1689 int __init apic_force_enable(unsigned long addr)
1690 {
1691 	u32 h, l;
1692 
1693 	if (disable_apic)
1694 		return -1;
1695 
1696 	/*
1697 	 * Some BIOSes disable the local APIC in the APIC_BASE
1698 	 * MSR. This can only be done in software for Intel P6 or later
1699 	 * and AMD K7 (Model > 1) or later.
1700 	 */
1701 	if (boot_cpu_data.x86 >= 6) {
1702 		rdmsr(MSR_IA32_APICBASE, l, h);
1703 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1704 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1705 			l &= ~MSR_IA32_APICBASE_BASE;
1706 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1707 			wrmsr(MSR_IA32_APICBASE, l, h);
1708 			enabled_via_apicbase = 1;
1709 		}
1710 	}
1711 	return apic_verify();
1712 }
1713 
1714 /*
1715  * Detect and initialize APIC
1716  */
1717 static int __init detect_init_APIC(void)
1718 {
1719 	/* Disabled by kernel option? */
1720 	if (disable_apic)
1721 		return -1;
1722 
1723 	switch (boot_cpu_data.x86_vendor) {
1724 	case X86_VENDOR_AMD:
1725 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1726 		    (boot_cpu_data.x86 >= 15))
1727 			break;
1728 		goto no_apic;
1729 	case X86_VENDOR_INTEL:
1730 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1731 		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
1732 			break;
1733 		goto no_apic;
1734 	default:
1735 		goto no_apic;
1736 	}
1737 
1738 	if (!cpu_has_apic) {
1739 		/*
1740 		 * Over-ride BIOS and try to enable the local APIC only if
1741 		 * "lapic" specified.
1742 		 */
1743 		if (!force_enable_local_apic) {
1744 			pr_info("Local APIC disabled by BIOS -- "
1745 				"you can enable it with \"lapic\"\n");
1746 			return -1;
1747 		}
1748 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1749 			return -1;
1750 	} else {
1751 		if (apic_verify())
1752 			return -1;
1753 	}
1754 
1755 	apic_pm_activate();
1756 
1757 	return 0;
1758 
1759 no_apic:
1760 	pr_info("No local APIC present or hardware disabled\n");
1761 	return -1;
1762 }
1763 #endif
1764 
1765 /**
1766  * init_apic_mappings - initialize APIC mappings
1767  */
1768 void __init init_apic_mappings(void)
1769 {
1770 	unsigned int new_apicid;
1771 
1772 	if (x2apic_mode) {
1773 		boot_cpu_physical_apicid = read_apic_id();
1774 		return;
1775 	}
1776 
1777 	/* If no local APIC can be found return early */
1778 	if (!smp_found_config && detect_init_APIC()) {
1779 		/* lets NOP'ify apic operations */
1780 		pr_info("APIC: disable apic facility\n");
1781 		apic_disable();
1782 	} else {
1783 		apic_phys = mp_lapic_addr;
1784 
1785 		/*
1786 		 * acpi lapic path already maps that address in
1787 		 * acpi_register_lapic_address()
1788 		 */
1789 		if (!acpi_lapic && !smp_found_config)
1790 			register_lapic_address(apic_phys);
1791 	}
1792 
1793 	/*
1794 	 * Fetch the APIC ID of the BSP in case we have a
1795 	 * default configuration (or the MP table is broken).
1796 	 */
1797 	new_apicid = read_apic_id();
1798 	if (boot_cpu_physical_apicid != new_apicid) {
1799 		boot_cpu_physical_apicid = new_apicid;
1800 		/*
1801 		 * yeah -- we lie about apic_version
1802 		 * in case if apic was disabled via boot option
1803 		 * but it's not a problem for SMP compiled kernel
1804 		 * since smp_sanity_check is prepared for such a case
1805 		 * and disable smp mode
1806 		 */
1807 		apic_version[new_apicid] =
1808 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1809 	}
1810 }
1811 
1812 void __init register_lapic_address(unsigned long address)
1813 {
1814 	mp_lapic_addr = address;
1815 
1816 	if (!x2apic_mode) {
1817 		set_fixmap_nocache(FIX_APIC_BASE, address);
1818 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1819 			    APIC_BASE, mp_lapic_addr);
1820 	}
1821 	if (boot_cpu_physical_apicid == -1U) {
1822 		boot_cpu_physical_apicid  = read_apic_id();
1823 		apic_version[boot_cpu_physical_apicid] =
1824 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1825 	}
1826 }
1827 
1828 /*
1829  * This initializes the IO-APIC and APIC hardware if this is
1830  * a UP kernel.
