xref: /openbmc/linux/arch/x86/kernel/apic/apic.c (revision 7211ec63)
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 /*
181  * Debug level, exported for io_apic.c
182  */
183 unsigned int apic_verbosity;
184 
185 int pic_mode;
186 
187 /* Have we found an MP table */
188 int smp_found_config;
189 
190 static struct resource lapic_resource = {
191 	.name = "Local APIC",
192 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
193 };
194 
195 unsigned int lapic_timer_frequency = 0;
196 
197 static void apic_pm_activate(void);
198 
199 static unsigned long apic_phys;
200 
201 /*
202  * Get the LAPIC version
203  */
204 static inline int lapic_get_version(void)
205 {
206 	return GET_APIC_VERSION(apic_read(APIC_LVR));
207 }
208 
209 /*
210  * Check, if the APIC is integrated or a separate chip
211  */
212 static inline int lapic_is_integrated(void)
213 {
214 #ifdef CONFIG_X86_64
215 	return 1;
216 #else
217 	return APIC_INTEGRATED(lapic_get_version());
218 #endif
219 }
220 
221 /*
222  * Check, whether this is a modern or a first generation APIC
223  */
224 static int modern_apic(void)
225 {
226 	/* AMD systems use old APIC versions, so check the CPU */
227 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
228 	    boot_cpu_data.x86 >= 0xf)
229 		return 1;
230 	return lapic_get_version() >= 0x14;
231 }
232 
233 /*
234  * right after this call apic become NOOP driven
235  * so apic->write/read doesn't do anything
236  */
237 static void __init apic_disable(void)
238 {
239 	pr_info("APIC: switched to apic NOOP\n");
240 	apic = &apic_noop;
241 }
242 
243 void native_apic_wait_icr_idle(void)
244 {
245 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
246 		cpu_relax();
247 }
248 
249 u32 native_safe_apic_wait_icr_idle(void)
250 {
251 	u32 send_status;
252 	int timeout;
253 
254 	timeout = 0;
255 	do {
256 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
257 		if (!send_status)
258 			break;
259 		inc_irq_stat(icr_read_retry_count);
260 		udelay(100);
261 	} while (timeout++ < 1000);
262 
263 	return send_status;
264 }
265 
266 void native_apic_icr_write(u32 low, u32 id)
267 {
268 	unsigned long flags;
269 
270 	local_irq_save(flags);
271 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
272 	apic_write(APIC_ICR, low);
273 	local_irq_restore(flags);
274 }
275 
276 u64 native_apic_icr_read(void)
277 {
278 	u32 icr1, icr2;
279 
280 	icr2 = apic_read(APIC_ICR2);
281 	icr1 = apic_read(APIC_ICR);
282 
283 	return icr1 | ((u64)icr2 << 32);
284 }
285 
286 #ifdef CONFIG_X86_32
287 /**
288  * get_physical_broadcast - Get number of physical broadcast IDs
289  */
290 int get_physical_broadcast(void)
291 {
292 	return modern_apic() ? 0xff : 0xf;
293 }
294 #endif
295 
296 /**
297  * lapic_get_maxlvt - get the maximum number of local vector table entries
298  */
299 int lapic_get_maxlvt(void)
300 {
301 	unsigned int v;
302 
303 	v = apic_read(APIC_LVR);
304 	/*
305 	 * - we always have APIC integrated on 64bit mode
306 	 * - 82489DXs do not report # of LVT entries
307 	 */
308 	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
309 }
310 
311 /*
312  * Local APIC timer
313  */
314 
315 /* Clock divisor */
316 #define APIC_DIVISOR 16
317 #define TSC_DIVISOR  8
318 
319 /*
320  * This function sets up the local APIC timer, with a timeout of
321  * 'clocks' APIC bus clock. During calibration we actually call
322  * this function twice on the boot CPU, once with a bogus timeout
323  * value, second time for real. The other (noncalibrating) CPUs
324  * call this function only once, with the real, calibrated value.
325  *
326  * We do reads before writes even if unnecessary, to get around the
327  * P5 APIC double write bug.
328  */
329 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
330 {
331 	unsigned int lvtt_value, tmp_value;
332 
333 	lvtt_value = LOCAL_TIMER_VECTOR;
334 	if (!oneshot)
335 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
336 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
337 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
338 
339 	if (!lapic_is_integrated())
340 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
341 
342 	if (!irqen)
343 		lvtt_value |= APIC_LVT_MASKED;
344 
345 	apic_write(APIC_LVTT, lvtt_value);
346 
347 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
348 		/*
349 		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
350 		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
351 		 * According to Intel, MFENCE can do the serialization here.
352 		 */
353 		asm volatile("mfence" : : : "memory");
354 
355 		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
356 		return;
357 	}
358 
359 	/*
360 	 * Divide PICLK by 16
361 	 */
362 	tmp_value = apic_read(APIC_TDCR);
363 	apic_write(APIC_TDCR,
364 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
365 		APIC_TDR_DIV_16);
366 
367 	if (!oneshot)
368 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
369 }
370 
371 /*
372  * Setup extended LVT, AMD specific
373  *
374  * Software should use the LVT offsets the BIOS provides.  The offsets
375  * are determined by the subsystems using it like those for MCE
376  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
377  * are supported. Beginning with family 10h at least 4 offsets are
378  * available.
379  *
380  * Since the offsets must be consistent for all cores, we keep track
381  * of the LVT offsets in software and reserve the offset for the same
382  * vector also to be used on other cores. An offset is freed by
383  * setting the entry to APIC_EILVT_MASKED.
384  *
385  * If the BIOS is right, there should be no conflicts. Otherwise a
386  * "[Firmware Bug]: ..." error message is generated. However, if
387  * software does not properly determines the offsets, it is not
388  * necessarily a BIOS bug.
389  */
390 
391 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
392 
393 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
394 {
395 	return (old & APIC_EILVT_MASKED)
396 		|| (new == APIC_EILVT_MASKED)
397 		|| ((new & ~APIC_EILVT_MASKED) == old);
398 }
399 
400 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
401 {
402 	unsigned int rsvd, vector;
403 
404 	if (offset >= APIC_EILVT_NR_MAX)
405 		return ~0;
406 
407 	rsvd = atomic_read(&eilvt_offsets[offset]);
408 	do {
409 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
410 		if (vector && !eilvt_entry_is_changeable(vector, new))
411 			/* may not change if vectors are different */
412 			return rsvd;
413 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
414 	} while (rsvd != new);
415 
416 	rsvd &= ~APIC_EILVT_MASKED;
417 	if (rsvd && rsvd != vector)
418 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
419 			offset, rsvd);
420 
421 	return new;
422 }
423 
424 /*
425  * If mask=1, the LVT entry does not generate interrupts while mask=0
426  * enables the vector. See also the BKDGs. Must be called with
427  * preemption disabled.
428  */
429 
430 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
431 {
432 	unsigned long reg = APIC_EILVTn(offset);
433 	unsigned int new, old, reserved;
434 
435 	new = (mask << 16) | (msg_type << 8) | vector;
436 	old = apic_read(reg);
437 	reserved = reserve_eilvt_offset(offset, new);
438 
439 	if (reserved != new) {
440 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
441 		       "vector 0x%x, but the register is already in use for "
442 		       "vector 0x%x on another cpu\n",
443 		       smp_processor_id(), reg, offset, new, reserved);
444 		return -EINVAL;
445 	}
446 
447 	if (!eilvt_entry_is_changeable(old, new)) {
448 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
449 		       "vector 0x%x, but the register is already in use for "
450 		       "vector 0x%x on this cpu\n",
451 		       smp_processor_id(), reg, offset, new, old);
452 		return -EBUSY;
453 	}
454 
455 	apic_write(reg, new);
456 
457 	return 0;
458 }
459 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
460 
461 /*
462  * Program the next event, relative to now
463  */
464 static int lapic_next_event(unsigned long delta,
465 			    struct clock_event_device *evt)
466 {
467 	apic_write(APIC_TMICT, delta);
468 	return 0;
469 }
470 
471 static int lapic_next_deadline(unsigned long delta,
472 			       struct clock_event_device *evt)
473 {
474 	u64 tsc;
475 
476 	tsc = rdtsc();
477 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
478 	return 0;
479 }
480 
481 static int lapic_timer_shutdown(struct clock_event_device *evt)
482 {
483 	unsigned int v;
484 
485 	/* Lapic used as dummy for broadcast ? */
486 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
487 		return 0;
488 
489 	v = apic_read(APIC_LVTT);
490 	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
491 	apic_write(APIC_LVTT, v);
492 	apic_write(APIC_TMICT, 0);
493 	return 0;
494 }
495 
496 static inline int
497 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
498 {
499 	/* Lapic used as dummy for broadcast ? */
500 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
501 		return 0;
502 
503 	__setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
504 	return 0;
505 }
506 
507 static int lapic_timer_set_periodic(struct clock_event_device *evt)
508 {
509 	return lapic_timer_set_periodic_oneshot(evt, false);
510 }
511 
512 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
513 {
514 	return lapic_timer_set_periodic_oneshot(evt, true);
515 }
516 
517 /*
518  * Local APIC timer broadcast function
519  */
520 static void lapic_timer_broadcast(const struct cpumask *mask)
521 {
522 #ifdef CONFIG_SMP
523 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
524 #endif
525 }
526 
527 
528 /*
529  * The local apic timer can be used for any function which is CPU local.
