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
59unsigned int num_processors;
60
61unsigned disabled_cpus;
62
63/* Processor that is doing the boot up */
64unsigned int boot_cpu_physical_apicid = -1U;
65EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
66
67/*
68 * The highest APIC ID seen during enumeration.
69 */
70static unsigned int max_physical_apicid;
71
72/*
73 * Bitmask of physically existing CPUs:
74 */
75physid_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 */
82static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
83
84/*
85 * Map cpu index to physical APIC ID
86 */
87DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
88DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
89EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
90EXPORT_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 */
100DEFINE_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 */
103static 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 */
113static 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
121static 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 */
135static int force_enable_local_apic __initdata;
136
137/* Control whether x2APIC mode is enabled or not */
138static bool nox2apic __initdata;
139
140/*
141 * APIC command line parameters
142 */
143static int __init parse_lapic(char *arg)
144{
145	if (config_enabled(CONFIG_X86_32) && !arg)
146		force_enable_local_apic = 1;
147	else if (arg && !strncmp(arg, "notscdeadline", 13))
148		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
149	return 0;
150}
151early_param("lapic", parse_lapic);
152
153#ifdef CONFIG_X86_64
154static int apic_calibrate_pmtmr __initdata;
155static __init int setup_apicpmtimer(char *s)
156{
157	apic_calibrate_pmtmr = 1;
158	notsc_setup(NULL);
159	return 0;
160}
161__setup("apicpmtimer", setup_apicpmtimer);
162#endif
163
164int x2apic_mode;
165#ifdef CONFIG_X86_X2APIC
166/* x2apic enabled before OS handover */
167int x2apic_preenabled;
168static int x2apic_disabled;
169static int __init setup_nox2apic(char *str)
170{
171	if (x2apic_enabled()) {
172		int apicid = native_apic_msr_read(APIC_ID);
173
174		if (apicid >= 255) {
175			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
176				   apicid);
177			return 0;
178		}
179
180		pr_warning("x2apic already enabled. will disable it\n");
181	} else
182		setup_clear_cpu_cap(X86_FEATURE_X2APIC);
183
184	nox2apic = true;
185
186	return 0;
187}
188early_param("nox2apic", setup_nox2apic);
189#endif
190
191unsigned long mp_lapic_addr;
192int disable_apic;
193/* Disable local APIC timer from the kernel commandline or via dmi quirk */
194static int disable_apic_timer __initdata;
195/* Local APIC timer works in C2 */
196int local_apic_timer_c2_ok;
197EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
198
199int first_system_vector = 0xfe;
200
201/*
202 * Debug level, exported for io_apic.c
203 */
204unsigned int apic_verbosity;
205
206int pic_mode;
207
208/* Have we found an MP table */
209int smp_found_config;
210
211static struct resource lapic_resource = {
212	.name = "Local APIC",
213	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
214};
215
216unsigned int lapic_timer_frequency = 0;
217
218static void apic_pm_activate(void);
219
220static unsigned long apic_phys;
221
222/*
223 * Get the LAPIC version
224 */
225static inline int lapic_get_version(void)
226{
227	return GET_APIC_VERSION(apic_read(APIC_LVR));
228}
229
230/*
231 * Check, if the APIC is integrated or a separate chip
232 */
233static inline int lapic_is_integrated(void)
234{
235#ifdef CONFIG_X86_64
236	return 1;
237#else
238	return APIC_INTEGRATED(lapic_get_version());
239#endif
240}
241
242/*
243 * Check, whether this is a modern or a first generation APIC
244 */
245static int modern_apic(void)
246{
247	/* AMD systems use old APIC versions, so check the CPU */
248	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
249	    boot_cpu_data.x86 >= 0xf)
250		return 1;
251	return lapic_get_version() >= 0x14;
252}
253
254/*
255 * right after this call apic become NOOP driven
256 * so apic->write/read doesn't do anything
257 */
258static void __init apic_disable(void)
259{
260	pr_info("APIC: switched to apic NOOP\n");
261	apic = &apic_noop;
262}
263
264void native_apic_wait_icr_idle(void)
265{
266	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
267		cpu_relax();
268}
269
270u32 native_safe_apic_wait_icr_idle(void)
271{
272	u32 send_status;
273	int timeout;
274
275	timeout = 0;
276	do {
277		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
278		if (!send_status)
279			break;
280		inc_irq_stat(icr_read_retry_count);
281		udelay(100);
282	} while (timeout++ < 1000);
283
284	return send_status;
285}
286
287void native_apic_icr_write(u32 low, u32 id)
288{
289	unsigned long flags;
290
291	local_irq_save(flags);
292	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
293	apic_write(APIC_ICR, low);
294	local_irq_restore(flags);
295}
296
297u64 native_apic_icr_read(void)
298{
299	u32 icr1, icr2;
300
301	icr2 = apic_read(APIC_ICR2);
302	icr1 = apic_read(APIC_ICR);
303
304	return icr1 | ((u64)icr2 << 32);
305}
306
307#ifdef CONFIG_X86_32
308/**
309 * get_physical_broadcast - Get number of physical broadcast IDs
310 */
311int get_physical_broadcast(void)
312{
313	return modern_apic() ? 0xff : 0xf;
314}
315#endif
316
317/**
318 * lapic_get_maxlvt - get the maximum number of local vector table entries
319 */
320int lapic_get_maxlvt(void)
321{
322	unsigned int v;
323
324	v = apic_read(APIC_LVR);
325	/*
326	 * - we always have APIC integrated on 64bit mode
327	 * - 82489DXs do not report # of LVT entries
328	 */
329	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
330}
331
332/*
333 * Local APIC timer
334 */
335
336/* Clock divisor */
337#define APIC_DIVISOR 16
338#define TSC_DIVISOR  32
339
340/*
341 * This function sets up the local APIC timer, with a timeout of
342 * 'clocks' APIC bus clock. During calibration we actually call
343 * this function twice on the boot CPU, once with a bogus timeout
344 * value, second time for real. The other (noncalibrating) CPUs
345 * call this function only once, with the real, calibrated value.
346 *
347 * We do reads before writes even if unnecessary, to get around the
348 * P5 APIC double write bug.
349 */
350static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
351{
352	unsigned int lvtt_value, tmp_value;
353
354	lvtt_value = LOCAL_TIMER_VECTOR;
355	if (!oneshot)
356		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
357	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
358		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
359
360	if (!lapic_is_integrated())
361		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
362
363	if (!irqen)
364		lvtt_value |= APIC_LVT_MASKED;
365
366	apic_write(APIC_LVTT, lvtt_value);
367
368	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
369		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
370		return;
371	}
372
373	/*
374	 * Divide PICLK by 16
375	 */
376	tmp_value = apic_read(APIC_TDCR);
377	apic_write(APIC_TDCR,
378		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
379		APIC_TDR_DIV_16);
380
381	if (!oneshot)
382		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
383}
384
385/*
386 * Setup extended LVT, AMD specific
387 *
388 * Software should use the LVT offsets the BIOS provides.  The offsets
389 * are determined by the subsystems using it like those for MCE
390 * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
391 * are supported. Beginning with family 10h at least 4 offsets are
392 * available.