1831  */
1832 int apic_version[MAX_LOCAL_APIC];
1833 
1834 int __init APIC_init_uniprocessor(void)
1835 {
1836 	if (disable_apic) {
1837 		pr_info("Apic disabled\n");
1838 		return -1;
1839 	}
1840 #ifdef CONFIG_X86_64
1841 	if (!cpu_has_apic) {
1842 		disable_apic = 1;
1843 		pr_info("Apic disabled by BIOS\n");
1844 		return -1;
1845 	}
1846 #else
1847 	if (!smp_found_config && !cpu_has_apic)
1848 		return -1;
1849 
1850 	/*
1851 	 * Complain if the BIOS pretends there is one.
1852 	 */
1853 	if (!cpu_has_apic &&
1854 	    APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1855 		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1856 			boot_cpu_physical_apicid);
1857 		return -1;
1858 	}
1859 #endif
1860 
1861 	default_setup_apic_routing();
1862 
1863 	verify_local_APIC();
1864 	connect_bsp_APIC();
1865 
1866 #ifdef CONFIG_X86_64
1867 	apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1868 #else
1869 	/*
1870 	 * Hack: In case of kdump, after a crash, kernel might be booting
1871 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1872 	 * might be zero if read from MP tables. Get it from LAPIC.
1873 	 */
1874 # ifdef CONFIG_CRASH_DUMP
1875 	boot_cpu_physical_apicid = read_apic_id();
1876 # endif
1877 #endif
1878 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1879 	setup_local_APIC();
1880 
1881 #ifdef CONFIG_X86_IO_APIC
1882 	/*
1883 	 * Now enable IO-APICs, actually call clear_IO_APIC
1884 	 * We need clear_IO_APIC before enabling error vector
1885 	 */
1886 	if (!skip_ioapic_setup && nr_ioapics)
1887 		enable_IO_APIC();
1888 #endif
1889 
1890 	bsp_end_local_APIC_setup();
1891 
1892 #ifdef CONFIG_X86_IO_APIC
1893 	if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1894 		setup_IO_APIC();
1895 	else {
1896 		nr_ioapics = 0;
1897 	}
1898 #endif
1899 
1900 	x86_init.timers.setup_percpu_clockev();
1901 	return 0;
1902 }
1903 
1904 /*
1905  * Local APIC interrupts
1906  */
1907 
1908 /*
1909  * This interrupt should _never_ happen with our APIC/SMP architecture
1910  */
1911 void smp_spurious_interrupt(struct pt_regs *regs)
1912 {
1913 	u32 v;
1914 
1915 	irq_enter();
1916 	exit_idle();
1917 	/*
1918 	 * Check if this really is a spurious interrupt and ACK it
1919 	 * if it is a vectored one.  Just in case...
1920 	 * Spurious interrupts should not be ACKed.
1921 	 */
1922 	v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1923 	if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1924 		ack_APIC_irq();
1925 
1926 	inc_irq_stat(irq_spurious_count);
1927 
1928 	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1929 	pr_info("spurious APIC interrupt on CPU#%d, "
1930 		"should never happen.\n", smp_processor_id());
1931 	irq_exit();
1932 }
1933 
1934 /*
1935  * This interrupt should never happen with our APIC/SMP architecture
1936  */
1937 void smp_error_interrupt(struct pt_regs *regs)
1938 {
1939 	u32 v0, v1;
1940 	u32 i = 0;
1941 	static const char * const error_interrupt_reason[] = {
1942 		"Send CS error",		/* APIC Error Bit 0 */
1943 		"Receive CS error",		/* APIC Error Bit 1 */
1944 		"Send accept error",		/* APIC Error Bit 2 */
1945 		"Receive accept error",		/* APIC Error Bit 3 */
1946 		"Redirectable IPI",		/* APIC Error Bit 4 */
1947 		"Send illegal vector",		/* APIC Error Bit 5 */
1948 		"Received illegal vector",	/* APIC Error Bit 6 */
1949 		"Illegal register address",	/* APIC Error Bit 7 */
1950 	};
1951 
1952 	irq_enter();
1953 	exit_idle();
1954 	/* First tickle the hardware, only then report what went on. -- REW */
1955 	v0 = apic_read(APIC_ESR);
1956 	apic_write(APIC_ESR, 0);
1957 	v1 = apic_read(APIC_ESR);
1958 	ack_APIC_irq();
1959 	atomic_inc(&irq_err_count);
1960 
1961 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1962 		    smp_processor_id(), v0 , v1);
1963 
1964 	v1 = v1 & 0xff;
1965 	while (v1) {
1966 		if (v1 & 0x1)
1967 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1968 		i++;
1969 		v1 >>= 1;
1970 	}
1971 
1972 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
1973 
1974 	irq_exit();
1975 }
1976 
1977 /**
1978  * connect_bsp_APIC - attach the APIC to the interrupt system
1979  */
1980 void __init connect_bsp_APIC(void)
1981 {
1982 #ifdef CONFIG_X86_32
1983 	if (pic_mode) {
1984 		/*
1985 		 * Do not trust the local APIC being empty at bootup.
