xref: /openbmc/linux/arch/x86/kernel/apic/apic.c (revision 9f69e8a7)
1 /*
2  *	Local APIC handling, local APIC timers
3  *
4  *	(c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
5  *
6  *	Fixes
7  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs;
8  *					thanks to Eric Gilmore
9  *					and Rolf G. Tews
10  *					for testing these extensively.
11  *	Maciej W. Rozycki	:	Various updates and fixes.
12  *	Mikael Pettersson	:	Power Management for UP-APIC.
13  *	Pavel Machek and
14  *	Mikael Pettersson	:	PM converted to driver model.
15  */
16 
17 #include <linux/perf_event.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/mc146818rtc.h>
20 #include <linux/acpi_pmtmr.h>
21 #include <linux/clockchips.h>
22 #include <linux/interrupt.h>
23 #include <linux/bootmem.h>
24 #include <linux/ftrace.h>
25 #include <linux/ioport.h>
26 #include <linux/module.h>
27 #include <linux/syscore_ops.h>
28 #include <linux/delay.h>
29 #include <linux/timex.h>
30 #include <linux/i8253.h>
31 #include <linux/dmar.h>
32 #include <linux/init.h>
33 #include <linux/cpu.h>
34 #include <linux/dmi.h>
35 #include <linux/smp.h>
36 #include <linux/mm.h>
37 
38 #include <asm/trace/irq_vectors.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/perf_event.h>
41 #include <asm/x86_init.h>
42 #include <asm/pgalloc.h>
43 #include <linux/atomic.h>
44 #include <asm/mpspec.h>
45 #include <asm/i8259.h>
46 #include <asm/proto.h>
47 #include <asm/apic.h>
48 #include <asm/io_apic.h>
49 #include <asm/desc.h>
50 #include <asm/hpet.h>
51 #include <asm/idle.h>
52 #include <asm/mtrr.h>
53 #include <asm/time.h>
54 #include <asm/smp.h>
55 #include <asm/mce.h>
56 #include <asm/tsc.h>
57 #include <asm/hypervisor.h>
58 
59 unsigned int num_processors;
60 
61 unsigned disabled_cpus;
62 
63 /* Processor that is doing the boot up */
64 unsigned int boot_cpu_physical_apicid = -1U;
65 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
66 
67 /*
68  * The highest APIC ID seen during enumeration.
69  */
70 static unsigned int max_physical_apicid;
71 
72 /*
73  * Bitmask of physically existing CPUs:
74  */
75 physid_mask_t phys_cpu_present_map;
76 
77 /*
78  * Processor to be disabled specified by kernel parameter
79  * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
80  * avoid undefined behaviour caused by sending INIT from AP to BSP.
81  */
82 static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
83 
84 /*
85  * Map cpu index to physical APIC ID
86  */
87 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
88 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
89 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
90 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
91 
92 #ifdef CONFIG_X86_32
93 
94 /*
95  * On x86_32, the mapping between cpu and logical apicid may vary
96  * depending on apic in use.  The following early percpu variable is
97  * used for the mapping.  This is where the behaviors of x86_64 and 32
98  * actually diverge.  Let's keep it ugly for now.
99  */
100 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
101 
102 /* Local APIC was disabled by the BIOS and enabled by the kernel */
103 static int enabled_via_apicbase;
104 
105 /*
106  * Handle interrupt mode configuration register (IMCR).
107  * This register controls whether the interrupt signals
108  * that reach the BSP come from the master PIC or from the
109  * local APIC. Before entering Symmetric I/O Mode, either
110  * the BIOS or the operating system must switch out of
111  * PIC Mode by changing the IMCR.
112  */
113 static inline void imcr_pic_to_apic(void)
114 {
115 	/* select IMCR register */
116 	outb(0x70, 0x22);
117 	/* NMI and 8259 INTR go through APIC */
118 	outb(0x01, 0x23);
119 }
120 
121 static inline void imcr_apic_to_pic(void)
122 {
123 	/* select IMCR register */
124 	outb(0x70, 0x22);
125 	/* NMI and 8259 INTR go directly to BSP */
126 	outb(0x00, 0x23);
127 }
128 #endif
129 
130 /*
131  * Knob to control our willingness to enable the local APIC.
132  *
133  * +1=force-enable
134  */
135 static int force_enable_local_apic __initdata;
136 
137 /*
138  * APIC command line parameters
139  */
140 static int __init parse_lapic(char *arg)
141 {
142 	if (config_enabled(CONFIG_X86_32) && !arg)
143 		force_enable_local_apic = 1;
144 	else if (arg && !strncmp(arg, "notscdeadline", 13))
145 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
146 	return 0;
147 }
148 early_param("lapic", parse_lapic);
149 
150 #ifdef CONFIG_X86_64
151 static int apic_calibrate_pmtmr __initdata;
152 static __init int setup_apicpmtimer(char *s)
153 {
154 	apic_calibrate_pmtmr = 1;
155 	notsc_setup(NULL);
156 	return 0;
157 }
158 __setup("apicpmtimer", setup_apicpmtimer);
159 #endif
160 
161 unsigned long mp_lapic_addr;
162 int disable_apic;
163 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
164 static int disable_apic_timer __initdata;
165 /* Local APIC timer works in C2 */
166 int local_apic_timer_c2_ok;
167 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
168 
169 int first_system_vector = FIRST_SYSTEM_VECTOR;
170 
171 /*
172  * Debug level, exported for io_apic.c
173  */
174 unsigned int apic_verbosity;
175 
176 int pic_mode;
177 
178 /* Have we found an MP table */
179 int smp_found_config;
180 
181 static struct resource lapic_resource = {
182 	.name = "Local APIC",
183 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
184 };
185 
186 unsigned int lapic_timer_frequency = 0;
187 
188 static void apic_pm_activate(void);
189 
190 static unsigned long apic_phys;
191 
192 /*
193  * Get the LAPIC version
194  */
195 static inline int lapic_get_version(void)
196 {
197 	return GET_APIC_VERSION(apic_read(APIC_LVR));
198 }
199 
200 /*
201  * Check, if the APIC is integrated or a separate chip
202  */
203 static inline int lapic_is_integrated(void)
204 {
205 #ifdef CONFIG_X86_64
206 	return 1;
207 #else
208 	return APIC_INTEGRATED(lapic_get_version());
209 #endif
210 }
211 
212 /*
213  * Check, whether this is a modern or a first generation APIC
214  */
215 static int modern_apic(void)
216 {
217 	/* AMD systems use old APIC versions, so check the CPU */
218 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
219 	    boot_cpu_data.x86 >= 0xf)
220 		return 1;
221 	return lapic_get_version() >= 0x14;
222 }
223 
224 /*
225  * right after this call apic become NOOP driven
226  * so apic->write/read doesn't do anything
227  */
228 static void __init apic_disable(void)
229 {
230 	pr_info("APIC: switched to apic NOOP\n");
231 	apic = &apic_noop;
232 }
233 
234 void native_apic_wait_icr_idle(void)
235 {
236 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
237 		cpu_relax();
238 }
239 
240 u32 native_safe_apic_wait_icr_idle(void)
241 {
242 	u32 send_status;
243 	int timeout;
244 
245 	timeout = 0;
246 	do {
247 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
248 		if (!send_status)
249 			break;
250 		inc_irq_stat(icr_read_retry_count);
251 		udelay(100);
252 	} while (timeout++ < 1000);
253 
254 	return send_status;
255 }
256 
257 void native_apic_icr_write(u32 low, u32 id)
258 {
259 	unsigned long flags;
260 
261 	local_irq_save(flags);
262 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
263 	apic_write(APIC_ICR, low);
264 	local_irq_restore(flags);
265 }
266 
267 u64 native_apic_icr_read(void)
268 {
269 	u32 icr1, icr2;
270 
271 	icr2 = apic_read(APIC_ICR2);
272 	icr1 = apic_read(APIC_ICR);
273 
274 	return icr1 | ((u64)icr2 << 32);
275 }
276 
277 #ifdef CONFIG_X86_32
278 /**
279  * get_physical_broadcast - Get number of physical broadcast IDs
280  */
281 int get_physical_broadcast(void)
282 {
283 	return modern_apic() ? 0xff : 0xf;
284 }
285 #endif
286 
287 /**
288  * lapic_get_maxlvt - get the maximum number of local vector table entries
289  */
290 int lapic_get_maxlvt(void)
291 {
292 	unsigned int v;
293 
294 	v = apic_read(APIC_LVR);
295 	/*
296 	 * - we always have APIC integrated on 64bit mode
297 	 * - 82489DXs do not report # of LVT entries
298 	 */
299 	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
300 }
301 
302 /*
303  * Local APIC timer
304  */
305 
306 /* Clock divisor */
307 #define APIC_DIVISOR 16
308 #define TSC_DIVISOR  32
309 
310 /*
311  * This function sets up the local APIC timer, with a timeout of
312  * 'clocks' APIC bus clock. During calibration we actually call
313  * this function twice on the boot CPU, once with a bogus timeout
314  * value, second time for real. The other (noncalibrating) CPUs
315  * call this function only once, with the real, calibrated value.
316  *
317  * We do reads before writes even if unnecessary, to get around the
318  * P5 APIC double write bug.