393 *
394 * Since the offsets must be consistent for all cores, we keep track
395 * of the LVT offsets in software and reserve the offset for the same
396 * vector also to be used on other cores. An offset is freed by
397 * setting the entry to APIC_EILVT_MASKED.
398 *
399 * If the BIOS is right, there should be no conflicts. Otherwise a
400 * "[Firmware Bug]: ..." error message is generated. However, if
401 * software does not properly determines the offsets, it is not
402 * necessarily a BIOS bug.
403 */
404
405static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
406
407static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
408{
409	return (old & APIC_EILVT_MASKED)
410		|| (new == APIC_EILVT_MASKED)
411		|| ((new & ~APIC_EILVT_MASKED) == old);
412}
413
414static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
415{
416	unsigned int rsvd, vector;
417
418	if (offset >= APIC_EILVT_NR_MAX)
419		return ~0;
420
421	rsvd = atomic_read(&eilvt_offsets[offset]);
422	do {
423		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
424		if (vector && !eilvt_entry_is_changeable(vector, new))
425			/* may not change if vectors are different */
426			return rsvd;
427		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
428	} while (rsvd != new);
429
430	rsvd &= ~APIC_EILVT_MASKED;
431	if (rsvd && rsvd != vector)
432		pr_info("LVT offset %d assigned for vector 0x%02x\n",
433			offset, rsvd);
434
435	return new;
436}
437
438/*
439 * If mask=1, the LVT entry does not generate interrupts while mask=0
440 * enables the vector. See also the BKDGs. Must be called with
441 * preemption disabled.
442 */
443
444int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
445{
446	unsigned long reg = APIC_EILVTn(offset);
447	unsigned int new, old, reserved;
448
449	new = (mask << 16) | (msg_type << 8) | vector;
450	old = apic_read(reg);
451	reserved = reserve_eilvt_offset(offset, new);
452
453	if (reserved != new) {
454		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
455		       "vector 0x%x, but the register is already in use for "
456		       "vector 0x%x on another cpu\n",
457		       smp_processor_id(), reg, offset, new, reserved);
458		return -EINVAL;
459	}
460
461	if (!eilvt_entry_is_changeable(old, new)) {
462		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
463		       "vector 0x%x, but the register is already in use for "
464		       "vector 0x%x on this cpu\n",
465		       smp_processor_id(), reg, offset, new, old);
466		return -EBUSY;
467	}
468
469	apic_write(reg, new);
470
471	return 0;
472}
473EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
474
475/*
476 * Program the next event, relative to now
477 */
478static int lapic_next_event(unsigned long delta,
479			    struct clock_event_device *evt)
480{
481	apic_write(APIC_TMICT, delta);
482	return 0;
483}
484
485static int lapic_next_deadline(unsigned long delta,
486			       struct clock_event_device *evt)
487{
488	u64 tsc;
489
490	rdtscll(tsc);
491	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
492	return 0;
493}
494
495/*
496 * Setup the lapic timer in periodic or oneshot mode
497 */
498static void lapic_timer_setup(enum clock_event_mode mode,
499			      struct clock_event_device *evt)
500{
501	unsigned long flags;
502	unsigned int v;
503
504	/* Lapic used as dummy for broadcast ? */
505	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
506		return;
507
508	local_irq_save(flags);
509
510	switch (mode) {
511	case CLOCK_EVT_MODE_PERIODIC:
512	case CLOCK_EVT_MODE_ONESHOT:
513		__setup_APIC_LVTT(lapic_timer_frequency,
514				  mode != CLOCK_EVT_MODE_PERIODIC, 1);
515		break;
516	case CLOCK_EVT_MODE_UNUSED:
517	case CLOCK_EVT_MODE_SHUTDOWN:
518		v = apic_read(APIC_LVTT);
519		v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
520		apic_write(APIC_LVTT, v);
521		apic_write(APIC_TMICT, 0);
522		break;
523	case CLOCK_EVT_MODE_RESUME:
524		/* Nothing to do here */
525		break;
526	}
527
528	local_irq_restore(flags);
529}
530
531/*
532 * Local APIC timer broadcast function
533 */
534static void lapic_timer_broadcast(const struct cpumask *mask)
535{
536#ifdef CONFIG_SMP
537	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
538#endif
539}
540
541
542/*
543 * The local apic timer can be used for any function which is CPU local.
544 */
545static struct clock_event_device lapic_clockevent = {
546	.name		= "lapic",
547	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
548			| CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
549	.shift		= 32,
550	.set_mode	= lapic_timer_setup,
551	.set_next_event	= lapic_next_event,
552	.broadcast	= lapic_timer_broadcast,
553	.rating		= 100,
554	.irq		= -1,
555};
556static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
557
558/*
559 * Setup the local APIC timer for this CPU. Copy the initialized values
560 * of the boot CPU and register the clock event in the framework.
561 */
562static void setup_APIC_timer(void)
563{
564	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
565
566	if (this_cpu_has(X86_FEATURE_ARAT)) {
567		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
568		/* Make LAPIC timer preferrable over percpu HPET */
569		lapic_clockevent.rating = 150;
570	}
571
572	memcpy(levt, &lapic_clockevent, sizeof(*levt));
573	levt->cpumask = cpumask_of(smp_processor_id());
574
575	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
576		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
577				    CLOCK_EVT_FEAT_DUMMY);
578		levt->set_next_event = lapic_next_deadline;
579		clockevents_config_and_register(levt,
580						(tsc_khz / TSC_DIVISOR) * 1000,
581						0xF, ~0UL);
582	} else
583		clockevents_register_device(levt);
584}
585
586/*
587 * In this functions we calibrate APIC bus clocks to the external timer.
588 *
589 * We want to do the calibration only once since we want to have local timer
590 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
591 * frequency.
592 *
593 * This was previously done by reading the PIT/HPET and waiting for a wrap
594 * around to find out, that a tick has elapsed. I have a box, where the PIT
595 * readout is broken, so it never gets out of the wait loop again. This was
596 * also reported by others.
597 *
598 * Monitoring the jiffies value is inaccurate and the clockevents
599 * infrastructure allows us to do a simple substitution of the interrupt
600 * handler.
601 *
602 * The calibration routine also uses the pm_timer when possible, as the PIT
603 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
604 * back to normal later in the boot process).
605 */
606
607#define LAPIC_CAL_LOOPS		(HZ/10)
608
609static __initdata int lapic_cal_loops = -1;
610static __initdata long lapic_cal_t1, lapic_cal_t2;
611static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
612static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
613static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
614
615/*
616 * Temporary interrupt handler.
