xref: /linux/arch/x86/kernel/apic/apic.c (revision 507e190946297c34a27d9366b0661d5e506fdd03)
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/export.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/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 #include <asm/cpu_device_id.h>
58 #include <asm/intel-family.h>
59 
60 unsigned int num_processors;
61 
62 unsigned disabled_cpus;
63 
64 /* Processor that is doing the boot up */
65 unsigned int boot_cpu_physical_apicid = -1U;
66 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
67 
68 u8 boot_cpu_apic_version;
69 
70 /*
71  * The highest APIC ID seen during enumeration.
72  */
73 static unsigned int max_physical_apicid;
74 
75 /*
76  * Bitmask of physically existing CPUs:
77  */
78 physid_mask_t phys_cpu_present_map;
79 
80 /*
81  * Processor to be disabled specified by kernel parameter
82  * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
83  * avoid undefined behaviour caused by sending INIT from AP to BSP.
84  */
85 static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
86 
87 /*
88  * This variable controls which CPUs receive external NMIs.  By default,
89  * external NMIs are delivered only to the BSP.
90  */
91 static int apic_extnmi = APIC_EXTNMI_BSP;
92 
93 /*
94  * Map cpu index to physical APIC ID
95  */
96 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
97 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
98 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
99 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
100 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
101 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
102 
103 #ifdef CONFIG_X86_32
104 
105 /*
106  * On x86_32, the mapping between cpu and logical apicid may vary
107  * depending on apic in use.  The following early percpu variable is
108  * used for the mapping.  This is where the behaviors of x86_64 and 32
109  * actually diverge.  Let's keep it ugly for now.
110  */
111 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
112 
113 /* Local APIC was disabled by the BIOS and enabled by the kernel */
114 static int enabled_via_apicbase;
115 
116 /*
117  * Handle interrupt mode configuration register (IMCR).
118  * This register controls whether the interrupt signals
119  * that reach the BSP come from the master PIC or from the
120  * local APIC. Before entering Symmetric I/O Mode, either
121  * the BIOS or the operating system must switch out of
122  * PIC Mode by changing the IMCR.
123  */
124 static inline void imcr_pic_to_apic(void)
125 {
126 	/* select IMCR register */
127 	outb(0x70, 0x22);
128 	/* NMI and 8259 INTR go through APIC */
129 	outb(0x01, 0x23);
130 }
131 
132 static inline void imcr_apic_to_pic(void)
133 {
134 	/* select IMCR register */
135 	outb(0x70, 0x22);
136 	/* NMI and 8259 INTR go directly to BSP */
137 	outb(0x00, 0x23);
138 }
139 #endif
140 
141 /*
142  * Knob to control our willingness to enable the local APIC.
143  *
144  * +1=force-enable
145  */
146 static int force_enable_local_apic __initdata;
147 
148 /*
149  * APIC command line parameters
150  */
151 static int __init parse_lapic(char *arg)
152 {
153 	if (IS_ENABLED(CONFIG_X86_32) && !arg)
154 		force_enable_local_apic = 1;
155 	else if (arg && !strncmp(arg, "notscdeadline", 13))
156 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
157 	return 0;
158 }
159 early_param("lapic", parse_lapic);
160 
161 #ifdef CONFIG_X86_64
162 static int apic_calibrate_pmtmr __initdata;
163 static __init int setup_apicpmtimer(char *s)
164 {
165 	apic_calibrate_pmtmr = 1;
166 	notsc_setup(NULL);
167 	return 0;
168 }
169 __setup("apicpmtimer", setup_apicpmtimer);
170 #endif
171 
172 unsigned long mp_lapic_addr;
173 int disable_apic;
174 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
175 static int disable_apic_timer __initdata;
176 /* Local APIC timer works in C2 */
177 int local_apic_timer_c2_ok;
178 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
179 
180 int first_system_vector = FIRST_SYSTEM_VECTOR;
181 
182 /*
183  * Debug level, exported for io_apic.c
184  */
185 unsigned int apic_verbosity;
186 
187 int pic_mode;
188 
189 /* Have we found an MP table */
190 int smp_found_config;
191 
192 static struct resource lapic_resource = {
193 	.name = "Local APIC",
194 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
195 };
196 
197 unsigned int lapic_timer_frequency = 0;
198 
199 static void apic_pm_activate(void);
200 
201 static unsigned long apic_phys;
202 
203 /*
204  * Get the LAPIC version
205  */
206 static inline int lapic_get_version(void)
207 {
208 	return GET_APIC_VERSION(apic_read(APIC_LVR));
209 }
210 
211 /*
212  * Check, if the APIC is integrated or a separate chip
213  */
214 static inline int lapic_is_integrated(void)
215 {
216 #ifdef CONFIG_X86_64
217 	return 1;
218 #else
219 	return APIC_INTEGRATED(lapic_get_version());
220 #endif
221 }
222 
223 /*
224  * Check, whether this is a modern or a first generation APIC
225  */
226 static int modern_apic(void)
227 {
228 	/* AMD systems use old APIC versions, so check the CPU */
229 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
230 	    boot_cpu_data.x86 >= 0xf)
231 		return 1;
232 	return lapic_get_version() >= 0x14;
233 }
234 
235 /*
236  * right after this call apic become NOOP driven
237  * so apic->write/read doesn't do anything
238  */
239 static void __init apic_disable(void)
240 {
241 	pr_info("APIC: switched to apic NOOP\n");
242 	apic = &apic_noop;
243 }
244 
245 void native_apic_wait_icr_idle(void)
246 {
247 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
248 		cpu_relax();
249 }
250 
251 u32 native_safe_apic_wait_icr_idle(void)
252 {
253 	u32 send_status;
254 	int timeout;
255 
256 	timeout = 0;
257 	do {
258 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
259 		if (!send_status)
260 			break;
261 		inc_irq_stat(icr_read_retry_count);
262 		udelay(100);
263 	} while (timeout++ < 1000);
264 
265 	return send_status;
266 }
267 
268 void native_apic_icr_write(u32 low, u32 id)
269 {
270 	unsigned long flags;
271 
272 	local_irq_save(flags);
273 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
274 	apic_write(APIC_ICR, low);
275 	local_irq_restore(flags);
276 }
277 
278 u64 native_apic_icr_read(void)
279 {
280 	u32 icr1, icr2;
281 
282 	icr2 = apic_read(APIC_ICR2);
283 	icr1 = apic_read(APIC_ICR);
284 
285 	return icr1 | ((u64)icr2 << 32);
286 }
287 
288 #ifdef CONFIG_X86_32
289 /**
290  * get_physical_broadcast - Get number of physical broadcast IDs
291  */
292 int get_physical_broadcast(void)
293 {
294 	return modern_apic() ? 0xff : 0xf;
295 }
296 #endif
297 
298 /**
299  * lapic_get_maxlvt - get the maximum number of local vector table entries
300  */
301 int lapic_get_maxlvt(void)
302 {
303 	unsigned int v;
304 
305 	v = apic_read(APIC_LVR);
306 	/*
307 	 * - we always have APIC integrated on 64bit mode
308 	 * - 82489DXs do not report # of LVT entries
309 	 */
310 	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
311 }
312 
313 /*
314  * Local APIC timer
315  */
316 
317 /* Clock divisor */
318 #define APIC_DIVISOR 16
319 #define TSC_DIVISOR  8
320 
321 /*
322  * This function sets up the local APIC timer, with a timeout of
323  * 'clocks' APIC bus clock. During calibration we actually call
324  * this function twice on the boot CPU, once with a bogus timeout
325  * value, second time for real. The other (noncalibrating) CPUs
326  * call this function only once, with the real, calibrated value.
327  *
328  * We do reads before writes even if unnecessary, to get around the
329  * P5 APIC double write bug.
330  */
331 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
332 {
333 	unsigned int lvtt_value, tmp_value;
334 
335 	lvtt_value = LOCAL_TIMER_VECTOR;
336 	if (!oneshot)
337 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
338 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
339 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
340 
341 	if (!lapic_is_integrated())
342 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
343 
344 	if (!irqen)
345 		lvtt_value |= APIC_LVT_MASKED;
346 
347 	apic_write(APIC_LVTT, lvtt_value);
348 
349 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
350 		/*
351 		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
352 		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
353 		 * According to Intel, MFENCE can do the serialization here.
354 		 */
355 		asm volatile("mfence" : : : "memory");
356 
357 		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
358 		return;
359 	}
360 
361 	/*
362 	 * Divide PICLK by 16
363 	 */
364 	tmp_value = apic_read(APIC_TDCR);
365 	apic_write(APIC_TDCR,
366 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
367 		APIC_TDR_DIV_16);
368 
369 	if (!oneshot)
370 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
371 }
372 
373 /*
374  * Setup extended LVT, AMD specific
375  *
376  * Software should use the LVT offsets the BIOS provides.  The offsets
377  * are determined by the subsystems using it like those for MCE
378  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
379  * are supported. Beginning with family 10h at least 4 offsets are
380  * available.
381  *
382  * Since the offsets must be consistent for all cores, we keep track
383  * of the LVT offsets in software and reserve the offset for the same
384  * vector also to be used on other cores. An offset is freed by
385  * setting the entry to APIC_EILVT_MASKED.
386  *
387  * If the BIOS is right, there should be no conflicts. Otherwise a
388  * "[Firmware Bug]: ..." error message is generated. However, if
389  * software does not properly determines the offsets, it is not
390  * necessarily a BIOS bug.
391  */
392 
393 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
394 
395 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
396 {
397 	return (old & APIC_EILVT_MASKED)
398 		|| (new == APIC_EILVT_MASKED)
399 		|| ((new & ~APIC_EILVT_MASKED) == old);
400 }
401 
402 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
403 {
404 	unsigned int rsvd, vector;
405 
406 	if (offset >= APIC_EILVT_NR_MAX)
407 		return ~0;
408 
409 	rsvd = atomic_read(&eilvt_offsets[offset]);
410 	do {
411 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
412 		if (vector && !eilvt_entry_is_changeable(vector, new))
413 			/* may not change if vectors are different */
414 			return rsvd;
415 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
416 	} while (rsvd != new);
417 
418 	rsvd &= ~APIC_EILVT_MASKED;
419 	if (rsvd && rsvd != vector)
420 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
421 			offset, rsvd);
422 
423 	return new;
424 }
425 
426 /*
427  * If mask=1, the LVT entry does not generate interrupts while mask=0
428  * enables the vector. See also the BKDGs. Must be called with
429  * preemption disabled.