530  */
531 static struct clock_event_device lapic_clockevent = {
532 	.name				= "lapic",
533 	.features			= CLOCK_EVT_FEAT_PERIODIC |
534 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
535 					  | CLOCK_EVT_FEAT_DUMMY,
536 	.shift				= 32,
537 	.set_state_shutdown		= lapic_timer_shutdown,
538 	.set_state_periodic		= lapic_timer_set_periodic,
539 	.set_state_oneshot		= lapic_timer_set_oneshot,
540 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
541 	.set_next_event			= lapic_next_event,
542 	.broadcast			= lapic_timer_broadcast,
543 	.rating				= 100,
544 	.irq				= -1,
545 };
546 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
547 
548 #define DEADLINE_MODEL_MATCH_FUNC(model, func)	\
549 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
550 
551 #define DEADLINE_MODEL_MATCH_REV(model, rev)	\
552 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
553 
554 static u32 hsx_deadline_rev(void)
555 {
556 	switch (boot_cpu_data.x86_mask) {
557 	case 0x02: return 0x3a; /* EP */
558 	case 0x04: return 0x0f; /* EX */
559 	}
560 
561 	return ~0U;
562 }
563 
564 static u32 bdx_deadline_rev(void)
565 {
566 	switch (boot_cpu_data.x86_mask) {
567 	case 0x02: return 0x00000011;
568 	case 0x03: return 0x0700000e;
569 	case 0x04: return 0x0f00000c;
570 	case 0x05: return 0x0e000003;
571 	}
572 
573 	return ~0U;
574 }
575 
576 static const struct x86_cpu_id deadline_match[] = {
577 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X,	hsx_deadline_rev),
578 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X,	0x0b000020),
579 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_XEON_D,	bdx_deadline_rev),
580 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_X,	0x02000014),
581 
582 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE,	0x22),
583 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_ULT,	0x20),
584 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_GT3E,	0x17),
585 
586 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_CORE,	0x25),
587 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_GT3E,	0x17),
588 
589 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE,	0xb2),
590 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP,	0xb2),
591 
592 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE,	0x52),
593 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP,	0x52),
594 
595 	{},
596 };
597 
598 static void apic_check_deadline_errata(void)
599 {
600 	const struct x86_cpu_id *m;
601 	u32 rev;
602 
603 	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
604 		return;
605 
606 	m = x86_match_cpu(deadline_match);
607 	if (!m)
608 		return;
609 
610 	/*
611 	 * Function pointers will have the MSB set due to address layout,
612 	 * immediate revisions will not.
613 	 */
614 	if ((long)m->driver_data < 0)
615 		rev = ((u32 (*)(void))(m->driver_data))();
616 	else
617 		rev = (u32)m->driver_data;
618 
619 	if (boot_cpu_data.microcode >= rev)
620 		return;
621 
622 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
623 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
624 	       "please update microcode to version: 0x%x (or later)\n", rev);
625 }
626 
627 /*
628  * Setup the local APIC timer for this CPU. Copy the initialized values
629  * of the boot CPU and register the clock event in the framework.
630  */
631 static void setup_APIC_timer(void)
632 {
633 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
634 
635 	if (this_cpu_has(X86_FEATURE_ARAT)) {
636 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
637 		/* Make LAPIC timer preferrable over percpu HPET */
638 		lapic_clockevent.rating = 150;
639 	}
640 
641 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
642 	levt->cpumask = cpumask_of(smp_processor_id());
643 
644 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
645 		levt->name = "lapic-deadline";
646 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
647 				    CLOCK_EVT_FEAT_DUMMY);
648 		levt->set_next_event = lapic_next_deadline;
649 		clockevents_config_and_register(levt,
650 						tsc_khz * (1000 / TSC_DIVISOR),
651 						0xF, ~0UL);
652 	} else
653 		clockevents_register_device(levt);
654 }
655 
656 /*
657  * Install the updated TSC frequency from recalibration at the TSC
658  * deadline clockevent devices.
659  */
660 static void __lapic_update_tsc_freq(void *info)
661 {
662 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
663 
664 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
665 		return;
666 
667 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
668 }
669 
670 void lapic_update_tsc_freq(void)
671 {
672 	/*
673 	 * The clockevent device's ->mult and ->shift can both be
674 	 * changed. In order to avoid races, schedule the frequency
675 	 * update code on each CPU.
676 	 */
677 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
678 }
679 
680 /*
681  * In this functions we calibrate APIC bus clocks to the external timer.
682  *
683  * We want to do the calibration only once since we want to have local timer
684  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
685  * frequency.
686  *
687  * This was previously done by reading the PIT/HPET and waiting for a wrap
688  * around to find out, that a tick has elapsed. I have a box, where the PIT
689  * readout is broken, so it never gets out of the wait loop again. This was
690  * also reported by others.
691  *
692  * Monitoring the jiffies value is inaccurate and the clockevents
693  * infrastructure allows us to do a simple substitution of the interrupt
694  * handler.
695  *
696  * The calibration routine also uses the pm_timer when possible, as the PIT
697  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
698  * back to normal later in the boot process).
699  */
700 
701 #define LAPIC_CAL_LOOPS		(HZ/10)
702 
703 static __initdata int lapic_cal_loops = -1;
704 static __initdata long lapic_cal_t1, lapic_cal_t2;
705 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
706 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
707 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
708 
709 /*
710  * Temporary interrupt handler.
711  */
712 static void __init lapic_cal_handler(struct clock_event_device *dev)
713 {
714 	unsigned long long tsc = 0;
715 	long tapic = apic_read(APIC_TMCCT);
716 	unsigned long pm = acpi_pm_read_early();
717 
718 	if (boot_cpu_has(X86_FEATURE_TSC))
719 		tsc = rdtsc();
720 
721 	switch (lapic_cal_loops++) {
722 	case 0:
723 		lapic_cal_t1 = tapic;
724 		lapic_cal_tsc1 = tsc;
725 		lapic_cal_pm1 = pm;
726 		lapic_cal_j1 = jiffies;
727 		break;
728 
729 	case LAPIC_CAL_LOOPS:
730 		lapic_cal_t2 = tapic;
731 		lapic_cal_tsc2 = tsc;
732 		if (pm < lapic_cal_pm1)
733 			pm += ACPI_PM_OVRRUN;
734 		lapic_cal_pm2 = pm;
735 		lapic_cal_j2 = jiffies;
736 		break;
737 	}
738 }
739 
740 static int __init
741 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
742 {
743 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
744 	const long pm_thresh = pm_100ms / 100;
745 	unsigned long mult;
746 	u64 res;
747 
748 #ifndef CONFIG_X86_PM_TIMER
749 	return -1;
750 #endif
751 
752 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
753 
754 	/* Check, if the PM timer is available */
755 	if (!deltapm)
756 		return -1;
757 
758 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
759 
760 	if (deltapm > (pm_100ms - pm_thresh) &&
761 	    deltapm < (pm_100ms + pm_thresh)) {
762 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
763 		return 0;
764 	}
765 
766 	res = (((u64)deltapm) *  mult) >> 22;
767 	do_div(res, 1000000);
768 	pr_warning("APIC calibration not consistent "
769 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
770 
771 	/* Correct the lapic counter value */
772 	res = (((u64)(*delta)) * pm_100ms);
773 	do_div(res, deltapm);
774 	pr_info("APIC delta adjusted to PM-Timer: "
775 		"%lu (%ld)\n", (unsigned long)res, *delta);
776 	*delta = (long)res;
777 
778 	/* Correct the tsc counter value */
779 	if (boot_cpu_has(X86_FEATURE_TSC)) {
780 		res = (((u64)(*deltatsc)) * pm_100ms);
781 		do_div(res, deltapm);
782 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
783 					  "PM-Timer: %lu (%ld)\n",
784 					(unsigned long)res, *deltatsc);
785 		*deltatsc = (long)res;
786 	}
787 
788 	return 0;
789 }
790 
791 static int __init calibrate_APIC_clock(void)
792 {
793 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
794 	void (*real_handler)(struct clock_event_device *dev);
795 	unsigned long deltaj;
796 	long delta, deltatsc;
797 	int pm_referenced = 0;
798 
799 	/**
800 	 * check if lapic timer has already been calibrated by platform
801 	 * specific routine, such as tsc calibration code. if so, we just fill
802 	 * in the clockevent structure and return.