319  */
320 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
321 {
322 	unsigned int lvtt_value, tmp_value;
323 
324 	lvtt_value = LOCAL_TIMER_VECTOR;
325 	if (!oneshot)
326 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
327 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
328 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
329 
330 	if (!lapic_is_integrated())
331 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
332 
333 	if (!irqen)
334 		lvtt_value |= APIC_LVT_MASKED;
335 
336 	apic_write(APIC_LVTT, lvtt_value);
337 
338 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
339 		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
340 		return;
341 	}
342 
343 	/*
344 	 * Divide PICLK by 16
345 	 */
346 	tmp_value = apic_read(APIC_TDCR);
347 	apic_write(APIC_TDCR,
348 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
349 		APIC_TDR_DIV_16);
350 
351 	if (!oneshot)
352 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
353 }
354 
355 /*
356  * Setup extended LVT, AMD specific
357  *
358  * Software should use the LVT offsets the BIOS provides.  The offsets
359  * are determined by the subsystems using it like those for MCE
360  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
361  * are supported. Beginning with family 10h at least 4 offsets are
362  * available.
363  *
364  * Since the offsets must be consistent for all cores, we keep track
365  * of the LVT offsets in software and reserve the offset for the same
366  * vector also to be used on other cores. An offset is freed by
367  * setting the entry to APIC_EILVT_MASKED.
368  *
369  * If the BIOS is right, there should be no conflicts. Otherwise a
370  * "[Firmware Bug]: ..." error message is generated. However, if
371  * software does not properly determines the offsets, it is not
372  * necessarily a BIOS bug.
373  */
374 
375 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
376 
377 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
378 {
379 	return (old & APIC_EILVT_MASKED)
380 		|| (new == APIC_EILVT_MASKED)
381 		|| ((new & ~APIC_EILVT_MASKED) == old);
382 }
383 
384 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
385 {
386 	unsigned int rsvd, vector;
387 
388 	if (offset >= APIC_EILVT_NR_MAX)
389 		return ~0;
390 
391 	rsvd = atomic_read(&eilvt_offsets[offset]);
392 	do {
393 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
394 		if (vector && !eilvt_entry_is_changeable(vector, new))
395 			/* may not change if vectors are different */
396 			return rsvd;
397 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
398 	} while (rsvd != new);
399 
400 	rsvd &= ~APIC_EILVT_MASKED;
401 	if (rsvd && rsvd != vector)
402 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
403 			offset, rsvd);
404 
405 	return new;
406 }
407 
408 /*
409  * If mask=1, the LVT entry does not generate interrupts while mask=0
410  * enables the vector. See also the BKDGs. Must be called with
411  * preemption disabled.
412  */
413 
414 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
415 {
416 	unsigned long reg = APIC_EILVTn(offset);
417 	unsigned int new, old, reserved;
418 
419 	new = (mask << 16) | (msg_type << 8) | vector;
420 	old = apic_read(reg);
421 	reserved = reserve_eilvt_offset(offset, new);
422 
423 	if (reserved != new) {
424 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
425 		       "vector 0x%x, but the register is already in use for "
426 		       "vector 0x%x on another cpu\n",
427 		       smp_processor_id(), reg, offset, new, reserved);
428 		return -EINVAL;
429 	}
430 
431 	if (!eilvt_entry_is_changeable(old, new)) {
432 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
433 		       "vector 0x%x, but the register is already in use for "
434 		       "vector 0x%x on this cpu\n",
435 		       smp_processor_id(), reg, offset, new, old);
436 		return -EBUSY;
437 	}
438 
439 	apic_write(reg, new);
440 
441 	return 0;
442 }
443 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
444 
445 /*
446  * Program the next event, relative to now
447  */
448 static int lapic_next_event(unsigned long delta,
449 			    struct clock_event_device *evt)
450 {
451 	apic_write(APIC_TMICT, delta);
452 	return 0;
453 }
454 
455 static int lapic_next_deadline(unsigned long delta,
456 			       struct clock_event_device *evt)
457 {
458 	u64 tsc;
459 
460 	rdtscll(tsc);
461 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
462 	return 0;
463 }
464 
465 /*
466  * Setup the lapic timer in periodic or oneshot mode
467  */
468 static void lapic_timer_setup(enum clock_event_mode mode,
469 			      struct clock_event_device *evt)
470 {
471 	unsigned long flags;
472 	unsigned int v;
473 
474 	/* Lapic used as dummy for broadcast ? */
475 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
476 		return;
477 
478 	local_irq_save(flags);
479 
480 	switch (mode) {
481 	case CLOCK_EVT_MODE_PERIODIC:
482 	case CLOCK_EVT_MODE_ONESHOT:
483 		__setup_APIC_LVTT(lapic_timer_frequency,
484 				  mode != CLOCK_EVT_MODE_PERIODIC, 1);
485 		break;
486 	case CLOCK_EVT_MODE_UNUSED:
487 	case CLOCK_EVT_MODE_SHUTDOWN:
488 		v = apic_read(APIC_LVTT);
489 		v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
490 		apic_write(APIC_LVTT, v);
491 		apic_write(APIC_TMICT, 0);
492 		break;
493 	case CLOCK_EVT_MODE_RESUME:
494 		/* Nothing to do here */
495 		break;
496 	}
497 
498 	local_irq_restore(flags);
499 }
500 
501 /*
502  * Local APIC timer broadcast function
503  */
504 static void lapic_timer_broadcast(const struct cpumask *mask)
505 {
506 #ifdef CONFIG_SMP
507 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
508 #endif
509 }
510 
511 
512 /*
513  * The local apic timer can be used for any function which is CPU local.
514  */
515 static struct clock_event_device lapic_clockevent = {
516 	.name		= "lapic",
517 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
518 			| CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
519 	.shift		= 32,
520 	.set_mode	= lapic_timer_setup,
521 	.set_next_event	= lapic_next_event,
522 	.broadcast	= lapic_timer_broadcast,
523 	.rating		= 100,
524 	.irq		= -1,
525 };
526 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
527 
528 /*
529  * Setup the local APIC timer for this CPU. Copy the initialized values
530  * of the boot CPU and register the clock event in the framework.
531  */
532 static void setup_APIC_timer(void)
533 {
534 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
535 
536 	if (this_cpu_has(X86_FEATURE_ARAT)) {
537 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
538 		/* Make LAPIC timer preferrable over percpu HPET */
539 		lapic_clockevent.rating = 150;
540 	}
541 
542 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
543 	levt->cpumask = cpumask_of(smp_processor_id());
544 
545 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
546 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
547 				    CLOCK_EVT_FEAT_DUMMY);
548 		levt->set_next_event = lapic_next_deadline;
549 		clockevents_config_and_register(levt,
550 						(tsc_khz / TSC_DIVISOR) * 1000,
551 						0xF, ~0UL);
552 	} else
553 		clockevents_register_device(levt);
554 }
555 
556 /*
557  * In this functions we calibrate APIC bus clocks to the external timer.
558  *
559  * We want to do the calibration only once since we want to have local timer
560  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
561  * frequency.
562  *
563  * This was previously done by reading the PIT/HPET and waiting for a wrap
564  * around to find out, that a tick has elapsed. I have a box, where the PIT
565  * readout is broken, so it never gets out of the wait loop again. This was
566  * also reported by others.
567  *
568  * Monitoring the jiffies value is inaccurate and the clockevents
569  * infrastructure allows us to do a simple substitution of the interrupt
570  * handler.
571  *
572  * The calibration routine also uses the pm_timer when possible, as the PIT
573  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
574  * back to normal later in the boot process).
575  */
576 
577 #define LAPIC_CAL_LOOPS		(HZ/10)
578 
579 static __initdata int lapic_cal_loops = -1;
580 static __initdata long lapic_cal_t1, lapic_cal_t2;
581 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
582 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
583 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
584 
585 /*
586  * Temporary interrupt handler.
587  */
588 static void __init lapic_cal_handler(struct clock_event_device *dev)
589 {
590 	unsigned long long tsc = 0;
591 	long tapic = apic_read(APIC_TMCCT);
592 	unsigned long pm = acpi_pm_read_early();
593 
594 	if (cpu_has_tsc)
595 		rdtscll(tsc);
596 
597 	switch (lapic_cal_loops++) {
598 	case 0:
599 		lapic_cal_t1 = tapic;
600 		lapic_cal_tsc1 = tsc;
601 		lapic_cal_pm1 = pm;
602 		lapic_cal_j1 = jiffies;
603 		break;
604 
605 	case LAPIC_CAL_LOOPS:
606 		lapic_cal_t2 = tapic;
607 		lapic_cal_tsc2 = tsc;
608 		if (pm < lapic_cal_pm1)
609 			pm += ACPI_PM_OVRRUN;
610 		lapic_cal_pm2 = pm;
611 		lapic_cal_j2 = jiffies;
612 		break;
613 	}
614 }
615 
616 static int __init
617 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
618 {
619 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
620 	const long pm_thresh = pm_100ms / 100;
621 	unsigned long mult;
622 	u64 res;
623 
624 #ifndef CONFIG_X86_PM_TIMER
625 	return -1;
626 #endif
627 
628 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
629 
630 	/* Check, if the PM timer is available */
631 	if (!deltapm)
632 		return -1;
633 
634 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
635 
636 	if (deltapm > (pm_100ms - pm_thresh) &&
637 	    deltapm < (pm_100ms + pm_thresh)) {
638 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
639 		return 0;
640 	}
641 
642 	res = (((u64)deltapm) *  mult) >> 22;
643 	do_div(res, 1000000);
644 	pr_warning("APIC calibration not consistent "
645 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
646 
647 	/* Correct the lapic counter value */
648 	res = (((u64)(*delta)) * pm_100ms);
649 	do_div(res, deltapm);
650 	pr_info("APIC delta adjusted to PM-Timer: "
651 		"%lu (%ld)\n", (unsigned long)res, *delta);
652 	*delta = (long)res;
653 
654 	/* Correct the tsc counter value */
655 	if (cpu_has_tsc) {
656 		res = (((u64)(*deltatsc)) * pm_100ms);
657 		do_div(res, deltapm);
658 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
659 					  "PM-Timer: %lu (%ld)\n",
660 					(unsigned long)res, *deltatsc);
661 		*deltatsc = (long)res;
662 	}
663 
664 	return 0;
665 }
666 
667 static int __init calibrate_APIC_clock(void)
668 {
669 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
670 	void (*real_handler)(struct clock_event_device *dev);
671 	unsigned long deltaj;
672 	long delta, deltatsc;
673 	int pm_referenced = 0;
674 
675 	/**
676 	 * check if lapic timer has already been calibrated by platform
677 	 * specific routine, such as tsc calibration code. if so, we just fill
678 	 * in the clockevent structure and return.
