xref: /openbmc/linux/arch/x86/kernel/apic/apic.c (revision 089a49b6)
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/trace/irq_vectors.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/perf_event.h>
41 #include <asm/x86_init.h>
42 #include <asm/pgalloc.h>
43 #include <linux/atomic.h>
44 #include <asm/mpspec.h>
45 #include <asm/i8259.h>
46 #include <asm/proto.h>
47 #include <asm/apic.h>
48 #include <asm/io_apic.h>
49 #include <asm/desc.h>
50 #include <asm/hpet.h>
51 #include <asm/idle.h>
52 #include <asm/mtrr.h>
53 #include <asm/time.h>
54 #include <asm/smp.h>
55 #include <asm/mce.h>
56 #include <asm/tsc.h>
57 #include <asm/hypervisor.h>
58 
59 unsigned int num_processors;
60 
61 unsigned disabled_cpus;
62 
63 /* Processor that is doing the boot up */
64 unsigned int boot_cpu_physical_apicid = -1U;
65 
66 /*
67  * The highest APIC ID seen during enumeration.
68  */
69 unsigned int max_physical_apicid;
70 
71 /*
72  * Bitmask of physically existing CPUs:
73  */
74 physid_mask_t phys_cpu_present_map;
75 
76 /*
77  * Map cpu index to physical APIC ID
78  */
79 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
80 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
81 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
82 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
83 
84 #ifdef CONFIG_X86_32
85 
86 /*
87  * On x86_32, the mapping between cpu and logical apicid may vary
88  * depending on apic in use.  The following early percpu variable is
89  * used for the mapping.  This is where the behaviors of x86_64 and 32
90  * actually diverge.  Let's keep it ugly for now.
91  */
92 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
93 
94 /* Local APIC was disabled by the BIOS and enabled by the kernel */
95 static int enabled_via_apicbase;
96 
97 /*
98  * Handle interrupt mode configuration register (IMCR).
99  * This register controls whether the interrupt signals
100  * that reach the BSP come from the master PIC or from the
101  * local APIC. Before entering Symmetric I/O Mode, either
102  * the BIOS or the operating system must switch out of
103  * PIC Mode by changing the IMCR.
104  */
105 static inline void imcr_pic_to_apic(void)
106 {
107 	/* select IMCR register */
108 	outb(0x70, 0x22);
109 	/* NMI and 8259 INTR go through APIC */
110 	outb(0x01, 0x23);
111 }
112 
113 static inline void imcr_apic_to_pic(void)
114 {
115 	/* select IMCR register */
116 	outb(0x70, 0x22);
117 	/* NMI and 8259 INTR go directly to BSP */
118 	outb(0x00, 0x23);
119 }
120 #endif
121 
122 /*
123  * Knob to control our willingness to enable the local APIC.
124  *
125  * +1=force-enable
126  */
127 static int force_enable_local_apic __initdata;
128 /*
129  * APIC command line parameters
130  */
131 static int __init parse_lapic(char *arg)
132 {
133 	if (config_enabled(CONFIG_X86_32) && !arg)
134 		force_enable_local_apic = 1;
135 	else if (arg && !strncmp(arg, "notscdeadline", 13))
136 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
137 	return 0;
138 }
139 early_param("lapic", parse_lapic);
140 
141 #ifdef CONFIG_X86_64
142 static int apic_calibrate_pmtmr __initdata;
143 static __init int setup_apicpmtimer(char *s)
144 {
145 	apic_calibrate_pmtmr = 1;
146 	notsc_setup(NULL);
147 	return 0;
148 }
149 __setup("apicpmtimer", setup_apicpmtimer);
150 #endif
151 
152 int x2apic_mode;
153 #ifdef CONFIG_X86_X2APIC
154 /* x2apic enabled before OS handover */
155 int x2apic_preenabled;
156 static int x2apic_disabled;
157 static int nox2apic;
158 static __init int setup_nox2apic(char *str)
159 {
160 	if (x2apic_enabled()) {
161 		int apicid = native_apic_msr_read(APIC_ID);
162 
163 		if (apicid >= 255) {
164 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
165 				   apicid);
166 			return 0;
167 		}
168 
169 		pr_warning("x2apic already enabled. will disable it\n");
170 	} else
171 		setup_clear_cpu_cap(X86_FEATURE_X2APIC);
172 
173 	nox2apic = 1;
174 
175 	return 0;
176 }
177 early_param("nox2apic", setup_nox2apic);
178 #endif
179 
180 unsigned long mp_lapic_addr;
181 int disable_apic;
182 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
183 static int disable_apic_timer __initdata;
184 /* Local APIC timer works in C2 */
185 int local_apic_timer_c2_ok;
186 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
187 
188 int first_system_vector = 0xfe;
189 
190 /*
191  * Debug level, exported for io_apic.c
192  */
193 unsigned int apic_verbosity;
194 
195 int pic_mode;
196 
197 /* Have we found an MP table */
198 int smp_found_config;
199 
200 static struct resource lapic_resource = {
201 	.name = "Local APIC",
202 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
203 };
204 
205 unsigned int lapic_timer_frequency = 0;
206 
207 static void apic_pm_activate(void);
208 
209 static unsigned long apic_phys;
210 
211 /*
212  * Get the LAPIC version
213  */
214 static inline int lapic_get_version(void)
215 {
216 	return GET_APIC_VERSION(apic_read(APIC_LVR));
217 }
218 
219 /*
220  * Check, if the APIC is integrated or a separate chip
221  */
222 static inline int lapic_is_integrated(void)
223 {
224 #ifdef CONFIG_X86_64
225 	return 1;
226 #else
227 	return APIC_INTEGRATED(lapic_get_version());
228 #endif
229 }
230 
231 /*
232  * Check, whether this is a modern or a first generation APIC
233  */
234 static int modern_apic(void)
235 {
236 	/* AMD systems use old APIC versions, so check the CPU */
237 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
238 	    boot_cpu_data.x86 >= 0xf)
239 		return 1;
240 	return lapic_get_version() >= 0x14;
241 }
242 
243 /*
244  * right after this call apic become NOOP driven
245  * so apic->write/read doesn't do anything
246  */
247 static void __init apic_disable(void)
248 {
249 	pr_info("APIC: switched to apic NOOP\n");
250 	apic = &apic_noop;
251 }
252 
253 void native_apic_wait_icr_idle(void)
254 {
255 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
256 		cpu_relax();
257 }
258 
259 u32 native_safe_apic_wait_icr_idle(void)
260 {
261 	u32 send_status;
262 	int timeout;
263 
264 	timeout = 0;
265 	do {
266 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
267 		if (!send_status)
268 			break;
269 		inc_irq_stat(icr_read_retry_count);
270 		udelay(100);
271 	} while (timeout++ < 1000);
272 
273 	return send_status;
274 }
275 
276 void native_apic_icr_write(u32 low, u32 id)
277 {
278 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
279 	apic_write(APIC_ICR, low);
280 }
281 
282 u64 native_apic_icr_read(void)
283 {
284 	u32 icr1, icr2;
285 
286 	icr2 = apic_read(APIC_ICR2);
287 	icr1 = apic_read(APIC_ICR);
288 
289 	return icr1 | ((u64)icr2 << 32);
290 }
291 
292 #ifdef CONFIG_X86_32
293 /**
294  * get_physical_broadcast - Get number of physical broadcast IDs
295  */
296 int get_physical_broadcast(void)
297 {
298 	return modern_apic() ? 0xff : 0xf;
299 }
300 #endif
301 
302 /**
303  * lapic_get_maxlvt - get the maximum number of local vector table entries
304  */
305 int lapic_get_maxlvt(void)
306 {
307 	unsigned int v;
308 
309 	v = apic_read(APIC_LVR);
310 	/*
311 	 * - we always have APIC integrated on 64bit mode
312 	 * - 82489DXs do not report # of LVT entries
313 	 */
314 	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
315 }
316 
317 /*
318  * Local APIC timer
319  */
320 
321 /* Clock divisor */
322 #define APIC_DIVISOR 16
323 #define TSC_DIVISOR  32
324 
325 /*
326  * This function sets up the local APIC timer, with a timeout of
327  * 'clocks' APIC bus clock. During calibration we actually call
328  * this function twice on the boot CPU, once with a bogus timeout
329  * value, second time for real. The other (noncalibrating) CPUs
330  * call this function only once, with the real, calibrated value.
331  *
332  * We do reads before writes even if unnecessary, to get around the
333  * P5 APIC double write bug.
334  */
335 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
336 {
337 	unsigned int lvtt_value, tmp_value;
338 
339 	lvtt_value = LOCAL_TIMER_VECTOR;
340 	if (!oneshot)
341 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
342 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
343 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
344 
345 	if (!lapic_is_integrated())
346 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
347 
348 	if (!irqen)
349 		lvtt_value |= APIC_LVT_MASKED;
350 
351 	apic_write(APIC_LVTT, lvtt_value);
352 
353 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
354 		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
355 		return;
356 	}
357 
358 	/*
359 	 * Divide PICLK by 16
360 	 */
361 	tmp_value = apic_read(APIC_TDCR);
362 	apic_write(APIC_TDCR,
363 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
364 		APIC_TDR_DIV_16);
365 
366 	if (!oneshot)
367 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
368 }
369 
370 /*
371  * Setup extended LVT, AMD specific
372  *
373  * Software should use the LVT offsets the BIOS provides.  The offsets
374  * are determined by the subsystems using it like those for MCE
375  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
376  * are supported. Beginning with family 10h at least 4 offsets are
377  * available.