803 	 */
804 
805 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
806 		return 0;
807 	} else if (lapic_timer_frequency) {
808 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
809 				lapic_timer_frequency);
810 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
811 					TICK_NSEC, lapic_clockevent.shift);
812 		lapic_clockevent.max_delta_ns =
813 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
814 		lapic_clockevent.max_delta_ticks = 0x7FFFFF;
815 		lapic_clockevent.min_delta_ns =
816 			clockevent_delta2ns(0xF, &lapic_clockevent);
817 		lapic_clockevent.min_delta_ticks = 0xF;
818 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
819 		return 0;
820 	}
821 
822 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
823 		    "calibrating APIC timer ...\n");
824 
825 	local_irq_disable();
826 
827 	/* Replace the global interrupt handler */
828 	real_handler = global_clock_event->event_handler;
829 	global_clock_event->event_handler = lapic_cal_handler;
830 
831 	/*
832 	 * Setup the APIC counter to maximum. There is no way the lapic
833 	 * can underflow in the 100ms detection time frame
834 	 */
835 	__setup_APIC_LVTT(0xffffffff, 0, 0);
836 
837 	/* Let the interrupts run */
838 	local_irq_enable();
839 
840 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
841 		cpu_relax();
842 
843 	local_irq_disable();
844 
845 	/* Restore the real event handler */
846 	global_clock_event->event_handler = real_handler;
847 
848 	/* Build delta t1-t2 as apic timer counts down */
849 	delta = lapic_cal_t1 - lapic_cal_t2;
850 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
851 
852 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
853 
854 	/* we trust the PM based calibration if possible */
855 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
856 					&delta, &deltatsc);
857 
858 	/* Calculate the scaled math multiplication factor */
859 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
860 				       lapic_clockevent.shift);
861 	lapic_clockevent.max_delta_ns =
862 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
863 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
864 	lapic_clockevent.min_delta_ns =
865 		clockevent_delta2ns(0xF, &lapic_clockevent);
866 	lapic_clockevent.min_delta_ticks = 0xF;
867 
868 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
869 
870 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
871 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
872 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
873 		    lapic_timer_frequency);
874 
875 	if (boot_cpu_has(X86_FEATURE_TSC)) {
876 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
877 			    "%ld.%04ld MHz.\n",
878 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
879 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
880 	}
881 
882 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
883 		    "%u.%04u MHz.\n",
884 		    lapic_timer_frequency / (1000000 / HZ),
885 		    lapic_timer_frequency % (1000000 / HZ));
886 
887 	/*
888 	 * Do a sanity check on the APIC calibration result
889 	 */
890 	if (lapic_timer_frequency < (1000000 / HZ)) {
891 		local_irq_enable();
892 		pr_warning("APIC frequency too slow, disabling apic timer\n");
893 		return -1;
894 	}
895 
896 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
897 
898 	/*
899 	 * PM timer calibration failed or not turned on
900 	 * so lets try APIC timer based calibration
901 	 */
902 	if (!pm_referenced) {
903 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
904 
905 		/*
906 		 * Setup the apic timer manually
907 		 */
908 		levt->event_handler = lapic_cal_handler;
909 		lapic_timer_set_periodic(levt);
910 		lapic_cal_loops = -1;
911 
912 		/* Let the interrupts run */
913 		local_irq_enable();
914 
915 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
916 			cpu_relax();
917 
918 		/* Stop the lapic timer */
919 		local_irq_disable();
920 		lapic_timer_shutdown(levt);
921 
922 		/* Jiffies delta */
923 		deltaj = lapic_cal_j2 - lapic_cal_j1;
924 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
925 
926 		/* Check, if the jiffies result is consistent */
927 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
928 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
929 		else
930 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
931 	}
932 	local_irq_enable();
933 
934 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
935 		pr_warning("APIC timer disabled due to verification failure\n");
936 			return -1;
937 	}
938 
939 	return 0;
940 }
941 
942 /*
943  * Setup the boot APIC
944  *
945  * Calibrate and verify the result.
946  */
947 void __init setup_boot_APIC_clock(void)
948 {
949 	/*
950 	 * The local apic timer can be disabled via the kernel
951 	 * commandline or from the CPU detection code. Register the lapic
952 	 * timer as a dummy clock event source on SMP systems, so the
953 	 * broadcast mechanism is used. On UP systems simply ignore it.
954 	 */
955 	if (disable_apic_timer) {
956 		pr_info("Disabling APIC timer\n");
957 		/* No broadcast on UP ! */
958 		if (num_possible_cpus() > 1) {
959 			lapic_clockevent.mult = 1;
960 			setup_APIC_timer();
961 		}
962 		return;
963 	}
964 
965 	if (calibrate_APIC_clock()) {
966 		/* No broadcast on UP ! */
967 		if (num_possible_cpus() > 1)
968 			setup_APIC_timer();
969 		return;
970 	}
971 
972 	/*
973 	 * If nmi_watchdog is set to IO_APIC, we need the
974 	 * PIT/HPET going.  Otherwise register lapic as a dummy
975 	 * device.
976 	 */
977 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
978 
979 	/* Setup the lapic or request the broadcast */
980 	setup_APIC_timer();
981 	amd_e400_c1e_apic_setup();
982 }
983 
984 void setup_secondary_APIC_clock(void)
985 {
986 	setup_APIC_timer();
987 	amd_e400_c1e_apic_setup();
988 }
989 
990 /*
991  * The guts of the apic timer interrupt
992  */
993 static void local_apic_timer_interrupt(void)
994 {
995 	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
996 
997 	/*
998 	 * Normally we should not be here till LAPIC has been initialized but
999 	 * in some cases like kdump, its possible that there is a pending LAPIC
1000 	 * timer interrupt from previous kernel's context and is delivered in
1001 	 * new kernel the moment interrupts are enabled.
1002 	 *
1003 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1004 	 * its possible that when we get here evt->event_handler is NULL.
1005 	 * Check for event_handler being NULL and discard the interrupt as
1006 	 * spurious.
1007 	 */
1008 	if (!evt->event_handler) {
1009 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1010 			   smp_processor_id());
1011 		/* Switch it off */
1012 		lapic_timer_shutdown(evt);
1013 		return;
1014 	}
1015 
1016 	/*
1017 	 * the NMI deadlock-detector uses this.
1018 	 */
1019 	inc_irq_stat(apic_timer_irqs);
1020 
1021 	evt->event_handler(evt);
1022 }
1023 
1024 /*
1025  * Local APIC timer interrupt. This is the most natural way for doing
1026  * local interrupts, but local timer interrupts can be emulated by
1027  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1028  *
1029  * [ if a single-CPU system runs an SMP kernel then we call the local
1030  *   interrupt as well. Thus we cannot inline the local irq ... ]
1031  */
1032 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1033 {
1034 	struct pt_regs *old_regs = set_irq_regs(regs);
1035 
1036 	/*
1037 	 * NOTE! We'd better ACK the irq immediately,
1038 	 * because timer handling can be slow.
1039 	 *
1040 	 * update_process_times() expects us to have done irq_enter().
1041 	 * Besides, if we don't timer interrupts ignore the global
1042 	 * interrupt lock, which is the WrongThing (tm) to do.
1043 	 */
1044 	entering_ack_irq();
1045 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1046 	local_apic_timer_interrupt();
1047 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1048 	exiting_irq();
1049 
1050 	set_irq_regs(old_regs);
1051 }
1052 
1053 int setup_profiling_timer(unsigned int multiplier)
1054 {
1055 	return -EINVAL;
1056 }
1057 
1058 /*
1059  * Local APIC start and shutdown
1060  */
1061 
1062 /**
1063  * clear_local_APIC - shutdown the local APIC
1064  *
1065  * This is called, when a CPU is disabled and before rebooting, so the state of
1066  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1067  * leftovers during boot.
1068  */
1069 void clear_local_APIC(void)
1070 {
1071 	int maxlvt;
1072 	u32 v;
1073 
1074 	/* APIC hasn't been mapped yet */
1075 	if (!x2apic_mode && !apic_phys)
1076 		return;
1077 
1078 	maxlvt = lapic_get_maxlvt();
1079 	/*
1080 	 * Masking an LVT entry can trigger a local APIC error
1081 	 * if the vector is zero. Mask LVTERR first to prevent this.
1082 	 */
1083 	if (maxlvt >= 3) {
1084 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1085 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1086 	}
1087 	/*
1088 	 * Careful: we have to set masks only first to deassert
1089 	 * any level-triggered sources.
1090 	 */
1091 	v = apic_read(APIC_LVTT);
1092 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1093 	v = apic_read(APIC_LVT0);
1094 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1095 	v = apic_read(APIC_LVT1);
1096 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1097 	if (maxlvt >= 4) {
1098 		v = apic_read(APIC_LVTPC);
1099 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1100 	}
1101 
1102 	/* lets not touch this if we didn't frob it */
1103 #ifdef CONFIG_X86_THERMAL_VECTOR
1104 	if (maxlvt >= 5) {
1105 		v = apic_read(APIC_LVTTHMR);
1106 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1107 	}
1108 #endif
1109 #ifdef CONFIG_X86_MCE_INTEL
1110 	if (maxlvt >= 6) {
1111 		v = apic_read(APIC_LVTCMCI);
1112 		if (!(v & APIC_LVT_MASKED))
1113 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1114 	}
1115 #endif
1116 
1117 	/*
1118 	 * Clean APIC state for other OSs:
1119 	 */
1120 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1121 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1122 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1123 	if (maxlvt >= 3)
1124 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1125 	if (maxlvt >= 4)
1126 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1127 
1128 	/* Integrated APIC (!82489DX) ? */
1129 	if (lapic_is_integrated()) {
1130 		if (maxlvt > 3)
1131 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1132 			apic_write(APIC_ESR, 0);
1133 		apic_read(APIC_ESR);
1134 	}
1135 }
1136 
1137 /**
1138  * disable_local_APIC - clear and disable the local APIC
1139  */
1140 void disable_local_APIC(void)
1141 {
1142 	unsigned int value;
1143 
1144 	/* APIC hasn't been mapped yet */
1145 	if (!x2apic_mode && !apic_phys)
1146 		return;
1147 
1148 	clear_local_APIC();
1149 
1150 	/*
1151 	 * Disable APIC (implies clearing of registers
1152 	 * for 82489DX!).
1153 	 */
1154 	value = apic_read(APIC_SPIV);
1155 	value &= ~APIC_SPIV_APIC_ENABLED;
1156 	apic_write(APIC_SPIV, value);
1157 
1158 #ifdef CONFIG_X86_32
1159 	/*
1160 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1161 	 * restore the disabled state.