617 */
618static void __init lapic_cal_handler(struct clock_event_device *dev)
619{
620	unsigned long long tsc = 0;
621	long tapic = apic_read(APIC_TMCCT);
622	unsigned long pm = acpi_pm_read_early();
623
624	if (cpu_has_tsc)
625		rdtscll(tsc);
626
627	switch (lapic_cal_loops++) {
628	case 0:
629		lapic_cal_t1 = tapic;
630		lapic_cal_tsc1 = tsc;
631		lapic_cal_pm1 = pm;
632		lapic_cal_j1 = jiffies;
633		break;
634
635	case LAPIC_CAL_LOOPS:
636		lapic_cal_t2 = tapic;
637		lapic_cal_tsc2 = tsc;
638		if (pm < lapic_cal_pm1)
639			pm += ACPI_PM_OVRRUN;
640		lapic_cal_pm2 = pm;
641		lapic_cal_j2 = jiffies;
642		break;
643	}
644}
645
646static int __init
647calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
648{
649	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
650	const long pm_thresh = pm_100ms / 100;
651	unsigned long mult;
652	u64 res;
653
654#ifndef CONFIG_X86_PM_TIMER
655	return -1;
656#endif
657
658	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
659
660	/* Check, if the PM timer is available */
661	if (!deltapm)
662		return -1;
663
664	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
665
666	if (deltapm > (pm_100ms - pm_thresh) &&
667	    deltapm < (pm_100ms + pm_thresh)) {
668		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
669		return 0;
670	}
671
672	res = (((u64)deltapm) *  mult) >> 22;
673	do_div(res, 1000000);
674	pr_warning("APIC calibration not consistent "
675		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
676
677	/* Correct the lapic counter value */
678	res = (((u64)(*delta)) * pm_100ms);
679	do_div(res, deltapm);
680	pr_info("APIC delta adjusted to PM-Timer: "
681		"%lu (%ld)\n", (unsigned long)res, *delta);
682	*delta = (long)res;
683
684	/* Correct the tsc counter value */
685	if (cpu_has_tsc) {
686		res = (((u64)(*deltatsc)) * pm_100ms);
687		do_div(res, deltapm);
688		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
689					  "PM-Timer: %lu (%ld)\n",
690					(unsigned long)res, *deltatsc);
691		*deltatsc = (long)res;
692	}
693
694	return 0;
695}
696
697static int __init calibrate_APIC_clock(void)
698{
699	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
700	void (*real_handler)(struct clock_event_device *dev);
701	unsigned long deltaj;
702	long delta, deltatsc;
703	int pm_referenced = 0;
704
705	/**
706	 * check if lapic timer has already been calibrated by platform
707	 * specific routine, such as tsc calibration code. if so, we just fill
708	 * in the clockevent structure and return.
709	 */
710
711	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
712		return 0;
713	} else if (lapic_timer_frequency) {
714		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
715				lapic_timer_frequency);
716		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
717					TICK_NSEC, lapic_clockevent.shift);
718		lapic_clockevent.max_delta_ns =
719			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
720		lapic_clockevent.min_delta_ns =
721			clockevent_delta2ns(0xF, &lapic_clockevent);
722		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
723		return 0;
724	}
725
726	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
727		    "calibrating APIC timer ...\n");
728
729	local_irq_disable();
730
731	/* Replace the global interrupt handler */
732	real_handler = global_clock_event->event_handler;
733	global_clock_event->event_handler = lapic_cal_handler;
734
735	/*
736	 * Setup the APIC counter to maximum. There is no way the lapic
737	 * can underflow in the 100ms detection time frame
738	 */
739	__setup_APIC_LVTT(0xffffffff, 0, 0);
740
741	/* Let the interrupts run */
742	local_irq_enable();
743
744	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
745		cpu_relax();
746
747	local_irq_disable();
748
749	/* Restore the real event handler */
750	global_clock_event->event_handler = real_handler;
751
752	/* Build delta t1-t2 as apic timer counts down */
753	delta = lapic_cal_t1 - lapic_cal_t2;
754	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
755
756	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
757
758	/* we trust the PM based calibration if possible */
759	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
760					&delta, &deltatsc);
761
762	/* Calculate the scaled math multiplication factor */
763	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
764				       lapic_clockevent.shift);
765	lapic_clockevent.max_delta_ns =
766		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
767	lapic_clockevent.min_delta_ns =
768		clockevent_delta2ns(0xF, &lapic_clockevent);
769
770	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
771
772	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
773	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
774	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
775		    lapic_timer_frequency);
776
777	if (cpu_has_tsc) {
778		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
779			    "%ld.%04ld MHz.\n",
780			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
781			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
782	}
783
784	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
785		    "%u.%04u MHz.\n",
786		    lapic_timer_frequency / (1000000 / HZ),
787		    lapic_timer_frequency % (1000000 / HZ));
788
789	/*
790	 * Do a sanity check on the APIC calibration result
791	 */
792	if (lapic_timer_frequency < (1000000 / HZ)) {
793		local_irq_enable();
794		pr_warning("APIC frequency too slow, disabling apic timer\n");
795		return -1;
796	}
797
798	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
799
800	/*
801	 * PM timer calibration failed or not turned on
802	 * so lets try APIC timer based calibration
803	 */
804	if (!pm_referenced) {
805		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
806
807		/*
808		 * Setup the apic timer manually
809		 */
810		levt->event_handler = lapic_cal_handler;
811		lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
812		lapic_cal_loops = -1;
813
814		/* Let the interrupts run */
815		local_irq_enable();
816
817		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
818			cpu_relax();
819
820		/* Stop the lapic timer */
821		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
822
823		/* Jiffies delta */
824		deltaj = lapic_cal_j2 - lapic_cal_j1;
825		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
826
827		/* Check, if the jiffies result is consistent */
828		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
829			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
830		else
831			levt->features |= CLOCK_EVT_FEAT_DUMMY;
832	} else
833		local_irq_enable();
834
835	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
836		pr_warning("APIC timer disabled due to verification failure\n");
837			return -1;
838	}
839
840	return 0;
841}
842
843/*
844 * Setup the boot APIC
845 *
846 * Calibrate and verify the result.
847 */
848void __init setup_boot_APIC_clock(void)
849{
850	/*
851	 * The local apic timer can be disabled via the kernel
852	 * commandline or from the CPU detection code. Register the lapic
853	 * timer as a dummy clock event source on SMP systems, so the
854	 * broadcast mechanism is used. On UP systems simply ignore it.
855	 */
856	if (disable_apic_timer) {
857		pr_info("Disabling APIC timer\n");
858		/* No broadcast on UP ! */
859		if (num_possible_cpus() > 1) {
860			lapic_clockevent.mult = 1;
861			setup_APIC_timer();
862		}
863		return;
864	}
865
866	if (calibrate_APIC_clock()) {
867		/* No broadcast on UP ! */
868		if (num_possible_cpus() > 1)
869			setup_APIC_timer();
870		return;
871	}
872
873	/*
874	 * If nmi_watchdog is set to IO_APIC, we need the
875	 * PIT/HPET going.  Otherwise register lapic as a dummy
876	 * device.