430  */
431 
432 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
433 {
434 	unsigned long reg = APIC_EILVTn(offset);
435 	unsigned int new, old, reserved;
436 
437 	new = (mask << 16) | (msg_type << 8) | vector;
438 	old = apic_read(reg);
439 	reserved = reserve_eilvt_offset(offset, new);
440 
441 	if (reserved != new) {
442 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
443 		       "vector 0x%x, but the register is already in use for "
444 		       "vector 0x%x on another cpu\n",
445 		       smp_processor_id(), reg, offset, new, reserved);
446 		return -EINVAL;
447 	}
448 
449 	if (!eilvt_entry_is_changeable(old, new)) {
450 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
451 		       "vector 0x%x, but the register is already in use for "
452 		       "vector 0x%x on this cpu\n",
453 		       smp_processor_id(), reg, offset, new, old);
454 		return -EBUSY;
455 	}
456 
457 	apic_write(reg, new);
458 
459 	return 0;
460 }
461 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
462 
463 /*
464  * Program the next event, relative to now
465  */
466 static int lapic_next_event(unsigned long delta,
467 			    struct clock_event_device *evt)
468 {
469 	apic_write(APIC_TMICT, delta);
470 	return 0;
471 }
472 
473 static int lapic_next_deadline(unsigned long delta,
474 			       struct clock_event_device *evt)
475 {
476 	u64 tsc;
477 
478 	tsc = rdtsc();
479 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
480 	return 0;
481 }
482 
483 static int lapic_timer_shutdown(struct clock_event_device *evt)
484 {
485 	unsigned int v;
486 
487 	/* Lapic used as dummy for broadcast ? */
488 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
489 		return 0;
490 
491 	v = apic_read(APIC_LVTT);
492 	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
493 	apic_write(APIC_LVTT, v);
494 	apic_write(APIC_TMICT, 0);
495 	return 0;
496 }
497 
498 static inline int
499 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
500 {
501 	/* Lapic used as dummy for broadcast ? */
502 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
503 		return 0;
504 
505 	__setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
506 	return 0;
507 }
508 
509 static int lapic_timer_set_periodic(struct clock_event_device *evt)
510 {
511 	return lapic_timer_set_periodic_oneshot(evt, false);
512 }
513 
514 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
515 {
516 	return lapic_timer_set_periodic_oneshot(evt, true);
517 }
518 
519 /*
520  * Local APIC timer broadcast function
521  */
522 static void lapic_timer_broadcast(const struct cpumask *mask)
523 {
524 #ifdef CONFIG_SMP
525 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
526 #endif
527 }
528 
529 
530 /*
531  * The local apic timer can be used for any function which is CPU local.
532  */
533 static struct clock_event_device lapic_clockevent = {
534 	.name				= "lapic",
535 	.features			= CLOCK_EVT_FEAT_PERIODIC |
536 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
537 					  | CLOCK_EVT_FEAT_DUMMY,
538 	.shift				= 32,
539 	.set_state_shutdown		= lapic_timer_shutdown,
540 	.set_state_periodic		= lapic_timer_set_periodic,
541 	.set_state_oneshot		= lapic_timer_set_oneshot,
542 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
543 	.set_next_event			= lapic_next_event,
544 	.broadcast			= lapic_timer_broadcast,
545 	.rating				= 100,
546 	.irq				= -1,
547 };
548 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
549 
550 #define DEADLINE_MODEL_MATCH_FUNC(model, func)	\
551 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
552 
553 #define DEADLINE_MODEL_MATCH_REV(model, rev)	\
554 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
555 
556 static u32 hsx_deadline_rev(void)
557 {
558 	switch (boot_cpu_data.x86_mask) {
559 	case 0x02: return 0x3a; /* EP */
560 	case 0x04: return 0x0f; /* EX */
561 	}
562 
563 	return ~0U;
564 }
565 
566 static u32 bdx_deadline_rev(void)
567 {
568 	switch (boot_cpu_data.x86_mask) {
569 	case 0x02: return 0x00000011;
570 	case 0x03: return 0x0700000e;
571 	case 0x04: return 0x0f00000c;
572 	case 0x05: return 0x0e000003;
573 	}
574 
575 	return ~0U;
576 }
577 
578 static const struct x86_cpu_id deadline_match[] = {
579 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X,	hsx_deadline_rev),
580 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X,	0x0b000020),
581 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_XEON_D,	bdx_deadline_rev),
582 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_X,	0x02000014),
583 
584 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE,	0x22),
585 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_ULT,	0x20),
586 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_GT3E,	0x17),
587 
588 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_CORE,	0x25),
589 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_GT3E,	0x17),
590 
591 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE,	0xb2),
592 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP,	0xb2),
593 
594 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE,	0x52),
595 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP,	0x52),
596 
597 	{},
598 };
599 
600 static void apic_check_deadline_errata(void)
601 {
602 	const struct x86_cpu_id *m = x86_match_cpu(deadline_match);
603 	u32 rev;
604 
605 	if (!m)
606 		return;
607 
608 	/*
609 	 * Function pointers will have the MSB set due to address layout,
610 	 * immediate revisions will not.
611 	 */
612 	if ((long)m->driver_data < 0)
613 		rev = ((u32 (*)(void))(m->driver_data))();
614 	else
615 		rev = (u32)m->driver_data;
616 
617 	if (boot_cpu_data.microcode >= rev)
618 		return;
619 
620 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
621 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
622 	       "please update microcode to version: 0x%x (or later)\n", rev);
623 }
624 
625 /*
626  * Setup the local APIC timer for this CPU. Copy the initialized values
627  * of the boot CPU and register the clock event in the framework.
628  */
629 static void setup_APIC_timer(void)
630 {
631 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
632 
633 	if (this_cpu_has(X86_FEATURE_ARAT)) {
634 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
635 		/* Make LAPIC timer preferrable over percpu HPET */
636 		lapic_clockevent.rating = 150;
637 	}
638 
639 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
640 	levt->cpumask = cpumask_of(smp_processor_id());
641 
642 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
643 		levt->name = "lapic-deadline";
644 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
645 				    CLOCK_EVT_FEAT_DUMMY);
646 		levt->set_next_event = lapic_next_deadline;
647 		clockevents_config_and_register(levt,
648 						tsc_khz * (1000 / TSC_DIVISOR),
649 						0xF, ~0UL);
650 	} else
651 		clockevents_register_device(levt);
652 }
653 
654 /*
655  * Install the updated TSC frequency from recalibration at the TSC
656  * deadline clockevent devices.
657  */
658 static void __lapic_update_tsc_freq(void *info)
659 {
660 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
661 
662 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
663 		return;
664 
665 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
666 }
667 
668 void lapic_update_tsc_freq(void)
669 {
670 	/*
671 	 * The clockevent device's ->mult and ->shift can both be
672 	 * changed. In order to avoid races, schedule the frequency
673 	 * update code on each CPU.
674 	 */
675 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
676 }
677 
678 /*
679  * In this functions we calibrate APIC bus clocks to the external timer.
680  *
681  * We want to do the calibration only once since we want to have local timer
682  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
683  * frequency.
684  *
685  * This was previously done by reading the PIT/HPET and waiting for a wrap
686  * around to find out, that a tick has elapsed. I have a box, where the PIT
687  * readout is broken, so it never gets out of the wait loop again. This was
688  * also reported by others.
689  *
690  * Monitoring the jiffies value is inaccurate and the clockevents
691  * infrastructure allows us to do a simple substitution of the interrupt
692  * handler.
693  *
694  * The calibration routine also uses the pm_timer when possible, as the PIT
695  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
696  * back to normal later in the boot process).
697  */
698 
699 #define LAPIC_CAL_LOOPS		(HZ/10)
700 
701 static __initdata int lapic_cal_loops = -1;
702 static __initdata long lapic_cal_t1, lapic_cal_t2;
703 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
704 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
705 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
706 
707 /*
708  * Temporary interrupt handler.
709  */
710 static void __init lapic_cal_handler(struct clock_event_device *dev)
711 {
712 	unsigned long long tsc = 0;
713 	long tapic = apic_read(APIC_TMCCT);
714 	unsigned long pm = acpi_pm_read_early();
715 
716 	if (boot_cpu_has(X86_FEATURE_TSC))
717 		tsc = rdtsc();
718 
719 	switch (lapic_cal_loops++) {
720 	case 0:
721 		lapic_cal_t1 = tapic;
722 		lapic_cal_tsc1 = tsc;
723 		lapic_cal_pm1 = pm;
724 		lapic_cal_j1 = jiffies;
725 		break;
726 
727 	case LAPIC_CAL_LOOPS:
728 		lapic_cal_t2 = tapic;
729 		lapic_cal_tsc2 = tsc;
730 		if (pm < lapic_cal_pm1)
731 			pm += ACPI_PM_OVRRUN;
732 		lapic_cal_pm2 = pm;
733 		lapic_cal_j2 = jiffies;
734 		break;
735 	}
736 }
737 
738 static int __init
739 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
740 {
741 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
742 	const long pm_thresh = pm_100ms / 100;
743 	unsigned long mult;
744 	u64 res;
745 
746 #ifndef CONFIG_X86_PM_TIMER
747 	return -1;
748 #endif
749 
750 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
751 
752 	/* Check, if the PM timer is available */
753 	if (!deltapm)
754 		return -1;
755 
756 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
757 
758 	if (deltapm > (pm_100ms - pm_thresh) &&
759 	    deltapm < (pm_100ms + pm_thresh)) {
760 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
761 		return 0;
762 	}
763 
764 	res = (((u64)deltapm) *  mult) >> 22;
765 	do_div(res, 1000000);
766 	pr_warning("APIC calibration not consistent "
767 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
768 
769 	/* Correct the lapic counter value */
770 	res = (((u64)(*delta)) * pm_100ms);
771 	do_div(res, deltapm);
772 	pr_info("APIC delta adjusted to PM-Timer: "
773 		"%lu (%ld)\n", (unsigned long)res, *delta);
774 	*delta = (long)res;
775 
776 	/* Correct the tsc counter value */
777 	if (boot_cpu_has(X86_FEATURE_TSC)) {
778 		res = (((u64)(*deltatsc)) * pm_100ms);
779 		do_div(res, deltapm);
780 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
781 					  "PM-Timer: %lu (%ld)\n",
782 					(unsigned long)res, *deltatsc);
783 		*deltatsc = (long)res;
784 	}
785 
786 	return 0;
787 }
788 
789 static int __init calibrate_APIC_clock(void)
790 {
791 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
792 	void (*real_handler)(struct clock_event_device *dev);
793 	unsigned long deltaj;
794 	long delta, deltatsc;
795 	int pm_referenced = 0;
796 
797 	/**
798 	 * check if lapic timer has already been calibrated by platform
799 	 * specific routine, such as tsc calibration code. if so, we just fill
800 	 * in the clockevent structure and return.