1162 	 */
1163 	if (enabled_via_apicbase) {
1164 		unsigned int l, h;
1165 
1166 		rdmsr(MSR_IA32_APICBASE, l, h);
1167 		l &= ~MSR_IA32_APICBASE_ENABLE;
1168 		wrmsr(MSR_IA32_APICBASE, l, h);
1169 	}
1170 #endif
1171 }
1172 
1173 /*
1174  * If Linux enabled the LAPIC against the BIOS default disable it down before
1175  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1176  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1177  * for the case where Linux didn't enable the LAPIC.
1178  */
1179 void lapic_shutdown(void)
1180 {
1181 	unsigned long flags;
1182 
1183 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1184 		return;
1185 
1186 	local_irq_save(flags);
1187 
1188 #ifdef CONFIG_X86_32
1189 	if (!enabled_via_apicbase)
1190 		clear_local_APIC();
1191 	else
1192 #endif
1193 		disable_local_APIC();
1194 
1195 
1196 	local_irq_restore(flags);
1197 }
1198 
1199 /**
1200  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1201  */
1202 void __init sync_Arb_IDs(void)
1203 {
1204 	/*
1205 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1206 	 * needed on AMD.
1207 	 */
1208 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1209 		return;
1210 
1211 	/*
1212 	 * Wait for idle.
1213 	 */
1214 	apic_wait_icr_idle();
1215 
1216 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1217 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1218 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1219 }
1220 
1221 /*
1222  * An initial setup of the virtual wire mode.
1223  */
1224 void __init init_bsp_APIC(void)
1225 {
1226 	unsigned int value;
1227 
1228 	/*
1229 	 * Don't do the setup now if we have a SMP BIOS as the
1230 	 * through-I/O-APIC virtual wire mode might be active.
1231 	 */
1232 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1233 		return;
1234 
1235 	/*
1236 	 * Do not trust the local APIC being empty at bootup.
1237 	 */
1238 	clear_local_APIC();
1239 
1240 	/*
1241 	 * Enable APIC.
1242 	 */
1243 	value = apic_read(APIC_SPIV);
1244 	value &= ~APIC_VECTOR_MASK;
1245 	value |= APIC_SPIV_APIC_ENABLED;
1246 
1247 #ifdef CONFIG_X86_32
1248 	/* This bit is reserved on P4/Xeon and should be cleared */
1249 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1250 	    (boot_cpu_data.x86 == 15))
1251 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1252 	else
1253 #endif
1254 		value |= APIC_SPIV_FOCUS_DISABLED;
1255 	value |= SPURIOUS_APIC_VECTOR;
1256 	apic_write(APIC_SPIV, value);
1257 
1258 	/*
1259 	 * Set up the virtual wire mode.
1260 	 */
1261 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1262 	value = APIC_DM_NMI;
1263 	if (!lapic_is_integrated())		/* 82489DX */
1264 		value |= APIC_LVT_LEVEL_TRIGGER;
1265 	if (apic_extnmi == APIC_EXTNMI_NONE)
1266 		value |= APIC_LVT_MASKED;
1267 	apic_write(APIC_LVT1, value);
1268 }
1269 
1270 static void lapic_setup_esr(void)
1271 {
1272 	unsigned int oldvalue, value, maxlvt;
1273 
1274 	if (!lapic_is_integrated()) {
1275 		pr_info("No ESR for 82489DX.\n");
1276 		return;
1277 	}
1278 
1279 	if (apic->disable_esr) {
1280 		/*
1281 		 * Something untraceable is creating bad interrupts on
1282 		 * secondary quads ... for the moment, just leave the
1283 		 * ESR disabled - we can't do anything useful with the
1284 		 * errors anyway - mbligh
1285 		 */
1286 		pr_info("Leaving ESR disabled.\n");
1287 		return;
1288 	}
1289 
1290 	maxlvt = lapic_get_maxlvt();
1291 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1292 		apic_write(APIC_ESR, 0);
1293 	oldvalue = apic_read(APIC_ESR);
1294 
1295 	/* enables sending errors */
1296 	value = ERROR_APIC_VECTOR;
1297 	apic_write(APIC_LVTERR, value);
1298 
1299 	/*
1300 	 * spec says clear errors after enabling vector.
1301 	 */
1302 	if (maxlvt > 3)
1303 		apic_write(APIC_ESR, 0);
1304 	value = apic_read(APIC_ESR);
1305 	if (value != oldvalue)
1306 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1307 			"vector: 0x%08x  after: 0x%08x\n",
1308 			oldvalue, value);
1309 }
1310 
1311 /**
1312  * setup_local_APIC - setup the local APIC
1313  *
1314  * Used to setup local APIC while initializing BSP or bringing up APs.
1315  * Always called with preemption disabled.
1316  */
1317 void setup_local_APIC(void)
1318 {
1319 	int cpu = smp_processor_id();
1320 	unsigned int value, queued;
1321 	int i, j, acked = 0;
1322 	unsigned long long tsc = 0, ntsc;
1323 	long long max_loops = cpu_khz ? cpu_khz : 1000000;
1324 
1325 	if (boot_cpu_has(X86_FEATURE_TSC))
1326 		tsc = rdtsc();
1327 
1328 	if (disable_apic) {
1329 		disable_ioapic_support();
1330 		return;
1331 	}
1332 
1333 #ifdef CONFIG_X86_32
1334 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1335 	if (lapic_is_integrated() && apic->disable_esr) {
1336 		apic_write(APIC_ESR, 0);
1337 		apic_write(APIC_ESR, 0);
1338 		apic_write(APIC_ESR, 0);
1339 		apic_write(APIC_ESR, 0);
1340 	}
1341 #endif
1342 	perf_events_lapic_init();
1343 
1344 	/*
1345 	 * Double-check whether this APIC is really registered.
1346 	 * This is meaningless in clustered apic mode, so we skip it.
1347 	 */
1348 	BUG_ON(!apic->apic_id_registered());
1349 
1350 	/*
1351 	 * Intel recommends to set DFR, LDR and TPR before enabling
1352 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1353 	 * document number 292116).  So here it goes...
1354 	 */
1355 	apic->init_apic_ldr();
1356 
1357 #ifdef CONFIG_X86_32
1358 	/*
1359 	 * APIC LDR is initialized.  If logical_apicid mapping was
1360 	 * initialized during get_smp_config(), make sure it matches the
1361 	 * actual value.
1362 	 */
1363 	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1364 	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1365 	/* always use the value from LDR */
1366 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1367 		logical_smp_processor_id();
1368 #endif
1369 
1370 	/*
1371 	 * Set Task Priority to 'accept all'. We never change this
1372 	 * later on.
1373 	 */
1374 	value = apic_read(APIC_TASKPRI);
1375 	value &= ~APIC_TPRI_MASK;
1376 	apic_write(APIC_TASKPRI, value);
1377 
1378 	/*
1379 	 * After a crash, we no longer service the interrupts and a pending
1380 	 * interrupt from previous kernel might still have ISR bit set.
1381 	 *
1382 	 * Most probably by now CPU has serviced that pending interrupt and
1383 	 * it might not have done the ack_APIC_irq() because it thought,
1384 	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1385 	 * does not clear the ISR bit and cpu thinks it has already serivced
1386 	 * the interrupt. Hence a vector might get locked. It was noticed
1387 	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1388 	 */
1389 	do {
1390 		queued = 0;
1391 		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1392 			queued |= apic_read(APIC_IRR + i*0x10);
1393 
1394 		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1395 			value = apic_read(APIC_ISR + i*0x10);
1396 			for (j = 31; j >= 0; j--) {
1397 				if (value & (1<<j)) {
1398 					ack_APIC_irq();
1399 					acked++;
1400 				}
1401 			}
1402 		}
1403 		if (acked > 256) {
1404 			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1405 			       acked);
1406 			break;
1407 		}
1408 		if (queued) {
1409 			if (boot_cpu_has(X86_FEATURE_TSC) && cpu_khz) {
1410 				ntsc = rdtsc();
1411 				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1412 			} else
1413 				max_loops--;
1414 		}
1415 	} while (queued && max_loops > 0);
1416 	WARN_ON(max_loops <= 0);
1417 
1418 	/*
1419 	 * Now that we are all set up, enable the APIC
1420 	 */
1421 	value = apic_read(APIC_SPIV);
1422 	value &= ~APIC_VECTOR_MASK;
1423 	/*
1424 	 * Enable APIC
1425 	 */
1426 	value |= APIC_SPIV_APIC_ENABLED;
1427 
1428 #ifdef CONFIG_X86_32
1429 	/*
1430 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1431 	 * certain networking cards. If high frequency interrupts are
1432 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1433 	 * entry is masked/unmasked at a high rate as well then sooner or
1434 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1435 	 * from the device. If focus CPU is disabled then the hang goes
1436 	 * away, oh well :-(
1437 	 *
1438 	 * [ This bug can be reproduced easily with a level-triggered
1439 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1440 	 *   BX chipset. ]
1441 	 */
1442 	/*
1443 	 * Actually disabling the focus CPU check just makes the hang less
1444 	 * frequent as it makes the interrupt distributon model be more
1445 	 * like LRU than MRU (the short-term load is more even across CPUs).
1446 	 */
1447 
1448 	/*
1449 	 * - enable focus processor (bit==0)
1450 	 * - 64bit mode always use processor focus
1451 	 *   so no need to set it
1452 	 */
1453 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1454 #endif
1455 
1456 	/*
1457 	 * Set spurious IRQ vector
1458 	 */
1459 	value |= SPURIOUS_APIC_VECTOR;
1460 	apic_write(APIC_SPIV, value);
1461 
1462 	/*
1463 	 * Set up LVT0, LVT1:
1464 	 *
1465 	 * set up through-local-APIC on the BP's LINT0. This is not
1466 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1467 	 * we delegate interrupts to the 8259A.