679 	 */
680 
681 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
682 		return 0;
683 	} else if (lapic_timer_frequency) {
684 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
685 				lapic_timer_frequency);
686 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
687 					TICK_NSEC, lapic_clockevent.shift);
688 		lapic_clockevent.max_delta_ns =
689 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
690 		lapic_clockevent.min_delta_ns =
691 			clockevent_delta2ns(0xF, &lapic_clockevent);
692 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
693 		return 0;
694 	}
695 
696 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
697 		    "calibrating APIC timer ...\n");
698 
699 	local_irq_disable();
700 
701 	/* Replace the global interrupt handler */
702 	real_handler = global_clock_event->event_handler;
703 	global_clock_event->event_handler = lapic_cal_handler;
704 
705 	/*
706 	 * Setup the APIC counter to maximum. There is no way the lapic
707 	 * can underflow in the 100ms detection time frame
708 	 */
709 	__setup_APIC_LVTT(0xffffffff, 0, 0);
710 
711 	/* Let the interrupts run */
712 	local_irq_enable();
713 
714 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
715 		cpu_relax();
716 
717 	local_irq_disable();
718 
719 	/* Restore the real event handler */
720 	global_clock_event->event_handler = real_handler;
721 
722 	/* Build delta t1-t2 as apic timer counts down */
723 	delta = lapic_cal_t1 - lapic_cal_t2;
724 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
725 
726 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
727 
728 	/* we trust the PM based calibration if possible */
729 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
730 					&delta, &deltatsc);
731 
732 	/* Calculate the scaled math multiplication factor */
733 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
734 				       lapic_clockevent.shift);
735 	lapic_clockevent.max_delta_ns =
736 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
737 	lapic_clockevent.min_delta_ns =
738 		clockevent_delta2ns(0xF, &lapic_clockevent);
739 
740 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
741 
742 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
743 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
744 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
745 		    lapic_timer_frequency);
746 
747 	if (cpu_has_tsc) {
748 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
749 			    "%ld.%04ld MHz.\n",
750 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
751 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
752 	}
753 
754 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
755 		    "%u.%04u MHz.\n",
756 		    lapic_timer_frequency / (1000000 / HZ),
757 		    lapic_timer_frequency % (1000000 / HZ));
758 
759 	/*
760 	 * Do a sanity check on the APIC calibration result
761 	 */
762 	if (lapic_timer_frequency < (1000000 / HZ)) {
763 		local_irq_enable();
764 		pr_warning("APIC frequency too slow, disabling apic timer\n");
765 		return -1;
766 	}
767 
768 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
769 
770 	/*
771 	 * PM timer calibration failed or not turned on
772 	 * so lets try APIC timer based calibration
773 	 */
774 	if (!pm_referenced) {
775 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
776 
777 		/*
778 		 * Setup the apic timer manually
779 		 */
780 		levt->event_handler = lapic_cal_handler;
781 		lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
782 		lapic_cal_loops = -1;
783 
784 		/* Let the interrupts run */
785 		local_irq_enable();
786 
787 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
788 			cpu_relax();
789 
790 		/* Stop the lapic timer */
791 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
792 
793 		/* Jiffies delta */
794 		deltaj = lapic_cal_j2 - lapic_cal_j1;
795 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
796 
797 		/* Check, if the jiffies result is consistent */
798 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
799 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
800 		else
801 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
802 	} else
803 		local_irq_enable();
804 
805 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
806 		pr_warning("APIC timer disabled due to verification failure\n");
807 			return -1;
808 	}
809 
810 	return 0;
811 }
812 
813 /*
814  * Setup the boot APIC
815  *
816  * Calibrate and verify the result.
817  */
818 void __init setup_boot_APIC_clock(void)
819 {
820 	/*
821 	 * The local apic timer can be disabled via the kernel
822 	 * commandline or from the CPU detection code. Register the lapic
823 	 * timer as a dummy clock event source on SMP systems, so the
824 	 * broadcast mechanism is used. On UP systems simply ignore it.
825 	 */
826 	if (disable_apic_timer) {
827 		pr_info("Disabling APIC timer\n");
828 		/* No broadcast on UP ! */
829 		if (num_possible_cpus() > 1) {
830 			lapic_clockevent.mult = 1;
831 			setup_APIC_timer();
832 		}
833 		return;
834 	}
835 
836 	if (calibrate_APIC_clock()) {
837 		/* No broadcast on UP ! */
838 		if (num_possible_cpus() > 1)
839 			setup_APIC_timer();
840 		return;
841 	}
842 
843 	/*
844 	 * If nmi_watchdog is set to IO_APIC, we need the
845 	 * PIT/HPET going.  Otherwise register lapic as a dummy
846 	 * device.
847 	 */
848 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
849 
850 	/* Setup the lapic or request the broadcast */
851 	setup_APIC_timer();
852 }
853 
854 void setup_secondary_APIC_clock(void)
855 {
856 	setup_APIC_timer();
857 }
858 
859 /*
860  * The guts of the apic timer interrupt
861  */
862 static void local_apic_timer_interrupt(void)
863 {
864 	int cpu = smp_processor_id();
865 	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
866 
867 	/*
868 	 * Normally we should not be here till LAPIC has been initialized but
869 	 * in some cases like kdump, its possible that there is a pending LAPIC
870 	 * timer interrupt from previous kernel's context and is delivered in
871 	 * new kernel the moment interrupts are enabled.
872 	 *
873 	 * Interrupts are enabled early and LAPIC is setup much later, hence
874 	 * its possible that when we get here evt->event_handler is NULL.
875 	 * Check for event_handler being NULL and discard the interrupt as
876 	 * spurious.
877 	 */
878 	if (!evt->event_handler) {
879 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
880 		/* Switch it off */
881 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
882 		return;
883 	}
884 
885 	/*
886 	 * the NMI deadlock-detector uses this.
887 	 */
888 	inc_irq_stat(apic_timer_irqs);
889 
890 	evt->event_handler(evt);
891 }
892 
893 /*
894  * Local APIC timer interrupt. This is the most natural way for doing
895  * local interrupts, but local timer interrupts can be emulated by
896  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
897  *
898  * [ if a single-CPU system runs an SMP kernel then we call the local
899  *   interrupt as well. Thus we cannot inline the local irq ... ]
900  */
901 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
902 {
903 	struct pt_regs *old_regs = set_irq_regs(regs);
904 
905 	/*
906 	 * NOTE! We'd better ACK the irq immediately,
907 	 * because timer handling can be slow.
908 	 *
909 	 * update_process_times() expects us to have done irq_enter().
910 	 * Besides, if we don't timer interrupts ignore the global
911 	 * interrupt lock, which is the WrongThing (tm) to do.
912 	 */
913 	entering_ack_irq();
914 	local_apic_timer_interrupt();
915 	exiting_irq();
916 
917 	set_irq_regs(old_regs);
918 }
919 
920 __visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
921 {
922 	struct pt_regs *old_regs = set_irq_regs(regs);
923 
924 	/*
925 	 * NOTE! We'd better ACK the irq immediately,
926 	 * because timer handling can be slow.
927 	 *
928 	 * update_process_times() expects us to have done irq_enter().
929 	 * Besides, if we don't timer interrupts ignore the global
930 	 * interrupt lock, which is the WrongThing (tm) to do.
931 	 */
932 	entering_ack_irq();
933 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
934 	local_apic_timer_interrupt();
935 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
936 	exiting_irq();
937 
938 	set_irq_regs(old_regs);
939 }
940 
941 int setup_profiling_timer(unsigned int multiplier)
942 {
943 	return -EINVAL;
944 }
945 
946 /*
947  * Local APIC start and shutdown
948  */
949 
950 /**
951  * clear_local_APIC - shutdown the local APIC
952  *
953  * This is called, when a CPU is disabled and before rebooting, so the state of
954  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
955  * leftovers during boot.
956  */
957 void clear_local_APIC(void)
958 {
959 	int maxlvt;
960 	u32 v;
961 
962 	/* APIC hasn't been mapped yet */
963 	if (!x2apic_mode && !apic_phys)
964 		return;
965 
966 	maxlvt = lapic_get_maxlvt();
967 	/*
968 	 * Masking an LVT entry can trigger a local APIC error
969 	 * if the vector is zero. Mask LVTERR first to prevent this.
970 	 */
971 	if (maxlvt >= 3) {
972 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
973 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
974 	}
975 	/*
976 	 * Careful: we have to set masks only first to deassert
977 	 * any level-triggered sources.