1986 		 */
1987 		clear_local_APIC();
1988 		/*
1989 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
1990 		 * local APIC to INT and NMI lines.
1991 		 */
1992 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1993 				"enabling APIC mode.\n");
1994 		imcr_pic_to_apic();
1995 	}
1996 #endif
1997 	if (apic->enable_apic_mode)
1998 		apic->enable_apic_mode();
1999 }
2000 
2001 /**
2002  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2003  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2004  *
2005  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2006  * APIC is disabled.
2007  */
2008 void disconnect_bsp_APIC(int virt_wire_setup)
2009 {
2010 	unsigned int value;
2011 
2012 #ifdef CONFIG_X86_32
2013 	if (pic_mode) {
2014 		/*
2015 		 * Put the board back into PIC mode (has an effect only on
2016 		 * certain older boards).  Note that APIC interrupts, including
2017 		 * IPIs, won't work beyond this point!  The only exception are
2018 		 * INIT IPIs.
2019 		 */
2020 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2021 				"entering PIC mode.\n");
2022 		imcr_apic_to_pic();
2023 		return;
2024 	}
2025 #endif
2026 
2027 	/* Go back to Virtual Wire compatibility mode */
2028 
2029 	/* For the spurious interrupt use vector F, and enable it */
2030 	value = apic_read(APIC_SPIV);
2031 	value &= ~APIC_VECTOR_MASK;
2032 	value |= APIC_SPIV_APIC_ENABLED;
2033 	value |= 0xf;
2034 	apic_write(APIC_SPIV, value);
2035 
2036 	if (!virt_wire_setup) {
2037 		/*
2038 		 * For LVT0 make it edge triggered, active high,
2039 		 * external and enabled
2040 		 */
2041 		value = apic_read(APIC_LVT0);
2042 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2043 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2044 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2045 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2046 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2047 		apic_write(APIC_LVT0, value);
2048 	} else {
2049 		/* Disable LVT0 */
2050 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2051 	}
2052 
2053 	/*
2054 	 * For LVT1 make it edge triggered, active high,
2055 	 * nmi and enabled
2056 	 */
2057 	value = apic_read(APIC_LVT1);
2058 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2059 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2060 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2061 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2062 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2063 	apic_write(APIC_LVT1, value);
2064 }
2065 
2066 void __cpuinit generic_processor_info(int apicid, int version)
2067 {
2068 	int cpu, max = nr_cpu_ids;
2069 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2070 				phys_cpu_present_map);
2071 
2072 	/*
2073 	 * If boot cpu has not been detected yet, then only allow upto
2074 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2075 	 */
2076 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2077 	    apicid != boot_cpu_physical_apicid) {
2078 		int thiscpu = max + disabled_cpus - 1;
2079 
2080 		pr_warning(
2081 			"ACPI: NR_CPUS/possible_cpus limit of %i almost"
2082 			" reached. Keeping one slot for boot cpu."
2083 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2084 
2085 		disabled_cpus++;
2086 		return;
2087 	}
2088 
2089 	if (num_processors >= nr_cpu_ids) {
2090 		int thiscpu = max + disabled_cpus;
2091 
2092 		pr_warning(
2093 			"ACPI: NR_CPUS/possible_cpus limit of %i reached."
2094 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2095 
2096 		disabled_cpus++;
2097 		return;
2098 	}
2099 
2100 	num_processors++;
2101 	if (apicid == boot_cpu_physical_apicid) {
2102 		/*
2103 		 * x86_bios_cpu_apicid is required to have processors listed
2104 		 * in same order as logical cpu numbers. Hence the first
2105 		 * entry is BSP, and so on.
2106 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2107 		 * for BSP.