378  *
379  * Since the offsets must be consistent for all cores, we keep track
380  * of the LVT offsets in software and reserve the offset for the same
381  * vector also to be used on other cores. An offset is freed by
382  * setting the entry to APIC_EILVT_MASKED.
383  *
384  * If the BIOS is right, there should be no conflicts. Otherwise a
385  * "[Firmware Bug]: ..." error message is generated. However, if
386  * software does not properly determines the offsets, it is not
387  * necessarily a BIOS bug.
388  */
389 
390 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
391 
392 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
393 {
394 	return (old & APIC_EILVT_MASKED)
395 		|| (new == APIC_EILVT_MASKED)
396 		|| ((new & ~APIC_EILVT_MASKED) == old);
397 }
398 
399 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
400 {
401 	unsigned int rsvd, vector;
402 
403 	if (offset >= APIC_EILVT_NR_MAX)
404 		return ~0;
405 
406 	rsvd = atomic_read(&eilvt_offsets[offset]);
407 	do {
408 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
409 		if (vector && !eilvt_entry_is_changeable(vector, new))
410 			/* may not change if vectors are different */
411 			return rsvd;
412 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
413 	} while (rsvd != new);
414 
415 	rsvd &= ~APIC_EILVT_MASKED;
416 	if (rsvd && rsvd != vector)
417 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
418 			offset, rsvd);
419 
420 	return new;
421 }
422 
423 /*
424  * If mask=1, the LVT entry does not generate interrupts while mask=0
425  * enables the vector. See also the BKDGs. Must be called with
426  * preemption disabled.
427  */
428 
429 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
430 {
431 	unsigned long reg = APIC_EILVTn(offset);
432 	unsigned int new, old, reserved;
433 
434 	new = (mask << 16) | (msg_type << 8) | vector;
435 	old = apic_read(reg);
436 	reserved = reserve_eilvt_offset(offset, new);
437 
438 	if (reserved != new) {
439 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
440 		       "vector 0x%x, but the register is already in use for "
441 		       "vector 0x%x on another cpu\n",
442 		       smp_processor_id(), reg, offset, new, reserved);
443 		return -EINVAL;
444 	}
445 
446 	if (!eilvt_entry_is_changeable(old, new)) {
447 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
448 		       "vector 0x%x, but the register is already in use for "
449 		       "vector 0x%x on this cpu\n",
450 		       smp_processor_id(), reg, offset, new, old);
451 		return -EBUSY;
452 	}
453 
454 	apic_write(reg, new);
455 
456 	return 0;
457 }
458 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
459 
460 /*
461  * Program the next event, relative to now
462  */
463 static int lapic_next_event(unsigned long delta,
464 			    struct clock_event_device *evt)
465 {
466 	apic_write(APIC_TMICT, delta);
467 	return 0;
468 }
469 
470 static int lapic_next_deadline(unsigned long delta,
471 			       struct clock_event_device *evt)
472 {
473 	u64 tsc;
474 
475 	rdtscll(tsc);
476 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
477 	return 0;
478 }
479 
480 /*
481  * Setup the lapic timer in periodic or oneshot mode
482  */
483 static void lapic_timer_setup(enum clock_event_mode mode,
484 			      struct clock_event_device *evt)
485 {
486 	unsigned long flags;
487 	unsigned int v;
488 
489 	/* Lapic used as dummy for broadcast ? */
490 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
491 		return;
492 
493 	local_irq_save(flags);
494 
495 	switch (mode) {
496 	case CLOCK_EVT_MODE_PERIODIC:
497 	case CLOCK_EVT_MODE_ONESHOT:
498 		__setup_APIC_LVTT(lapic_timer_frequency,
499 				  mode != CLOCK_EVT_MODE_PERIODIC, 1);
500 		break;
501 	case CLOCK_EVT_MODE_UNUSED:
502 	case CLOCK_EVT_MODE_SHUTDOWN:
503 		v = apic_read(APIC_LVTT);
504 		v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
505 		apic_write(APIC_LVTT, v);
506 		apic_write(APIC_TMICT, 0);
507 		break;
508 	case CLOCK_EVT_MODE_RESUME:
509 		/* Nothing to do here */
510 		break;
511 	}
512 
513 	local_irq_restore(flags);
514 }
515 
516 /*
517  * Local APIC timer broadcast function
518  */
519 static void lapic_timer_broadcast(const struct cpumask *mask)
520 {
521 #ifdef CONFIG_SMP
522 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
523 #endif
524 }
525 
526 
527 /*
528  * The local apic timer can be used for any function which is CPU local.
529  */
530 static struct clock_event_device lapic_clockevent = {
531 	.name		= "lapic",
532 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
533 			| CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
534 	.shift		= 32,
535 	.set_mode	= lapic_timer_setup,
536 	.set_next_event	= lapic_next_event,
537 	.broadcast	= lapic_timer_broadcast,
538 	.rating		= 100,
539 	.irq		= -1,
540 };
541 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
542 
543 /*
544  * Setup the local APIC timer for this CPU. Copy the initialized values
545  * of the boot CPU and register the clock event in the framework.
546  */
547 static void setup_APIC_timer(void)
548 {
549 	struct clock_event_device *levt = &__get_cpu_var(lapic_events);
550 
551 	if (this_cpu_has(X86_FEATURE_ARAT)) {
552 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
553 		/* Make LAPIC timer preferrable over percpu HPET */
554 		lapic_clockevent.rating = 150;
555 	}
556 
557 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
558 	levt->cpumask = cpumask_of(smp_processor_id());
559 
560 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
561 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
562 				    CLOCK_EVT_FEAT_DUMMY);
563 		levt->set_next_event = lapic_next_deadline;
564 		clockevents_config_and_register(levt,
565 						(tsc_khz / TSC_DIVISOR) * 1000,
566 						0xF, ~0UL);
567 	} else
568 		clockevents_register_device(levt);
569 }
570 
571 /*
572  * In this functions we calibrate APIC bus clocks to the external timer.
573  *
574  * We want to do the calibration only once since we want to have local timer
575  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
576  * frequency.
577  *
578  * This was previously done by reading the PIT/HPET and waiting for a wrap
579  * around to find out, that a tick has elapsed. I have a box, where the PIT
580  * readout is broken, so it never gets out of the wait loop again. This was
581  * also reported by others.
582  *
583  * Monitoring the jiffies value is inaccurate and the clockevents
584  * infrastructure allows us to do a simple substitution of the interrupt
585  * handler.
586  *
587  * The calibration routine also uses the pm_timer when possible, as the PIT
588  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
589  * back to normal later in the boot process).
590  */
591 
592 #define LAPIC_CAL_LOOPS		(HZ/10)
593 
594 static __initdata int lapic_cal_loops = -1;
595 static __initdata long lapic_cal_t1, lapic_cal_t2;
596 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
597 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
598 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
599 
600 /*
601  * Temporary interrupt handler.
602  */
603 static void __init lapic_cal_handler(struct clock_event_device *dev)
604 {
605 	unsigned long long tsc = 0;
606 	long tapic = apic_read(APIC_TMCCT);
607 	unsigned long pm = acpi_pm_read_early();
608 
609 	if (cpu_has_tsc)
610 		rdtscll(tsc);
611 
612 	switch (lapic_cal_loops++) {
613 	case 0:
614 		lapic_cal_t1 = tapic;
615 		lapic_cal_tsc1 = tsc;
616 		lapic_cal_pm1 = pm;
617 		lapic_cal_j1 = jiffies;
618 		break;
619 
620 	case LAPIC_CAL_LOOPS:
621 		lapic_cal_t2 = tapic;
622 		lapic_cal_tsc2 = tsc;
623 		if (pm < lapic_cal_pm1)
624 			pm += ACPI_PM_OVRRUN;
625 		lapic_cal_pm2 = pm;
626 		lapic_cal_j2 = jiffies;
627 		break;
628 	}
629 }
630 
631 static int __init
632 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
633 {
634 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
635 	const long pm_thresh = pm_100ms / 100;
636 	unsigned long mult;
637 	u64 res;
638 
639 #ifndef CONFIG_X86_PM_TIMER
640 	return -1;
641 #endif
642 
643 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
644 
645 	/* Check, if the PM timer is available */
646 	if (!deltapm)
647 		return -1;
648 
649 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
650 
651 	if (deltapm > (pm_100ms - pm_thresh) &&
652 	    deltapm < (pm_100ms + pm_thresh)) {
653 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
654 		return 0;
655 	}
656 
657 	res = (((u64)deltapm) *  mult) >> 22;
658 	do_div(res, 1000000);
659 	pr_warning("APIC calibration not consistent "
660 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
661 
662 	/* Correct the lapic counter value */
663 	res = (((u64)(*delta)) * pm_100ms);
664 	do_div(res, deltapm);
665 	pr_info("APIC delta adjusted to PM-Timer: "
666 		"%lu (%ld)\n", (unsigned long)res, *delta);
667 	*delta = (long)res;
668 
669 	/* Correct the tsc counter value */
670 	if (cpu_has_tsc) {
671 		res = (((u64)(*deltatsc)) * pm_100ms);
672 		do_div(res, deltapm);
673 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
674 					  "PM-Timer: %lu (%ld)\n",
675 					(unsigned long)res, *deltatsc);
676 		*deltatsc = (long)res;
677 	}
678 
679 	return 0;
680 }
681 
682 static int __init calibrate_APIC_clock(void)
683 {
684 	struct clock_event_device *levt = &__get_cpu_var(lapic_events);
685 	void (*real_handler)(struct clock_event_device *dev);
686 	unsigned long deltaj;
687 	long delta, deltatsc;
688 	int pm_referenced = 0;
689 
690 	/**
691 	 * check if lapic timer has already been calibrated by platform
692 	 * specific routine, such as tsc calibration code. if so, we just fill
693 	 * in the clockevent structure and return.