978 	 */
979 	v = apic_read(APIC_LVTT);
980 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
981 	v = apic_read(APIC_LVT0);
982 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
983 	v = apic_read(APIC_LVT1);
984 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
985 	if (maxlvt >= 4) {
986 		v = apic_read(APIC_LVTPC);
987 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
988 	}
989 
990 	/* lets not touch this if we didn't frob it */
991 #ifdef CONFIG_X86_THERMAL_VECTOR
992 	if (maxlvt >= 5) {
993 		v = apic_read(APIC_LVTTHMR);
994 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
995 	}
996 #endif
997 #ifdef CONFIG_X86_MCE_INTEL
998 	if (maxlvt >= 6) {
999 		v = apic_read(APIC_LVTCMCI);
1000 		if (!(v & APIC_LVT_MASKED))
1001 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1002 	}
1003 #endif
1004 
1005 	/*
1006 	 * Clean APIC state for other OSs:
1007 	 */
1008 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1009 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1010 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1011 	if (maxlvt >= 3)
1012 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1013 	if (maxlvt >= 4)
1014 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1015 
1016 	/* Integrated APIC (!82489DX) ? */
1017 	if (lapic_is_integrated()) {
1018 		if (maxlvt > 3)
1019 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1020 			apic_write(APIC_ESR, 0);
1021 		apic_read(APIC_ESR);
1022 	}
1023 }
1024 
1025 /**
1026  * disable_local_APIC - clear and disable the local APIC
1027  */
1028 void disable_local_APIC(void)
1029 {
1030 	unsigned int value;
1031 
1032 	/* APIC hasn't been mapped yet */
1033 	if (!x2apic_mode && !apic_phys)
1034 		return;
1035 
1036 	clear_local_APIC();
1037 
1038 	/*
1039 	 * Disable APIC (implies clearing of registers
1040 	 * for 82489DX!).
1041 	 */
1042 	value = apic_read(APIC_SPIV);
1043 	value &= ~APIC_SPIV_APIC_ENABLED;
1044 	apic_write(APIC_SPIV, value);
1045 
1046 #ifdef CONFIG_X86_32
1047 	/*
1048 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1049 	 * restore the disabled state.
1050 	 */
1051 	if (enabled_via_apicbase) {
1052 		unsigned int l, h;
1053 
1054 		rdmsr(MSR_IA32_APICBASE, l, h);
1055 		l &= ~MSR_IA32_APICBASE_ENABLE;
1056 		wrmsr(MSR_IA32_APICBASE, l, h);
1057 	}
1058 #endif
1059 }
1060 
1061 /*
1062  * If Linux enabled the LAPIC against the BIOS default disable it down before
1063  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1064  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1065  * for the case where Linux didn't enable the LAPIC.
1066  */
1067 void lapic_shutdown(void)
1068 {
1069 	unsigned long flags;
1070 
1071 	if (!cpu_has_apic && !apic_from_smp_config())
1072 		return;
1073 
1074 	local_irq_save(flags);
1075 
1076 #ifdef CONFIG_X86_32
1077 	if (!enabled_via_apicbase)
1078 		clear_local_APIC();
1079 	else
1080 #endif
1081 		disable_local_APIC();
1082 
1083 
1084 	local_irq_restore(flags);
1085 }
1086 
1087 /**
1088  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1089  */
1090 void __init sync_Arb_IDs(void)
1091 {
1092 	/*
1093 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1094 	 * needed on AMD.
1095 	 */
1096 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1097 		return;
1098 
1099 	/*
1100 	 * Wait for idle.
1101 	 */
1102 	apic_wait_icr_idle();
1103 
1104 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1105 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1106 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1107 }
1108 
1109 /*
1110  * An initial setup of the virtual wire mode.
1111  */
1112 void __init init_bsp_APIC(void)
1113 {
1114 	unsigned int value;
1115 
1116 	/*
1117 	 * Don't do the setup now if we have a SMP BIOS as the
1118 	 * through-I/O-APIC virtual wire mode might be active.
1119 	 */
1120 	if (smp_found_config || !cpu_has_apic)
1121 		return;
1122 
1123 	/*
1124 	 * Do not trust the local APIC being empty at bootup.
1125 	 */
1126 	clear_local_APIC();
1127 
1128 	/*
1129 	 * Enable APIC.
1130 	 */
1131 	value = apic_read(APIC_SPIV);
1132 	value &= ~APIC_VECTOR_MASK;
1133 	value |= APIC_SPIV_APIC_ENABLED;
1134 
1135 #ifdef CONFIG_X86_32
1136 	/* This bit is reserved on P4/Xeon and should be cleared */
1137 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1138 	    (boot_cpu_data.x86 == 15))
1139 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1140 	else
1141 #endif
1142 		value |= APIC_SPIV_FOCUS_DISABLED;
1143 	value |= SPURIOUS_APIC_VECTOR;
1144 	apic_write(APIC_SPIV, value);
1145 
1146 	/*
1147 	 * Set up the virtual wire mode.
1148 	 */
1149 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1150 	value = APIC_DM_NMI;
1151 	if (!lapic_is_integrated())		/* 82489DX */
1152 		value |= APIC_LVT_LEVEL_TRIGGER;
1153 	apic_write(APIC_LVT1, value);
1154 }
1155 
1156 static void lapic_setup_esr(void)
1157 {
1158 	unsigned int oldvalue, value, maxlvt;
1159 
1160 	if (!lapic_is_integrated()) {
1161 		pr_info("No ESR for 82489DX.\n");
1162 		return;
1163 	}
1164 
1165 	if (apic->disable_esr) {
1166 		/*
1167 		 * Something untraceable is creating bad interrupts on
1168 		 * secondary quads ... for the moment, just leave the
1169 		 * ESR disabled - we can't do anything useful with the
1170 		 * errors anyway - mbligh
1171 		 */
1172 		pr_info("Leaving ESR disabled.\n");
1173 		return;
1174 	}
1175 
1176 	maxlvt = lapic_get_maxlvt();
1177 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1178 		apic_write(APIC_ESR, 0);
1179 	oldvalue = apic_read(APIC_ESR);
1180 
1181 	/* enables sending errors */
1182 	value = ERROR_APIC_VECTOR;
1183 	apic_write(APIC_LVTERR, value);
1184 
1185 	/*
1186 	 * spec says clear errors after enabling vector.
1187 	 */
1188 	if (maxlvt > 3)
1189 		apic_write(APIC_ESR, 0);
1190 	value = apic_read(APIC_ESR);
1191 	if (value != oldvalue)
1192 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1193 			"vector: 0x%08x  after: 0x%08x\n",
1194 			oldvalue, value);
1195 }
1196 
1197 /**
1198  * setup_local_APIC - setup the local APIC
1199  *
1200  * Used to setup local APIC while initializing BSP or bringin up APs.
1201  * Always called with preemption disabled.
1202  */
1203 void setup_local_APIC(void)
1204 {
1205 	int cpu = smp_processor_id();
1206 	unsigned int value, queued;
1207 	int i, j, acked = 0;
1208 	unsigned long long tsc = 0, ntsc;
1209 	long long max_loops = cpu_khz ? cpu_khz : 1000000;
1210 
1211 	if (cpu_has_tsc)
1212 		rdtscll(tsc);
1213 
1214 	if (disable_apic) {
1215 		disable_ioapic_support();
1216 		return;
1217 	}
1218 
1219 #ifdef CONFIG_X86_32
1220 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1221 	if (lapic_is_integrated() && apic->disable_esr) {
1222 		apic_write(APIC_ESR, 0);
1223 		apic_write(APIC_ESR, 0);
1224 		apic_write(APIC_ESR, 0);
1225 		apic_write(APIC_ESR, 0);
1226 	}
1227 #endif
1228 	perf_events_lapic_init();
1229 
1230 	/*
1231 	 * Double-check whether this APIC is really registered.
1232 	 * This is meaningless in clustered apic mode, so we skip it.
1233 	 */
1234 	BUG_ON(!apic->apic_id_registered());
1235 
1236 	/*
1237 	 * Intel recommends to set DFR, LDR and TPR before enabling
1238 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1239 	 * document number 292116).  So here it goes...
1240 	 */
1241 	apic->init_apic_ldr();
1242 
1243 #ifdef CONFIG_X86_32
1244 	/*
1245 	 * APIC LDR is initialized.  If logical_apicid mapping was
1246 	 * initialized during get_smp_config(), make sure it matches the
1247 	 * actual value.
1248 	 */
1249 	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1250 	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1251 	/* always use the value from LDR */
1252 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1253 		logical_smp_processor_id();
1254 #endif
1255 
1256 	/*
1257 	 * Set Task Priority to 'accept all'. We never change this
1258 	 * later on.
1259 	 */
1260 	value = apic_read(APIC_TASKPRI);
1261 	value &= ~APIC_TPRI_MASK;
1262 	apic_write(APIC_TASKPRI, value);
1263 
1264 	/*
1265 	 * After a crash, we no longer service the interrupts and a pending
1266 	 * interrupt from previous kernel might still have ISR bit set.