877	 */
878	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
879
880	/* Setup the lapic or request the broadcast */
881	setup_APIC_timer();
882}
883
884void setup_secondary_APIC_clock(void)
885{
886	setup_APIC_timer();
887}
888
889/*
890 * The guts of the apic timer interrupt
891 */
892static void local_apic_timer_interrupt(void)
893{
894	int cpu = smp_processor_id();
895	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
896
897	/*
898	 * Normally we should not be here till LAPIC has been initialized but
899	 * in some cases like kdump, its possible that there is a pending LAPIC
900	 * timer interrupt from previous kernel's context and is delivered in
901	 * new kernel the moment interrupts are enabled.
902	 *
903	 * Interrupts are enabled early and LAPIC is setup much later, hence
904	 * its possible that when we get here evt->event_handler is NULL.
905	 * Check for event_handler being NULL and discard the interrupt as
906	 * spurious.
907	 */
908	if (!evt->event_handler) {
909		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
910		/* Switch it off */
911		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
912		return;
913	}
914
915	/*
916	 * the NMI deadlock-detector uses this.
917	 */
918	inc_irq_stat(apic_timer_irqs);
919
920	evt->event_handler(evt);
921}
922
923/*
924 * Local APIC timer interrupt. This is the most natural way for doing
925 * local interrupts, but local timer interrupts can be emulated by
926 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
927 *
928 * [ if a single-CPU system runs an SMP kernel then we call the local
929 *   interrupt as well. Thus we cannot inline the local irq ... ]
930 */
931__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
932{
933	struct pt_regs *old_regs = set_irq_regs(regs);
934
935	/*
936	 * NOTE! We'd better ACK the irq immediately,
937	 * because timer handling can be slow.
938	 *
939	 * update_process_times() expects us to have done irq_enter().
940	 * Besides, if we don't timer interrupts ignore the global
941	 * interrupt lock, which is the WrongThing (tm) to do.
942	 */
943	entering_ack_irq();
944	local_apic_timer_interrupt();
945	exiting_irq();
946
947	set_irq_regs(old_regs);
948}
949
950__visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
951{
952	struct pt_regs *old_regs = set_irq_regs(regs);
953
954	/*
955	 * NOTE! We'd better ACK the irq immediately,
956	 * because timer handling can be slow.
957	 *
958	 * update_process_times() expects us to have done irq_enter().
959	 * Besides, if we don't timer interrupts ignore the global
960	 * interrupt lock, which is the WrongThing (tm) to do.
961	 */
962	entering_ack_irq();
963	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
964	local_apic_timer_interrupt();
965	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
966	exiting_irq();
967
968	set_irq_regs(old_regs);
969}
970
971int setup_profiling_timer(unsigned int multiplier)
972{
973	return -EINVAL;
974}
975
976/*
977 * Local APIC start and shutdown
978 */
979
980/**
981 * clear_local_APIC - shutdown the local APIC
982 *
983 * This is called, when a CPU is disabled and before rebooting, so the state of
984 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
985 * leftovers during boot.
986 */
987void clear_local_APIC(void)
988{
989	int maxlvt;
990	u32 v;
991
992	/* APIC hasn't been mapped yet */
993	if (!x2apic_mode && !apic_phys)
994		return;
995
996	maxlvt = lapic_get_maxlvt();
997	/*
998	 * Masking an LVT entry can trigger a local APIC error
999	 * if the vector is zero. Mask LVTERR first to prevent this.
1000	 */
1001	if (maxlvt >= 3) {
1002		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1003		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1004	}
1005	/*
1006	 * Careful: we have to set masks only first to deassert
1007	 * any level-triggered sources.
1008	 */
1009	v = apic_read(APIC_LVTT);
1010	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1011	v = apic_read(APIC_LVT0);
1012	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1013	v = apic_read(APIC_LVT1);
1014	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1015	if (maxlvt >= 4) {
1016		v = apic_read(APIC_LVTPC);
1017		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1018	}
1019
1020	/* lets not touch this if we didn't frob it */
1021#ifdef CONFIG_X86_THERMAL_VECTOR
1022	if (maxlvt >= 5) {
1023		v = apic_read(APIC_LVTTHMR);
1024		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1025	}
1026#endif
1027#ifdef CONFIG_X86_MCE_INTEL
1028	if (maxlvt >= 6) {
1029		v = apic_read(APIC_LVTCMCI);
1030		if (!(v & APIC_LVT_MASKED))
1031			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1032	}
1033#endif
1034
1035	/*
1036	 * Clean APIC state for other OSs:
1037	 */
1038	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1039	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1040	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1041	if (maxlvt >= 3)
1042		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1043	if (maxlvt >= 4)
1044		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1045
1046	/* Integrated APIC (!82489DX) ? */
1047	if (lapic_is_integrated()) {
1048		if (maxlvt > 3)
1049			/* Clear ESR due to Pentium errata 3AP and 11AP */
1050			apic_write(APIC_ESR, 0);
1051		apic_read(APIC_ESR);
1052	}
1053}
1054
1055/**
1056 * disable_local_APIC - clear and disable the local APIC
1057 */
1058void disable_local_APIC(void)
1059{
1060	unsigned int value;
1061
1062	/* APIC hasn't been mapped yet */
1063	if (!x2apic_mode && !apic_phys)
1064		return;
1065
1066	clear_local_APIC();
1067
1068	/*
1069	 * Disable APIC (implies clearing of registers
1070	 * for 82489DX!).
1071	 */
1072	value = apic_read(APIC_SPIV);
1073	value &= ~APIC_SPIV_APIC_ENABLED;
1074	apic_write(APIC_SPIV, value);
1075
1076#ifdef CONFIG_X86_32
1077	/*
1078	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1079	 * restore the disabled state.
1080	 */
1081	if (enabled_via_apicbase) {
1082		unsigned int l, h;
1083
1084		rdmsr(MSR_IA32_APICBASE, l, h);
1085		l &= ~MSR_IA32_APICBASE_ENABLE;
1086		wrmsr(MSR_IA32_APICBASE, l, h);
1087	}
1088#endif
1089}
1090
1091/*
1092 * If Linux enabled the LAPIC against the BIOS default disable it down before
1093 * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1094 * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1095 * for the case where Linux didn't enable the LAPIC.
1096 */
1097void lapic_shutdown(void)
1098{
1099	unsigned long flags;
1100
1101	if (!cpu_has_apic && !apic_from_smp_config())
1102		return;
1103
1104	local_irq_save(flags);
1105
1106#ifdef CONFIG_X86_32
1107	if (!enabled_via_apicbase)
1108		clear_local_APIC();
1109	else
1110#endif
1111		disable_local_APIC();
1112
1113
1114	local_irq_restore(flags);
1115}
1116
1117/*
1118 * This is to verify that we're looking at a real local APIC.