801 	 */
802 
803 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
804 		return 0;
805 	} else if (lapic_timer_frequency) {
806 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
807 				lapic_timer_frequency);
808 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
809 					TICK_NSEC, lapic_clockevent.shift);
810 		lapic_clockevent.max_delta_ns =
811 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
812 		lapic_clockevent.max_delta_ticks = 0x7FFFFF;
813 		lapic_clockevent.min_delta_ns =
814 			clockevent_delta2ns(0xF, &lapic_clockevent);
815 		lapic_clockevent.min_delta_ticks = 0xF;
816 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
817 		return 0;
818 	}
819 
820 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
821 		    "calibrating APIC timer ...\n");
822 
823 	local_irq_disable();
824 
825 	/* Replace the global interrupt handler */
826 	real_handler = global_clock_event->event_handler;
827 	global_clock_event->event_handler = lapic_cal_handler;
828 
829 	/*
830 	 * Setup the APIC counter to maximum. There is no way the lapic
831 	 * can underflow in the 100ms detection time frame
832 	 */
833 	__setup_APIC_LVTT(0xffffffff, 0, 0);
834 
835 	/* Let the interrupts run */
836 	local_irq_enable();
837 
838 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
839 		cpu_relax();
840 
841 	local_irq_disable();
842 
843 	/* Restore the real event handler */
844 	global_clock_event->event_handler = real_handler;
845 
846 	/* Build delta t1-t2 as apic timer counts down */
847 	delta = lapic_cal_t1 - lapic_cal_t2;
848 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
849 
850 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
851 
852 	/* we trust the PM based calibration if possible */
853 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
854 					&delta, &deltatsc);
855 
856 	/* Calculate the scaled math multiplication factor */
857 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
858 				       lapic_clockevent.shift);
859 	lapic_clockevent.max_delta_ns =
860 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
861 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
862 	lapic_clockevent.min_delta_ns =
863 		clockevent_delta2ns(0xF, &lapic_clockevent);
864 	lapic_clockevent.min_delta_ticks = 0xF;
865 
866 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
867 
868 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
869 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
870 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
871 		    lapic_timer_frequency);
872 
873 	if (boot_cpu_has(X86_FEATURE_TSC)) {
874 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
875 			    "%ld.%04ld MHz.\n",
876 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
877 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
878 	}
879 
880 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
881 		    "%u.%04u MHz.\n",
882 		    lapic_timer_frequency / (1000000 / HZ),
883 		    lapic_timer_frequency % (1000000 / HZ));
884 
885 	/*
886 	 * Do a sanity check on the APIC calibration result
887 	 */
888 	if (lapic_timer_frequency < (1000000 / HZ)) {
889 		local_irq_enable();
890 		pr_warning("APIC frequency too slow, disabling apic timer\n");
891 		return -1;
892 	}
893 
894 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
895 
896 	/*
897 	 * PM timer calibration failed or not turned on
898 	 * so lets try APIC timer based calibration
899 	 */
900 	if (!pm_referenced) {
901 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
902 
903 		/*
904 		 * Setup the apic timer manually
905 		 */
906 		levt->event_handler = lapic_cal_handler;
907 		lapic_timer_set_periodic(levt);
908 		lapic_cal_loops = -1;
909 
910 		/* Let the interrupts run */
911 		local_irq_enable();
912 
913 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
914 			cpu_relax();
915 
916 		/* Stop the lapic timer */
917 		local_irq_disable();
918 		lapic_timer_shutdown(levt);
919 
920 		/* Jiffies delta */
921 		deltaj = lapic_cal_j2 - lapic_cal_j1;
922 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
923 
924 		/* Check, if the jiffies result is consistent */
925 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
926 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
927 		else
928 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
929 	}
930 	local_irq_enable();
931 
932 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
933 		pr_warning("APIC timer disabled due to verification failure\n");
934 			return -1;
935 	}
936 
937 	return 0;
938 }
939 
940 /*
941  * Setup the boot APIC
942  *
943  * Calibrate and verify the result.
944  */
945 void __init setup_boot_APIC_clock(void)
946 {
947 	/*
948 	 * The local apic timer can be disabled via the kernel
949 	 * commandline or from the CPU detection code. Register the lapic
950 	 * timer as a dummy clock event source on SMP systems, so the
951 	 * broadcast mechanism is used. On UP systems simply ignore it.
952 	 */
953 	if (disable_apic_timer) {
954 		pr_info("Disabling APIC timer\n");
955 		/* No broadcast on UP ! */
956 		if (num_possible_cpus() > 1) {
957 			lapic_clockevent.mult = 1;
958 			setup_APIC_timer();
959 		}
960 		return;
961 	}
962 
963 	if (calibrate_APIC_clock()) {
964 		/* No broadcast on UP ! */
965 		if (num_possible_cpus() > 1)
966 			setup_APIC_timer();
967 		return;
968 	}
969 
970 	/*
971 	 * If nmi_watchdog is set to IO_APIC, we need the
972 	 * PIT/HPET going.  Otherwise register lapic as a dummy
973 	 * device.
974 	 */
975 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
976 
977 	/* Setup the lapic or request the broadcast */
978 	setup_APIC_timer();
979 	amd_e400_c1e_apic_setup();
980 }
981 
982 void setup_secondary_APIC_clock(void)
983 {
984 	setup_APIC_timer();
985 	amd_e400_c1e_apic_setup();
986 }
987 
988 /*
989  * The guts of the apic timer interrupt
990  */
991 static void local_apic_timer_interrupt(void)
992 {
993 	int cpu = smp_processor_id();
994 	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
995 
996 	/*
997 	 * Normally we should not be here till LAPIC has been initialized but
998 	 * in some cases like kdump, its possible that there is a pending LAPIC
999 	 * timer interrupt from previous kernel's context and is delivered in
1000 	 * new kernel the moment interrupts are enabled.
1001 	 *
1002 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1003 	 * its possible that when we get here evt->event_handler is NULL.
1004 	 * Check for event_handler being NULL and discard the interrupt as
1005 	 * spurious.
1006 	 */
1007 	if (!evt->event_handler) {
1008 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
1009 		/* Switch it off */
1010 		lapic_timer_shutdown(evt);
1011 		return;
1012 	}
1013 
1014 	/*
1015 	 * the NMI deadlock-detector uses this.
1016 	 */
1017 	inc_irq_stat(apic_timer_irqs);
1018 
1019 	evt->event_handler(evt);
1020 }
1021 
1022 /*
1023  * Local APIC timer interrupt. This is the most natural way for doing
1024  * local interrupts, but local timer interrupts can be emulated by
1025  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1026  *
1027  * [ if a single-CPU system runs an SMP kernel then we call the local
1028  *   interrupt as well. Thus we cannot inline the local irq ... ]
1029  */
1030 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1031 {
1032 	struct pt_regs *old_regs = set_irq_regs(regs);
1033 
1034 	/*
1035 	 * NOTE! We'd better ACK the irq immediately,
1036 	 * because timer handling can be slow.
1037 	 *
1038 	 * update_process_times() expects us to have done irq_enter().
1039 	 * Besides, if we don't timer interrupts ignore the global
1040 	 * interrupt lock, which is the WrongThing (tm) to do.
1041 	 */
1042 	entering_ack_irq();
1043 	local_apic_timer_interrupt();
1044 	exiting_irq();
1045 
1046 	set_irq_regs(old_regs);
1047 }
1048 
1049 __visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
1050 {
1051 	struct pt_regs *old_regs = set_irq_regs(regs);
1052 
1053 	/*
1054 	 * NOTE! We'd better ACK the irq immediately,
1055 	 * because timer handling can be slow.
1056 	 *
1057 	 * update_process_times() expects us to have done irq_enter().
1058 	 * Besides, if we don't timer interrupts ignore the global
1059 	 * interrupt lock, which is the WrongThing (tm) to do.
1060 	 */
1061 	entering_ack_irq();
1062 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1063 	local_apic_timer_interrupt();
1064 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1065 	exiting_irq();
1066 
1067 	set_irq_regs(old_regs);
1068 }
1069 
1070 int setup_profiling_timer(unsigned int multiplier)
1071 {
1072 	return -EINVAL;
1073 }
1074 
1075 /*
1076  * Local APIC start and shutdown
1077  */
1078 
1079 /**
1080  * clear_local_APIC - shutdown the local APIC
1081  *
1082  * This is called, when a CPU is disabled and before rebooting, so the state of
1083  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1084  * leftovers during boot.
1085  */
1086 void clear_local_APIC(void)
1087 {
1088 	int maxlvt;
1089 	u32 v;
1090 
1091 	/* APIC hasn't been mapped yet */
1092 	if (!x2apic_mode && !apic_phys)
1093 		return;
1094 
1095 	maxlvt = lapic_get_maxlvt();
1096 	/*
1097 	 * Masking an LVT entry can trigger a local APIC error
1098 	 * if the vector is zero. Mask LVTERR first to prevent this.
1099 	 */
1100 	if (maxlvt >= 3) {
1101 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1102 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1103 	}
1104 	/*
1105 	 * Careful: we have to set masks only first to deassert
1106 	 * any level-triggered sources.