1468 	 */
1469 	/*
1470 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1471 	 */
1472 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1473 	if (!cpu && (pic_mode || !value)) {
1474 		value = APIC_DM_EXTINT;
1475 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1476 	} else {
1477 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1478 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1479 	}
1480 	apic_write(APIC_LVT0, value);
1481 
1482 	/*
1483 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1484 	 * modified by apic_extnmi= boot option.
1485 	 */
1486 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1487 	    apic_extnmi == APIC_EXTNMI_ALL)
1488 		value = APIC_DM_NMI;
1489 	else
1490 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1491 	if (!lapic_is_integrated())		/* 82489DX */
1492 		value |= APIC_LVT_LEVEL_TRIGGER;
1493 	apic_write(APIC_LVT1, value);
1494 
1495 #ifdef CONFIG_X86_MCE_INTEL
1496 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1497 	if (!cpu)
1498 		cmci_recheck();
1499 #endif
1500 }
1501 
1502 static void end_local_APIC_setup(void)
1503 {
1504 	lapic_setup_esr();
1505 
1506 #ifdef CONFIG_X86_32
1507 	{
1508 		unsigned int value;
1509 		/* Disable the local apic timer */
1510 		value = apic_read(APIC_LVTT);
1511 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1512 		apic_write(APIC_LVTT, value);
1513 	}
1514 #endif
1515 
1516 	apic_pm_activate();
1517 }
1518 
1519 /*
1520  * APIC setup function for application processors. Called from smpboot.c
1521  */
1522 void apic_ap_setup(void)
1523 {
1524 	setup_local_APIC();
1525 	end_local_APIC_setup();
1526 }
1527 
1528 #ifdef CONFIG_X86_X2APIC
1529 int x2apic_mode;
1530 
1531 enum {
1532 	X2APIC_OFF,
1533 	X2APIC_ON,
1534 	X2APIC_DISABLED,
1535 };
1536 static int x2apic_state;
1537 
1538 static void __x2apic_disable(void)
1539 {
1540 	u64 msr;
1541 
1542 	if (!boot_cpu_has(X86_FEATURE_APIC))
1543 		return;
1544 
1545 	rdmsrl(MSR_IA32_APICBASE, msr);
1546 	if (!(msr & X2APIC_ENABLE))
1547 		return;
1548 	/* Disable xapic and x2apic first and then reenable xapic mode */
1549 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1550 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1551 	printk_once(KERN_INFO "x2apic disabled\n");
1552 }
1553 
1554 static void __x2apic_enable(void)
1555 {
1556 	u64 msr;
1557 
1558 	rdmsrl(MSR_IA32_APICBASE, msr);
1559 	if (msr & X2APIC_ENABLE)
1560 		return;
1561 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1562 	printk_once(KERN_INFO "x2apic enabled\n");
1563 }
1564 
1565 static int __init setup_nox2apic(char *str)
1566 {
1567 	if (x2apic_enabled()) {
1568 		int apicid = native_apic_msr_read(APIC_ID);
1569 
1570 		if (apicid >= 255) {
1571 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1572 				   apicid);
1573 			return 0;
1574 		}
1575 		pr_warning("x2apic already enabled.\n");
1576 		__x2apic_disable();
1577 	}
1578 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1579 	x2apic_state = X2APIC_DISABLED;
1580 	x2apic_mode = 0;
1581 	return 0;
1582 }
1583 early_param("nox2apic", setup_nox2apic);
1584 
1585 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
1586 void x2apic_setup(void)
1587 {
1588 	/*
1589 	 * If x2apic is not in ON state, disable it if already enabled
1590 	 * from BIOS.
1591 	 */
1592 	if (x2apic_state != X2APIC_ON) {
1593 		__x2apic_disable();
1594 		return;
1595 	}
1596 	__x2apic_enable();
1597 }
1598 
1599 static __init void x2apic_disable(void)
1600 {
1601 	u32 x2apic_id, state = x2apic_state;
1602 
1603 	x2apic_mode = 0;
1604 	x2apic_state = X2APIC_DISABLED;
1605 
1606 	if (state != X2APIC_ON)
1607 		return;
1608 
1609 	x2apic_id = read_apic_id();
1610 	if (x2apic_id >= 255)
1611 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1612 
1613 	__x2apic_disable();
1614 	register_lapic_address(mp_lapic_addr);
1615 }
1616 
1617 static __init void x2apic_enable(void)
1618 {
1619 	if (x2apic_state != X2APIC_OFF)
1620 		return;
1621 
1622 	x2apic_mode = 1;
1623 	x2apic_state = X2APIC_ON;
1624 	__x2apic_enable();
1625 }
1626 
1627 static __init void try_to_enable_x2apic(int remap_mode)
1628 {
1629 	if (x2apic_state == X2APIC_DISABLED)
1630 		return;
1631 
1632 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1633 		/* IR is required if there is APIC ID > 255 even when running
1634 		 * under KVM
1635 		 */
1636 		if (max_physical_apicid > 255 ||
1637 		    !hypervisor_x2apic_available()) {
1638 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1639 			x2apic_disable();
1640 			return;
1641 		}
1642 
1643 		/*
1644 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1645 		 * only in physical mode
1646 		 */
1647 		x2apic_phys = 1;
1648 	}
1649 	x2apic_enable();
1650 }
1651 
1652 void __init check_x2apic(void)
1653 {
1654 	if (x2apic_enabled()) {
1655 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1656 		x2apic_mode = 1;
1657 		x2apic_state = X2APIC_ON;
1658 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1659 		x2apic_state = X2APIC_DISABLED;
1660 	}
1661 }
1662 #else /* CONFIG_X86_X2APIC */
1663 static int __init validate_x2apic(void)
1664 {
1665 	if (!apic_is_x2apic_enabled())
1666 		return 0;
1667 	/*
1668 	 * Checkme: Can we simply turn off x2apic here instead of panic?
1669 	 */
1670 	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1671 }
1672 early_initcall(validate_x2apic);
1673 
1674 static inline void try_to_enable_x2apic(int remap_mode) { }
1675 static inline void __x2apic_enable(void) { }
1676 #endif /* !CONFIG_X86_X2APIC */
1677 
1678 void __init enable_IR_x2apic(void)
1679 {
1680 	unsigned long flags;
1681 	int ret, ir_stat;
1682 
1683 	if (skip_ioapic_setup) {
1684 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1685 		return;
1686 	}
1687 
1688 	ir_stat = irq_remapping_prepare();
1689 	if (ir_stat < 0 && !x2apic_supported())
1690 		return;
1691 
1692 	ret = save_ioapic_entries();
1693 	if (ret) {
1694 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1695 		return;
1696 	}
1697 
1698 	local_irq_save(flags);
1699 	legacy_pic->mask_all();
1700 	mask_ioapic_entries();
1701 
1702 	/* If irq_remapping_prepare() succeeded, try to enable it */
1703 	if (ir_stat >= 0)
1704 		ir_stat = irq_remapping_enable();
1705 	/* ir_stat contains the remap mode or an error code */
1706 	try_to_enable_x2apic(ir_stat);
1707 
1708 	if (ir_stat < 0)
1709 		restore_ioapic_entries();
1710 	legacy_pic->restore_mask();
1711 	local_irq_restore(flags);
1712 }
1713 
1714 #ifdef CONFIG_X86_64
1715 /*
1716  * Detect and enable local APICs on non-SMP boards.
1717  * Original code written by Keir Fraser.
1718  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1719  * not correctly set up (usually the APIC timer won't work etc.)
1720  */
1721 static int __init detect_init_APIC(void)
1722 {
1723 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1724 		pr_info("No local APIC present\n");
1725 		return -1;
1726 	}
1727 
1728 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1729 	return 0;
1730 }
1731 #else
1732 
1733 static int __init apic_verify(void)
1734 {
1735 	u32 features, h, l;
1736 
1737 	/*
1738 	 * The APIC feature bit should now be enabled
1739 	 * in `cpuid'
1740 	 */
1741 	features = cpuid_edx(1);
1742 	if (!(features & (1 << X86_FEATURE_APIC))) {
1743 		pr_warning("Could not enable APIC!\n");
1744 		return -1;
1745 	}
1746 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1747 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1748 
1749 	/* The BIOS may have set up the APIC at some other address */
1750 	if (boot_cpu_data.x86 >= 6) {
1751 		rdmsr(MSR_IA32_APICBASE, l, h);
1752 		if (l & MSR_IA32_APICBASE_ENABLE)
1753 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1754 	}
1755 
1756 	pr_info("Found and enabled local APIC!\n");
1757 	return 0;
1758 }
1759 
1760 int __init apic_force_enable(unsigned long addr)
1761 {
1762 	u32 h, l;
1763 
1764 	if (disable_apic)
1765 		return -1;
1766 
1767 	/*
1768 	 * Some BIOSes disable the local APIC in the APIC_BASE
1769 	 * MSR. This can only be done in software for Intel P6 or later
1770 	 * and AMD K7 (Model > 1) or later.