1119 * Check these against your board if the CPUs aren't getting
1120 * started for no apparent reason.
1121 */
1122int __init verify_local_APIC(void)
1123{
1124	unsigned int reg0, reg1;
1125
1126	/*
1127	 * The version register is read-only in a real APIC.
1128	 */
1129	reg0 = apic_read(APIC_LVR);
1130	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1131	apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1132	reg1 = apic_read(APIC_LVR);
1133	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1134
1135	/*
1136	 * The two version reads above should print the same
1137	 * numbers.  If the second one is different, then we
1138	 * poke at a non-APIC.
1139	 */
1140	if (reg1 != reg0)
1141		return 0;
1142
1143	/*
1144	 * Check if the version looks reasonably.
1145	 */
1146	reg1 = GET_APIC_VERSION(reg0);
1147	if (reg1 == 0x00 || reg1 == 0xff)
1148		return 0;
1149	reg1 = lapic_get_maxlvt();
1150	if (reg1 < 0x02 || reg1 == 0xff)
1151		return 0;
1152
1153	/*
1154	 * The ID register is read/write in a real APIC.
1155	 */
1156	reg0 = apic_read(APIC_ID);
1157	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1158	apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1159	reg1 = apic_read(APIC_ID);
1160	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1161	apic_write(APIC_ID, reg0);
1162	if (reg1 != (reg0 ^ apic->apic_id_mask))
1163		return 0;
1164
1165	/*
1166	 * The next two are just to see if we have sane values.
1167	 * They're only really relevant if we're in Virtual Wire
1168	 * compatibility mode, but most boxes are anymore.
1169	 */
1170	reg0 = apic_read(APIC_LVT0);
1171	apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1172	reg1 = apic_read(APIC_LVT1);
1173	apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1174
1175	return 1;
1176}
1177
1178/**
1179 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1180 */
1181void __init sync_Arb_IDs(void)
1182{
1183	/*
1184	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1185	 * needed on AMD.
1186	 */
1187	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1188		return;
1189
1190	/*
1191	 * Wait for idle.
1192	 */
1193	apic_wait_icr_idle();
1194
1195	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1196	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1197			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1198}
1199
1200/*
1201 * An initial setup of the virtual wire mode.
1202 */
1203void __init init_bsp_APIC(void)
1204{
1205	unsigned int value;
1206
1207	/*
1208	 * Don't do the setup now if we have a SMP BIOS as the
1209	 * through-I/O-APIC virtual wire mode might be active.
1210	 */
1211	if (smp_found_config || !cpu_has_apic)
1212		return;
1213
1214	/*
1215	 * Do not trust the local APIC being empty at bootup.
1216	 */
1217	clear_local_APIC();
1218
1219	/*
1220	 * Enable APIC.
1221	 */
1222	value = apic_read(APIC_SPIV);
1223	value &= ~APIC_VECTOR_MASK;
1224	value |= APIC_SPIV_APIC_ENABLED;
1225
1226#ifdef CONFIG_X86_32
1227	/* This bit is reserved on P4/Xeon and should be cleared */
1228	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1229	    (boot_cpu_data.x86 == 15))
1230		value &= ~APIC_SPIV_FOCUS_DISABLED;
1231	else
1232#endif
1233		value |= APIC_SPIV_FOCUS_DISABLED;
1234	value |= SPURIOUS_APIC_VECTOR;
1235	apic_write(APIC_SPIV, value);
1236
1237	/*
1238	 * Set up the virtual wire mode.
1239	 */
1240	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1241	value = APIC_DM_NMI;
1242	if (!lapic_is_integrated())		/* 82489DX */
1243		value |= APIC_LVT_LEVEL_TRIGGER;
1244	apic_write(APIC_LVT1, value);
1245}
1246
1247static void lapic_setup_esr(void)
1248{
1249	unsigned int oldvalue, value, maxlvt;
1250
1251	if (!lapic_is_integrated()) {
1252		pr_info("No ESR for 82489DX.\n");
1253		return;
1254	}
1255
1256	if (apic->disable_esr) {
1257		/*
1258		 * Something untraceable is creating bad interrupts on
1259		 * secondary quads ... for the moment, just leave the
1260		 * ESR disabled - we can't do anything useful with the
1261		 * errors anyway - mbligh
1262		 */
1263		pr_info("Leaving ESR disabled.\n");
1264		return;
1265	}
1266
1267	maxlvt = lapic_get_maxlvt();
1268	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1269		apic_write(APIC_ESR, 0);
1270	oldvalue = apic_read(APIC_ESR);
1271
1272	/* enables sending errors */
1273	value = ERROR_APIC_VECTOR;
1274	apic_write(APIC_LVTERR, value);
1275
1276	/*
1277	 * spec says clear errors after enabling vector.
1278	 */
1279	if (maxlvt > 3)
1280		apic_write(APIC_ESR, 0);
1281	value = apic_read(APIC_ESR);
1282	if (value != oldvalue)
1283		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1284			"vector: 0x%08x  after: 0x%08x\n",
1285			oldvalue, value);
1286}
1287
1288/**
1289 * setup_local_APIC - setup the local APIC
1290 *
1291 * Used to setup local APIC while initializing BSP or bringin up APs.
1292 * Always called with preemption disabled.
1293 */
1294void setup_local_APIC(void)
1295{
1296	int cpu = smp_processor_id();
1297	unsigned int value, queued;
1298	int i, j, acked = 0;
1299	unsigned long long tsc = 0, ntsc;
1300	long long max_loops = cpu_khz ? cpu_khz : 1000000;
1301
1302	if (cpu_has_tsc)
1303		rdtscll(tsc);
1304
1305	if (disable_apic) {
1306		disable_ioapic_support();
1307		return;
1308	}
1309
1310#ifdef CONFIG_X86_32
1311	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1312	if (lapic_is_integrated() && apic->disable_esr) {
1313		apic_write(APIC_ESR, 0);
1314		apic_write(APIC_ESR, 0);
1315		apic_write(APIC_ESR, 0);
1316		apic_write(APIC_ESR, 0);
1317	}
1318#endif
1319	perf_events_lapic_init();
1320
1321	/*
1322	 * Double-check whether this APIC is really registered.
1323	 * This is meaningless in clustered apic mode, so we skip it.
1324	 */
1325	BUG_ON(!apic->apic_id_registered());
1326
1327	/*
1328	 * Intel recommends to set DFR, LDR and TPR before enabling
1329	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1330	 * document number 292116).  So here it goes...
1331	 */
1332	apic->init_apic_ldr();
1333
1334#ifdef CONFIG_X86_32
1335	/*
1336	 * APIC LDR is initialized.  If logical_apicid mapping was
1337	 * initialized during get_smp_config(), make sure it matches the
1338	 * actual value.