1107 	 */
1108 	v = apic_read(APIC_LVTT);
1109 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1110 	v = apic_read(APIC_LVT0);
1111 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1112 	v = apic_read(APIC_LVT1);
1113 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1114 	if (maxlvt >= 4) {
1115 		v = apic_read(APIC_LVTPC);
1116 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1117 	}
1118 
1119 	/* lets not touch this if we didn't frob it */
1120 #ifdef CONFIG_X86_THERMAL_VECTOR
1121 	if (maxlvt >= 5) {
1122 		v = apic_read(APIC_LVTTHMR);
1123 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1124 	}
1125 #endif
1126 #ifdef CONFIG_X86_MCE_INTEL
1127 	if (maxlvt >= 6) {
1128 		v = apic_read(APIC_LVTCMCI);
1129 		if (!(v & APIC_LVT_MASKED))
1130 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1131 	}
1132 #endif
1133 
1134 	/*
1135 	 * Clean APIC state for other OSs:
1136 	 */
1137 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1138 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1139 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1140 	if (maxlvt >= 3)
1141 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1142 	if (maxlvt >= 4)
1143 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1144 
1145 	/* Integrated APIC (!82489DX) ? */
1146 	if (lapic_is_integrated()) {
1147 		if (maxlvt > 3)
1148 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1149 			apic_write(APIC_ESR, 0);
1150 		apic_read(APIC_ESR);
1151 	}
1152 }
1153 
1154 /**
1155  * disable_local_APIC - clear and disable the local APIC
1156  */
1157 void disable_local_APIC(void)
1158 {
1159 	unsigned int value;
1160 
1161 	/* APIC hasn't been mapped yet */
1162 	if (!x2apic_mode && !apic_phys)
1163 		return;
1164 
1165 	clear_local_APIC();
1166 
1167 	/*
1168 	 * Disable APIC (implies clearing of registers
1169 	 * for 82489DX!).
1170 	 */
1171 	value = apic_read(APIC_SPIV);
1172 	value &= ~APIC_SPIV_APIC_ENABLED;
1173 	apic_write(APIC_SPIV, value);
1174 
1175 #ifdef CONFIG_X86_32
1176 	/*
1177 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1178 	 * restore the disabled state.
1179 	 */
1180 	if (enabled_via_apicbase) {
1181 		unsigned int l, h;
1182 
1183 		rdmsr(MSR_IA32_APICBASE, l, h);
1184 		l &= ~MSR_IA32_APICBASE_ENABLE;
1185 		wrmsr(MSR_IA32_APICBASE, l, h);
1186 	}
1187 #endif
1188 }
1189 
1190 /*
1191  * If Linux enabled the LAPIC against the BIOS default disable it down before
1192  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1193  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1194  * for the case where Linux didn't enable the LAPIC.
1195  */
1196 void lapic_shutdown(void)
1197 {
1198 	unsigned long flags;
1199 
1200 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1201 		return;
1202 
1203 	local_irq_save(flags);
1204 
1205 #ifdef CONFIG_X86_32
1206 	if (!enabled_via_apicbase)
1207 		clear_local_APIC();
1208 	else
1209 #endif
1210 		disable_local_APIC();
1211 
1212 
1213 	local_irq_restore(flags);
1214 }
1215 
1216 /**
1217  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1218  */
1219 void __init sync_Arb_IDs(void)
1220 {
1221 	/*
1222 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1223 	 * needed on AMD.
1224 	 */
1225 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1226 		return;
1227 
1228 	/*
1229 	 * Wait for idle.
1230 	 */
1231 	apic_wait_icr_idle();
1232 
1233 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1234 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1235 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1236 }
1237 
1238 /*
1239  * An initial setup of the virtual wire mode.
1240  */
1241 void __init init_bsp_APIC(void)
1242 {
1243 	unsigned int value;
1244 
1245 	/*
1246 	 * Don't do the setup now if we have a SMP BIOS as the
1247 	 * through-I/O-APIC virtual wire mode might be active.
1248 	 */
1249 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1250 		return;
1251 
1252 	/*
1253 	 * Do not trust the local APIC being empty at bootup.
1254 	 */
1255 	clear_local_APIC();
1256 
1257 	/*
1258 	 * Enable APIC.
1259 	 */
1260 	value = apic_read(APIC_SPIV);
1261 	value &= ~APIC_VECTOR_MASK;
1262 	value |= APIC_SPIV_APIC_ENABLED;
1263 
1264 #ifdef CONFIG_X86_32
1265 	/* This bit is reserved on P4/Xeon and should be cleared */
1266 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1267 	    (boot_cpu_data.x86 == 15))
1268 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1269 	else
1270 #endif
1271 		value |= APIC_SPIV_FOCUS_DISABLED;
1272 	value |= SPURIOUS_APIC_VECTOR;
1273 	apic_write(APIC_SPIV, value);
1274 
1275 	/*
1276 	 * Set up the virtual wire mode.
1277 	 */
1278 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1279 	value = APIC_DM_NMI;
1280 	if (!lapic_is_integrated())		/* 82489DX */
1281 		value |= APIC_LVT_LEVEL_TRIGGER;
1282 	if (apic_extnmi == APIC_EXTNMI_NONE)
1283 		value |= APIC_LVT_MASKED;
1284 	apic_write(APIC_LVT1, value);
1285 }
1286 
1287 static void lapic_setup_esr(void)
1288 {
1289 	unsigned int oldvalue, value, maxlvt;
1290 
1291 	if (!lapic_is_integrated()) {
1292 		pr_info("No ESR for 82489DX.\n");
1293 		return;
1294 	}
1295 
1296 	if (apic->disable_esr) {
1297 		/*
1298 		 * Something untraceable is creating bad interrupts on
1299 		 * secondary quads ... for the moment, just leave the
1300 		 * ESR disabled - we can't do anything useful with the
1301 		 * errors anyway - mbligh
1302 		 */
1303 		pr_info("Leaving ESR disabled.\n");
1304 		return;
1305 	}
1306 
1307 	maxlvt = lapic_get_maxlvt();
1308 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1309 		apic_write(APIC_ESR, 0);
1310 	oldvalue = apic_read(APIC_ESR);
1311 
1312 	/* enables sending errors */
1313 	value = ERROR_APIC_VECTOR;
1314 	apic_write(APIC_LVTERR, value);
1315 
1316 	/*
1317 	 * spec says clear errors after enabling vector.
1318 	 */
1319 	if (maxlvt > 3)
1320 		apic_write(APIC_ESR, 0);
1321 	value = apic_read(APIC_ESR);
1322 	if (value != oldvalue)
1323 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1324 			"vector: 0x%08x  after: 0x%08x\n",
1325 			oldvalue, value);
1326 }
1327 
1328 /**
1329  * setup_local_APIC - setup the local APIC
1330  *
1331  * Used to setup local APIC while initializing BSP or bringing up APs.
1332  * Always called with preemption disabled.
1333  */
1334 void setup_local_APIC(void)
1335 {
1336 	int cpu = smp_processor_id();
1337 	unsigned int value, queued;
1338 	int i, j, acked = 0;
1339 	unsigned long long tsc = 0, ntsc;
1340 	long long max_loops = cpu_khz ? cpu_khz : 1000000;
1341 
1342 	if (boot_cpu_has(X86_FEATURE_TSC))
1343 		tsc = rdtsc();
1344 
1345 	if (disable_apic) {
1346 		disable_ioapic_support();
1347 		return;
1348 	}
1349 
1350 #ifdef CONFIG_X86_32
1351 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1352 	if (lapic_is_integrated() && apic->disable_esr) {
1353 		apic_write(APIC_ESR, 0);
1354 		apic_write(APIC_ESR, 0);
1355 		apic_write(APIC_ESR, 0);
1356 		apic_write(APIC_ESR, 0);
1357 	}
1358 #endif
1359 	perf_events_lapic_init();
1360 
1361 	/*
1362 	 * Double-check whether this APIC is really registered.
1363 	 * This is meaningless in clustered apic mode, so we skip it.
1364 	 */
1365 	BUG_ON(!apic->apic_id_registered());
1366 
1367 	/*
1368 	 * Intel recommends to set DFR, LDR and TPR before enabling
1369 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1370 	 * document number 292116).  So here it goes...
1371 	 */
1372 	apic->init_apic_ldr();
1373 
1374 #ifdef CONFIG_X86_32
1375 	/*
1376 	 * APIC LDR is initialized.  If logical_apicid mapping was
1377 	 * initialized during get_smp_config(), make sure it matches the
1378 	 * actual value.
1379 	 */
1380 	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1381 	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1382 	/* always use the value from LDR */
1383 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1384 		logical_smp_processor_id();
1385 #endif
1386 
1387 	/*
1388 	 * Set Task Priority to 'accept all'. We never change this
1389 	 * later on.
1390 	 */
1391 	value = apic_read(APIC_TASKPRI);
1392 	value &= ~APIC_TPRI_MASK;
1393 	apic_write(APIC_TASKPRI, value);
1394 
1395 	/*
1396 	 * After a crash, we no longer service the interrupts and a pending
1397 	 * interrupt from previous kernel might still have ISR bit set.