1771 	 */
1772 	if (boot_cpu_data.x86 >= 6) {
1773 		rdmsr(MSR_IA32_APICBASE, l, h);
1774 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1775 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1776 			l &= ~MSR_IA32_APICBASE_BASE;
1777 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1778 			wrmsr(MSR_IA32_APICBASE, l, h);
1779 			enabled_via_apicbase = 1;
1780 		}
1781 	}
1782 	return apic_verify();
1783 }
1784 
1785 /*
1786  * Detect and initialize APIC
1787  */
1788 static int __init detect_init_APIC(void)
1789 {
1790 	/* Disabled by kernel option? */
1791 	if (disable_apic)
1792 		return -1;
1793 
1794 	switch (boot_cpu_data.x86_vendor) {
1795 	case X86_VENDOR_AMD:
1796 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1797 		    (boot_cpu_data.x86 >= 15))
1798 			break;
1799 		goto no_apic;
1800 	case X86_VENDOR_INTEL:
1801 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1802 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
1803 			break;
1804 		goto no_apic;
1805 	default:
1806 		goto no_apic;
1807 	}
1808 
1809 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1810 		/*
1811 		 * Over-ride BIOS and try to enable the local APIC only if
1812 		 * "lapic" specified.
1813 		 */
1814 		if (!force_enable_local_apic) {
1815 			pr_info("Local APIC disabled by BIOS -- "
1816 				"you can enable it with \"lapic\"\n");
1817 			return -1;
1818 		}
1819 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1820 			return -1;
1821 	} else {
1822 		if (apic_verify())
1823 			return -1;
1824 	}
1825 
1826 	apic_pm_activate();
1827 
1828 	return 0;
1829 
1830 no_apic:
1831 	pr_info("No local APIC present or hardware disabled\n");
1832 	return -1;
1833 }
1834 #endif
1835 
1836 /**
1837  * init_apic_mappings - initialize APIC mappings
1838  */
1839 void __init init_apic_mappings(void)
1840 {
1841 	unsigned int new_apicid;
1842 
1843 	apic_check_deadline_errata();
1844 
1845 	if (x2apic_mode) {
1846 		boot_cpu_physical_apicid = read_apic_id();
1847 		return;
1848 	}
1849 
1850 	/* If no local APIC can be found return early */
1851 	if (!smp_found_config && detect_init_APIC()) {
1852 		/* lets NOP'ify apic operations */
1853 		pr_info("APIC: disable apic facility\n");
1854 		apic_disable();
1855 	} else {
1856 		apic_phys = mp_lapic_addr;
1857 
1858 		/*
1859 		 * If the system has ACPI MADT tables or MP info, the LAPIC
1860 		 * address is already registered.
1861 		 */
1862 		if (!acpi_lapic && !smp_found_config)
1863 			register_lapic_address(apic_phys);
1864 	}
1865 
1866 	/*
1867 	 * Fetch the APIC ID of the BSP in case we have a
1868 	 * default configuration (or the MP table is broken).
1869 	 */
1870 	new_apicid = read_apic_id();
1871 	if (boot_cpu_physical_apicid != new_apicid) {
1872 		boot_cpu_physical_apicid = new_apicid;
1873 		/*
1874 		 * yeah -- we lie about apic_version
1875 		 * in case if apic was disabled via boot option
1876 		 * but it's not a problem for SMP compiled kernel
1877 		 * since smp_sanity_check is prepared for such a case
1878 		 * and disable smp mode
1879 		 */
1880 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1881 	}
1882 }
1883 
1884 void __init register_lapic_address(unsigned long address)
1885 {
1886 	mp_lapic_addr = address;
1887 
1888 	if (!x2apic_mode) {
1889 		set_fixmap_nocache(FIX_APIC_BASE, address);
1890 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1891 			    APIC_BASE, address);
1892 	}
1893 	if (boot_cpu_physical_apicid == -1U) {
1894 		boot_cpu_physical_apicid  = read_apic_id();
1895 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1896 	}
1897 }
1898 
1899 /*
1900  * Local APIC interrupts
1901  */
1902 
1903 /*
1904  * This interrupt should _never_ happen with our APIC/SMP architecture
1905  */
1906 __visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
1907 {
1908 	u8 vector = ~regs->orig_ax;
1909 	u32 v;
1910 
1911 	entering_irq();
1912 	trace_spurious_apic_entry(vector);
1913 
1914 	/*
1915 	 * Check if this really is a spurious interrupt and ACK it
1916 	 * if it is a vectored one.  Just in case...
1917 	 * Spurious interrupts should not be ACKed.
1918 	 */
1919 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
1920 	if (v & (1 << (vector & 0x1f)))
1921 		ack_APIC_irq();
1922 
1923 	inc_irq_stat(irq_spurious_count);
1924 
1925 	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1926 	pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
1927 		"should never happen.\n", vector, smp_processor_id());
1928 
1929 	trace_spurious_apic_exit(vector);
1930 	exiting_irq();
1931 }
1932 
1933 /*
1934  * This interrupt should never happen with our APIC/SMP architecture
1935  */
1936 __visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
1937 {
1938 	static const char * const error_interrupt_reason[] = {
1939 		"Send CS error",		/* APIC Error Bit 0 */
1940 		"Receive CS error",		/* APIC Error Bit 1 */
1941 		"Send accept error",		/* APIC Error Bit 2 */
1942 		"Receive accept error",		/* APIC Error Bit 3 */
1943 		"Redirectable IPI",		/* APIC Error Bit 4 */
1944 		"Send illegal vector",		/* APIC Error Bit 5 */
1945 		"Received illegal vector",	/* APIC Error Bit 6 */
1946 		"Illegal register address",	/* APIC Error Bit 7 */
1947 	};
1948 	u32 v, i = 0;
1949 
1950 	entering_irq();
1951 	trace_error_apic_entry(ERROR_APIC_VECTOR);
1952 
1953 	/* First tickle the hardware, only then report what went on. -- REW */
1954 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
1955 		apic_write(APIC_ESR, 0);
1956 	v = apic_read(APIC_ESR);
1957 	ack_APIC_irq();
1958 	atomic_inc(&irq_err_count);
1959 
1960 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
1961 		    smp_processor_id(), v);
1962 
1963 	v &= 0xff;
1964 	while (v) {
1965 		if (v & 0x1)
1966 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1967 		i++;
1968 		v >>= 1;
1969 	}
1970 
1971 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
1972 
1973 	trace_error_apic_exit(ERROR_APIC_VECTOR);
1974 	exiting_irq();
1975 }
1976 
1977 /**
1978  * connect_bsp_APIC - attach the APIC to the interrupt system
1979  */
1980 static void __init connect_bsp_APIC(void)
1981 {
1982 #ifdef CONFIG_X86_32
1983 	if (pic_mode) {
1984 		/*
1985 		 * Do not trust the local APIC being empty at bootup.
1986 		 */
1987 		clear_local_APIC();
1988 		/*
1989 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
1990 		 * local APIC to INT and NMI lines.
1991 		 */
1992 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1993 				"enabling APIC mode.\n");
1994 		imcr_pic_to_apic();
1995 	}
1996 #endif
1997 }
1998 
1999 /**
2000  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2001  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2002  *
2003  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2004  * APIC is disabled.
2005  */
2006 void disconnect_bsp_APIC(int virt_wire_setup)
2007 {
2008 	unsigned int value;
2009 
2010 #ifdef CONFIG_X86_32
2011 	if (pic_mode) {
2012 		/*
2013 		 * Put the board back into PIC mode (has an effect only on
2014 		 * certain older boards).  Note that APIC interrupts, including
2015 		 * IPIs, won't work beyond this point!  The only exception are
2016 		 * INIT IPIs.
2017 		 */
2018 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2019 				"entering PIC mode.\n");
2020 		imcr_apic_to_pic();
2021 		return;
2022 	}
2023 #endif
2024 
2025 	/* Go back to Virtual Wire compatibility mode */
2026 
2027 	/* For the spurious interrupt use vector F, and enable it */
2028 	value = apic_read(APIC_SPIV);
2029 	value &= ~APIC_VECTOR_MASK;
2030 	value |= APIC_SPIV_APIC_ENABLED;
2031 	value |= 0xf;
2032 	apic_write(APIC_SPIV, value);
2033 
2034 	if (!virt_wire_setup) {
2035 		/*
2036 		 * For LVT0 make it edge triggered, active high,
2037 		 * external and enabled
2038 		 */
2039 		value = apic_read(APIC_LVT0);
2040 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2041 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2042 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2043 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2044 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2045 		apic_write(APIC_LVT0, value);
2046 	} else {
2047 		/* Disable LVT0 */
2048 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2049 	}
2050 
2051 	/*
2052 	 * For LVT1 make it edge triggered, active high,
2053 	 * nmi and enabled
2054 	 */
2055 	value = apic_read(APIC_LVT1);
2056 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2057 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2058 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2059 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2060 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2061 	apic_write(APIC_LVT1, value);
2062 }
2063 
2064 /*
2065  * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2066  * contiguously, it equals to current allocated max logical CPU ID plus 1.
2067  * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2068  * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2069  *
2070  * NOTE: Reserve 0 for BSP.
2071  */
2072 static int nr_logical_cpuids = 1;
2073 
2074 /*
2075  * Used to store mapping between logical CPU IDs and APIC IDs.
2076  */
2077 static int cpuid_to_apicid[] = {
2078 	[0 ... NR_CPUS - 1] = -1,
2079 };
2080 
2081 /*
2082  * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2083  * and cpuid_to_apicid[] synchronized.
2084  */
2085 static int allocate_logical_cpuid(int apicid)
2086 {
2087 	int i;
2088 
2089 	/*
2090 	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2091 	 * check if the kernel has allocated a cpuid for it.