1339	 */
1340	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1341	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1342	/* always use the value from LDR */
1343	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1344		logical_smp_processor_id();
1345#endif
1346
1347	/*
1348	 * Set Task Priority to 'accept all'. We never change this
1349	 * later on.
1350	 */
1351	value = apic_read(APIC_TASKPRI);
1352	value &= ~APIC_TPRI_MASK;
1353	apic_write(APIC_TASKPRI, value);
1354
1355	/*
1356	 * After a crash, we no longer service the interrupts and a pending
1357	 * interrupt from previous kernel might still have ISR bit set.
1358	 *
1359	 * Most probably by now CPU has serviced that pending interrupt and
1360	 * it might not have done the ack_APIC_irq() because it thought,
1361	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1362	 * does not clear the ISR bit and cpu thinks it has already serivced
1363	 * the interrupt. Hence a vector might get locked. It was noticed
1364	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1365	 */
1366	do {
1367		queued = 0;
1368		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1369			queued |= apic_read(APIC_IRR + i*0x10);
1370
1371		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1372			value = apic_read(APIC_ISR + i*0x10);
1373			for (j = 31; j >= 0; j--) {
1374				if (value & (1<<j)) {
1375					ack_APIC_irq();
1376					acked++;
1377				}
1378			}
1379		}
1380		if (acked > 256) {
1381			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1382			       acked);
1383			break;
1384		}
1385		if (queued) {
1386			if (cpu_has_tsc && cpu_khz) {
1387				rdtscll(ntsc);
1388				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1389			} else
1390				max_loops--;
1391		}
1392	} while (queued && max_loops > 0);
1393	WARN_ON(max_loops <= 0);
1394
1395	/*
1396	 * Now that we are all set up, enable the APIC
1397	 */
1398	value = apic_read(APIC_SPIV);
1399	value &= ~APIC_VECTOR_MASK;
1400	/*
1401	 * Enable APIC
1402	 */
1403	value |= APIC_SPIV_APIC_ENABLED;
1404
1405#ifdef CONFIG_X86_32
1406	/*
1407	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1408	 * certain networking cards. If high frequency interrupts are
1409	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1410	 * entry is masked/unmasked at a high rate as well then sooner or
1411	 * later IOAPIC line gets 'stuck', no more interrupts are received
1412	 * from the device. If focus CPU is disabled then the hang goes
1413	 * away, oh well :-(
1414	 *
1415	 * [ This bug can be reproduced easily with a level-triggered
1416	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1417	 *   BX chipset. ]
1418	 */
1419	/*
1420	 * Actually disabling the focus CPU check just makes the hang less
1421	 * frequent as it makes the interrupt distributon model be more
1422	 * like LRU than MRU (the short-term load is more even across CPUs).
1423	 * See also the comment in end_level_ioapic_irq().  --macro
1424	 */
1425
1426	/*
1427	 * - enable focus processor (bit==0)
1428	 * - 64bit mode always use processor focus
1429	 *   so no need to set it
1430	 */
1431	value &= ~APIC_SPIV_FOCUS_DISABLED;
1432#endif
1433
1434	/*
1435	 * Set spurious IRQ vector
1436	 */
1437	value |= SPURIOUS_APIC_VECTOR;
1438	apic_write(APIC_SPIV, value);
1439
1440	/*
1441	 * Set up LVT0, LVT1:
1442	 *
1443	 * set up through-local-APIC on the BP's LINT0. This is not
1444	 * strictly necessary in pure symmetric-IO mode, but sometimes
1445	 * we delegate interrupts to the 8259A.
1446	 */
1447	/*
1448	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1449	 */
1450	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1451	if (!cpu && (pic_mode || !value)) {
1452		value = APIC_DM_EXTINT;
1453		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1454	} else {
1455		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1456		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1457	}
1458	apic_write(APIC_LVT0, value);
1459
1460	/*
1461	 * only the BP should see the LINT1 NMI signal, obviously.
1462	 */
1463	if (!cpu)
1464		value = APIC_DM_NMI;
1465	else
1466		value = APIC_DM_NMI | APIC_LVT_MASKED;
1467	if (!lapic_is_integrated())		/* 82489DX */
1468		value |= APIC_LVT_LEVEL_TRIGGER;
1469	apic_write(APIC_LVT1, value);
1470
1471#ifdef CONFIG_X86_MCE_INTEL
1472	/* Recheck CMCI information after local APIC is up on CPU #0 */
1473	if (!cpu)
1474		cmci_recheck();
1475#endif
1476}
1477
1478void end_local_APIC_setup(void)
1479{
1480	lapic_setup_esr();
1481
1482#ifdef CONFIG_X86_32
1483	{
1484		unsigned int value;
1485		/* Disable the local apic timer */
1486		value = apic_read(APIC_LVTT);
1487		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1488		apic_write(APIC_LVTT, value);
1489	}
1490#endif
1491
1492	apic_pm_activate();
1493}
1494
1495void __init bsp_end_local_APIC_setup(void)
1496{
1497	end_local_APIC_setup();
1498
1499	/*
1500	 * Now that local APIC setup is completed for BP, configure the fault
1501	 * handling for interrupt remapping.