2108 		 */
2109 		cpu = 0;
2110 	} else
2111 		cpu = cpumask_next_zero(-1, cpu_present_mask);
2112 
2113 	/*
2114 	 * Validate version
2115 	 */
2116 	if (version == 0x0) {
2117 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2118 			   cpu, apicid);
2119 		version = 0x10;
2120 	}
2121 	apic_version[apicid] = version;
2122 
2123 	if (version != apic_version[boot_cpu_physical_apicid]) {
2124 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2125 			apic_version[boot_cpu_physical_apicid], cpu, version);
2126 	}
2127 
2128 	physid_set(apicid, phys_cpu_present_map);
2129 	if (apicid > max_physical_apicid)
2130 		max_physical_apicid = apicid;
2131 
2132 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2133 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2134 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2135 #endif
2136 #ifdef CONFIG_X86_32
2137 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2138 		apic->x86_32_early_logical_apicid(cpu);
2139 #endif
2140 	set_cpu_possible(cpu, true);
2141 	set_cpu_present(cpu, true);
2142 }
2143 
2144 int hard_smp_processor_id(void)
2145 {
2146 	return read_apic_id();
2147 }
2148 
2149 void default_init_apic_ldr(void)
2150 {
2151 	unsigned long val;
2152 
2153 	apic_write(APIC_DFR, APIC_DFR_VALUE);
2154 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2155 	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2156 	apic_write(APIC_LDR, val);
2157 }
2158 
2159 int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2160 				   const struct cpumask *andmask,
2161 				   unsigned int *apicid)
2162 {
2163 	unsigned int cpu;
2164 
2165 	for_each_cpu_and(cpu, cpumask, andmask) {
2166 		if (cpumask_test_cpu(cpu, cpu_online_mask))
2167 			break;
2168 	}
2169 
2170 	if (likely(cpu < nr_cpu_ids)) {
2171 		*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2172 		return 0;
2173 	}
2174 
2175 	return -EINVAL;
2176 }
2177 
2178 /*
2179  * Override the generic EOI implementation with an optimized version.
2180  * Only called during early boot when only one CPU is active and with
2181  * interrupts disabled, so we know this does not race with actual APIC driver
2182  * use.
2183  */
2184 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2185 {
2186 	struct apic **drv;
2187 
2188 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2189 		/* Should happen once for each apic */
2190 		WARN_ON((*drv)->eoi_write == eoi_write);
2191 		(*drv)->eoi_write = eoi_write;
2192 	}
2193 }
2194 
2195 /*
2196  * Power management
2197  */
2198 #ifdef CONFIG_PM
2199 
2200 static struct {
2201 	/*
2202 	 * 'active' is true if the local APIC was enabled by us and
2203 	 * not the BIOS; this signifies that we are also responsible
2204 	 * for disabling it before entering apm/acpi suspend
2205 	 */
2206 	int active;
2207 	/* r/w apic fields */
2208 	unsigned int apic_id;
2209 	unsigned int apic_taskpri;
2210 	unsigned int apic_ldr;
2211 	unsigned int apic_dfr;
2212 	unsigned int apic_spiv;
2213 	unsigned int apic_lvtt;
2214 	unsigned int apic_lvtpc;
2215 	unsigned int apic_lvt0;
2216 	unsigned int apic_lvt1;
2217 	unsigned int apic_lvterr;
2218 	unsigned int apic_tmict;
2219 	unsigned int apic_tdcr;
2220 	unsigned int apic_thmr;
2221 } apic_pm_state;
2222 
2223 static int lapic_suspend(void)
2224 {
2225 	unsigned long flags;
2226 	int maxlvt;
2227 
2228 	if (!apic_pm_state.active)
2229 		return 0;
2230 
2231 	maxlvt = lapic_get_maxlvt();
2232 
2233 	apic_pm_state.apic_id = apic_read(APIC_ID);
2234 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2235 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2236 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2237 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2238 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2239 	if (maxlvt >= 4)
2240 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2241 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2242 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2243 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2244 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2245 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2246 #ifdef CONFIG_X86_THERMAL_VECTOR
2247 	if (maxlvt >= 5)
2248 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2249 #endif
2250 
2251 	local_irq_save(flags);
2252 	disable_local_APIC();
2253 
2254 	if (irq_remapping_enabled)
2255 		irq_remapping_disable();
2256 
2257 	local_irq_restore(flags);
2258 	return 0;
2259 }
2260 
2261 static void lapic_resume(void)
2262 {
2263 	unsigned int l, h;
2264 	unsigned long flags;
2265 	int maxlvt;
2266 
2267 	if (!apic_pm_state.active)
2268 		return;
2269 
2270 	local_irq_save(flags);
2271 	if (irq_remapping_enabled) {
2272 		/*
2273 		 * IO-APIC and PIC have their own resume routines.