2092 	 */
2093 	for (i = 0; i < nr_logical_cpuids; i++) {
2094 		if (cpuid_to_apicid[i] == apicid)
2095 			return i;
2096 	}
2097 
2098 	/* Allocate a new cpuid. */
2099 	if (nr_logical_cpuids >= nr_cpu_ids) {
2100 		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2101 			     "Processor %d/0x%x and the rest are ignored.\n",
2102 			     nr_cpu_ids, nr_logical_cpuids, apicid);
2103 		return -EINVAL;
2104 	}
2105 
2106 	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2107 	return nr_logical_cpuids++;
2108 }
2109 
2110 int 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 	 * boot_cpu_physical_apicid is designed to have the apicid
2118 	 * returned by read_apic_id(), i.e, the apicid of the
2119 	 * currently booting-up processor. However, on some platforms,
2120 	 * it is temporarily modified by the apicid reported as BSP
2121 	 * through MP table. Concretely:
2122 	 *
2123 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2124 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2125 	 *
2126 	 * This function is executed with the modified
2127 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2128 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2129 	 *
2130 	 * Since fixing handling of boot_cpu_physical_apicid requires
2131 	 * another discussion and tests on each platform, we leave it
2132 	 * for now and here we use read_apic_id() directly in this
2133 	 * function, generic_processor_info().
2134 	 */
2135 	if (disabled_cpu_apicid != BAD_APICID &&
2136 	    disabled_cpu_apicid != read_apic_id() &&
2137 	    disabled_cpu_apicid == apicid) {
2138 		int thiscpu = num_processors + disabled_cpus;
2139 
2140 		pr_warning("APIC: Disabling requested cpu."
2141 			   " Processor %d/0x%x ignored.\n",
2142 			   thiscpu, apicid);
2143 
2144 		disabled_cpus++;
2145 		return -ENODEV;
2146 	}
2147 
2148 	/*
2149 	 * If boot cpu has not been detected yet, then only allow upto
2150 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2151 	 */
2152 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2153 	    apicid != boot_cpu_physical_apicid) {
2154 		int thiscpu = max + disabled_cpus - 1;
2155 
2156 		pr_warning(
2157 			"APIC: NR_CPUS/possible_cpus limit of %i almost"
2158 			" reached. Keeping one slot for boot cpu."
2159 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2160 
2161 		disabled_cpus++;
2162 		return -ENODEV;
2163 	}
2164 
2165 	if (num_processors >= nr_cpu_ids) {
2166 		int thiscpu = max + disabled_cpus;
2167 
2168 		pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2169 			   "reached. Processor %d/0x%x ignored.\n",
2170 			   max, thiscpu, apicid);
2171 
2172 		disabled_cpus++;
2173 		return -EINVAL;
2174 	}
2175 
2176 	if (apicid == boot_cpu_physical_apicid) {
2177 		/*
2178 		 * x86_bios_cpu_apicid is required to have processors listed
2179 		 * in same order as logical cpu numbers. Hence the first
2180 		 * entry is BSP, and so on.
2181 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2182 		 * for BSP.
2183 		 */
2184 		cpu = 0;
2185 
2186 		/* Logical cpuid 0 is reserved for BSP. */
2187 		cpuid_to_apicid[0] = apicid;
2188 	} else {
2189 		cpu = allocate_logical_cpuid(apicid);
2190 		if (cpu < 0) {
2191 			disabled_cpus++;
2192 			return -EINVAL;
2193 		}
2194 	}
2195 
2196 	/*
2197 	 * Validate version
2198 	 */
2199 	if (version == 0x0) {
2200 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2201 			   cpu, apicid);
2202 		version = 0x10;
2203 	}
2204 
2205 	if (version != boot_cpu_apic_version) {
2206 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2207 			boot_cpu_apic_version, cpu, version);
2208 	}
2209 
2210 	if (apicid > max_physical_apicid)
2211 		max_physical_apicid = apicid;
2212 
2213 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2214 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2215 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2216 #endif
2217 #ifdef CONFIG_X86_32
2218 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2219 		apic->x86_32_early_logical_apicid(cpu);
2220 #endif
2221 	set_cpu_possible(cpu, true);
2222 	physid_set(apicid, phys_cpu_present_map);
2223 	set_cpu_present(cpu, true);
2224 	num_processors++;
2225 
2226 	return cpu;
2227 }
2228 
2229 int hard_smp_processor_id(void)
2230 {
2231 	return read_apic_id();
2232 }
2233 
2234 void default_init_apic_ldr(void)
2235 {
2236 	unsigned long val;
2237 
2238 	apic_write(APIC_DFR, APIC_DFR_VALUE);
2239 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2240 	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2241 	apic_write(APIC_LDR, val);
2242 }
2243 
2244 int default_cpu_mask_to_apicid(const struct cpumask *mask,
2245 			       struct irq_data *irqdata,
2246 			       unsigned int *apicid)
2247 {
2248 	unsigned int cpu = cpumask_first(mask);
2249 
2250 	if (cpu >= nr_cpu_ids)
2251 		return -EINVAL;
2252 	*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2253 	irq_data_update_effective_affinity(irqdata, cpumask_of(cpu));
2254 	return 0;
2255 }
2256 
2257 int flat_cpu_mask_to_apicid(const struct cpumask *mask,
2258 			    struct irq_data *irqdata,
2259 			    unsigned int *apicid)
2260 
2261 {
2262 	struct cpumask *effmsk = irq_data_get_effective_affinity_mask(irqdata);
2263 	unsigned long cpu_mask = cpumask_bits(mask)[0] & APIC_ALL_CPUS;
2264 
2265 	if (!cpu_mask)
2266 		return -EINVAL;
2267 	*apicid = (unsigned int)cpu_mask;
2268 	cpumask_bits(effmsk)[0] = cpu_mask;
2269 	return 0;
2270 }
2271 
2272 /*
2273  * Override the generic EOI implementation with an optimized version.
2274  * Only called during early boot when only one CPU is active and with
2275  * interrupts disabled, so we know this does not race with actual APIC driver
2276  * use.
2277  */
2278 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2279 {
2280 	struct apic **drv;
2281 
2282 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2283 		/* Should happen once for each apic */
2284 		WARN_ON((*drv)->eoi_write == eoi_write);
2285 		(*drv)->native_eoi_write = (*drv)->eoi_write;
2286 		(*drv)->eoi_write = eoi_write;
2287 	}
2288 }
2289 
2290 static void __init apic_bsp_up_setup(void)
2291 {
2292 #ifdef CONFIG_X86_64
2293 	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2294 #else
2295 	/*
2296 	 * Hack: In case of kdump, after a crash, kernel might be booting
2297 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2298 	 * might be zero if read from MP tables. Get it from LAPIC.
2299 	 */
2300 # ifdef CONFIG_CRASH_DUMP
2301 	boot_cpu_physical_apicid = read_apic_id();
2302 # endif
2303 #endif
2304 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2305 }
2306 
2307 /**
2308  * apic_bsp_setup - Setup function for local apic and io-apic
2309  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2310  *
2311  * Returns:
2312  * apic_id of BSP APIC
2313  */
2314 int __init apic_bsp_setup(bool upmode)
2315 {
2316 	int id;
2317 
2318 	connect_bsp_APIC();
2319 	if (upmode)
2320 		apic_bsp_up_setup();
2321 	setup_local_APIC();
2322 
2323 	if (x2apic_mode)
2324 		id = apic_read(APIC_LDR);
2325 	else
2326 		id = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
2327 
2328 	enable_IO_APIC();
2329 	end_local_APIC_setup();
2330 	irq_remap_enable_fault_handling();
2331 	setup_IO_APIC();
2332 	/* Setup local timer */
2333 	x86_init.timers.setup_percpu_clockev();
2334 	return id;
2335 }
2336 
2337 /*
2338  * This initializes the IO-APIC and APIC hardware if this is
2339  * a UP kernel.
2340  */
2341 int __init APIC_init_uniprocessor(void)
2342 {
2343 	if (disable_apic) {
2344 		pr_info("Apic disabled\n");
2345 		return -1;
2346 	}
2347 #ifdef CONFIG_X86_64
2348 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2349 		disable_apic = 1;
2350 		pr_info("Apic disabled by BIOS\n");
2351 		return -1;
2352 	}
2353 #else
2354 	if (!smp_found_config && !boot_cpu_has(X86_FEATURE_APIC))
2355 		return -1;
2356 
2357 	/*
2358 	 * Complain if the BIOS pretends there is one.