1267 	 *
1268 	 * Most probably by now CPU has serviced that pending interrupt and
1269 	 * it might not have done the ack_APIC_irq() because it thought,
1270 	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1271 	 * does not clear the ISR bit and cpu thinks it has already serivced
1272 	 * the interrupt. Hence a vector might get locked. It was noticed
1273 	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1274 	 */
1275 	do {
1276 		queued = 0;
1277 		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1278 			queued |= apic_read(APIC_IRR + i*0x10);
1279 
1280 		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1281 			value = apic_read(APIC_ISR + i*0x10);
1282 			for (j = 31; j >= 0; j--) {
1283 				if (value & (1<<j)) {
1284 					ack_APIC_irq();
1285 					acked++;
1286 				}
1287 			}
1288 		}
1289 		if (acked > 256) {
1290 			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1291 			       acked);
1292 			break;
1293 		}
1294 		if (queued) {
1295 			if (cpu_has_tsc && cpu_khz) {
1296 				rdtscll(ntsc);
1297 				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1298 			} else
1299 				max_loops--;
1300 		}
1301 	} while (queued && max_loops > 0);
1302 	WARN_ON(max_loops <= 0);
1303 
1304 	/*
1305 	 * Now that we are all set up, enable the APIC
1306 	 */
1307 	value = apic_read(APIC_SPIV);
1308 	value &= ~APIC_VECTOR_MASK;
1309 	/*
1310 	 * Enable APIC
1311 	 */
1312 	value |= APIC_SPIV_APIC_ENABLED;
1313 
1314 #ifdef CONFIG_X86_32
1315 	/*
1316 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1317 	 * certain networking cards. If high frequency interrupts are
1318 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1319 	 * entry is masked/unmasked at a high rate as well then sooner or
1320 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1321 	 * from the device. If focus CPU is disabled then the hang goes
1322 	 * away, oh well :-(
1323 	 *
1324 	 * [ This bug can be reproduced easily with a level-triggered
1325 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1326 	 *   BX chipset. ]
1327 	 */
1328 	/*
1329 	 * Actually disabling the focus CPU check just makes the hang less
1330 	 * frequent as it makes the interrupt distributon model be more
1331 	 * like LRU than MRU (the short-term load is more even across CPUs).
1332 	 * See also the comment in end_level_ioapic_irq().  --macro
1333 	 */
1334 
1335 	/*
1336 	 * - enable focus processor (bit==0)
1337 	 * - 64bit mode always use processor focus
1338 	 *   so no need to set it
1339 	 */
1340 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1341 #endif
1342 
1343 	/*
1344 	 * Set spurious IRQ vector
1345 	 */
1346 	value |= SPURIOUS_APIC_VECTOR;
1347 	apic_write(APIC_SPIV, value);
1348 
1349 	/*
1350 	 * Set up LVT0, LVT1:
1351 	 *
1352 	 * set up through-local-APIC on the BP's LINT0. This is not
1353 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1354 	 * we delegate interrupts to the 8259A.
1355 	 */
1356 	/*
1357 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1358 	 */
1359 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1360 	if (!cpu && (pic_mode || !value)) {
1361 		value = APIC_DM_EXTINT;
1362 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1363 	} else {
1364 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1365 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1366 	}
1367 	apic_write(APIC_LVT0, value);
1368 
1369 	/*
1370 	 * only the BP should see the LINT1 NMI signal, obviously.
1371 	 */
1372 	if (!cpu)
1373 		value = APIC_DM_NMI;
1374 	else
1375 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1376 	if (!lapic_is_integrated())		/* 82489DX */
1377 		value |= APIC_LVT_LEVEL_TRIGGER;
1378 	apic_write(APIC_LVT1, value);
1379 
1380 #ifdef CONFIG_X86_MCE_INTEL
1381 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1382 	if (!cpu)
1383 		cmci_recheck();
1384 #endif
1385 }
1386 
1387 static void end_local_APIC_setup(void)
1388 {
1389 	lapic_setup_esr();
1390 
1391 #ifdef CONFIG_X86_32
1392 	{
1393 		unsigned int value;
1394 		/* Disable the local apic timer */
1395 		value = apic_read(APIC_LVTT);
1396 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1397 		apic_write(APIC_LVTT, value);
1398 	}
1399 #endif
1400 
1401 	apic_pm_activate();
1402 }
1403 
1404 /*
1405  * APIC setup function for application processors. Called from smpboot.c
1406  */
1407 void apic_ap_setup(void)
1408 {
1409 	setup_local_APIC();
1410 	end_local_APIC_setup();
1411 }
1412 
1413 #ifdef CONFIG_X86_X2APIC
1414 int x2apic_mode;
1415 
1416 enum {
1417 	X2APIC_OFF,
1418 	X2APIC_ON,
1419 	X2APIC_DISABLED,
1420 };
1421 static int x2apic_state;
1422 
1423 static inline void __x2apic_disable(void)
1424 {
1425 	u64 msr;
1426 
1427 	if (cpu_has_apic)
1428 		return;
1429 
1430 	rdmsrl(MSR_IA32_APICBASE, msr);
1431 	if (!(msr & X2APIC_ENABLE))
1432 		return;
1433 	/* Disable xapic and x2apic first and then reenable xapic mode */
1434 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1435 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1436 	printk_once(KERN_INFO "x2apic disabled\n");
1437 }
1438 
1439 static inline void __x2apic_enable(void)
1440 {
1441 	u64 msr;
1442 
1443 	rdmsrl(MSR_IA32_APICBASE, msr);
1444 	if (msr & X2APIC_ENABLE)
1445 		return;
1446 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1447 	printk_once(KERN_INFO "x2apic enabled\n");
1448 }
1449 
1450 static int __init setup_nox2apic(char *str)
1451 {
1452 	if (x2apic_enabled()) {
1453 		int apicid = native_apic_msr_read(APIC_ID);
1454 
1455 		if (apicid >= 255) {
1456 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1457 				   apicid);
1458 			return 0;
1459 		}
1460 		pr_warning("x2apic already enabled.\n");
1461 		__x2apic_disable();
1462 	}
1463 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1464 	x2apic_state = X2APIC_DISABLED;
1465 	x2apic_mode = 0;
1466 	return 0;
1467 }
1468 early_param("nox2apic", setup_nox2apic);
1469 
1470 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
1471 void x2apic_setup(void)
1472 {
1473 	/*
1474 	 * If x2apic is not in ON state, disable it if already enabled
1475 	 * from BIOS.
1476 	 */
1477 	if (x2apic_state != X2APIC_ON) {
1478 		__x2apic_disable();
1479 		return;
1480 	}
1481 	__x2apic_enable();
1482 }
1483 
1484 static __init void x2apic_disable(void)
1485 {
1486 	u32 x2apic_id;
1487 
1488 	if (x2apic_state != X2APIC_ON)
1489 		goto out;
1490 
1491 	x2apic_id = read_apic_id();
1492 	if (x2apic_id >= 255)
1493 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1494 
1495 	__x2apic_disable();
1496 	register_lapic_address(mp_lapic_addr);
1497 out:
1498 	x2apic_state = X2APIC_DISABLED;
1499 	x2apic_mode = 0;
1500 }
1501 
1502 static __init void x2apic_enable(void)
1503 {
1504 	if (x2apic_state != X2APIC_OFF)
1505 		return;
1506 
1507 	x2apic_mode = 1;
1508 	x2apic_state = X2APIC_ON;
1509 	__x2apic_enable();
1510 }
1511 
1512 static __init void try_to_enable_x2apic(int remap_mode)
1513 {
1514 	if (x2apic_state == X2APIC_DISABLED)
1515 		return;
1516 
1517 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1518 		/* IR is required if there is APIC ID > 255 even when running
1519 		 * under KVM
1520 		 */
1521 		if (max_physical_apicid > 255 ||
1522 		    !hypervisor_x2apic_available()) {
1523 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1524 			x2apic_disable();
1525 			return;
1526 		}
1527 
1528 		/*
1529 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1530 		 * only in physical mode
1531 		 */
1532 		x2apic_phys = 1;
1533 	}
1534 	x2apic_enable();
1535 }
1536 
1537 void __init check_x2apic(void)
1538 {
1539 	if (x2apic_enabled()) {
1540 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1541 		x2apic_mode = 1;
1542 		x2apic_state = X2APIC_ON;
1543 	} else if (!cpu_has_x2apic) {
1544 		x2apic_state = X2APIC_DISABLED;
1545 	}
1546 }
1547 #else /* CONFIG_X86_X2APIC */
1548 static int __init validate_x2apic(void)
1549 {
1550 	if (!apic_is_x2apic_enabled())
1551 		return 0;
1552 	/*
1553 	 * Checkme: Can we simply turn off x2apic here instead of panic?
1554 	 */
1555 	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1556 }
1557 early_initcall(validate_x2apic);
1558 
1559 static inline void try_to_enable_x2apic(int remap_mode) { }
1560 static inline void __x2apic_enable(void) { }
1561 #endif /* !CONFIG_X86_X2APIC */
1562 
1563 static int __init try_to_enable_IR(void)
1564 {
1565 #ifdef CONFIG_X86_IO_APIC
1566 	if (!x2apic_enabled() && skip_ioapic_setup) {
1567 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1568 		return -1;
1569 	}
1570 #endif
1571 	return irq_remapping_enable();
1572 }
1573 
1574 void __init enable_IR_x2apic(void)
1575 {
1576 	unsigned long flags;
1577 	int ret, ir_stat;
1578 
1579 	ir_stat = irq_remapping_prepare();
1580 	if (ir_stat < 0 && !x2apic_supported())
1581 		return;
1582 
1583 	ret = save_ioapic_entries();
1584 	if (ret) {
1585 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1586 		return;
1587 	}
1588 
1589 	local_irq_save(flags);
1590 	legacy_pic->mask_all();
1591 	mask_ioapic_entries();
1592 
1593 	/* If irq_remapping_prepare() succeded, try to enable it */
1594 	if (ir_stat >= 0)
1595 		ir_stat = try_to_enable_IR();
1596 	/* ir_stat contains the remap mode or an error code */
1597 	try_to_enable_x2apic(ir_stat);
1598 
1599 	if (ir_stat < 0)
1600 		restore_ioapic_entries();
1601 	legacy_pic->restore_mask();
1602 	local_irq_restore(flags);
1603 }
1604 
1605 #ifdef CONFIG_X86_64
1606 /*
1607  * Detect and enable local APICs on non-SMP boards.