1502	 */
1503	irq_remap_enable_fault_handling();
1504
1505}
1506
1507#ifdef CONFIG_X86_X2APIC
1508/*
1509 * Need to disable xapic and x2apic at the same time and then enable xapic mode
1510 */
1511static inline void __disable_x2apic(u64 msr)
1512{
1513	wrmsrl(MSR_IA32_APICBASE,
1514	       msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1515	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1516}
1517
1518static __init void disable_x2apic(void)
1519{
1520	u64 msr;
1521
1522	if (!cpu_has_x2apic)
1523		return;
1524
1525	rdmsrl(MSR_IA32_APICBASE, msr);
1526	if (msr & X2APIC_ENABLE) {
1527		u32 x2apic_id = read_apic_id();
1528
1529		if (x2apic_id >= 255)
1530			panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1531
1532		pr_info("Disabling x2apic\n");
1533		__disable_x2apic(msr);
1534
1535		if (nox2apic) {
1536			clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1537			setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1538		}
1539
1540		x2apic_disabled = 1;
1541		x2apic_mode = 0;
1542
1543		register_lapic_address(mp_lapic_addr);
1544	}
1545}
1546
1547void check_x2apic(void)
1548{
1549	if (x2apic_enabled()) {
1550		pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1551		x2apic_preenabled = x2apic_mode = 1;
1552	}
1553}
1554
1555void enable_x2apic(void)
1556{
1557	u64 msr;
1558
1559	rdmsrl(MSR_IA32_APICBASE, msr);
1560	if (x2apic_disabled) {
1561		__disable_x2apic(msr);
1562		return;
1563	}
1564
1565	if (!x2apic_mode)
1566		return;
1567
1568	if (!(msr & X2APIC_ENABLE)) {
1569		printk_once(KERN_INFO "Enabling x2apic\n");
1570		wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1571	}
1572}
1573#endif /* CONFIG_X86_X2APIC */
1574
1575int __init enable_IR(void)
1576{
1577#ifdef CONFIG_IRQ_REMAP
1578	if (!irq_remapping_supported()) {
1579		pr_debug("intr-remapping not supported\n");
1580		return -1;
1581	}
1582
1583	if (!x2apic_preenabled && skip_ioapic_setup) {
1584		pr_info("Skipped enabling intr-remap because of skipping "
1585			"io-apic setup\n");
1586		return -1;
1587	}
1588
1589	return irq_remapping_enable();
1590#endif
1591	return -1;
1592}
1593
1594void __init enable_IR_x2apic(void)
1595{
1596	unsigned long flags;
1597	int ret, x2apic_enabled = 0;
1598	int hardware_init_ret;
1599
1600	/* Make sure irq_remap_ops are initialized */
1601	setup_irq_remapping_ops();
1602
1603	hardware_init_ret = irq_remapping_prepare();
1604	if (hardware_init_ret && !x2apic_supported())
1605		return;
1606
1607	ret = save_ioapic_entries();
1608	if (ret) {
1609		pr_info("Saving IO-APIC state failed: %d\n", ret);
1610		return;
1611	}
1612
1613	local_irq_save(flags);
1614	legacy_pic->mask_all();
1615	mask_ioapic_entries();
1616
1617	if (x2apic_preenabled && nox2apic)
1618		disable_x2apic();
1619
1620	if (hardware_init_ret)
1621		ret = -1;
1622	else
1623		ret = enable_IR();
1624
1625	if (!x2apic_supported())
1626		goto skip_x2apic;
1627
1628	if (ret < 0) {
1629		/* IR is required if there is APIC ID > 255 even when running
1630		 * under KVM
1631		 */
1632		if (max_physical_apicid > 255 ||
1633		    !hypervisor_x2apic_available()) {
1634			if (x2apic_preenabled)
1635				disable_x2apic();
1636			goto skip_x2apic;
1637		}
1638		/*
1639		 * without IR all CPUs can be addressed by IOAPIC/MSI
1640		 * only in physical mode
1641		 */
1642		x2apic_force_phys();
1643	}
1644
1645	if (ret == IRQ_REMAP_XAPIC_MODE) {
1646		pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1647		goto skip_x2apic;
1648	}
1649
1650	x2apic_enabled = 1;
1651
1652	if (x2apic_supported() && !x2apic_mode) {
1653		x2apic_mode = 1;
1654		enable_x2apic();
1655		pr_info("Enabled x2apic\n");
1656	}
1657
1658skip_x2apic:
1659	if (ret < 0) /* IR enabling failed */
1660		restore_ioapic_entries();
1661	legacy_pic->restore_mask();
1662	local_irq_restore(flags);
1663}
1664
1665#ifdef CONFIG_X86_64
1666/*
1667 * Detect and enable local APICs on non-SMP boards.
1668 * Original code written by Keir Fraser.
1669 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1670 * not correctly set up (usually the APIC timer won't work etc.)
1671 */
1672static int __init detect_init_APIC(void)
1673{
1674	if (!cpu_has_apic) {
1675		pr_info("No local APIC present\n");
1676		return -1;
1677	}
1678
1679	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1680	return 0;
1681}
1682#else
1683
1684static int __init apic_verify(void)
1685{
1686	u32 features, h, l;
1687
1688	/*
1689	 * The APIC feature bit should now be enabled
1690	 * in `cpuid'
1691	 */
1692	features = cpuid_edx(1);
1693	if (!(features & (1 << X86_FEATURE_APIC))) {
1694		pr_warning("Could not enable APIC!\n");
1695		return -1;
1696	}
1697	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1698	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1699
1700	/* The BIOS may have set up the APIC at some other address */
1701	if (boot_cpu_data.x86 >= 6) {
1702		rdmsr(MSR_IA32_APICBASE, l, h);
1703		if (l & MSR_IA32_APICBASE_ENABLE)
1704			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1705	}
1706
1707	pr_info("Found and enabled local APIC!\n");
1708	return 0;
1709}
1710
1711int __init apic_force_enable(unsigned long addr)
1712{
1713	u32 h, l;
1714
1715	if (disable_apic)
1716		return -1;
1717
1718	/*
1719	 * Some BIOSes disable the local APIC in the APIC_BASE
1720	 * MSR. This can only be done in software for Intel P6 or later
1721	 * and AMD K7 (Model > 1) or later.
1722	 */
1723	if (boot_cpu_data.x86 >= 6) {
1724		rdmsr(MSR_IA32_APICBASE, l, h);
1725		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1726			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1727			l &= ~MSR_IA32_APICBASE_BASE;
1728			l |= MSR_IA32_APICBASE_ENABLE | addr;
1729			wrmsr(MSR_IA32_APICBASE, l, h);
1730			enabled_via_apicbase = 1;
1731		}
1732	}
1733	return apic_verify();
1734}
1735
1736/*
1737 * Detect and initialize APIC
1738 */
1739static int __init detect_init_APIC(void)
1740{
1741	/* Disabled by kernel option? */
1742	if (disable_apic)
1743		return -1;
1744
1745	switch (boot_cpu_data.x86_vendor) {
1746	case X86_VENDOR_AMD:
1747		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1748		    (boot_cpu_data.x86 >= 15))
1749			break;
1750		goto no_apic;
1751	case X86_VENDOR_INTEL:
1752		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1753		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
1754			break;
1755		goto no_apic;
1756	default:
1757		goto no_apic;
1758	}
1759
1760	if (!cpu_has_apic) {
1761		/*
1762		 * Over-ride BIOS and try to enable the local APIC only if
1763		 * "lapic" specified.
1764		 */
1765		if (!force_enable_local_apic) {
1766			pr_info("Local APIC disabled by BIOS -- "
1767				"you can enable it with \"lapic\"\n");
1768			return -1;
1769		}
1770		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1771			return -1;
1772	} else {
1773		if (apic_verify())
1774			return -1;
1775	}
1776
1777	apic_pm_activate();
1778
1779	return 0;
1780
1781no_apic:
1782	pr_info("No local APIC present or hardware disabled\n");
1783	return -1;
1784}
1785#endif
1786
1787/**
1788 * init_apic_mappings - initialize APIC mappings
1789 */
1790void __init init_apic_mappings(void)
1791{
1792	unsigned int new_apicid;
1793
1794	if (x2apic_mode) {
1795		boot_cpu_physical_apicid = read_apic_id();
1796		return;
1797	}
1798
1799	/* If no local APIC can be found return early */
1800	if (!smp_found_config && detect_init_APIC()) {
1801		/* lets NOP'ify apic operations */
1802		pr_info("APIC: disable apic facility\n");
1803		apic_disable();
1804	} else {
1805		apic_phys = mp_lapic_addr;
1806
1807		/*
1808		 * acpi lapic path already maps that address in
1809		 * acpi_register_lapic_address()
1810		 */
1811		if (!acpi_lapic && !smp_found_config)
1812			register_lapic_address(apic_phys);
1813	}
1814
1815	/*
1816	 * Fetch the APIC ID of the BSP in case we have a
1817	 * default configuration (or the MP table is broken).