2359 	 */
2360 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
2361 	    APIC_INTEGRATED(boot_cpu_apic_version)) {
2362 		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
2363 			boot_cpu_physical_apicid);
2364 		return -1;
2365 	}
2366 #endif
2367 
2368 	if (!smp_found_config)
2369 		disable_ioapic_support();
2370 
2371 	default_setup_apic_routing();
2372 	apic_bsp_setup(true);
2373 	return 0;
2374 }
2375 
2376 #ifdef CONFIG_UP_LATE_INIT
2377 void __init up_late_init(void)
2378 {
2379 	APIC_init_uniprocessor();
2380 }
2381 #endif
2382 
2383 /*
2384  * Power management
2385  */
2386 #ifdef CONFIG_PM
2387 
2388 static struct {
2389 	/*
2390 	 * 'active' is true if the local APIC was enabled by us and
2391 	 * not the BIOS; this signifies that we are also responsible
2392 	 * for disabling it before entering apm/acpi suspend
2393 	 */
2394 	int active;
2395 	/* r/w apic fields */
2396 	unsigned int apic_id;
2397 	unsigned int apic_taskpri;
2398 	unsigned int apic_ldr;
2399 	unsigned int apic_dfr;
2400 	unsigned int apic_spiv;
2401 	unsigned int apic_lvtt;
2402 	unsigned int apic_lvtpc;
2403 	unsigned int apic_lvt0;
2404 	unsigned int apic_lvt1;
2405 	unsigned int apic_lvterr;
2406 	unsigned int apic_tmict;
2407 	unsigned int apic_tdcr;
2408 	unsigned int apic_thmr;
2409 	unsigned int apic_cmci;
2410 } apic_pm_state;
2411 
2412 static int lapic_suspend(void)
2413 {
2414 	unsigned long flags;
2415 	int maxlvt;
2416 
2417 	if (!apic_pm_state.active)
2418 		return 0;
2419 
2420 	maxlvt = lapic_get_maxlvt();
2421 
2422 	apic_pm_state.apic_id = apic_read(APIC_ID);
2423 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2424 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2425 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2426 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2427 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2428 	if (maxlvt >= 4)
2429 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2430 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2431 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2432 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2433 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2434 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2435 #ifdef CONFIG_X86_THERMAL_VECTOR
2436 	if (maxlvt >= 5)
2437 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2438 #endif
2439 #ifdef CONFIG_X86_MCE_INTEL
2440 	if (maxlvt >= 6)
2441 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2442 #endif
2443 
2444 	local_irq_save(flags);
2445 	disable_local_APIC();
2446 
2447 	irq_remapping_disable();
2448 
2449 	local_irq_restore(flags);
2450 	return 0;
2451 }
2452 
2453 static void lapic_resume(void)
2454 {
2455 	unsigned int l, h;
2456 	unsigned long flags;
2457 	int maxlvt;
2458 
2459 	if (!apic_pm_state.active)
2460 		return;
2461 
2462 	local_irq_save(flags);
2463 
2464 	/*
2465 	 * IO-APIC and PIC have their own resume routines.
2466 	 * We just mask them here to make sure the interrupt
2467 	 * subsystem is completely quiet while we enable x2apic
2468 	 * and interrupt-remapping.
2469 	 */
2470 	mask_ioapic_entries();
2471 	legacy_pic->mask_all();
2472 
2473 	if (x2apic_mode) {
2474 		__x2apic_enable();
2475 	} else {
2476 		/*
2477 		 * Make sure the APICBASE points to the right address
2478 		 *
2479 		 * FIXME! This will be wrong if we ever support suspend on
2480 		 * SMP! We'll need to do this as part of the CPU restore!
2481 		 */
2482 		if (boot_cpu_data.x86 >= 6) {
2483 			rdmsr(MSR_IA32_APICBASE, l, h);
2484 			l &= ~MSR_IA32_APICBASE_BASE;
2485 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2486 			wrmsr(MSR_IA32_APICBASE, l, h);
2487 		}
2488 	}
2489 
2490 	maxlvt = lapic_get_maxlvt();
2491 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2492 	apic_write(APIC_ID, apic_pm_state.apic_id);
2493 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2494 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2495 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2496 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2497 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2498 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2499 #ifdef CONFIG_X86_THERMAL_VECTOR
2500 	if (maxlvt >= 5)
2501 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2502 #endif
2503 #ifdef CONFIG_X86_MCE_INTEL
2504 	if (maxlvt >= 6)
2505 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2506 #endif
2507 	if (maxlvt >= 4)
2508 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2509 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2510 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2511 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2512 	apic_write(APIC_ESR, 0);
2513 	apic_read(APIC_ESR);
2514 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2515 	apic_write(APIC_ESR, 0);
2516 	apic_read(APIC_ESR);
2517 
2518 	irq_remapping_reenable(x2apic_mode);
2519 
2520 	local_irq_restore(flags);
2521 }
2522 
2523 /*
2524  * This device has no shutdown method - fully functioning local APICs
2525  * are needed on every CPU up until machine_halt/restart/poweroff.
2526  */
2527 
2528 static struct syscore_ops lapic_syscore_ops = {
2529 	.resume		= lapic_resume,
2530 	.suspend	= lapic_suspend,
2531 };
2532 
2533 static void apic_pm_activate(void)
2534 {
2535 	apic_pm_state.active = 1;
2536 }
2537 
2538 static int __init init_lapic_sysfs(void)
2539 {
2540 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2541 	if (boot_cpu_has(X86_FEATURE_APIC))
2542 		register_syscore_ops(&lapic_syscore_ops);
2543 
2544 	return 0;
2545 }
2546 
2547 /* local apic needs to resume before other devices access its registers. */
2548 core_initcall(init_lapic_sysfs);
2549 
2550 #else	/* CONFIG_PM */
2551 
2552 static void apic_pm_activate(void) { }
2553 
2554 #endif	/* CONFIG_PM */
2555 
2556 #ifdef CONFIG_X86_64
2557 
2558 static int multi_checked;
2559 static int multi;
2560 
2561 static int set_multi(const struct dmi_system_id *d)
2562 {
2563 	if (multi)
2564 		return 0;
2565 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2566 	multi = 1;
2567 	return 0;
2568 }
2569 
2570 static const struct dmi_system_id multi_dmi_table[] = {
2571 	{
2572 		.callback = set_multi,
2573 		.ident = "IBM System Summit2",
2574 		.matches = {
2575 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2576 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2577 		},
2578 	},
2579 	{}
2580 };
2581 
2582 static void dmi_check_multi(void)
2583 {
2584 	if (multi_checked)
2585 		return;
2586 
2587 	dmi_check_system(multi_dmi_table);
2588 	multi_checked = 1;
2589 }
2590 
2591 /*
2592  * apic_is_clustered_box() -- Check if we can expect good TSC
2593  *
2594  * Thus far, the major user of this is IBM's Summit2 series:
2595  * Clustered boxes may have unsynced TSC problems if they are
2596  * multi-chassis.
2597  * Use DMI to check them
2598  */
2599 int apic_is_clustered_box(void)
2600 {
2601 	dmi_check_multi();
2602 	return multi;
2603 }
2604 #endif
2605 
2606 /*
2607  * APIC command line parameters
2608  */
2609 static int __init setup_disableapic(char *arg)
2610 {
2611 	disable_apic = 1;
2612 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2613 	return 0;
2614 }
2615 early_param("disableapic", setup_disableapic);
2616 
2617 /* same as disableapic, for compatibility */
2618 static int __init setup_nolapic(char *arg)
2619 {
2620 	return setup_disableapic(arg);
2621 }
2622 early_param("nolapic", setup_nolapic);
2623 
2624 static int __init parse_lapic_timer_c2_ok(char *arg)
2625 {
2626 	local_apic_timer_c2_ok = 1;
2627 	return 0;
2628 }
2629 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2630 
2631 static int __init parse_disable_apic_timer(char *arg)
2632 {
2633 	disable_apic_timer = 1;
2634 	return 0;
2635 }
2636 early_param("noapictimer", parse_disable_apic_timer);
2637 
2638 static int __init parse_nolapic_timer(char *arg)
2639 {
2640 	disable_apic_timer = 1;
2641 	return 0;
2642 }
2643 early_param("nolapic_timer", parse_nolapic_timer);
2644 
2645 static int __init apic_set_verbosity(char *arg)
2646 {
2647 	if (!arg)  {
2648 #ifdef CONFIG_X86_64
2649 		skip_ioapic_setup = 0;
2650 		return 0;
2651 #endif
2652 		return -EINVAL;
2653 	}
2654 
2655 	if (strcmp("debug", arg) == 0)
2656 		apic_verbosity = APIC_DEBUG;
2657 	else if (strcmp("verbose", arg) == 0)
2658 		apic_verbosity = APIC_VERBOSE;
2659 	else {
2660 		pr_warning("APIC Verbosity level %s not recognised"
2661 			" use apic=verbose or apic=debug\n", arg);
2662 		return -EINVAL;
2663 	}
2664 
2665 	return 0;
2666 }
2667 early_param("apic", apic_set_verbosity);
2668 
2669 static int __init lapic_insert_resource(void)
2670 {
2671 	if (!apic_phys)
2672 		return -1;
2673 
2674 	/* Put local APIC into the resource map. */
2675 	lapic_resource.start = apic_phys;
2676 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2677 	insert_resource(&iomem_resource, &lapic_resource);
2678 
2679 	return 0;
2680 }
2681 
2682 /*
2683  * need call insert after e820__reserve_resources()
2684  * that is using request_resource
2685  */
2686 late_initcall(lapic_insert_resource);
2687 
2688 static int __init apic_set_disabled_cpu_apicid(char *arg)
2689 {
2690 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2691 		return -EINVAL;
2692 
2693 	return 0;
2694 }
2695 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2696 
2697 static int __init apic_set_extnmi(char *arg)
2698 {
2699 	if (!arg)
2700 		return -EINVAL;
2701 
2702 	if (!strncmp("all", arg, 3))
2703 		apic_extnmi = APIC_EXTNMI_ALL;
2704 	else if (!strncmp("none", arg, 4))
2705 		apic_extnmi = APIC_EXTNMI_NONE;
2706 	else if (!strncmp("bsp", arg, 3))
2707 		apic_extnmi = APIC_EXTNMI_BSP;
2708 	else {
2709 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2710 		return -EINVAL;
2711 	}
2712 
2713 	return 0;
2714 }
2715 early_param("apic_extnmi", apic_set_extnmi);
2716