1398 	 *
1399 	 * Most probably by now CPU has serviced that pending interrupt and
1400 	 * it might not have done the ack_APIC_irq() because it thought,
1401 	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1402 	 * does not clear the ISR bit and cpu thinks it has already serivced
1403 	 * the interrupt. Hence a vector might get locked. It was noticed
1404 	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1405 	 */
1406 	do {
1407 		queued = 0;
1408 		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1409 			queued |= apic_read(APIC_IRR + i*0x10);
1410 
1411 		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1412 			value = apic_read(APIC_ISR + i*0x10);
1413 			for (j = 31; j >= 0; j--) {
1414 				if (value & (1<<j)) {
1415 					ack_APIC_irq();
1416 					acked++;
1417 				}
1418 			}
1419 		}
1420 		if (acked > 256) {
1421 			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1422 			       acked);
1423 			break;
1424 		}
1425 		if (queued) {
1426 			if (boot_cpu_has(X86_FEATURE_TSC) && cpu_khz) {
1427 				ntsc = rdtsc();
1428 				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1429 			} else
1430 				max_loops--;
1431 		}
1432 	} while (queued && max_loops > 0);
1433 	WARN_ON(max_loops <= 0);
1434 
1435 	/*
1436 	 * Now that we are all set up, enable the APIC
1437 	 */
1438 	value = apic_read(APIC_SPIV);
1439 	value &= ~APIC_VECTOR_MASK;
1440 	/*
1441 	 * Enable APIC
1442 	 */
1443 	value |= APIC_SPIV_APIC_ENABLED;
1444 
1445 #ifdef CONFIG_X86_32
1446 	/*
1447 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1448 	 * certain networking cards. If high frequency interrupts are
1449 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1450 	 * entry is masked/unmasked at a high rate as well then sooner or
1451 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1452 	 * from the device. If focus CPU is disabled then the hang goes
1453 	 * away, oh well :-(
1454 	 *
1455 	 * [ This bug can be reproduced easily with a level-triggered
1456 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1457 	 *   BX chipset. ]
1458 	 */
1459 	/*
1460 	 * Actually disabling the focus CPU check just makes the hang less
1461 	 * frequent as it makes the interrupt distributon model be more
1462 	 * like LRU than MRU (the short-term load is more even across CPUs).
1463 	 */
1464 
1465 	/*
1466 	 * - enable focus processor (bit==0)
1467 	 * - 64bit mode always use processor focus
1468 	 *   so no need to set it
1469 	 */
1470 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1471 #endif
1472 
1473 	/*
1474 	 * Set spurious IRQ vector
1475 	 */
1476 	value |= SPURIOUS_APIC_VECTOR;
1477 	apic_write(APIC_SPIV, value);
1478 
1479 	/*
1480 	 * Set up LVT0, LVT1:
1481 	 *
1482 	 * set up through-local-APIC on the BP's LINT0. This is not
1483 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1484 	 * we delegate interrupts to the 8259A.
1485 	 */
1486 	/*
1487 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1488 	 */
1489 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1490 	if (!cpu && (pic_mode || !value)) {
1491 		value = APIC_DM_EXTINT;
1492 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1493 	} else {
1494 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1495 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1496 	}
1497 	apic_write(APIC_LVT0, value);
1498 
1499 	/*
1500 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1501 	 * modified by apic_extnmi= boot option.
1502 	 */
1503 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1504 	    apic_extnmi == APIC_EXTNMI_ALL)
1505 		value = APIC_DM_NMI;
1506 	else
1507 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1508 	if (!lapic_is_integrated())		/* 82489DX */
1509 		value |= APIC_LVT_LEVEL_TRIGGER;
1510 	apic_write(APIC_LVT1, value);
1511 
1512 #ifdef CONFIG_X86_MCE_INTEL
1513 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1514 	if (!cpu)
1515 		cmci_recheck();
1516 #endif
1517 }
1518 
1519 static void end_local_APIC_setup(void)
1520 {
1521 	lapic_setup_esr();
1522 
1523 #ifdef CONFIG_X86_32
1524 	{
1525 		unsigned int value;
1526 		/* Disable the local apic timer */
1527 		value = apic_read(APIC_LVTT);
1528 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1529 		apic_write(APIC_LVTT, value);
1530 	}
1531 #endif
1532 
1533 	apic_pm_activate();
1534 }
1535 
1536 /*
1537  * APIC setup function for application processors. Called from smpboot.c
1538  */
1539 void apic_ap_setup(void)
1540 {
1541 	setup_local_APIC();
1542 	end_local_APIC_setup();
1543 }
1544 
1545 #ifdef CONFIG_X86_X2APIC
1546 int x2apic_mode;
1547 
1548 enum {
1549 	X2APIC_OFF,
1550 	X2APIC_ON,
1551 	X2APIC_DISABLED,
1552 };
1553 static int x2apic_state;
1554 
1555 static void __x2apic_disable(void)
1556 {
1557 	u64 msr;
1558 
1559 	if (!boot_cpu_has(X86_FEATURE_APIC))
1560 		return;
1561 
1562 	rdmsrl(MSR_IA32_APICBASE, msr);
1563 	if (!(msr & X2APIC_ENABLE))
1564 		return;
1565 	/* Disable xapic and x2apic first and then reenable xapic mode */
1566 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1567 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1568 	printk_once(KERN_INFO "x2apic disabled\n");
1569 }
1570 
1571 static void __x2apic_enable(void)
1572 {
1573 	u64 msr;
1574 
1575 	rdmsrl(MSR_IA32_APICBASE, msr);
1576 	if (msr & X2APIC_ENABLE)
1577 		return;
1578 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1579 	printk_once(KERN_INFO "x2apic enabled\n");
1580 }
1581 
1582 static int __init setup_nox2apic(char *str)
1583 {
1584 	if (x2apic_enabled()) {
1585 		int apicid = native_apic_msr_read(APIC_ID);
1586 
1587 		if (apicid >= 255) {
1588 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1589 				   apicid);
1590 			return 0;
1591 		}
1592 		pr_warning("x2apic already enabled.\n");
1593 		__x2apic_disable();
1594 	}
1595 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1596 	x2apic_state = X2APIC_DISABLED;
1597 	x2apic_mode = 0;
1598 	return 0;
1599 }
1600 early_param("nox2apic", setup_nox2apic);
1601 
1602 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
1603 void x2apic_setup(void)
1604 {
1605 	/*
1606 	 * If x2apic is not in ON state, disable it if already enabled
1607 	 * from BIOS.
1608 	 */
1609 	if (x2apic_state != X2APIC_ON) {
1610 		__x2apic_disable();
1611 		return;
1612 	}
1613 	__x2apic_enable();
1614 }
1615 
1616 static __init void x2apic_disable(void)
1617 {
1618 	u32 x2apic_id, state = x2apic_state;
1619 
1620 	x2apic_mode = 0;
1621 	x2apic_state = X2APIC_DISABLED;
1622 
1623 	if (state != X2APIC_ON)
1624 		return;
1625 
1626 	x2apic_id = read_apic_id();
1627 	if (x2apic_id >= 255)
1628 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1629 
1630 	__x2apic_disable();
1631 	register_lapic_address(mp_lapic_addr);
1632 }
1633 
1634 static __init void x2apic_enable(void)
1635 {
1636 	if (x2apic_state != X2APIC_OFF)
1637 		return;
1638 
1639 	x2apic_mode = 1;
1640 	x2apic_state = X2APIC_ON;
1641 	__x2apic_enable();
1642 }
1643 
1644 static __init void try_to_enable_x2apic(int remap_mode)
1645 {
1646 	if (x2apic_state == X2APIC_DISABLED)
1647 		return;
1648 
1649 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1650 		/* IR is required if there is APIC ID > 255 even when running
1651 		 * under KVM
1652 		 */
1653 		if (max_physical_apicid > 255 ||
1654 		    !hypervisor_x2apic_available()) {
1655 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1656 			x2apic_disable();
1657 			return;
1658 		}
1659 
1660 		/*
1661 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1662 		 * only in physical mode
1663 		 */
1664 		x2apic_phys = 1;
1665 	}
1666 	x2apic_enable();
1667 }
1668 
1669 void __init check_x2apic(void)
1670 {
1671 	if (x2apic_enabled()) {
1672 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1673 		x2apic_mode = 1;
1674 		x2apic_state = X2APIC_ON;
1675 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1676 		x2apic_state = X2APIC_DISABLED;
1677 	}
1678 }
1679 #else /* CONFIG_X86_X2APIC */
1680 static int __init validate_x2apic(void)
1681 {
1682 	if (!apic_is_x2apic_enabled())
1683 		return 0;
1684 	/*
1685 	 * Checkme: Can we simply turn off x2apic here instead of panic?
1686 	 */
1687 	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1688 }
1689 early_initcall(validate_x2apic);
1690 
1691 static inline void try_to_enable_x2apic(int remap_mode) { }
1692 static inline void __x2apic_enable(void) { }
1693 #endif /* !CONFIG_X86_X2APIC */
1694 
1695 void __init enable_IR_x2apic(void)
1696 {
1697 	unsigned long flags;
1698 	int ret, ir_stat;
1699 
1700 	if (skip_ioapic_setup) {
1701 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1702 		return;
1703 	}
1704 
1705 	ir_stat = irq_remapping_prepare();
1706 	if (ir_stat < 0 && !x2apic_supported())
1707 		return;
1708 
1709 	ret = save_ioapic_entries();
1710 	if (ret) {
1711 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1712 		return;
1713 	}
1714 
1715 	local_irq_save(flags);
1716 	legacy_pic->mask_all();
1717 	mask_ioapic_entries();
1718 
1719 	/* If irq_remapping_prepare() succeeded, try to enable it */
1720 	if (ir_stat >= 0)
1721 		ir_stat = irq_remapping_enable();
1722 	/* ir_stat contains the remap mode or an error code */
1723 	try_to_enable_x2apic(ir_stat);
1724 
1725 	if (ir_stat < 0)
1726 		restore_ioapic_entries();
1727 	legacy_pic->restore_mask();
1728 	local_irq_restore(flags);
1729 }
1730 
1731 #ifdef CONFIG_X86_64
1732 /*
1733  * Detect and enable local APICs on non-SMP boards.
1734  * Original code written by Keir Fraser.
1735  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1736  * not correctly set up (usually the APIC timer won't work etc.)