694 	 */
695 
696 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
697 		return 0;
698 	} else if (lapic_timer_frequency) {
699 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
700 				lapic_timer_frequency);
701 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
702 					TICK_NSEC, lapic_clockevent.shift);
703 		lapic_clockevent.max_delta_ns =
704 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
705 		lapic_clockevent.min_delta_ns =
706 			clockevent_delta2ns(0xF, &lapic_clockevent);
707 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
708 		return 0;
709 	}
710 
711 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
712 		    "calibrating APIC timer ...\n");
713 
714 	local_irq_disable();
715 
716 	/* Replace the global interrupt handler */
717 	real_handler = global_clock_event->event_handler;
718 	global_clock_event->event_handler = lapic_cal_handler;
719 
720 	/*
721 	 * Setup the APIC counter to maximum. There is no way the lapic
722 	 * can underflow in the 100ms detection time frame
723 	 */
724 	__setup_APIC_LVTT(0xffffffff, 0, 0);
725 
726 	/* Let the interrupts run */
727 	local_irq_enable();
728 
729 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
730 		cpu_relax();
731 
732 	local_irq_disable();
733 
734 	/* Restore the real event handler */
735 	global_clock_event->event_handler = real_handler;
736 
737 	/* Build delta t1-t2 as apic timer counts down */
738 	delta = lapic_cal_t1 - lapic_cal_t2;
739 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
740 
741 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
742 
743 	/* we trust the PM based calibration if possible */
744 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
745 					&delta, &deltatsc);
746 
747 	/* Calculate the scaled math multiplication factor */
748 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
749 				       lapic_clockevent.shift);
750 	lapic_clockevent.max_delta_ns =
751 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
752 	lapic_clockevent.min_delta_ns =
753 		clockevent_delta2ns(0xF, &lapic_clockevent);
754 
755 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
756 
757 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
758 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
759 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
760 		    lapic_timer_frequency);
761 
762 	if (cpu_has_tsc) {
763 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
764 			    "%ld.%04ld MHz.\n",
765 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
766 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
767 	}
768 
769 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
770 		    "%u.%04u MHz.\n",
771 		    lapic_timer_frequency / (1000000 / HZ),
772 		    lapic_timer_frequency % (1000000 / HZ));
773 
774 	/*
775 	 * Do a sanity check on the APIC calibration result
776 	 */
777 	if (lapic_timer_frequency < (1000000 / HZ)) {
778 		local_irq_enable();
779 		pr_warning("APIC frequency too slow, disabling apic timer\n");
780 		return -1;
781 	}
782 
783 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
784 
785 	/*
786 	 * PM timer calibration failed or not turned on
787 	 * so lets try APIC timer based calibration
788 	 */
789 	if (!pm_referenced) {
790 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
791 
792 		/*
793 		 * Setup the apic timer manually
794 		 */
795 		levt->event_handler = lapic_cal_handler;
796 		lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
797 		lapic_cal_loops = -1;
798 
799 		/* Let the interrupts run */
800 		local_irq_enable();
801 
802 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
803 			cpu_relax();
804 
805 		/* Stop the lapic timer */
806 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
807 
808 		/* Jiffies delta */
809 		deltaj = lapic_cal_j2 - lapic_cal_j1;
810 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
811 
812 		/* Check, if the jiffies result is consistent */
813 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
814 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
815 		else
816 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
817 	} else
818 		local_irq_enable();
819 
820 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
821 		pr_warning("APIC timer disabled due to verification failure\n");
822 			return -1;
823 	}
824 
825 	return 0;
826 }
827 
828 /*
829  * Setup the boot APIC
830  *
831  * Calibrate and verify the result.
832  */
833 void __init setup_boot_APIC_clock(void)
834 {
835 	/*
836 	 * The local apic timer can be disabled via the kernel
837 	 * commandline or from the CPU detection code. Register the lapic
838 	 * timer as a dummy clock event source on SMP systems, so the
839 	 * broadcast mechanism is used. On UP systems simply ignore it.
840 	 */
841 	if (disable_apic_timer) {
842 		pr_info("Disabling APIC timer\n");
843 		/* No broadcast on UP ! */
844 		if (num_possible_cpus() > 1) {
845 			lapic_clockevent.mult = 1;
846 			setup_APIC_timer();
847 		}
848 		return;
849 	}
850 
851 	if (calibrate_APIC_clock()) {
852 		/* No broadcast on UP ! */
853 		if (num_possible_cpus() > 1)
854 			setup_APIC_timer();
855 		return;
856 	}
857 
858 	/*
859 	 * If nmi_watchdog is set to IO_APIC, we need the
860 	 * PIT/HPET going.  Otherwise register lapic as a dummy
861 	 * device.
862 	 */
863 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
864 
865 	/* Setup the lapic or request the broadcast */
866 	setup_APIC_timer();
867 }
868 
869 void setup_secondary_APIC_clock(void)
870 {
871 	setup_APIC_timer();
872 }
873 
874 /*
875  * The guts of the apic timer interrupt
876  */
877 static void local_apic_timer_interrupt(void)
878 {
879 	int cpu = smp_processor_id();
880 	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
881 
882 	/*
883 	 * Normally we should not be here till LAPIC has been initialized but
884 	 * in some cases like kdump, its possible that there is a pending LAPIC
885 	 * timer interrupt from previous kernel's context and is delivered in
886 	 * new kernel the moment interrupts are enabled.
887 	 *
888 	 * Interrupts are enabled early and LAPIC is setup much later, hence
889 	 * its possible that when we get here evt->event_handler is NULL.
890 	 * Check for event_handler being NULL and discard the interrupt as
891 	 * spurious.
892 	 */
893 	if (!evt->event_handler) {
894 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
895 		/* Switch it off */
896 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
897 		return;
898 	}
899 
900 	/*
901 	 * the NMI deadlock-detector uses this.
902 	 */
903 	inc_irq_stat(apic_timer_irqs);
904 
905 	evt->event_handler(evt);
906 }
907 
908 /*
909  * Local APIC timer interrupt. This is the most natural way for doing
910  * local interrupts, but local timer interrupts can be emulated by
911  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
912  *
913  * [ if a single-CPU system runs an SMP kernel then we call the local
914  *   interrupt as well. Thus we cannot inline the local irq ... ]
915  */
916 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
917 {
918 	struct pt_regs *old_regs = set_irq_regs(regs);
919 
920 	/*
921 	 * NOTE! We'd better ACK the irq immediately,
922 	 * because timer handling can be slow.
923 	 *
924 	 * update_process_times() expects us to have done irq_enter().
925 	 * Besides, if we don't timer interrupts ignore the global
926 	 * interrupt lock, which is the WrongThing (tm) to do.
927 	 */
928 	entering_ack_irq();
929 	local_apic_timer_interrupt();
930 	exiting_irq();
931 
932 	set_irq_regs(old_regs);
933 }
934 
935 __visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
936 {
937 	struct pt_regs *old_regs = set_irq_regs(regs);
938 
939 	/*
940 	 * NOTE! We'd better ACK the irq immediately,
941 	 * because timer handling can be slow.
942 	 *
943 	 * update_process_times() expects us to have done irq_enter().
944 	 * Besides, if we don't timer interrupts ignore the global
945 	 * interrupt lock, which is the WrongThing (tm) to do.
946 	 */
947 	entering_ack_irq();
948 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
949 	local_apic_timer_interrupt();
950 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
951 	exiting_irq();
952 
953 	set_irq_regs(old_regs);
954 }
955 
956 int setup_profiling_timer(unsigned int multiplier)
957 {
958 	return -EINVAL;
959 }
960 
961 /*
962  * Local APIC start and shutdown
963  */
964 
965 /**
966  * clear_local_APIC - shutdown the local APIC
967  *
968  * This is called, when a CPU is disabled and before rebooting, so the state of
969  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
970  * leftovers during boot.
971  */
972 void clear_local_APIC(void)
973 {
974 	int maxlvt;
975 	u32 v;
976 
977 	/* APIC hasn't been mapped yet */
978 	if (!x2apic_mode && !apic_phys)
979 		return;
980 
981 	maxlvt = lapic_get_maxlvt();
982 	/*
983 	 * Masking an LVT entry can trigger a local APIC error
984 	 * if the vector is zero. Mask LVTERR first to prevent this.
985 	 */
986 	if (maxlvt >= 3) {
987 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
988 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
989 	}
990 	/*
991 	 * Careful: we have to set masks only first to deassert
992 	 * any level-triggered sources.
993 	 */
994 	v = apic_read(APIC_LVTT);
995 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
996 	v = apic_read(APIC_LVT0);
997 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
998 	v = apic_read(APIC_LVT1);
999 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1000 	if (maxlvt >= 4) {
1001 		v = apic_read(APIC_LVTPC);
1002 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1003 	}
1004 
1005 	/* lets not touch this if we didn't frob it */
1006 #ifdef CONFIG_X86_THERMAL_VECTOR
1007 	if (maxlvt >= 5) {
1008 		v = apic_read(APIC_LVTTHMR);
1009 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1010 	}
1011 #endif
1012 #ifdef CONFIG_X86_MCE_INTEL
1013 	if (maxlvt >= 6) {
1014 		v = apic_read(APIC_LVTCMCI);
1015 		if (!(v & APIC_LVT_MASKED))
1016 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1017 	}
1018 #endif
1019 
1020 	/*
1021 	 * Clean APIC state for other OSs:
1022 	 */
1023 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1024 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1025 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1026 	if (maxlvt >= 3)
1027 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1028 	if (maxlvt >= 4)
1029 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1030 
1031 	/* Integrated APIC (!82489DX) ? */
1032 	if (lapic_is_integrated()) {
1033 		if (maxlvt > 3)
1034 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1035 			apic_write(APIC_ESR, 0);
1036 		apic_read(APIC_ESR);
1037 	}
1038 }
1039 
1040 /**
1041  * disable_local_APIC - clear and disable the local APIC
1042  */
1043 void disable_local_APIC(void)
1044 {
1045 	unsigned int value;
1046 
1047 	/* APIC hasn't been mapped yet */
1048 	if (!x2apic_mode && !apic_phys)
1049 		return;
1050 
1051 	clear_local_APIC();
1052 
1053 	/*
1054 	 * Disable APIC (implies clearing of registers
1055 	 * for 82489DX!).