2274 		 * We just mask them here to make sure the interrupt
2275 		 * subsystem is completely quiet while we enable x2apic
2276 		 * and interrupt-remapping.
2277 		 */
2278 		mask_ioapic_entries();
2279 		legacy_pic->mask_all();
2280 	}
2281 
2282 	if (x2apic_mode)
2283 		enable_x2apic();
2284 	else {
2285 		/*
2286 		 * Make sure the APICBASE points to the right address
2287 		 *
2288 		 * FIXME! This will be wrong if we ever support suspend on
2289 		 * SMP! We'll need to do this as part of the CPU restore!
2290 		 */
2291 		if (boot_cpu_data.x86 >= 6) {
2292 			rdmsr(MSR_IA32_APICBASE, l, h);
2293 			l &= ~MSR_IA32_APICBASE_BASE;
2294 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2295 			wrmsr(MSR_IA32_APICBASE, l, h);
2296 		}
2297 	}
2298 
2299 	maxlvt = lapic_get_maxlvt();
2300 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2301 	apic_write(APIC_ID, apic_pm_state.apic_id);
2302 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2303 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2304 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2305 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2306 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2307 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2308 #if defined(CONFIG_X86_MCE_P4THERMAL) || defined(CONFIG_X86_MCE_INTEL)
2309 	if (maxlvt >= 5)
2310 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2311 #endif
2312 	if (maxlvt >= 4)
2313 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2314 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2315 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2316 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2317 	apic_write(APIC_ESR, 0);
2318 	apic_read(APIC_ESR);
2319 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2320 	apic_write(APIC_ESR, 0);
2321 	apic_read(APIC_ESR);
2322 
2323 	if (irq_remapping_enabled)
2324 		irq_remapping_reenable(x2apic_mode);
2325 
2326 	local_irq_restore(flags);
2327 }
2328 
2329 /*
2330  * This device has no shutdown method - fully functioning local APICs
2331  * are needed on every CPU up until machine_halt/restart/poweroff.
2332  */
2333 
2334 static struct syscore_ops lapic_syscore_ops = {
2335 	.resume		= lapic_resume,
2336 	.suspend	= lapic_suspend,
2337 };
2338 
2339 static void __cpuinit apic_pm_activate(void)
2340 {
2341 	apic_pm_state.active = 1;
2342 }
2343 
2344 static int __init init_lapic_sysfs(void)
2345 {
2346 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2347 	if (cpu_has_apic)
2348 		register_syscore_ops(&lapic_syscore_ops);
2349 
2350 	return 0;
2351 }
2352 
2353 /* local apic needs to resume before other devices access its registers. */
2354 core_initcall(init_lapic_sysfs);
2355 
2356 #else	/* CONFIG_PM */
2357 
2358 static void apic_pm_activate(void) { }
2359 
2360 #endif	/* CONFIG_PM */
2361 
2362 #ifdef CONFIG_X86_64
2363 
2364 static int __cpuinit apic_cluster_num(void)
2365 {
2366 	int i, clusters, zeros;
2367 	unsigned id;
2368 	u16 *bios_cpu_apicid;
2369 	DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2370 
2371 	bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2372 	bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2373 
2374 	for (i = 0; i < nr_cpu_ids; i++) {
2375 		/* are we being called early in kernel startup? */
2376 		if (bios_cpu_apicid) {
2377 			id = bios_cpu_apicid[i];
2378 		} else if (i < nr_cpu_ids) {
2379 			if (cpu_present(i))
2380 				id = per_cpu(x86_bios_cpu_apicid, i);
2381 			else
2382 				continue;
2383 		} else
2384 			break;
2385 
2386 		if (id != BAD_APICID)
2387 			__set_bit(APIC_CLUSTERID(id), clustermap);
2388 	}
2389 
2390 	/* Problem:  Partially populated chassis may not have CPUs in some of
2391 	 * the APIC clusters they have been allocated.  Only present CPUs have
2392 	 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2393 	 * Since clusters are allocated sequentially, count zeros only if
2394 	 * they are bounded by ones.