1737  */
1738 static int __init detect_init_APIC(void)
1739 {
1740 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1741 		pr_info("No local APIC present\n");
1742 		return -1;
1743 	}
1744 
1745 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1746 	return 0;
1747 }
1748 #else
1749 
1750 static int __init apic_verify(void)
1751 {
1752 	u32 features, h, l;
1753 
1754 	/*
1755 	 * The APIC feature bit should now be enabled
1756 	 * in `cpuid'
1757 	 */
1758 	features = cpuid_edx(1);
1759 	if (!(features & (1 << X86_FEATURE_APIC))) {
1760 		pr_warning("Could not enable APIC!\n");
1761 		return -1;
1762 	}
1763 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1764 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1765 
1766 	/* The BIOS may have set up the APIC at some other address */
1767 	if (boot_cpu_data.x86 >= 6) {
1768 		rdmsr(MSR_IA32_APICBASE, l, h);
1769 		if (l & MSR_IA32_APICBASE_ENABLE)
1770 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1771 	}
1772 
1773 	pr_info("Found and enabled local APIC!\n");
1774 	return 0;
1775 }
1776 
1777 int __init apic_force_enable(unsigned long addr)
1778 {
1779 	u32 h, l;
1780 
1781 	if (disable_apic)
1782 		return -1;
1783 
1784 	/*
1785 	 * Some BIOSes disable the local APIC in the APIC_BASE
1786 	 * MSR. This can only be done in software for Intel P6 or later
1787 	 * and AMD K7 (Model > 1) or later.
1788 	 */
1789 	if (boot_cpu_data.x86 >= 6) {
1790 		rdmsr(MSR_IA32_APICBASE, l, h);
1791 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1792 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1793 			l &= ~MSR_IA32_APICBASE_BASE;
1794 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1795 			wrmsr(MSR_IA32_APICBASE, l, h);
1796 			enabled_via_apicbase = 1;
1797 		}
1798 	}
1799 	return apic_verify();
1800 }
1801 
1802 /*
1803  * Detect and initialize APIC
1804  */
1805 static int __init detect_init_APIC(void)
1806 {
1807 	/* Disabled by kernel option? */
1808 	if (disable_apic)
1809 		return -1;
1810 
1811 	switch (boot_cpu_data.x86_vendor) {
1812 	case X86_VENDOR_AMD:
1813 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1814 		    (boot_cpu_data.x86 >= 15))
1815 			break;
1816 		goto no_apic;
1817 	case X86_VENDOR_INTEL:
1818 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1819 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
1820 			break;
1821 		goto no_apic;
1822 	default:
1823 		goto no_apic;
1824 	}
1825 
1826 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1827 		/*
1828 		 * Over-ride BIOS and try to enable the local APIC only if
1829 		 * "lapic" specified.
1830 		 */
1831 		if (!force_enable_local_apic) {
1832 			pr_info("Local APIC disabled by BIOS -- "
1833 				"you can enable it with \"lapic\"\n");
1834 			return -1;
1835 		}
1836 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1837 			return -1;
1838 	} else {
1839 		if (apic_verify())
1840 			return -1;
1841 	}
1842 
1843 	apic_pm_activate();
1844 
1845 	return 0;
1846 
1847 no_apic:
1848 	pr_info("No local APIC present or hardware disabled\n");
1849 	return -1;
1850 }
1851 #endif
1852 
1853 /**
1854  * init_apic_mappings - initialize APIC mappings
1855  */
1856 void __init init_apic_mappings(void)
1857 {
1858 	unsigned int new_apicid;
1859 
1860 	apic_check_deadline_errata();
1861 
1862 	if (x2apic_mode) {
1863 		boot_cpu_physical_apicid = read_apic_id();
1864 		return;
1865 	}
1866 
1867 	/* If no local APIC can be found return early */
1868 	if (!smp_found_config && detect_init_APIC()) {
1869 		/* lets NOP'ify apic operations */
1870 		pr_info("APIC: disable apic facility\n");
1871 		apic_disable();
1872 	} else {
1873 		apic_phys = mp_lapic_addr;
1874 
1875 		/*
1876 		 * If the system has ACPI MADT tables or MP info, the LAPIC
1877 		 * address is already registered.
1878 		 */
1879 		if (!acpi_lapic && !smp_found_config)
1880 			register_lapic_address(apic_phys);
1881 	}
1882 
1883 	/*
1884 	 * Fetch the APIC ID of the BSP in case we have a
1885 	 * default configuration (or the MP table is broken).
1886 	 */
1887 	new_apicid = read_apic_id();
1888 	if (boot_cpu_physical_apicid != new_apicid) {
1889 		boot_cpu_physical_apicid = new_apicid;
1890 		/*
1891 		 * yeah -- we lie about apic_version
1892 		 * in case if apic was disabled via boot option
1893 		 * but it's not a problem for SMP compiled kernel
1894 		 * since smp_sanity_check is prepared for such a case
1895 		 * and disable smp mode
1896 		 */
1897 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1898 	}
1899 }
1900 
1901 void __init register_lapic_address(unsigned long address)
1902 {
1903 	mp_lapic_addr = address;
1904 
1905 	if (!x2apic_mode) {
1906 		set_fixmap_nocache(FIX_APIC_BASE, address);
1907 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1908 			    APIC_BASE, address);
1909 	}
1910 	if (boot_cpu_physical_apicid == -1U) {
1911 		boot_cpu_physical_apicid  = read_apic_id();
1912 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1913 	}
1914 }
1915 
1916 /*
1917  * Local APIC interrupts
1918  */
1919 
1920 /*
1921  * This interrupt should _never_ happen with our APIC/SMP architecture
1922  */
1923 static void __smp_spurious_interrupt(u8 vector)
1924 {
1925 	u32 v;
1926 
1927 	/*
1928 	 * Check if this really is a spurious interrupt and ACK it
1929 	 * if it is a vectored one.  Just in case...
1930 	 * Spurious interrupts should not be ACKed.
1931 	 */
1932 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
1933 	if (v & (1 << (vector & 0x1f)))
1934 		ack_APIC_irq();
1935 
1936 	inc_irq_stat(irq_spurious_count);
1937 
1938 	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1939 	pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
1940 		"should never happen.\n", vector, smp_processor_id());
1941 }
1942 
1943 __visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
1944 {
1945 	entering_irq();
1946 	__smp_spurious_interrupt(~regs->orig_ax);
1947 	exiting_irq();
1948 }
1949 
1950 __visible void __irq_entry smp_trace_spurious_interrupt(struct pt_regs *regs)
1951 {
1952 	u8 vector = ~regs->orig_ax;
1953 
1954 	entering_irq();
1955 	trace_spurious_apic_entry(vector);
1956 	__smp_spurious_interrupt(vector);
1957 	trace_spurious_apic_exit(vector);
1958 	exiting_irq();
1959 }
1960 
1961 /*
1962  * This interrupt should never happen with our APIC/SMP architecture
1963  */
1964 static void __smp_error_interrupt(struct pt_regs *regs)
1965 {
1966 	u32 v;
1967 	u32 i = 0;
1968 	static const char * const error_interrupt_reason[] = {
1969 		"Send CS error",		/* APIC Error Bit 0 */
1970 		"Receive CS error",		/* APIC Error Bit 1 */
1971 		"Send accept error",		/* APIC Error Bit 2 */
1972 		"Receive accept error",		/* APIC Error Bit 3 */
1973 		"Redirectable IPI",		/* APIC Error Bit 4 */
1974 		"Send illegal vector",		/* APIC Error Bit 5 */
1975 		"Received illegal vector",	/* APIC Error Bit 6 */
1976 		"Illegal register address",	/* APIC Error Bit 7 */
1977 	};
1978 
1979 	/* First tickle the hardware, only then report what went on. -- REW */
1980 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
1981 		apic_write(APIC_ESR, 0);
1982 	v = apic_read(APIC_ESR);
1983 	ack_APIC_irq();
1984 	atomic_inc(&irq_err_count);
1985 
1986 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
1987 		    smp_processor_id(), v);
1988 
1989 	v &= 0xff;
1990 	while (v) {
1991 		if (v & 0x1)
1992 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1993 		i++;
1994 		v >>= 1;
1995 	}
1996 
1997 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
1998 
1999 }
2000 
2001 __visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2002 {
2003 	entering_irq();
2004 	__smp_error_interrupt(regs);
2005 	exiting_irq();
2006 }
2007 
2008 __visible void __irq_entry smp_trace_error_interrupt(struct pt_regs *regs)
2009 {
2010 	entering_irq();
2011 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2012 	__smp_error_interrupt(regs);
2013 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2014 	exiting_irq();
2015 }
2016 
2017 /**
2018  * connect_bsp_APIC - attach the APIC to the interrupt system
2019  */
2020 static void __init connect_bsp_APIC(void)
2021 {
2022 #ifdef CONFIG_X86_32
2023 	if (pic_mode) {
2024 		/*
2025 		 * Do not trust the local APIC being empty at bootup.
2026 		 */
2027 		clear_local_APIC();
2028 		/*
2029 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2030 		 * local APIC to INT and NMI lines.
2031 		 */
2032 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2033 				"enabling APIC mode.\n");
2034 		imcr_pic_to_apic();
2035 	}
2036 #endif
2037 }
2038 
2039 /**
2040  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2041  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2042  *
2043  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2044  * APIC is disabled.
2045  */
2046 void disconnect_bsp_APIC(int virt_wire_setup)
2047 {
2048 	unsigned int value;
2049 
2050 #ifdef CONFIG_X86_32
2051 	if (pic_mode) {
2052 		/*
2053 		 * Put the board back into PIC mode (has an effect only on
2054 		 * certain older boards).  Note that APIC interrupts, including
2055 		 * IPIs, won't work beyond this point!  The only exception are
2056 		 * INIT IPIs.
2057 		 */
2058 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2059 				"entering PIC mode.\n");
2060 		imcr_apic_to_pic();
2061 		return;
2062 	}
2063 #endif
2064 
2065 	/* Go back to Virtual Wire compatibility mode */
2066 
2067 	/* For the spurious interrupt use vector F, and enable it */
2068 	value = apic_read(APIC_SPIV);
2069 	value &= ~APIC_VECTOR_MASK;
2070 	value |= APIC_SPIV_APIC_ENABLED;
2071 	value |= 0xf;
2072 	apic_write(APIC_SPIV, value);
2073 
2074 	if (!virt_wire_setup) {
2075 		/*
2076 		 * For LVT0 make it edge triggered, active high,
2077 		 * external and enabled
2078 		 */
2079 		value = apic_read(APIC_LVT0);
2080 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2081 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2082 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2083 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2084 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2085 		apic_write(APIC_LVT0, value);
2086 	} else {
2087 		/* Disable LVT0 */
2088 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2089 	}
2090 
2091 	/*
2092 	 * For LVT1 make it edge triggered, active high,
2093 	 * nmi and enabled
2094 	 */
2095 	value = apic_read(APIC_LVT1);
2096 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2097 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2098 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2099 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2100 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2101 	apic_write(APIC_LVT1, value);
2102 }
2103 
2104 /*
2105  * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2106  * contiguously, it equals to current allocated max logical CPU ID plus 1.