1056 	 */
1057 	value = apic_read(APIC_SPIV);
1058 	value &= ~APIC_SPIV_APIC_ENABLED;
1059 	apic_write(APIC_SPIV, value);
1060 
1061 #ifdef CONFIG_X86_32
1062 	/*
1063 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1064 	 * restore the disabled state.
1065 	 */
1066 	if (enabled_via_apicbase) {
1067 		unsigned int l, h;
1068 
1069 		rdmsr(MSR_IA32_APICBASE, l, h);
1070 		l &= ~MSR_IA32_APICBASE_ENABLE;
1071 		wrmsr(MSR_IA32_APICBASE, l, h);
1072 	}
1073 #endif
1074 }
1075 
1076 /*
1077  * If Linux enabled the LAPIC against the BIOS default disable it down before
1078  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1079  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1080  * for the case where Linux didn't enable the LAPIC.
1081  */
1082 void lapic_shutdown(void)
1083 {
1084 	unsigned long flags;
1085 
1086 	if (!cpu_has_apic && !apic_from_smp_config())
1087 		return;
1088 
1089 	local_irq_save(flags);
1090 
1091 #ifdef CONFIG_X86_32
1092 	if (!enabled_via_apicbase)
1093 		clear_local_APIC();
1094 	else
1095 #endif
1096 		disable_local_APIC();
1097 
1098 
1099 	local_irq_restore(flags);
1100 }
1101 
1102 /*
1103  * This is to verify that we're looking at a real local APIC.
1104  * Check these against your board if the CPUs aren't getting
1105  * started for no apparent reason.
1106  */
1107 int __init verify_local_APIC(void)
1108 {
1109 	unsigned int reg0, reg1;
1110 
1111 	/*
1112 	 * The version register is read-only in a real APIC.
1113 	 */
1114 	reg0 = apic_read(APIC_LVR);
1115 	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1116 	apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1117 	reg1 = apic_read(APIC_LVR);
1118 	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1119 
1120 	/*
1121 	 * The two version reads above should print the same
1122 	 * numbers.  If the second one is different, then we
1123 	 * poke at a non-APIC.
1124 	 */
1125 	if (reg1 != reg0)
1126 		return 0;
1127 
1128 	/*
1129 	 * Check if the version looks reasonably.
1130 	 */
1131 	reg1 = GET_APIC_VERSION(reg0);
1132 	if (reg1 == 0x00 || reg1 == 0xff)
1133 		return 0;
1134 	reg1 = lapic_get_maxlvt();
1135 	if (reg1 < 0x02 || reg1 == 0xff)
1136 		return 0;
1137 
1138 	/*
1139 	 * The ID register is read/write in a real APIC.
1140 	 */
1141 	reg0 = apic_read(APIC_ID);
1142 	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1143 	apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1144 	reg1 = apic_read(APIC_ID);
1145 	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1146 	apic_write(APIC_ID, reg0);
1147 	if (reg1 != (reg0 ^ apic->apic_id_mask))
1148 		return 0;
1149 
1150 	/*
1151 	 * The next two are just to see if we have sane values.
1152 	 * They're only really relevant if we're in Virtual Wire
1153 	 * compatibility mode, but most boxes are anymore.
1154 	 */
1155 	reg0 = apic_read(APIC_LVT0);
1156 	apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1157 	reg1 = apic_read(APIC_LVT1);
1158 	apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1159 
1160 	return 1;
1161 }
1162 
1163 /**
1164  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1165  */
1166 void __init sync_Arb_IDs(void)
1167 {
1168 	/*
1169 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1170 	 * needed on AMD.
1171 	 */
1172 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1173 		return;
1174 
1175 	/*
1176 	 * Wait for idle.
1177 	 */
1178 	apic_wait_icr_idle();
1179 
1180 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1181 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1182 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1183 }
1184 
1185 /*
1186  * An initial setup of the virtual wire mode.
1187  */
1188 void __init init_bsp_APIC(void)
1189 {
1190 	unsigned int value;
1191 
1192 	/*
1193 	 * Don't do the setup now if we have a SMP BIOS as the
1194 	 * through-I/O-APIC virtual wire mode might be active.
1195 	 */
1196 	if (smp_found_config || !cpu_has_apic)
1197 		return;
1198 
1199 	/*
1200 	 * Do not trust the local APIC being empty at bootup.
1201 	 */
1202 	clear_local_APIC();
1203 
1204 	/*
1205 	 * Enable APIC.
1206 	 */
1207 	value = apic_read(APIC_SPIV);
1208 	value &= ~APIC_VECTOR_MASK;
1209 	value |= APIC_SPIV_APIC_ENABLED;
1210 
1211 #ifdef CONFIG_X86_32
1212 	/* This bit is reserved on P4/Xeon and should be cleared */
1213 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1214 	    (boot_cpu_data.x86 == 15))
1215 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1216 	else
1217 #endif
1218 		value |= APIC_SPIV_FOCUS_DISABLED;
1219 	value |= SPURIOUS_APIC_VECTOR;
1220 	apic_write(APIC_SPIV, value);
1221 
1222 	/*
1223 	 * Set up the virtual wire mode.
1224 	 */
1225 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1226 	value = APIC_DM_NMI;
1227 	if (!lapic_is_integrated())		/* 82489DX */
1228 		value |= APIC_LVT_LEVEL_TRIGGER;
1229 	apic_write(APIC_LVT1, value);
1230 }
1231 
1232 static void lapic_setup_esr(void)
1233 {
1234 	unsigned int oldvalue, value, maxlvt;
1235 
1236 	if (!lapic_is_integrated()) {
1237 		pr_info("No ESR for 82489DX.\n");
1238 		return;
1239 	}
1240 
1241 	if (apic->disable_esr) {
1242 		/*
1243 		 * Something untraceable is creating bad interrupts on
1244 		 * secondary quads ... for the moment, just leave the
1245 		 * ESR disabled - we can't do anything useful with the
1246 		 * errors anyway - mbligh
1247 		 */
1248 		pr_info("Leaving ESR disabled.\n");
1249 		return;
1250 	}
1251 
1252 	maxlvt = lapic_get_maxlvt();
1253 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1254 		apic_write(APIC_ESR, 0);
1255 	oldvalue = apic_read(APIC_ESR);
1256 
1257 	/* enables sending errors */
1258 	value = ERROR_APIC_VECTOR;
1259 	apic_write(APIC_LVTERR, value);
1260 
1261 	/*
1262 	 * spec says clear errors after enabling vector.
1263 	 */
1264 	if (maxlvt > 3)
1265 		apic_write(APIC_ESR, 0);
1266 	value = apic_read(APIC_ESR);
1267 	if (value != oldvalue)
1268 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1269 			"vector: 0x%08x  after: 0x%08x\n",
1270 			oldvalue, value);
1271 }
1272 
1273 /**
1274  * setup_local_APIC - setup the local APIC
1275  *
1276  * Used to setup local APIC while initializing BSP or bringin up APs.
1277  * Always called with preemption disabled.
1278  */
1279 void setup_local_APIC(void)
1280 {
1281 	int cpu = smp_processor_id();
1282 	unsigned int value, queued;
1283 	int i, j, acked = 0;
1284 	unsigned long long tsc = 0, ntsc;
1285 	long long max_loops = cpu_khz;
1286 
1287 	if (cpu_has_tsc)
1288 		rdtscll(tsc);
1289 
1290 	if (disable_apic) {
1291 		disable_ioapic_support();
1292 		return;
1293 	}
1294 
1295 #ifdef CONFIG_X86_32
1296 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1297 	if (lapic_is_integrated() && apic->disable_esr) {
1298 		apic_write(APIC_ESR, 0);
1299 		apic_write(APIC_ESR, 0);
1300 		apic_write(APIC_ESR, 0);
1301 		apic_write(APIC_ESR, 0);
1302 	}
1303 #endif
1304 	perf_events_lapic_init();
1305 
1306 	/*
1307 	 * Double-check whether this APIC is really registered.
1308 	 * This is meaningless in clustered apic mode, so we skip it.
1309 	 */
1310 	BUG_ON(!apic->apic_id_registered());
1311 
1312 	/*
1313 	 * Intel recommends to set DFR, LDR and TPR before enabling
1314 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1315 	 * document number 292116).  So here it goes...
1316 	 */
1317 	apic->init_apic_ldr();
1318 
1319 #ifdef CONFIG_X86_32
1320 	/*
1321 	 * APIC LDR is initialized.  If logical_apicid mapping was
1322 	 * initialized during get_smp_config(), make sure it matches the
1323 	 * actual value.