1608  * Original code written by Keir Fraser.
1609  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1610  * not correctly set up (usually the APIC timer won't work etc.)
1611  */
1612 static int __init detect_init_APIC(void)
1613 {
1614 	if (!cpu_has_apic) {
1615 		pr_info("No local APIC present\n");
1616 		return -1;
1617 	}
1618 
1619 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1620 	return 0;
1621 }
1622 #else
1623 
1624 static int __init apic_verify(void)
1625 {
1626 	u32 features, h, l;
1627 
1628 	/*
1629 	 * The APIC feature bit should now be enabled
1630 	 * in `cpuid'
1631 	 */
1632 	features = cpuid_edx(1);
1633 	if (!(features & (1 << X86_FEATURE_APIC))) {
1634 		pr_warning("Could not enable APIC!\n");
1635 		return -1;
1636 	}
1637 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1638 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1639 
1640 	/* The BIOS may have set up the APIC at some other address */
1641 	if (boot_cpu_data.x86 >= 6) {
1642 		rdmsr(MSR_IA32_APICBASE, l, h);
1643 		if (l & MSR_IA32_APICBASE_ENABLE)
1644 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1645 	}
1646 
1647 	pr_info("Found and enabled local APIC!\n");
1648 	return 0;
1649 }
1650 
1651 int __init apic_force_enable(unsigned long addr)
1652 {
1653 	u32 h, l;
1654 
1655 	if (disable_apic)
1656 		return -1;
1657 
1658 	/*
1659 	 * Some BIOSes disable the local APIC in the APIC_BASE
1660 	 * MSR. This can only be done in software for Intel P6 or later
1661 	 * and AMD K7 (Model > 1) or later.
1662 	 */
1663 	if (boot_cpu_data.x86 >= 6) {
1664 		rdmsr(MSR_IA32_APICBASE, l, h);
1665 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1666 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1667 			l &= ~MSR_IA32_APICBASE_BASE;
1668 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1669 			wrmsr(MSR_IA32_APICBASE, l, h);
1670 			enabled_via_apicbase = 1;
1671 		}
1672 	}
1673 	return apic_verify();
1674 }
1675 
1676 /*
1677  * Detect and initialize APIC
1678  */
1679 static int __init detect_init_APIC(void)
1680 {
1681 	/* Disabled by kernel option? */
1682 	if (disable_apic)
1683 		return -1;
1684 
1685 	switch (boot_cpu_data.x86_vendor) {
1686 	case X86_VENDOR_AMD:
1687 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1688 		    (boot_cpu_data.x86 >= 15))
1689 			break;
1690 		goto no_apic;
1691 	case X86_VENDOR_INTEL:
1692 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1693 		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
1694 			break;
1695 		goto no_apic;
1696 	default:
1697 		goto no_apic;
1698 	}
1699 
1700 	if (!cpu_has_apic) {
1701 		/*
1702 		 * Over-ride BIOS and try to enable the local APIC only if
1703 		 * "lapic" specified.
1704 		 */
1705 		if (!force_enable_local_apic) {
1706 			pr_info("Local APIC disabled by BIOS -- "
1707 				"you can enable it with \"lapic\"\n");
1708 			return -1;
1709 		}
1710 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1711 			return -1;
1712 	} else {
1713 		if (apic_verify())
1714 			return -1;
1715 	}
1716 
1717 	apic_pm_activate();
1718 
1719 	return 0;
1720 
1721 no_apic:
1722 	pr_info("No local APIC present or hardware disabled\n");
1723 	return -1;
1724 }
1725 #endif
1726 
1727 /**
1728  * init_apic_mappings - initialize APIC mappings
1729  */
1730 void __init init_apic_mappings(void)
1731 {
1732 	unsigned int new_apicid;
1733 
1734 	if (x2apic_mode) {
1735 		boot_cpu_physical_apicid = read_apic_id();
1736 		return;
1737 	}
1738 
1739 	/* If no local APIC can be found return early */
1740 	if (!smp_found_config && detect_init_APIC()) {
1741 		/* lets NOP'ify apic operations */
1742 		pr_info("APIC: disable apic facility\n");
1743 		apic_disable();
1744 	} else {
1745 		apic_phys = mp_lapic_addr;
1746 
1747 		/*
1748 		 * acpi lapic path already maps that address in
1749 		 * acpi_register_lapic_address()
1750 		 */
1751 		if (!acpi_lapic && !smp_found_config)
1752 			register_lapic_address(apic_phys);
1753 	}
1754 
1755 	/*
1756 	 * Fetch the APIC ID of the BSP in case we have a
1757 	 * default configuration (or the MP table is broken).
1758 	 */
1759 	new_apicid = read_apic_id();
1760 	if (boot_cpu_physical_apicid != new_apicid) {
1761 		boot_cpu_physical_apicid = new_apicid;
1762 		/*
1763 		 * yeah -- we lie about apic_version
1764 		 * in case if apic was disabled via boot option
1765 		 * but it's not a problem for SMP compiled kernel
1766 		 * since smp_sanity_check is prepared for such a case
1767 		 * and disable smp mode
1768 		 */
1769 		apic_version[new_apicid] =
1770 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1771 	}
1772 }
1773 
1774 void __init register_lapic_address(unsigned long address)
1775 {
1776 	mp_lapic_addr = address;
1777 
1778 	if (!x2apic_mode) {
1779 		set_fixmap_nocache(FIX_APIC_BASE, address);
1780 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1781 			    APIC_BASE, mp_lapic_addr);
1782 	}
1783 	if (boot_cpu_physical_apicid == -1U) {
1784 		boot_cpu_physical_apicid  = read_apic_id();
1785 		apic_version[boot_cpu_physical_apicid] =
1786 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1787 	}
1788 }
1789 
1790 int apic_version[MAX_LOCAL_APIC];
1791 
1792 /*
1793  * Local APIC interrupts
1794  */
1795 
1796 /*
1797  * This interrupt should _never_ happen with our APIC/SMP architecture
1798  */
1799 static inline void __smp_spurious_interrupt(u8 vector)
1800 {
1801 	u32 v;
1802 
1803 	/*
1804 	 * Check if this really is a spurious interrupt and ACK it
1805 	 * if it is a vectored one.  Just in case...
1806 	 * Spurious interrupts should not be ACKed.
1807 	 */
1808 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
1809 	if (v & (1 << (vector & 0x1f)))
1810 		ack_APIC_irq();
1811 
1812 	inc_irq_stat(irq_spurious_count);
1813 
1814 	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1815 	pr_info("spurious APIC interrupt through vector %02x on CPU#%d, "
1816 		"should never happen.\n", vector, smp_processor_id());
1817 }
1818 
1819 __visible void smp_spurious_interrupt(struct pt_regs *regs)
1820 {
1821 	entering_irq();
1822 	__smp_spurious_interrupt(~regs->orig_ax);
1823 	exiting_irq();
1824 }
1825 
1826 __visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1827 {
1828 	u8 vector = ~regs->orig_ax;
1829 
1830 	entering_irq();
1831 	trace_spurious_apic_entry(vector);
1832 	__smp_spurious_interrupt(vector);
1833 	trace_spurious_apic_exit(vector);
1834 	exiting_irq();
1835 }
1836 
1837 /*
1838  * This interrupt should never happen with our APIC/SMP architecture
1839  */
1840 static inline void __smp_error_interrupt(struct pt_regs *regs)
1841 {
1842 	u32 v;
1843 	u32 i = 0;
1844 	static const char * const error_interrupt_reason[] = {
1845 		"Send CS error",		/* APIC Error Bit 0 */
1846 		"Receive CS error",		/* APIC Error Bit 1 */
1847 		"Send accept error",		/* APIC Error Bit 2 */
1848 		"Receive accept error",		/* APIC Error Bit 3 */
1849 		"Redirectable IPI",		/* APIC Error Bit 4 */
1850 		"Send illegal vector",		/* APIC Error Bit 5 */
1851 		"Received illegal vector",	/* APIC Error Bit 6 */
1852 		"Illegal register address",	/* APIC Error Bit 7 */
1853 	};
1854 
1855 	/* First tickle the hardware, only then report what went on. -- REW */
1856 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
1857 		apic_write(APIC_ESR, 0);
1858 	v = apic_read(APIC_ESR);
1859 	ack_APIC_irq();
1860 	atomic_inc(&irq_err_count);
1861 
1862 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
1863 		    smp_processor_id(), v);
1864 
1865 	v &= 0xff;
1866 	while (v) {
1867 		if (v & 0x1)
1868 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1869 		i++;
1870 		v >>= 1;
1871 	}
1872 
1873 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
1874 
1875 }
1876 
1877 __visible void smp_error_interrupt(struct pt_regs *regs)
1878 {
1879 	entering_irq();
1880 	__smp_error_interrupt(regs);
1881 	exiting_irq();
1882 }
1883 
1884 __visible void smp_trace_error_interrupt(struct pt_regs *regs)
1885 {
1886 	entering_irq();
1887 	trace_error_apic_entry(ERROR_APIC_VECTOR);
1888 	__smp_error_interrupt(regs);
1889 	trace_error_apic_exit(ERROR_APIC_VECTOR);
1890 	exiting_irq();
1891 }
1892 
1893 /**
1894  * connect_bsp_APIC - attach the APIC to the interrupt system
1895  */
1896 static void __init connect_bsp_APIC(void)
1897 {
1898 #ifdef CONFIG_X86_32
1899 	if (pic_mode) {
1900 		/*
1901 		 * Do not trust the local APIC being empty at bootup.