2107  * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2108  * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2109  *
2110  * NOTE: Reserve 0 for BSP.
2111  */
2112 static int nr_logical_cpuids = 1;
2113 
2114 /*
2115  * Used to store mapping between logical CPU IDs and APIC IDs.
2116  */
2117 static int cpuid_to_apicid[] = {
2118 	[0 ... NR_CPUS - 1] = -1,
2119 };
2120 
2121 /*
2122  * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2123  * and cpuid_to_apicid[] synchronized.
2124  */
2125 static int allocate_logical_cpuid(int apicid)
2126 {
2127 	int i;
2128 
2129 	/*
2130 	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2131 	 * check if the kernel has allocated a cpuid for it.
2132 	 */
2133 	for (i = 0; i < nr_logical_cpuids; i++) {
2134 		if (cpuid_to_apicid[i] == apicid)
2135 			return i;
2136 	}
2137 
2138 	/* Allocate a new cpuid. */
2139 	if (nr_logical_cpuids >= nr_cpu_ids) {
2140 		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %i reached. "
2141 			     "Processor %d/0x%x and the rest are ignored.\n",
2142 			     nr_cpu_ids, nr_logical_cpuids, apicid);
2143 		return -EINVAL;
2144 	}
2145 
2146 	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2147 	return nr_logical_cpuids++;
2148 }
2149 
2150 int generic_processor_info(int apicid, int version)
2151 {
2152 	int cpu, max = nr_cpu_ids;
2153 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2154 				phys_cpu_present_map);
2155 
2156 	/*
2157 	 * boot_cpu_physical_apicid is designed to have the apicid
2158 	 * returned by read_apic_id(), i.e, the apicid of the
2159 	 * currently booting-up processor. However, on some platforms,
2160 	 * it is temporarily modified by the apicid reported as BSP
2161 	 * through MP table. Concretely:
2162 	 *
2163 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2164 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2165 	 *
2166 	 * This function is executed with the modified
2167 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2168 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2169 	 *
2170 	 * Since fixing handling of boot_cpu_physical_apicid requires
2171 	 * another discussion and tests on each platform, we leave it
2172 	 * for now and here we use read_apic_id() directly in this
2173 	 * function, __generic_processor_info().
2174 	 */
2175 	if (disabled_cpu_apicid != BAD_APICID &&
2176 	    disabled_cpu_apicid != read_apic_id() &&
2177 	    disabled_cpu_apicid == apicid) {
2178 		int thiscpu = num_processors + disabled_cpus;
2179 
2180 		pr_warning("APIC: Disabling requested cpu."
2181 			   " Processor %d/0x%x ignored.\n",
2182 			   thiscpu, apicid);
2183 
2184 		disabled_cpus++;
2185 		return -ENODEV;
2186 	}
2187 
2188 	/*
2189 	 * If boot cpu has not been detected yet, then only allow upto
2190 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2191 	 */
2192 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2193 	    apicid != boot_cpu_physical_apicid) {
2194 		int thiscpu = max + disabled_cpus - 1;
2195 
2196 		pr_warning(
2197 			"APIC: NR_CPUS/possible_cpus limit of %i almost"
2198 			" reached. Keeping one slot for boot cpu."
2199 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2200 
2201 		disabled_cpus++;
2202 		return -ENODEV;
2203 	}
2204 
2205 	if (num_processors >= nr_cpu_ids) {
2206 		int thiscpu = max + disabled_cpus;
2207 
2208 		pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2209 			   "reached. Processor %d/0x%x ignored.\n",
2210 			   max, thiscpu, apicid);
2211 
2212 		disabled_cpus++;
2213 		return -EINVAL;
2214 	}
2215 
2216 	if (apicid == boot_cpu_physical_apicid) {
2217 		/*
2218 		 * x86_bios_cpu_apicid is required to have processors listed
2219 		 * in same order as logical cpu numbers. Hence the first
2220 		 * entry is BSP, and so on.
2221 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2222 		 * for BSP.
2223 		 */
2224 		cpu = 0;
2225 
2226 		/* Logical cpuid 0 is reserved for BSP. */
2227 		cpuid_to_apicid[0] = apicid;
2228 	} else {
2229 		cpu = allocate_logical_cpuid(apicid);
2230 		if (cpu < 0) {
2231 			disabled_cpus++;
2232 			return -EINVAL;
2233 		}
2234 	}
2235 
2236 	/*
2237 	 * Validate version
2238 	 */
2239 	if (version == 0x0) {
2240 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2241 			   cpu, apicid);
2242 		version = 0x10;
2243 	}
2244 
2245 	if (version != boot_cpu_apic_version) {
2246 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2247 			boot_cpu_apic_version, cpu, version);
2248 	}
2249 
2250 	if (apicid > max_physical_apicid)
2251 		max_physical_apicid = apicid;
2252 
2253 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2254 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2255 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2256 #endif
2257 #ifdef CONFIG_X86_32
2258 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2259 		apic->x86_32_early_logical_apicid(cpu);
2260 #endif
2261 	set_cpu_possible(cpu, true);
2262 	physid_set(apicid, phys_cpu_present_map);
2263 	set_cpu_present(cpu, true);
2264 	num_processors++;
2265 
2266 	return cpu;
2267 }
2268 
2269 int hard_smp_processor_id(void)
2270 {
2271 	return read_apic_id();
2272 }
2273 
2274 void default_init_apic_ldr(void)
2275 {
2276 	unsigned long val;
2277 
2278 	apic_write(APIC_DFR, APIC_DFR_VALUE);
2279 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2280 	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2281 	apic_write(APIC_LDR, val);
2282 }
2283 
2284 int default_cpu_mask_to_apicid(const struct cpumask *mask,
2285 			       struct irq_data *irqdata,
2286 			       unsigned int *apicid)
2287 {
2288 	unsigned int cpu = cpumask_first(mask);
2289 
2290 	if (cpu >= nr_cpu_ids)
2291 		return -EINVAL;
2292 	*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2293 	irq_data_update_effective_affinity(irqdata, cpumask_of(cpu));
2294 	return 0;
2295 }
2296 
2297 int flat_cpu_mask_to_apicid(const struct cpumask *mask,
2298 			    struct irq_data *irqdata,
2299 			    unsigned int *apicid)
2300 
2301 {
2302 	struct cpumask *effmsk = irq_data_get_effective_affinity_mask(irqdata);
2303 	unsigned long cpu_mask = cpumask_bits(mask)[0] & APIC_ALL_CPUS;
2304 
2305 	if (!cpu_mask)
2306 		return -EINVAL;
2307 	*apicid = (unsigned int)cpu_mask;
2308 	cpumask_bits(effmsk)[0] = cpu_mask;
2309 	return 0;
2310 }
2311 
2312 /*
2313  * Override the generic EOI implementation with an optimized version.
2314  * Only called during early boot when only one CPU is active and with
2315  * interrupts disabled, so we know this does not race with actual APIC driver
2316  * use.
2317  */
2318 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2319 {
2320 	struct apic **drv;
2321 
2322 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2323 		/* Should happen once for each apic */
2324 		WARN_ON((*drv)->eoi_write == eoi_write);
2325 		(*drv)->native_eoi_write = (*drv)->eoi_write;
2326 		(*drv)->eoi_write = eoi_write;
2327 	}
2328 }
2329 
2330 static void __init apic_bsp_up_setup(void)
2331 {
2332 #ifdef CONFIG_X86_64
2333 	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2334 #else
2335 	/*
2336 	 * Hack: In case of kdump, after a crash, kernel might be booting
2337 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2338 	 * might be zero if read from MP tables. Get it from LAPIC.
2339 	 */
2340 # ifdef CONFIG_CRASH_DUMP
2341 	boot_cpu_physical_apicid = read_apic_id();
2342 # endif
2343 #endif
2344 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2345 }
2346 
2347 /**
2348  * apic_bsp_setup - Setup function for local apic and io-apic
2349  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2350  *
2351  * Returns:
2352  * apic_id of BSP APIC
2353  */
2354 int __init apic_bsp_setup(bool upmode)
2355 {
2356 	int id;
2357 
2358 	connect_bsp_APIC();
2359 	if (upmode)
2360 		apic_bsp_up_setup();
2361 	setup_local_APIC();
2362 
2363 	if (x2apic_mode)
2364 		id = apic_read(APIC_LDR);
2365 	else
2366 		id = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
2367 
2368 	enable_IO_APIC();
2369 	end_local_APIC_setup();
2370 	irq_remap_enable_fault_handling();
2371 	setup_IO_APIC();
2372 	/* Setup local timer */
2373 	x86_init.timers.setup_percpu_clockev();
2374 	return id;
2375 }
2376 
2377 /*
2378  * This initializes the IO-APIC and APIC hardware if this is
2379  * a UP kernel.
2380  */
2381 int __init APIC_init_uniprocessor(void)
2382 {
2383 	if (disable_apic) {
2384 		pr_info("Apic disabled\n");
2385 		return -1;
2386 	}
2387 #ifdef CONFIG_X86_64
2388 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2389 		disable_apic = 1;
2390 		pr_info("Apic disabled by BIOS\n");
2391 		return -1;
2392 	}
2393 #else
2394 	if (!smp_found_config && !boot_cpu_has(X86_FEATURE_APIC))
2395 		return -1;
2396 
2397 	/*
2398 	 * Complain if the BIOS pretends there is one.