1324 	 */
1325 	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1326 	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1327 	/* always use the value from LDR */
1328 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1329 		logical_smp_processor_id();
1330 
1331 	/*
1332 	 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1333 	 * node mapping during NUMA init.  Now that logical apicid is
1334 	 * guaranteed to be known, give it another chance.  This is already
1335 	 * a bit too late - percpu allocation has already happened without
1336 	 * proper NUMA affinity.
1337 	 */
1338 	if (apic->x86_32_numa_cpu_node)
1339 		set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1340 				   apic->x86_32_numa_cpu_node(cpu));
1341 #endif
1342 
1343 	/*
1344 	 * Set Task Priority to 'accept all'. We never change this
1345 	 * later on.
1346 	 */
1347 	value = apic_read(APIC_TASKPRI);
1348 	value &= ~APIC_TPRI_MASK;
1349 	apic_write(APIC_TASKPRI, value);
1350 
1351 	/*
1352 	 * After a crash, we no longer service the interrupts and a pending
1353 	 * interrupt from previous kernel might still have ISR bit set.
1354 	 *
1355 	 * Most probably by now CPU has serviced that pending interrupt and
1356 	 * it might not have done the ack_APIC_irq() because it thought,
1357 	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1358 	 * does not clear the ISR bit and cpu thinks it has already serivced
1359 	 * the interrupt. Hence a vector might get locked. It was noticed
1360 	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1361 	 */
1362 	do {
1363 		queued = 0;
1364 		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1365 			queued |= apic_read(APIC_IRR + i*0x10);
1366 
1367 		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1368 			value = apic_read(APIC_ISR + i*0x10);
1369 			for (j = 31; j >= 0; j--) {
1370 				if (value & (1<<j)) {
1371 					ack_APIC_irq();
1372 					acked++;
1373 				}
1374 			}
1375 		}
1376 		if (acked > 256) {
1377 			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1378 			       acked);
1379 			break;
1380 		}
1381 		if (queued) {
1382 			if (cpu_has_tsc) {
1383 				rdtscll(ntsc);
1384 				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1385 			} else
1386 				max_loops--;
1387 		}
1388 	} while (queued && max_loops > 0);
1389 	WARN_ON(max_loops <= 0);
1390 
1391 	/*
1392 	 * Now that we are all set up, enable the APIC
1393 	 */
1394 	value = apic_read(APIC_SPIV);
1395 	value &= ~APIC_VECTOR_MASK;
1396 	/*
1397 	 * Enable APIC
1398 	 */
1399 	value |= APIC_SPIV_APIC_ENABLED;
1400 
1401 #ifdef CONFIG_X86_32
1402 	/*
1403 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1404 	 * certain networking cards. If high frequency interrupts are
1405 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1406 	 * entry is masked/unmasked at a high rate as well then sooner or
1407 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1408 	 * from the device. If focus CPU is disabled then the hang goes
1409 	 * away, oh well :-(
1410 	 *
1411 	 * [ This bug can be reproduced easily with a level-triggered
1412 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1413 	 *   BX chipset. ]
1414 	 */
1415 	/*
1416 	 * Actually disabling the focus CPU check just makes the hang less
1417 	 * frequent as it makes the interrupt distributon model be more
1418 	 * like LRU than MRU (the short-term load is more even across CPUs).
1419 	 * See also the comment in end_level_ioapic_irq().  --macro
1420 	 */
1421 
1422 	/*
1423 	 * - enable focus processor (bit==0)
1424 	 * - 64bit mode always use processor focus
1425 	 *   so no need to set it
1426 	 */
1427 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1428 #endif
1429 
1430 	/*
1431 	 * Set spurious IRQ vector
1432 	 */
1433 	value |= SPURIOUS_APIC_VECTOR;
1434 	apic_write(APIC_SPIV, value);
1435 
1436 	/*
1437 	 * Set up LVT0, LVT1:
1438 	 *
1439 	 * set up through-local-APIC on the BP's LINT0. This is not
1440 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1441 	 * we delegate interrupts to the 8259A.
1442 	 */
1443 	/*
1444 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1445 	 */
1446 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1447 	if (!cpu && (pic_mode || !value)) {
1448 		value = APIC_DM_EXTINT;
1449 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1450 	} else {
1451 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1452 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1453 	}
1454 	apic_write(APIC_LVT0, value);
1455 
1456 	/*
1457 	 * only the BP should see the LINT1 NMI signal, obviously.
1458 	 */
1459 	if (!cpu)
1460 		value = APIC_DM_NMI;
1461 	else
1462 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1463 	if (!lapic_is_integrated())		/* 82489DX */
1464 		value |= APIC_LVT_LEVEL_TRIGGER;
1465 	apic_write(APIC_LVT1, value);
1466 
1467 #ifdef CONFIG_X86_MCE_INTEL
1468 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1469 	if (!cpu)
1470 		cmci_recheck();
1471 #endif
1472 }
1473 
1474 void end_local_APIC_setup(void)
1475 {
1476 	lapic_setup_esr();
1477 
1478 #ifdef CONFIG_X86_32
1479 	{
1480 		unsigned int value;
1481 		/* Disable the local apic timer */
1482 		value = apic_read(APIC_LVTT);
1483 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1484 		apic_write(APIC_LVTT, value);
1485 	}
1486 #endif
1487 
1488 	apic_pm_activate();
1489 }
1490 
1491 void __init bsp_end_local_APIC_setup(void)
1492 {
1493 	end_local_APIC_setup();
1494 
1495 	/*
1496 	 * Now that local APIC setup is completed for BP, configure the fault
1497 	 * handling for interrupt remapping.
1498 	 */
1499 	irq_remap_enable_fault_handling();
1500 
1501 }
1502 
1503 #ifdef CONFIG_X86_X2APIC
1504 /*
1505  * Need to disable xapic and x2apic at the same time and then enable xapic mode
1506  */
1507 static inline void __disable_x2apic(u64 msr)
1508 {
1509 	wrmsrl(MSR_IA32_APICBASE,
1510 	       msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1511 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1512 }
1513 
1514 static __init void disable_x2apic(void)
1515 {
1516 	u64 msr;
1517 
1518 	if (!cpu_has_x2apic)
1519 		return;
1520 
1521 	rdmsrl(MSR_IA32_APICBASE, msr);
1522 	if (msr & X2APIC_ENABLE) {
1523 		u32 x2apic_id = read_apic_id();
1524 
1525 		if (x2apic_id >= 255)
1526 			panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1527 
1528 		pr_info("Disabling x2apic\n");
1529 		__disable_x2apic(msr);
1530 
1531 		if (nox2apic) {
1532 			clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1533 			setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1534 		}
1535 
1536 		x2apic_disabled = 1;
1537 		x2apic_mode = 0;
1538 
1539 		register_lapic_address(mp_lapic_addr);
1540 	}
1541 }
1542 
1543 void check_x2apic(void)
1544 {
1545 	if (x2apic_enabled()) {
1546 		pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1547 		x2apic_preenabled = x2apic_mode = 1;
1548 	}
1549 }
1550 
1551 void enable_x2apic(void)
1552 {
1553 	u64 msr;
1554 
1555 	rdmsrl(MSR_IA32_APICBASE, msr);
1556 	if (x2apic_disabled) {
1557 		__disable_x2apic(msr);
1558 		return;
1559 	}
1560 
1561 	if (!x2apic_mode)
1562 		return;
1563 
1564 	if (!(msr & X2APIC_ENABLE)) {
1565 		printk_once(KERN_INFO "Enabling x2apic\n");
1566 		wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1567 	}
1568 }
1569 #endif /* CONFIG_X86_X2APIC */
1570 
1571 int __init enable_IR(void)
1572 {
1573 #ifdef CONFIG_IRQ_REMAP
1574 	if (!irq_remapping_supported()) {
1575 		pr_debug("intr-remapping not supported\n");
1576 		return -1;
1577 	}
1578 
1579 	if (!x2apic_preenabled && skip_ioapic_setup) {
1580 		pr_info("Skipped enabling intr-remap because of skipping "
1581 			"io-apic setup\n");
1582 		return -1;
1583 	}
1584 
1585 	return irq_remapping_enable();
1586 #endif
1587 	return -1;
1588 }
1589 
1590 void __init enable_IR_x2apic(void)
1591 {
1592 	unsigned long flags;
1593 	int ret, x2apic_enabled = 0;
1594 	int hardware_init_ret;
1595 
1596 	/* Make sure irq_remap_ops are initialized */
1597 	setup_irq_remapping_ops();
1598 
1599 	hardware_init_ret = irq_remapping_prepare();
1600 	if (hardware_init_ret && !x2apic_supported())
1601 		return;
1602 
1603 	ret = save_ioapic_entries();
1604 	if (ret) {
1605 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1606 		return;
1607 	}
1608 
1609 	local_irq_save(flags);
1610 	legacy_pic->mask_all();
1611 	mask_ioapic_entries();
1612 
1613 	if (x2apic_preenabled && nox2apic)
1614 		disable_x2apic();
1615 
1616 	if (hardware_init_ret)
1617 		ret = -1;
1618 	else
1619 		ret = enable_IR();
1620 
1621 	if (!x2apic_supported())
1622 		goto skip_x2apic;
1623 
1624 	if (ret < 0) {
1625 		/* IR is required if there is APIC ID > 255 even when running
1626 		 * under KVM
1627 		 */
1628 		if (max_physical_apicid > 255 ||
1629 		    !hypervisor_x2apic_available()) {
1630 			if (x2apic_preenabled)
1631 				disable_x2apic();
1632 			goto skip_x2apic;
1633 		}
1634 		/*
1635 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1636 		 * only in physical mode
1637 		 */
1638 		x2apic_force_phys();
1639 	}
1640 
1641 	if (ret == IRQ_REMAP_XAPIC_MODE) {
1642 		pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1643 		goto skip_x2apic;
1644 	}
1645 
1646 	x2apic_enabled = 1;
1647 
1648 	if (x2apic_supported() && !x2apic_mode) {
1649 		x2apic_mode = 1;
1650 		enable_x2apic();
1651 		pr_info("Enabled x2apic\n");
1652 	}
1653 
1654 skip_x2apic:
1655 	if (ret < 0) /* IR enabling failed */
1656 		restore_ioapic_entries();
1657 	legacy_pic->restore_mask();
1658 	local_irq_restore(flags);
1659 }
1660 
1661 #ifdef CONFIG_X86_64
1662 /*
1663  * Detect and enable local APICs on non-SMP boards.