1902 		 */
1903 		clear_local_APIC();
1904 		/*
1905 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
1906 		 * local APIC to INT and NMI lines.
1907 		 */
1908 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
1909 				"enabling APIC mode.\n");
1910 		imcr_pic_to_apic();
1911 	}
1912 #endif
1913 }
1914 
1915 /**
1916  * disconnect_bsp_APIC - detach the APIC from the interrupt system
1917  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
1918  *
1919  * Virtual wire mode is necessary to deliver legacy interrupts even when the
1920  * APIC is disabled.
1921  */
1922 void disconnect_bsp_APIC(int virt_wire_setup)
1923 {
1924 	unsigned int value;
1925 
1926 #ifdef CONFIG_X86_32
1927 	if (pic_mode) {
1928 		/*
1929 		 * Put the board back into PIC mode (has an effect only on
1930 		 * certain older boards).  Note that APIC interrupts, including
1931 		 * IPIs, won't work beyond this point!  The only exception are
1932 		 * INIT IPIs.
1933 		 */
1934 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
1935 				"entering PIC mode.\n");
1936 		imcr_apic_to_pic();
1937 		return;
1938 	}
1939 #endif
1940 
1941 	/* Go back to Virtual Wire compatibility mode */
1942 
1943 	/* For the spurious interrupt use vector F, and enable it */
1944 	value = apic_read(APIC_SPIV);
1945 	value &= ~APIC_VECTOR_MASK;
1946 	value |= APIC_SPIV_APIC_ENABLED;
1947 	value |= 0xf;
1948 	apic_write(APIC_SPIV, value);
1949 
1950 	if (!virt_wire_setup) {
1951 		/*
1952 		 * For LVT0 make it edge triggered, active high,
1953 		 * external and enabled
1954 		 */
1955 		value = apic_read(APIC_LVT0);
1956 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1957 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1958 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1959 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1960 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1961 		apic_write(APIC_LVT0, value);
1962 	} else {
1963 		/* Disable LVT0 */
1964 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
1965 	}
1966 
1967 	/*
1968 	 * For LVT1 make it edge triggered, active high,
1969 	 * nmi and enabled
1970 	 */
1971 	value = apic_read(APIC_LVT1);
1972 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1973 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1974 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1975 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1976 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1977 	apic_write(APIC_LVT1, value);
1978 }
1979 
1980 int generic_processor_info(int apicid, int version)
1981 {
1982 	int cpu, max = nr_cpu_ids;
1983 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
1984 				phys_cpu_present_map);
1985 
1986 	/*
1987 	 * boot_cpu_physical_apicid is designed to have the apicid
1988 	 * returned by read_apic_id(), i.e, the apicid of the
1989 	 * currently booting-up processor. However, on some platforms,
1990 	 * it is temporarily modified by the apicid reported as BSP
1991 	 * through MP table. Concretely:
1992 	 *
1993 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
1994 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
1995 	 *
1996 	 * This function is executed with the modified
1997 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
1998 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
1999 	 *
2000 	 * Since fixing handling of boot_cpu_physical_apicid requires
2001 	 * another discussion and tests on each platform, we leave it
2002 	 * for now and here we use read_apic_id() directly in this
2003 	 * function, generic_processor_info().
2004 	 */
2005 	if (disabled_cpu_apicid != BAD_APICID &&
2006 	    disabled_cpu_apicid != read_apic_id() &&
2007 	    disabled_cpu_apicid == apicid) {
2008 		int thiscpu = num_processors + disabled_cpus;
2009 
2010 		pr_warning("APIC: Disabling requested cpu."
2011 			   " Processor %d/0x%x ignored.\n",
2012 			   thiscpu, apicid);
2013 
2014 		disabled_cpus++;
2015 		return -ENODEV;
2016 	}
2017 
2018 	/*
2019 	 * If boot cpu has not been detected yet, then only allow upto
2020 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2021 	 */
2022 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2023 	    apicid != boot_cpu_physical_apicid) {
2024 		int thiscpu = max + disabled_cpus - 1;
2025 
2026 		pr_warning(
2027 			"ACPI: NR_CPUS/possible_cpus limit of %i almost"
2028 			" reached. Keeping one slot for boot cpu."
2029 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2030 
2031 		disabled_cpus++;
2032 		return -ENODEV;
2033 	}
2034 
2035 	if (num_processors >= nr_cpu_ids) {
2036 		int thiscpu = max + disabled_cpus;
2037 
2038 		pr_warning(
2039 			"ACPI: NR_CPUS/possible_cpus limit of %i reached."
2040 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2041 
2042 		disabled_cpus++;
2043 		return -EINVAL;
2044 	}
2045 
2046 	num_processors++;
2047 	if (apicid == boot_cpu_physical_apicid) {
2048 		/*
2049 		 * x86_bios_cpu_apicid is required to have processors listed
2050 		 * in same order as logical cpu numbers. Hence the first
2051 		 * entry is BSP, and so on.
2052 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2053 		 * for BSP.
2054 		 */
2055 		cpu = 0;
2056 	} else
2057 		cpu = cpumask_next_zero(-1, cpu_present_mask);
2058 
2059 	/*
2060 	 * Validate version
2061 	 */
2062 	if (version == 0x0) {
2063 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2064 			   cpu, apicid);
2065 		version = 0x10;
2066 	}
2067 	apic_version[apicid] = version;
2068 
2069 	if (version != apic_version[boot_cpu_physical_apicid]) {
2070 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2071 			apic_version[boot_cpu_physical_apicid], cpu, version);
2072 	}
2073 
2074 	physid_set(apicid, phys_cpu_present_map);
2075 	if (apicid > max_physical_apicid)
2076 		max_physical_apicid = apicid;
2077 
2078 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2079 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2080 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2081 #endif
2082 #ifdef CONFIG_X86_32
2083 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2084 		apic->x86_32_early_logical_apicid(cpu);
2085 #endif
2086 	set_cpu_possible(cpu, true);
2087 	set_cpu_present(cpu, true);
2088 
2089 	return cpu;
2090 }
2091 
2092 int hard_smp_processor_id(void)
2093 {
2094 	return read_apic_id();
2095 }
2096 
2097 void default_init_apic_ldr(void)
2098 {
2099 	unsigned long val;
2100 
2101 	apic_write(APIC_DFR, APIC_DFR_VALUE);
2102 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2103 	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2104 	apic_write(APIC_LDR, val);
2105 }
2106 
2107 int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2108 				   const struct cpumask *andmask,
2109 				   unsigned int *apicid)
2110 {
2111 	unsigned int cpu;
2112 
2113 	for_each_cpu_and(cpu, cpumask, andmask) {
2114 		if (cpumask_test_cpu(cpu, cpu_online_mask))
2115 			break;
2116 	}
2117 
2118 	if (likely(cpu < nr_cpu_ids)) {
2119 		*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2120 		return 0;
2121 	}
2122 
2123 	return -EINVAL;
2124 }
2125 
2126 /*
2127  * Override the generic EOI implementation with an optimized version.
2128  * Only called during early boot when only one CPU is active and with
2129  * interrupts disabled, so we know this does not race with actual APIC driver
2130  * use.
2131  */
2132 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2133 {
2134 	struct apic **drv;
2135 
2136 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2137 		/* Should happen once for each apic */
2138 		WARN_ON((*drv)->eoi_write == eoi_write);
2139 		(*drv)->eoi_write = eoi_write;
2140 	}
2141 }
2142 
2143 static void __init apic_bsp_up_setup(void)
2144 {
2145 #ifdef CONFIG_X86_64
2146 	apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
2147 #else
2148 	/*
2149 	 * Hack: In case of kdump, after a crash, kernel might be booting
2150 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2151 	 * might be zero if read from MP tables. Get it from LAPIC.
2152 	 */
2153 # ifdef CONFIG_CRASH_DUMP
2154 	boot_cpu_physical_apicid = read_apic_id();
2155 # endif
2156 #endif
2157 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2158 }
2159 
2160 /**
2161  * apic_bsp_setup - Setup function for local apic and io-apic
2162  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2163  *
2164  * Returns:
2165  * apic_id of BSP APIC
2166  */
2167 int __init apic_bsp_setup(bool upmode)
2168 {
2169 	int id;
2170 
2171 	connect_bsp_APIC();
2172 	if (upmode)
2173 		apic_bsp_up_setup();
2174 	setup_local_APIC();
2175 
2176 	if (x2apic_mode)
2177 		id = apic_read(APIC_LDR);
2178 	else
2179 		id = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
2180 
2181 	enable_IO_APIC();
2182 	end_local_APIC_setup();
2183 	irq_remap_enable_fault_handling();
2184 	setup_IO_APIC();
2185 	/* Setup local timer */
2186 	x86_init.timers.setup_percpu_clockev();
2187 	return id;
2188 }
2189 
2190 /*
2191  * This initializes the IO-APIC and APIC hardware if this is
2192  * a UP kernel.
2193  */
2194 int __init APIC_init_uniprocessor(void)
2195 {
2196 	if (disable_apic) {
2197 		pr_info("Apic disabled\n");
2198 		return -1;
2199 	}
2200 #ifdef CONFIG_X86_64
2201 	if (!cpu_has_apic) {
2202 		disable_apic = 1;
2203 		pr_info("Apic disabled by BIOS\n");
2204 		return -1;
2205 	}
2206 #else
2207 	if (!smp_found_config && !cpu_has_apic)
2208 		return -1;
2209 
2210 	/*
2211 	 * Complain if the BIOS pretends there is one.