2399 	 */
2400 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
2401 	    APIC_INTEGRATED(boot_cpu_apic_version)) {
2402 		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
2403 			boot_cpu_physical_apicid);
2404 		return -1;
2405 	}
2406 #endif
2407 
2408 	if (!smp_found_config)
2409 		disable_ioapic_support();
2410 
2411 	default_setup_apic_routing();
2412 	apic_bsp_setup(true);
2413 	return 0;
2414 }
2415 
2416 #ifdef CONFIG_UP_LATE_INIT
2417 void __init up_late_init(void)
2418 {
2419 	APIC_init_uniprocessor();
2420 }
2421 #endif
2422 
2423 /*
2424  * Power management
2425  */
2426 #ifdef CONFIG_PM
2427 
2428 static struct {
2429 	/*
2430 	 * 'active' is true if the local APIC was enabled by us and
2431 	 * not the BIOS; this signifies that we are also responsible
2432 	 * for disabling it before entering apm/acpi suspend
2433 	 */
2434 	int active;
2435 	/* r/w apic fields */
2436 	unsigned int apic_id;
2437 	unsigned int apic_taskpri;
2438 	unsigned int apic_ldr;
2439 	unsigned int apic_dfr;
2440 	unsigned int apic_spiv;
2441 	unsigned int apic_lvtt;
2442 	unsigned int apic_lvtpc;
2443 	unsigned int apic_lvt0;
2444 	unsigned int apic_lvt1;
2445 	unsigned int apic_lvterr;
2446 	unsigned int apic_tmict;
2447 	unsigned int apic_tdcr;
2448 	unsigned int apic_thmr;
2449 	unsigned int apic_cmci;
2450 } apic_pm_state;
2451 
2452 static int lapic_suspend(void)
2453 {
2454 	unsigned long flags;
2455 	int maxlvt;
2456 
2457 	if (!apic_pm_state.active)
2458 		return 0;
2459 
2460 	maxlvt = lapic_get_maxlvt();
2461 
2462 	apic_pm_state.apic_id = apic_read(APIC_ID);
2463 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2464 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2465 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2466 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2467 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2468 	if (maxlvt >= 4)
2469 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2470 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2471 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2472 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2473 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2474 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2475 #ifdef CONFIG_X86_THERMAL_VECTOR
2476 	if (maxlvt >= 5)
2477 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2478 #endif
2479 #ifdef CONFIG_X86_MCE_INTEL
2480 	if (maxlvt >= 6)
2481 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2482 #endif
2483 
2484 	local_irq_save(flags);
2485 	disable_local_APIC();
2486 
2487 	irq_remapping_disable();
2488 
2489 	local_irq_restore(flags);
2490 	return 0;
2491 }
2492 
2493 static void lapic_resume(void)
2494 {
2495 	unsigned int l, h;
2496 	unsigned long flags;
2497 	int maxlvt;
2498 
2499 	if (!apic_pm_state.active)
2500 		return;
2501 
2502 	local_irq_save(flags);
2503 
2504 	/*
2505 	 * IO-APIC and PIC have their own resume routines.
2506 	 * We just mask them here to make sure the interrupt
2507 	 * subsystem is completely quiet while we enable x2apic
2508 	 * and interrupt-remapping.
2509 	 */
2510 	mask_ioapic_entries();
2511 	legacy_pic->mask_all();
2512 
2513 	if (x2apic_mode) {
2514 		__x2apic_enable();
2515 	} else {
2516 		/*
2517 		 * Make sure the APICBASE points to the right address
2518 		 *
2519 		 * FIXME! This will be wrong if we ever support suspend on
2520 		 * SMP! We'll need to do this as part of the CPU restore!
2521 		 */
2522 		if (boot_cpu_data.x86 >= 6) {
2523 			rdmsr(MSR_IA32_APICBASE, l, h);
2524 			l &= ~MSR_IA32_APICBASE_BASE;
2525 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2526 			wrmsr(MSR_IA32_APICBASE, l, h);
2527 		}
2528 	}
2529 
2530 	maxlvt = lapic_get_maxlvt();
2531 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2532 	apic_write(APIC_ID, apic_pm_state.apic_id);
2533 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2534 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2535 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2536 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2537 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2538 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2539 #ifdef CONFIG_X86_THERMAL_VECTOR
2540 	if (maxlvt >= 5)
2541 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2542 #endif
2543 #ifdef CONFIG_X86_MCE_INTEL
2544 	if (maxlvt >= 6)
2545 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2546 #endif
2547 	if (maxlvt >= 4)
2548 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2549 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2550 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2551 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2552 	apic_write(APIC_ESR, 0);
2553 	apic_read(APIC_ESR);
2554 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2555 	apic_write(APIC_ESR, 0);
2556 	apic_read(APIC_ESR);
2557 
2558 	irq_remapping_reenable(x2apic_mode);
2559 
2560 	local_irq_restore(flags);
2561 }
2562 
2563 /*
2564  * This device has no shutdown method - fully functioning local APICs
2565  * are needed on every CPU up until machine_halt/restart/poweroff.
2566  */
2567 
2568 static struct syscore_ops lapic_syscore_ops = {
2569 	.resume		= lapic_resume,
2570 	.suspend	= lapic_suspend,
2571 };
2572 
2573 static void apic_pm_activate(void)
2574 {
2575 	apic_pm_state.active = 1;
2576 }
2577 
2578 static int __init init_lapic_sysfs(void)
2579 {
2580 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2581 	if (boot_cpu_has(X86_FEATURE_APIC))
2582 		register_syscore_ops(&lapic_syscore_ops);
2583 
2584 	return 0;
2585 }
2586 
2587 /* local apic needs to resume before other devices access its registers. */
2588 core_initcall(init_lapic_sysfs);
2589 
2590 #else	/* CONFIG_PM */
2591 
2592 static void apic_pm_activate(void) { }
2593 
2594 #endif	/* CONFIG_PM */
2595 
2596 #ifdef CONFIG_X86_64
2597 
2598 static int multi_checked;
2599 static int multi;
2600 
2601 static int set_multi(const struct dmi_system_id *d)
2602 {
2603 	if (multi)
2604 		return 0;
2605 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2606 	multi = 1;
2607 	return 0;
2608 }
2609 
2610 static const struct dmi_system_id multi_dmi_table[] = {
2611 	{
2612 		.callback = set_multi,
2613 		.ident = "IBM System Summit2",
2614 		.matches = {
2615 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2616 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2617 		},
2618 	},
2619 	{}
2620 };
2621 
2622 static void dmi_check_multi(void)
2623 {
2624 	if (multi_checked)
2625 		return;
2626 
2627 	dmi_check_system(multi_dmi_table);
2628 	multi_checked = 1;
2629 }
2630 
2631 /*
2632  * apic_is_clustered_box() -- Check if we can expect good TSC
2633  *
2634  * Thus far, the major user of this is IBM's Summit2 series:
2635  * Clustered boxes may have unsynced TSC problems if they are
2636  * multi-chassis.
2637  * Use DMI to check them
2638  */
2639 int apic_is_clustered_box(void)
2640 {
2641 	dmi_check_multi();
2642 	return multi;
2643 }
2644 #endif
2645 
2646 /*
2647  * APIC command line parameters
2648  */
2649 static int __init setup_disableapic(char *arg)
2650 {
2651 	disable_apic = 1;
2652 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2653 	return 0;
2654 }
2655 early_param("disableapic", setup_disableapic);
2656 
2657 /* same as disableapic, for compatibility */
2658 static int __init setup_nolapic(char *arg)
2659 {
2660 	return setup_disableapic(arg);
2661 }
2662 early_param("nolapic", setup_nolapic);
2663 
2664 static int __init parse_lapic_timer_c2_ok(char *arg)
2665 {
2666 	local_apic_timer_c2_ok = 1;
2667 	return 0;
2668 }
2669 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2670 
2671 static int __init parse_disable_apic_timer(char *arg)
2672 {
2673 	disable_apic_timer = 1;
2674 	return 0;
2675 }
2676 early_param("noapictimer", parse_disable_apic_timer);
2677 
2678 static int __init parse_nolapic_timer(char *arg)
2679 {
2680 	disable_apic_timer = 1;
2681 	return 0;
2682 }
2683 early_param("nolapic_timer", parse_nolapic_timer);
2684 
2685 static int __init apic_set_verbosity(char *arg)
2686 {
2687 	if (!arg)  {
2688 #ifdef CONFIG_X86_64
2689 		skip_ioapic_setup = 0;
2690 		return 0;
2691 #endif
2692 		return -EINVAL;
2693 	}
2694 
2695 	if (strcmp("debug", arg) == 0)
2696 		apic_verbosity = APIC_DEBUG;
2697 	else if (strcmp("verbose", arg) == 0)
2698 		apic_verbosity = APIC_VERBOSE;
2699 	else {
2700 		pr_warning("APIC Verbosity level %s not recognised"
2701 			" use apic=verbose or apic=debug\n", arg);
2702 		return -EINVAL;
2703 	}
2704 
2705 	return 0;
2706 }
2707 early_param("apic", apic_set_verbosity);
2708 
2709 static int __init lapic_insert_resource(void)
2710 {
2711 	if (!apic_phys)
2712 		return -1;
2713 
2714 	/* Put local APIC into the resource map. */
2715 	lapic_resource.start = apic_phys;
2716 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2717 	insert_resource(&iomem_resource, &lapic_resource);
2718 
2719 	return 0;
2720 }
2721 
2722 /*
2723  * need call insert after e820__reserve_resources()
2724  * that is using request_resource
2725  */
2726 late_initcall(lapic_insert_resource);
2727 
2728 static int __init apic_set_disabled_cpu_apicid(char *arg)
2729 {
2730 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2731 		return -EINVAL;
2732 
2733 	return 0;
2734 }
2735 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2736 
2737 static int __init apic_set_extnmi(char *arg)
2738 {
2739 	if (!arg)
2740 		return -EINVAL;
2741 
2742 	if (!strncmp("all", arg, 3))
2743 		apic_extnmi = APIC_EXTNMI_ALL;
2744 	else if (!strncmp("none", arg, 4))
2745 		apic_extnmi = APIC_EXTNMI_NONE;
2746 	else if (!strncmp("bsp", arg, 3))
2747 		apic_extnmi = APIC_EXTNMI_BSP;
2748 	else {
2749 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2750 		return -EINVAL;
2751 	}
2752 
2753 	return 0;
2754 }
2755 early_param("apic_extnmi", apic_set_extnmi);
2756