1664  * Original code written by Keir Fraser.
1665  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1666  * not correctly set up (usually the APIC timer won't work etc.)
1667  */
1668 static int __init detect_init_APIC(void)
1669 {
1670 	if (!cpu_has_apic) {
1671 		pr_info("No local APIC present\n");
1672 		return -1;
1673 	}
1674 
1675 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1676 	return 0;
1677 }
1678 #else
1679 
1680 static int __init apic_verify(void)
1681 {
1682 	u32 features, h, l;
1683 
1684 	/*
1685 	 * The APIC feature bit should now be enabled
1686 	 * in `cpuid'
1687 	 */
1688 	features = cpuid_edx(1);
1689 	if (!(features & (1 << X86_FEATURE_APIC))) {
1690 		pr_warning("Could not enable APIC!\n");
1691 		return -1;
1692 	}
1693 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1694 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1695 
1696 	/* The BIOS may have set up the APIC at some other address */
1697 	if (boot_cpu_data.x86 >= 6) {
1698 		rdmsr(MSR_IA32_APICBASE, l, h);
1699 		if (l & MSR_IA32_APICBASE_ENABLE)
1700 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1701 	}
1702 
1703 	pr_info("Found and enabled local APIC!\n");
1704 	return 0;
1705 }
1706 
1707 int __init apic_force_enable(unsigned long addr)
1708 {
1709 	u32 h, l;
1710 
1711 	if (disable_apic)
1712 		return -1;
1713 
1714 	/*
1715 	 * Some BIOSes disable the local APIC in the APIC_BASE
1716 	 * MSR. This can only be done in software for Intel P6 or later
1717 	 * and AMD K7 (Model > 1) or later.
1718 	 */
1719 	if (boot_cpu_data.x86 >= 6) {
1720 		rdmsr(MSR_IA32_APICBASE, l, h);
1721 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1722 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1723 			l &= ~MSR_IA32_APICBASE_BASE;
1724 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1725 			wrmsr(MSR_IA32_APICBASE, l, h);
1726 			enabled_via_apicbase = 1;
1727 		}
1728 	}
1729 	return apic_verify();
1730 }
1731 
1732 /*
1733  * Detect and initialize APIC
1734  */
1735 static int __init detect_init_APIC(void)
1736 {
1737 	/* Disabled by kernel option? */
1738 	if (disable_apic)
1739 		return -1;
1740 
1741 	switch (boot_cpu_data.x86_vendor) {
1742 	case X86_VENDOR_AMD:
1743 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1744 		    (boot_cpu_data.x86 >= 15))
1745 			break;
1746 		goto no_apic;
1747 	case X86_VENDOR_INTEL:
1748 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1749 		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
1750 			break;
1751 		goto no_apic;
1752 	default:
1753 		goto no_apic;
1754 	}
1755 
1756 	if (!cpu_has_apic) {
1757 		/*
1758 		 * Over-ride BIOS and try to enable the local APIC only if
1759 		 * "lapic" specified.
1760 		 */
1761 		if (!force_enable_local_apic) {
1762 			pr_info("Local APIC disabled by BIOS -- "
1763 				"you can enable it with \"lapic\"\n");
1764 			return -1;
1765 		}
1766 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1767 			return -1;
1768 	} else {
1769 		if (apic_verify())
1770 			return -1;
1771 	}
1772 
1773 	apic_pm_activate();
1774 
1775 	return 0;
1776 
1777 no_apic:
1778 	pr_info("No local APIC present or hardware disabled\n");
1779 	return -1;
1780 }
1781 #endif
1782 
1783 /**
1784  * init_apic_mappings - initialize APIC mappings
1785  */
1786 void __init init_apic_mappings(void)
1787 {
1788 	unsigned int new_apicid;
1789 
1790 	if (x2apic_mode) {
1791 		boot_cpu_physical_apicid = read_apic_id();
1792 		return;
1793 	}
1794 
1795 	/* If no local APIC can be found return early */
1796 	if (!smp_found_config && detect_init_APIC()) {
1797 		/* lets NOP'ify apic operations */
1798 		pr_info("APIC: disable apic facility\n");
1799 		apic_disable();
1800 	} else {
1801 		apic_phys = mp_lapic_addr;
1802 
1803 		/*
1804 		 * acpi lapic path already maps that address in
1805 		 * acpi_register_lapic_address()
1806 		 */
1807 		if (!acpi_lapic && !smp_found_config)
1808 			register_lapic_address(apic_phys);
1809 	}
1810 
1811 	/*
1812 	 * Fetch the APIC ID of the BSP in case we have a
1813 	 * default configuration (or the MP table is broken).
1814 	 */
1815 	new_apicid = read_apic_id();
1816 	if (boot_cpu_physical_apicid != new_apicid) {
1817 		boot_cpu_physical_apicid = new_apicid;
1818 		/*
1819 		 * yeah -- we lie about apic_version
1820 		 * in case if apic was disabled via boot option
1821 		 * but it's not a problem for SMP compiled kernel
1822 		 * since smp_sanity_check is prepared for such a case
1823 		 * and disable smp mode
1824 		 */
1825 		apic_version[new_apicid] =
1826 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1827 	}
1828 }
1829 
1830 void __init register_lapic_address(unsigned long address)
1831 {
1832 	mp_lapic_addr = address;
1833 
1834 	if (!x2apic_mode) {
1835 		set_fixmap_nocache(FIX_APIC_BASE, address);
1836 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1837 			    APIC_BASE, mp_lapic_addr);
1838 	}
1839 	if (boot_cpu_physical_apicid == -1U) {
1840 		boot_cpu_physical_apicid  = read_apic_id();
1841 		apic_version[boot_cpu_physical_apicid] =
1842 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1843 	}
1844 }
1845 
1846 /*
1847  * This initializes the IO-APIC and APIC hardware if this is
1848  * a UP kernel.
1849  */
1850 int apic_version[MAX_LOCAL_APIC];
1851 
1852 int __init APIC_init_uniprocessor(void)
1853 {
1854 	if (disable_apic) {
1855 		pr_info("Apic disabled\n");
1856 		return -1;
1857 	}
1858 #ifdef CONFIG_X86_64
1859 	if (!cpu_has_apic) {
1860 		disable_apic = 1;
1861 		pr_info("Apic disabled by BIOS\n");
1862 		return -1;
1863 	}
1864 #else
1865 	if (!smp_found_config && !cpu_has_apic)
1866 		return -1;
1867 
1868 	/*
1869 	 * Complain if the BIOS pretends there is one.
1870 	 */
1871 	if (!cpu_has_apic &&
1872 	    APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1873 		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1874 			boot_cpu_physical_apicid);
1875 		return -1;
1876 	}
1877 #endif
1878 
1879 	default_setup_apic_routing();
1880 
1881 	verify_local_APIC();
1882 	connect_bsp_APIC();
1883 
1884 #ifdef CONFIG_X86_64
1885 	apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1886 #else
1887 	/*
1888 	 * Hack: In case of kdump, after a crash, kernel might be booting
1889 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1890 	 * might be zero if read from MP tables. Get it from LAPIC.
1891 	 */
1892 # ifdef CONFIG_CRASH_DUMP
1893 	boot_cpu_physical_apicid = read_apic_id();
1894 # endif
1895 #endif
1896 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1897 	setup_local_APIC();
1898 
1899 #ifdef CONFIG_X86_IO_APIC
1900 	/*
1901 	 * Now enable IO-APICs, actually call clear_IO_APIC
1902 	 * We need clear_IO_APIC before enabling error vector
1903 	 */
1904 	if (!skip_ioapic_setup && nr_ioapics)
1905 		enable_IO_APIC();
1906 #endif
1907 
1908 	bsp_end_local_APIC_setup();
1909 
1910 #ifdef CONFIG_X86_IO_APIC
1911 	if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1912 		setup_IO_APIC();
1913 	else {
1914 		nr_ioapics = 0;
1915 	}
1916 #endif
1917 
1918 	x86_init.timers.setup_percpu_clockev();
1919 	return 0;
1920 }
1921 
1922 /*
1923  * Local APIC interrupts
1924  */
1925 
1926 /*
1927  * This interrupt should _never_ happen with our APIC/SMP architecture
1928  */
1929 static inline void __smp_spurious_interrupt(void)
1930 {
1931 	u32 v;
1932 
1933 	/*
1934 	 * Check if this really is a spurious interrupt and ACK it
1935 	 * if it is a vectored one.  Just in case...
1936 	 * Spurious interrupts should not be ACKed.