2212 	 */
2213 	if (!cpu_has_apic &&
2214 	    APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
2215 		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
2216 			boot_cpu_physical_apicid);
2217 		return -1;
2218 	}
2219 #endif
2220 
2221 	if (!smp_found_config)
2222 		disable_ioapic_support();
2223 
2224 	default_setup_apic_routing();
2225 	apic_bsp_setup(true);
2226 	return 0;
2227 }
2228 
2229 #ifdef CONFIG_UP_LATE_INIT
2230 void __init up_late_init(void)
2231 {
2232 	APIC_init_uniprocessor();
2233 }
2234 #endif
2235 
2236 /*
2237  * Power management
2238  */
2239 #ifdef CONFIG_PM
2240 
2241 static struct {
2242 	/*
2243 	 * 'active' is true if the local APIC was enabled by us and
2244 	 * not the BIOS; this signifies that we are also responsible
2245 	 * for disabling it before entering apm/acpi suspend
2246 	 */
2247 	int active;
2248 	/* r/w apic fields */
2249 	unsigned int apic_id;
2250 	unsigned int apic_taskpri;
2251 	unsigned int apic_ldr;
2252 	unsigned int apic_dfr;
2253 	unsigned int apic_spiv;
2254 	unsigned int apic_lvtt;
2255 	unsigned int apic_lvtpc;
2256 	unsigned int apic_lvt0;
2257 	unsigned int apic_lvt1;
2258 	unsigned int apic_lvterr;
2259 	unsigned int apic_tmict;
2260 	unsigned int apic_tdcr;
2261 	unsigned int apic_thmr;
2262 } apic_pm_state;
2263 
2264 static int lapic_suspend(void)
2265 {
2266 	unsigned long flags;
2267 	int maxlvt;
2268 
2269 	if (!apic_pm_state.active)
2270 		return 0;
2271 
2272 	maxlvt = lapic_get_maxlvt();
2273 
2274 	apic_pm_state.apic_id = apic_read(APIC_ID);
2275 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2276 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2277 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2278 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2279 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2280 	if (maxlvt >= 4)
2281 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2282 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2283 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2284 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2285 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2286 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2287 #ifdef CONFIG_X86_THERMAL_VECTOR
2288 	if (maxlvt >= 5)
2289 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2290 #endif
2291 
2292 	local_irq_save(flags);
2293 	disable_local_APIC();
2294 
2295 	irq_remapping_disable();
2296 
2297 	local_irq_restore(flags);
2298 	return 0;
2299 }
2300 
2301 static void lapic_resume(void)
2302 {
2303 	unsigned int l, h;
2304 	unsigned long flags;
2305 	int maxlvt;
2306 
2307 	if (!apic_pm_state.active)
2308 		return;
2309 
2310 	local_irq_save(flags);
2311 
2312 	/*
2313 	 * IO-APIC and PIC have their own resume routines.
2314 	 * We just mask them here to make sure the interrupt
2315 	 * subsystem is completely quiet while we enable x2apic
2316 	 * and interrupt-remapping.
2317 	 */
2318 	mask_ioapic_entries();
2319 	legacy_pic->mask_all();
2320 
2321 	if (x2apic_mode) {
2322 		__x2apic_enable();
2323 	} else {
2324 		/*
2325 		 * Make sure the APICBASE points to the right address
2326 		 *
2327 		 * FIXME! This will be wrong if we ever support suspend on
2328 		 * SMP! We'll need to do this as part of the CPU restore!
2329 		 */
2330 		if (boot_cpu_data.x86 >= 6) {
2331 			rdmsr(MSR_IA32_APICBASE, l, h);
2332 			l &= ~MSR_IA32_APICBASE_BASE;
2333 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2334 			wrmsr(MSR_IA32_APICBASE, l, h);
2335 		}
2336 	}
2337 
2338 	maxlvt = lapic_get_maxlvt();
2339 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2340 	apic_write(APIC_ID, apic_pm_state.apic_id);
2341 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2342 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2343 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2344 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2345 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2346 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2347 #if defined(CONFIG_X86_MCE_INTEL)
2348 	if (maxlvt >= 5)
2349 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2350 #endif
2351 	if (maxlvt >= 4)
2352 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2353 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2354 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2355 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2356 	apic_write(APIC_ESR, 0);
2357 	apic_read(APIC_ESR);
2358 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2359 	apic_write(APIC_ESR, 0);
2360 	apic_read(APIC_ESR);
2361 
2362 	irq_remapping_reenable(x2apic_mode);
2363 
2364 	local_irq_restore(flags);
2365 }
2366 
2367 /*
2368  * This device has no shutdown method - fully functioning local APICs
2369  * are needed on every CPU up until machine_halt/restart/poweroff.
2370  */
2371 
2372 static struct syscore_ops lapic_syscore_ops = {
2373 	.resume		= lapic_resume,
2374 	.suspend	= lapic_suspend,
2375 };
2376 
2377 static void apic_pm_activate(void)
2378 {
2379 	apic_pm_state.active = 1;
2380 }
2381 
2382 static int __init init_lapic_sysfs(void)
2383 {
2384 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2385 	if (cpu_has_apic)
2386 		register_syscore_ops(&lapic_syscore_ops);
2387 
2388 	return 0;
2389 }
2390 
2391 /* local apic needs to resume before other devices access its registers. */
2392 core_initcall(init_lapic_sysfs);
2393 
2394 #else	/* CONFIG_PM */
2395 
2396 static void apic_pm_activate(void) { }
2397 
2398 #endif	/* CONFIG_PM */
2399 
2400 #ifdef CONFIG_X86_64
2401 
2402 static int multi_checked;
2403 static int multi;
2404 
2405 static int set_multi(const struct dmi_system_id *d)
2406 {
2407 	if (multi)
2408 		return 0;
2409 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2410 	multi = 1;
2411 	return 0;
2412 }
2413 
2414 static const struct dmi_system_id multi_dmi_table[] = {
2415 	{
2416 		.callback = set_multi,
2417 		.ident = "IBM System Summit2",
2418 		.matches = {
2419 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2420 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2421 		},
2422 	},
2423 	{}
2424 };
2425 
2426 static void dmi_check_multi(void)
2427 {
2428 	if (multi_checked)
2429 		return;
2430 
2431 	dmi_check_system(multi_dmi_table);
2432 	multi_checked = 1;
2433 }
2434 
2435 /*
2436  * apic_is_clustered_box() -- Check if we can expect good TSC
2437  *
2438  * Thus far, the major user of this is IBM's Summit2 series:
2439  * Clustered boxes may have unsynced TSC problems if they are
2440  * multi-chassis.
2441  * Use DMI to check them
2442  */
2443 int apic_is_clustered_box(void)
2444 {
2445 	dmi_check_multi();
2446 	return multi;
2447 }
2448 #endif
2449 
2450 /*
2451  * APIC command line parameters
2452  */
2453 static int __init setup_disableapic(char *arg)
2454 {
2455 	disable_apic = 1;
2456 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2457 	return 0;
2458 }
2459 early_param("disableapic", setup_disableapic);
2460 
2461 /* same as disableapic, for compatibility */
2462 static int __init setup_nolapic(char *arg)
2463 {
2464 	return setup_disableapic(arg);
2465 }
2466 early_param("nolapic", setup_nolapic);
2467 
2468 static int __init parse_lapic_timer_c2_ok(char *arg)
2469 {
2470 	local_apic_timer_c2_ok = 1;
2471 	return 0;
2472 }
2473 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2474 
2475 static int __init parse_disable_apic_timer(char *arg)
2476 {
2477 	disable_apic_timer = 1;
2478 	return 0;
2479 }
2480 early_param("noapictimer", parse_disable_apic_timer);
2481 
2482 static int __init parse_nolapic_timer(char *arg)
2483 {
2484 	disable_apic_timer = 1;
2485 	return 0;
2486 }
2487 early_param("nolapic_timer", parse_nolapic_timer);
2488 
2489 static int __init apic_set_verbosity(char *arg)
2490 {
2491 	if (!arg)  {
2492 #ifdef CONFIG_X86_64
2493 		skip_ioapic_setup = 0;
2494 		return 0;
2495 #endif
2496 		return -EINVAL;
2497 	}
2498 
2499 	if (strcmp("debug", arg) == 0)
2500 		apic_verbosity = APIC_DEBUG;
2501 	else if (strcmp("verbose", arg) == 0)
2502 		apic_verbosity = APIC_VERBOSE;
2503 	else {
2504 		pr_warning("APIC Verbosity level %s not recognised"
2505 			" use apic=verbose or apic=debug\n", arg);
2506 		return -EINVAL;
2507 	}
2508 
2509 	return 0;
2510 }
2511 early_param("apic", apic_set_verbosity);
2512 
2513 static int __init lapic_insert_resource(void)
2514 {
2515 	if (!apic_phys)
2516 		return -1;
2517 
2518 	/* Put local APIC into the resource map. */
2519 	lapic_resource.start = apic_phys;
2520 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2521 	insert_resource(&iomem_resource, &lapic_resource);
2522 
2523 	return 0;
2524 }
2525 
2526 /*
2527  * need call insert after e820_reserve_resources()
2528  * that is using request_resource
2529  */
2530 late_initcall(lapic_insert_resource);
2531 
2532 static int __init apic_set_disabled_cpu_apicid(char *arg)
2533 {
2534 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2535 		return -EINVAL;
2536 
2537 	return 0;
2538 }
2539 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2540