2395 	 */
2396 	clusters = 0;
2397 	zeros = 0;
2398 	for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2399 		if (test_bit(i, clustermap)) {
2400 			clusters += 1 + zeros;
2401 			zeros = 0;
2402 		} else
2403 			++zeros;
2404 	}
2405 
2406 	return clusters;
2407 }
2408 
2409 static int __cpuinitdata multi_checked;
2410 static int __cpuinitdata multi;
2411 
2412 static int __cpuinit set_multi(const struct dmi_system_id *d)
2413 {
2414 	if (multi)
2415 		return 0;
2416 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2417 	multi = 1;
2418 	return 0;
2419 }
2420 
2421 static const __cpuinitconst struct dmi_system_id multi_dmi_table[] = {
2422 	{
2423 		.callback = set_multi,
2424 		.ident = "IBM System Summit2",
2425 		.matches = {
2426 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2427 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2428 		},
2429 	},
2430 	{}
2431 };
2432 
2433 static void __cpuinit dmi_check_multi(void)
2434 {
2435 	if (multi_checked)
2436 		return;
2437 
2438 	dmi_check_system(multi_dmi_table);
2439 	multi_checked = 1;
2440 }
2441 
2442 /*
2443  * apic_is_clustered_box() -- Check if we can expect good TSC
2444  *
2445  * Thus far, the major user of this is IBM's Summit2 series:
2446  * Clustered boxes may have unsynced TSC problems if they are
2447  * multi-chassis.
2448  * Use DMI to check them
2449  */
2450 __cpuinit int apic_is_clustered_box(void)
2451 {
2452 	dmi_check_multi();
2453 	if (multi)
2454 		return 1;
2455 
2456 	if (!is_vsmp_box())
2457 		return 0;
2458 
2459 	/*
2460 	 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2461 	 * not guaranteed to be synced between boards
2462 	 */
2463 	if (apic_cluster_num() > 1)
2464 		return 1;
2465 
2466 	return 0;
2467 }
2468 #endif
2469 
2470 /*
2471  * APIC command line parameters
2472  */
2473 static int __init setup_disableapic(char *arg)
2474 {
2475 	disable_apic = 1;
2476 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2477 	return 0;
2478 }
2479 early_param("disableapic", setup_disableapic);
2480 
2481 /* same as disableapic, for compatibility */
2482 static int __init setup_nolapic(char *arg)
2483 {
2484 	return setup_disableapic(arg);
2485 }
2486 early_param("nolapic", setup_nolapic);
2487 
2488 static int __init parse_lapic_timer_c2_ok(char *arg)
2489 {
2490 	local_apic_timer_c2_ok = 1;
2491 	return 0;
2492 }
2493 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2494 
2495 static int __init parse_disable_apic_timer(char *arg)
2496 {
2497 	disable_apic_timer = 1;
2498 	return 0;
2499 }
2500 early_param("noapictimer", parse_disable_apic_timer);
2501 
2502 static int __init parse_nolapic_timer(char *arg)
2503 {
2504 	disable_apic_timer = 1;
2505 	return 0;
2506 }
2507 early_param("nolapic_timer", parse_nolapic_timer);
2508 
2509 static int __init apic_set_verbosity(char *arg)
2510 {
2511 	if (!arg)  {
2512 #ifdef CONFIG_X86_64
2513 		skip_ioapic_setup = 0;
2514 		return 0;
2515 #endif
2516 		return -EINVAL;
2517 	}
2518 
2519 	if (strcmp("debug", arg) == 0)
2520 		apic_verbosity = APIC_DEBUG;
2521 	else if (strcmp("verbose", arg) == 0)
2522 		apic_verbosity = APIC_VERBOSE;
2523 	else {
2524 		pr_warning("APIC Verbosity level %s not recognised"
2525 			" use apic=verbose or apic=debug\n", arg);
2526 		return -EINVAL;
2527 	}
2528 
2529 	return 0;
2530 }
2531 early_param("apic", apic_set_verbosity);
2532 
2533 static int __init lapic_insert_resource(void)
2534 {
2535 	if (!apic_phys)
2536 		return -1;
2537 
2538 	/* Put local APIC into the resource map. */
2539 	lapic_resource.start = apic_phys;
2540 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2541 	insert_resource(&iomem_resource, &lapic_resource);
2542 
2543 	return 0;
2544 }
2545 
2546 /*
2547  * need call insert after e820_reserve_resources()
2548  * that is using request_resource
2549  */
2550 late_initcall(lapic_insert_resource);
2551