1937 	 */
1938 	v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1939 	if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1940 		ack_APIC_irq();
1941 
1942 	inc_irq_stat(irq_spurious_count);
1943 
1944 	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1945 	pr_info("spurious APIC interrupt on CPU#%d, "
1946 		"should never happen.\n", smp_processor_id());
1947 }
1948 
1949 __visible void smp_spurious_interrupt(struct pt_regs *regs)
1950 {
1951 	entering_irq();
1952 	__smp_spurious_interrupt();
1953 	exiting_irq();
1954 }
1955 
1956 __visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1957 {
1958 	entering_irq();
1959 	trace_spurious_apic_entry(SPURIOUS_APIC_VECTOR);
1960 	__smp_spurious_interrupt();
1961 	trace_spurious_apic_exit(SPURIOUS_APIC_VECTOR);
1962 	exiting_irq();
1963 }
1964 
1965 /*
1966  * This interrupt should never happen with our APIC/SMP architecture
1967  */
1968 static inline void __smp_error_interrupt(struct pt_regs *regs)
1969 {
1970 	u32 v0, v1;
1971 	u32 i = 0;
1972 	static const char * const error_interrupt_reason[] = {
1973 		"Send CS error",		/* APIC Error Bit 0 */
1974 		"Receive CS error",		/* APIC Error Bit 1 */
1975 		"Send accept error",		/* APIC Error Bit 2 */
1976 		"Receive accept error",		/* APIC Error Bit 3 */
1977 		"Redirectable IPI",		/* APIC Error Bit 4 */
1978 		"Send illegal vector",		/* APIC Error Bit 5 */
1979 		"Received illegal vector",	/* APIC Error Bit 6 */
1980 		"Illegal register address",	/* APIC Error Bit 7 */
1981 	};
1982 
1983 	/* First tickle the hardware, only then report what went on. -- REW */
1984 	v0 = apic_read(APIC_ESR);
1985 	apic_write(APIC_ESR, 0);
1986 	v1 = apic_read(APIC_ESR);
1987 	ack_APIC_irq();
1988 	atomic_inc(&irq_err_count);
1989 
1990 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1991 		    smp_processor_id(), v0 , v1);
1992 
1993 	v1 = v1 & 0xff;
1994 	while (v1) {
1995 		if (v1 & 0x1)
1996 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1997 		i++;
1998 		v1 >>= 1;
1999 	}
2000 
2001 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2002 
2003 }
2004 
2005 __visible void smp_error_interrupt(struct pt_regs *regs)
2006 {
2007 	entering_irq();
2008 	__smp_error_interrupt(regs);
2009 	exiting_irq();
2010 }
2011 
2012 __visible void smp_trace_error_interrupt(struct pt_regs *regs)
2013 {
2014 	entering_irq();
2015 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2016 	__smp_error_interrupt(regs);
2017 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2018 	exiting_irq();
2019 }
2020 
2021 /**
2022  * connect_bsp_APIC - attach the APIC to the interrupt system
2023  */
2024 void __init connect_bsp_APIC(void)
2025 {
2026 #ifdef CONFIG_X86_32
2027 	if (pic_mode) {
2028 		/*
2029 		 * Do not trust the local APIC being empty at bootup.
2030 		 */
2031 		clear_local_APIC();
2032 		/*
2033 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2034 		 * local APIC to INT and NMI lines.
2035 		 */
2036 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2037 				"enabling APIC mode.\n");
2038 		imcr_pic_to_apic();
2039 	}
2040 #endif
2041 	if (apic->enable_apic_mode)
2042 		apic->enable_apic_mode();
2043 }
2044 
2045 /**
2046  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2047  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2048  *
2049  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2050  * APIC is disabled.
2051  */
2052 void disconnect_bsp_APIC(int virt_wire_setup)
2053 {
2054 	unsigned int value;
2055 
2056 #ifdef CONFIG_X86_32
2057 	if (pic_mode) {
2058 		/*
2059 		 * Put the board back into PIC mode (has an effect only on
2060 		 * certain older boards).  Note that APIC interrupts, including
2061 		 * IPIs, won't work beyond this point!  The only exception are
2062 		 * INIT IPIs.
2063 		 */
2064 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2065 				"entering PIC mode.\n");
2066 		imcr_apic_to_pic();
2067 		return;
2068 	}
2069 #endif
2070 
2071 	/* Go back to Virtual Wire compatibility mode */
2072 
2073 	/* For the spurious interrupt use vector F, and enable it */
2074 	value = apic_read(APIC_SPIV);
2075 	value &= ~APIC_VECTOR_MASK;
2076 	value |= APIC_SPIV_APIC_ENABLED;
2077 	value |= 0xf;
2078 	apic_write(APIC_SPIV, value);
2079 
2080 	if (!virt_wire_setup) {
2081 		/*
2082 		 * For LVT0 make it edge triggered, active high,
2083 		 * external and enabled
2084 		 */
2085 		value = apic_read(APIC_LVT0);
2086 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2087 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2088 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2089 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2090 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2091 		apic_write(APIC_LVT0, value);
2092 	} else {
2093 		/* Disable LVT0 */
2094 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2095 	}
2096 
2097 	/*
2098 	 * For LVT1 make it edge triggered, active high,
2099 	 * nmi and enabled
2100 	 */
2101 	value = apic_read(APIC_LVT1);
2102 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2103 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2104 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2105 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2106 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2107 	apic_write(APIC_LVT1, value);
2108 }
2109 
2110 void generic_processor_info(int apicid, int version)
2111 {
2112 	int cpu, max = nr_cpu_ids;
2113 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2114 				phys_cpu_present_map);
2115 
2116 	/*
2117 	 * If boot cpu has not been detected yet, then only allow upto
2118 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2119 	 */
2120 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2121 	    apicid != boot_cpu_physical_apicid) {
2122 		int thiscpu = max + disabled_cpus - 1;
2123 
2124 		pr_warning(
2125 			"ACPI: NR_CPUS/possible_cpus limit of %i almost"
2126 			" reached. Keeping one slot for boot cpu."
2127 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2128 
2129 		disabled_cpus++;
2130 		return;
2131 	}
2132 
2133 	if (num_processors >= nr_cpu_ids) {
2134 		int thiscpu = max + disabled_cpus;
2135 
2136 		pr_warning(
2137 			"ACPI: NR_CPUS/possible_cpus limit of %i reached."
2138 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2139 
2140 		disabled_cpus++;
2141 		return;
2142 	}
2143 
2144 	num_processors++;
2145 	if (apicid == boot_cpu_physical_apicid) {
2146 		/*
2147 		 * x86_bios_cpu_apicid is required to have processors listed
2148 		 * in same order as logical cpu numbers. Hence the first
2149 		 * entry is BSP, and so on.
2150 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2151 		 * for BSP.
2152 		 */
2153 		cpu = 0;
2154 	} else
2155 		cpu = cpumask_next_zero(-1, cpu_present_mask);
2156 
2157 	/*
2158 	 * Validate version
2159 	 */
2160 	if (version == 0x0) {
2161 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2162 			   cpu, apicid);
2163 		version = 0x10;
2164 	}
2165 	apic_version[apicid] = version;
2166 
2167 	if (version != apic_version[boot_cpu_physical_apicid]) {
2168 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2169 			apic_version[boot_cpu_physical_apicid], cpu, version);
2170 	}
2171 
2172 	physid_set(apicid, phys_cpu_present_map);
2173 	if (apicid > max_physical_apicid)
2174 		max_physical_apicid = apicid;
2175 
2176 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2177 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2178 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2179 #endif
2180 #ifdef CONFIG_X86_32
2181 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2182 		apic->x86_32_early_logical_apicid(cpu);
2183 #endif
2184 	set_cpu_possible(cpu, true);
2185 	set_cpu_present(cpu, true);
2186 }
2187 
2188 int hard_smp_processor_id(void)
2189 {
2190 	return read_apic_id();
2191 }
2192 
2193 void default_init_apic_ldr(void)
2194 {
2195 	unsigned long val;
2196 
2197 	apic_write(APIC_DFR, APIC_DFR_VALUE);
2198 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2199 	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2200 	apic_write(APIC_LDR, val);
2201 }
2202 
2203 int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2204 				   const struct cpumask *andmask,
2205 				   unsigned int *apicid)
2206 {
2207 	unsigned int cpu;
2208 
2209 	for_each_cpu_and(cpu, cpumask, andmask) {
2210 		if (cpumask_test_cpu(cpu, cpu_online_mask))
2211 			break;
2212 	}
2213 
2214 	if (likely(cpu < nr_cpu_ids)) {
2215 		*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2216 		return 0;
2217 	}
2218 
2219 	return -EINVAL;
2220 }
2221 
2222 /*
2223  * Override the generic EOI implementation with an optimized version.
2224  * Only called during early boot when only one CPU is active and with
2225  * interrupts disabled, so we know this does not race with actual APIC driver
2226  * use.
2227  */
2228 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2229 {
2230 	struct apic **drv;
2231 
2232 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2233 		/* Should happen once for each apic */
2234 		WARN_ON((*drv)->eoi_write == eoi_write);
2235 		(*drv)->eoi_write = eoi_write;
2236 	}
2237 }
2238 
2239 /*
2240  * Power management
2241  */
2242 #ifdef CONFIG_PM
2243 
2244 static struct {
2245 	/*
2246 	 * 'active' is true if the local APIC was enabled by us and
2247 	 * not the BIOS; this signifies that we are also responsible
2248 	 * for disabling it before entering apm/acpi suspend
2249 	 */
2250 	int active;
2251 	/* r/w apic fields */
2252 	unsigned int apic_id;
2253 	unsigned int apic_taskpri;
2254 	unsigned int apic_ldr;
2255 	unsigned int apic_dfr;
2256 	unsigned int apic_spiv;
2257 	unsigned int apic_lvtt;
2258 	unsigned int apic_lvtpc;
2259 	unsigned int apic_lvt0;
2260 	unsigned int apic_lvt1;
2261 	unsigned int apic_lvterr;
2262 	unsigned int apic_tmict;
2263 	unsigned int apic_tdcr;
2264 	unsigned int apic_thmr;
2265 } apic_pm_state;
2266 
2267 static int lapic_suspend(void)
2268 {
2269 	unsigned long flags;
2270 	int maxlvt;
2271 
2272 	if (!apic_pm_state.active)
2273 		return 0;
2274 
2275 	maxlvt = lapic_get_maxlvt();
2276 
2277 	apic_pm_state.apic_id = apic_read(APIC_ID);
2278 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2279 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2280 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2281 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2282 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2283 	if (maxlvt >= 4)
2284 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2285 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2286 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2287 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2288 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2289 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2290 #ifdef CONFIG_X86_THERMAL_VECTOR
2291 	if (maxlvt >= 5)
2292 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2293 #endif
2294 
2295 	local_irq_save(flags);
2296 	disable_local_APIC();
2297 
2298 	irq_remapping_disable();
2299 
2300 	local_irq_restore(flags);
2301 	return 0;
2302 }
2303 
2304 static void lapic_resume(void)
2305 {
2306 	unsigned int l, h;
2307 	unsigned long flags;
2308 	int maxlvt;
2309 
2310 	if (!apic_pm_state.active)
2311 		return;
2312 
2313 	local_irq_save(flags);
2314 
2315 	/*
2316 	 * IO-APIC and PIC have their own resume routines.
2317 	 * We just mask them here to make sure the interrupt
2318 	 * subsystem is completely quiet while we enable x2apic
2319 	 * and interrupt-remapping.
2320 	 */
2321 	mask_ioapic_entries();
2322 	legacy_pic->mask_all();
2323 
2324 	if (x2apic_mode)
2325 		enable_x2apic();
2326 	else {
2327 		/*
2328 		 * Make sure the APICBASE points to the right address
2329 		 *
2330 		 * FIXME! This will be wrong if we ever support suspend on
2331 		 * SMP! We'll need to do this as part of the CPU restore!
2332 		 */
2333 		if (boot_cpu_data.x86 >= 6) {
2334 			rdmsr(MSR_IA32_APICBASE, l, h);
2335 			l &= ~MSR_IA32_APICBASE_BASE;
2336 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2337 			wrmsr(MSR_IA32_APICBASE, l, h);
2338 		}
2339 	}
2340 
2341 	maxlvt = lapic_get_maxlvt();
2342 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2343 	apic_write(APIC_ID, apic_pm_state.apic_id);
2344 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2345 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2346 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2347 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2348 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2349 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2350 #if defined(CONFIG_X86_MCE_INTEL)
2351 	if (maxlvt >= 5)
2352 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2353 #endif
2354 	if (maxlvt >= 4)
2355 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2356 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2357 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2358 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2359 	apic_write(APIC_ESR, 0);
2360 	apic_read(APIC_ESR);
2361 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2362 	apic_write(APIC_ESR, 0);
2363 	apic_read(APIC_ESR);
2364 
2365 	irq_remapping_reenable(x2apic_mode);
2366 
2367 	local_irq_restore(flags);
2368 }
2369 
2370 /*
2371  * This device has no shutdown method - fully functioning local APICs
2372  * are needed on every CPU up until machine_halt/restart/poweroff.
2373  */
2374 
2375 static struct syscore_ops lapic_syscore_ops = {
2376 	.resume		= lapic_resume,
2377 	.suspend	= lapic_suspend,
2378 };
2379 
2380 static void apic_pm_activate(void)
2381 {
2382 	apic_pm_state.active = 1;
2383 }
2384 
2385 static int __init init_lapic_sysfs(void)
2386 {
2387 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2388 	if (cpu_has_apic)
2389 		register_syscore_ops(&lapic_syscore_ops);
2390 
2391 	return 0;
2392 }
2393 
2394 /* local apic needs to resume before other devices access its registers. */
2395 core_initcall(init_lapic_sysfs);
2396 
2397 #else	/* CONFIG_PM */
2398 
2399 static void apic_pm_activate(void) { }
2400 
2401 #endif	/* CONFIG_PM */
2402 
2403 #ifdef CONFIG_X86_64
2404 
2405 static int apic_cluster_num(void)
2406 {
2407 	int i, clusters, zeros;
2408 	unsigned id;
2409 	u16 *bios_cpu_apicid;
2410 	DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2411 
2412 	bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2413 	bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2414 
2415 	for (i = 0; i < nr_cpu_ids; i++) {
2416 		/* are we being called early in kernel startup? */
2417 		if (bios_cpu_apicid) {
2418 			id = bios_cpu_apicid[i];
2419 		} else if (i < nr_cpu_ids) {
2420 			if (cpu_present(i))
2421 				id = per_cpu(x86_bios_cpu_apicid, i);
2422 			else
2423 				continue;
2424 		} else
2425 			break;
2426 
2427 		if (id != BAD_APICID)
2428 			__set_bit(APIC_CLUSTERID(id), clustermap);
2429 	}
2430 
2431 	/* Problem:  Partially populated chassis may not have CPUs in some of
2432 	 * the APIC clusters they have been allocated.  Only present CPUs have
2433 	 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2434 	 * Since clusters are allocated sequentially, count zeros only if
2435 	 * they are bounded by ones.
2436 	 */
2437 	clusters = 0;
2438 	zeros = 0;
2439 	for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2440 		if (test_bit(i, clustermap)) {
2441 			clusters += 1 + zeros;
2442 			zeros = 0;
2443 		} else
2444 			++zeros;
2445 	}
2446 
2447 	return clusters;
2448 }
2449 
2450 static int multi_checked;
2451 static int multi;
2452 
2453 static int set_multi(const struct dmi_system_id *d)
2454 {
2455 	if (multi)
2456 		return 0;
2457 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2458 	multi = 1;
2459 	return 0;
2460 }
2461 
2462 static const struct dmi_system_id multi_dmi_table[] = {
2463 	{
2464 		.callback = set_multi,
2465 		.ident = "IBM System Summit2",
2466 		.matches = {
2467 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2468 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2469 		},
2470 	},
2471 	{}
2472 };
2473 
2474 static void dmi_check_multi(void)
2475 {
2476 	if (multi_checked)
2477 		return;
2478 
2479 	dmi_check_system(multi_dmi_table);
2480 	multi_checked = 1;
2481 }
2482 
2483 /*
2484  * apic_is_clustered_box() -- Check if we can expect good TSC
2485  *
2486  * Thus far, the major user of this is IBM's Summit2 series:
2487  * Clustered boxes may have unsynced TSC problems if they are
2488  * multi-chassis.
2489  * Use DMI to check them
2490  */
2491 int apic_is_clustered_box(void)
2492 {
2493 	dmi_check_multi();
2494 	if (multi)
2495 		return 1;
2496 
2497 	if (!is_vsmp_box())
2498 		return 0;
2499 
2500 	/*
2501 	 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2502 	 * not guaranteed to be synced between boards
2503 	 */
2504 	if (apic_cluster_num() > 1)
2505 		return 1;
2506 
2507 	return 0;
2508 }
2509 #endif
2510 
2511 /*
2512  * APIC command line parameters
2513  */
2514 static int __init setup_disableapic(char *arg)
2515 {
2516 	disable_apic = 1;
2517 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2518 	return 0;
2519 }
2520 early_param("disableapic", setup_disableapic);
2521 
2522 /* same as disableapic, for compatibility */
2523 static int __init setup_nolapic(char *arg)
2524 {
2525 	return setup_disableapic(arg);
2526 }
2527 early_param("nolapic", setup_nolapic);
2528 
2529 static int __init parse_lapic_timer_c2_ok(char *arg)
2530 {
2531 	local_apic_timer_c2_ok = 1;
2532 	return 0;
2533 }
2534 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2535 
2536 static int __init parse_disable_apic_timer(char *arg)
2537 {
2538 	disable_apic_timer = 1;
2539 	return 0;
2540 }
2541 early_param("noapictimer", parse_disable_apic_timer);
2542 
2543 static int __init parse_nolapic_timer(char *arg)
2544 {
2545 	disable_apic_timer = 1;
2546 	return 0;
2547 }
2548 early_param("nolapic_timer", parse_nolapic_timer);
2549 
2550 static int __init apic_set_verbosity(char *arg)
2551 {
2552 	if (!arg)  {
2553 #ifdef CONFIG_X86_64
2554 		skip_ioapic_setup = 0;
2555 		return 0;
2556 #endif
2557 		return -EINVAL;
2558 	}
2559 
2560 	if (strcmp("debug", arg) == 0)
2561 		apic_verbosity = APIC_DEBUG;
2562 	else if (strcmp("verbose", arg) == 0)
2563 		apic_verbosity = APIC_VERBOSE;
2564 	else {
2565 		pr_warning("APIC Verbosity level %s not recognised"
2566 			" use apic=verbose or apic=debug\n", arg);
2567 		return -EINVAL;
2568 	}
2569 
2570 	return 0;
2571 }
2572 early_param("apic", apic_set_verbosity);
2573 
2574 static int __init lapic_insert_resource(void)
2575 {
2576 	if (!apic_phys)
2577 		return -1;
2578 
2579 	/* Put local APIC into the resource map. */
2580 	lapic_resource.start = apic_phys;
2581 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2582 	insert_resource(&iomem_resource, &lapic_resource);
2583 
2584 	return 0;
2585 }
2586 
2587 /*
2588  * need call insert after e820_reserve_resources()
2589  * that is using request_resource
2590  */
2591 late_initcall(lapic_insert_resource);
2592