1818	 */
1819	new_apicid = read_apic_id();
1820	if (boot_cpu_physical_apicid != new_apicid) {
1821		boot_cpu_physical_apicid = new_apicid;
1822		/*
1823		 * yeah -- we lie about apic_version
1824		 * in case if apic was disabled via boot option
1825		 * but it's not a problem for SMP compiled kernel
1826		 * since smp_sanity_check is prepared for such a case
1827		 * and disable smp mode
1828		 */
1829		apic_version[new_apicid] =
1830			 GET_APIC_VERSION(apic_read(APIC_LVR));
1831	}
1832}
1833
1834void __init register_lapic_address(unsigned long address)
1835{
1836	mp_lapic_addr = address;
1837
1838	if (!x2apic_mode) {
1839		set_fixmap_nocache(FIX_APIC_BASE, address);
1840		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1841			    APIC_BASE, mp_lapic_addr);
1842	}
1843	if (boot_cpu_physical_apicid == -1U) {
1844		boot_cpu_physical_apicid  = read_apic_id();
1845		apic_version[boot_cpu_physical_apicid] =
1846			 GET_APIC_VERSION(apic_read(APIC_LVR));
1847	}
1848}
1849
1850/*
1851 * This initializes the IO-APIC and APIC hardware if this is
1852 * a UP kernel.
1853 */
1854int apic_version[MAX_LOCAL_APIC];
1855
1856int __init APIC_init_uniprocessor(void)
1857{
1858	if (disable_apic) {
1859		pr_info("Apic disabled\n");
1860		return -1;
1861	}
1862#ifdef CONFIG_X86_64
1863	if (!cpu_has_apic) {
1864		disable_apic = 1;
1865		pr_info("Apic disabled by BIOS\n");
1866		return -1;
1867	}
1868#else
1869	if (!smp_found_config && !cpu_has_apic)
1870		return -1;
1871
1872	/*
1873	 * Complain if the BIOS pretends there is one.
1874	 */
1875	if (!cpu_has_apic &&
1876	    APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1877		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1878			boot_cpu_physical_apicid);
1879		return -1;
1880	}
1881#endif
1882
1883	default_setup_apic_routing();
1884
1885	verify_local_APIC();
1886	connect_bsp_APIC();
1887
1888#ifdef CONFIG_X86_64
1889	apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1890#else
1891	/*
1892	 * Hack: In case of kdump, after a crash, kernel might be booting
1893	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1894	 * might be zero if read from MP tables. Get it from LAPIC.
1895	 */
1896# ifdef CONFIG_CRASH_DUMP
1897	boot_cpu_physical_apicid = read_apic_id();
1898# endif
1899#endif
1900	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1901	setup_local_APIC();
1902
1903#ifdef CONFIG_X86_IO_APIC
1904	/*
1905	 * Now enable IO-APICs, actually call clear_IO_APIC
1906	 * We need clear_IO_APIC before enabling error vector
1907	 */
1908	if (!skip_ioapic_setup && nr_ioapics)
1909		enable_IO_APIC();
1910#endif
1911
1912	bsp_end_local_APIC_setup();
1913
1914#ifdef CONFIG_X86_IO_APIC
1915	if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1916		setup_IO_APIC();
1917	else {
1918		nr_ioapics = 0;
1919	}
1920#endif
1921
1922	x86_init.timers.setup_percpu_clockev();
1923	return 0;
1924}
1925
1926/*
1927 * Local APIC interrupts
1928 */
1929
1930/*
1931 * This interrupt should _never_ happen with our APIC/SMP architecture
1932 */
1933static inline void __smp_spurious_interrupt(void)
1934{
1935	u32 v;
1936
1937	/*
1938	 * Check if this really is a spurious interrupt and ACK it
1939	 * if it is a vectored one.  Just in case...
1940	 * Spurious interrupts should not be ACKed.
1941	 */
1942	v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1943	if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1944		ack_APIC_irq();
1945
1946	inc_irq_stat(irq_spurious_count);
1947
1948	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1949	pr_info("spurious APIC interrupt on CPU#%d, "
1950		"should never happen.\n", smp_processor_id());
1951}
1952
1953__visible void smp_spurious_interrupt(struct pt_regs *regs)
1954{
1955	entering_irq();
1956	__smp_spurious_interrupt();
1957	exiting_irq();
1958}
1959
1960__visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1961{
1962	entering_irq();
1963	trace_spurious_apic_entry(SPURIOUS_APIC_VECTOR);
1964	__smp_spurious_interrupt();
1965	trace_spurious_apic_exit(SPURIOUS_APIC_VECTOR);
1966	exiting_irq();
1967}
1968
1969/*
1970 * This interrupt should never happen with our APIC/SMP architecture
1971 */
1972static inline void __smp_error_interrupt(struct pt_regs *regs)
1973{
1974	u32 v;
1975	u32 i = 0;
1976	static const char * const error_interrupt_reason[] = {
1977		"Send CS error",		/* APIC Error Bit 0 */
1978		"Receive CS error",		/* APIC Error Bit 1 */
1979		"Send accept error",		/* APIC Error Bit 2 */
1980		"Receive accept error",		/* APIC Error Bit 3 */
1981		"Redirectable IPI",		/* APIC Error Bit 4 */
1982		"Send illegal vector",		/* APIC Error Bit 5 */
1983		"Received illegal vector",	/* APIC Error Bit 6 */
1984		"Illegal register address",	/* APIC Error Bit 7 */
1985	};
1986
1987	/* First tickle the hardware, only then report what went on. -- REW */
1988	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
1989		apic_write(APIC_ESR, 0);
1990	v = apic_read(APIC_ESR);
1991	ack_APIC_irq();
1992	atomic_inc(&irq_err_count);
1993
1994	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
1995		    smp_processor_id(), v);
1996
1997	v &= 0xff;
1998	while (v) {
1999		if (v & 0x1)
2000			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2001		i++;
2002		v >>= 1;
2003	}
2004
2005	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2006
2007}
2008
2009__visible void smp_error_interrupt(struct pt_regs *regs)
2010{
2011	entering_irq();
2012	__smp_error_interrupt(regs);
2013	exiting_irq();
2014}
2015
2016__visible void smp_trace_error_interrupt(struct pt_regs *regs)
2017{
2018	entering_irq();
2019	trace_error_apic_entry(ERROR_APIC_VECTOR);
2020	__smp_error_interrupt(regs);
2021	trace_error_apic_exit(ERROR_APIC_VECTOR);
2022	exiting_irq();
2023}
2024
2025/**
2026 * connect_bsp_APIC - attach the APIC to the interrupt system
2027 */
2028void __init connect_bsp_APIC(void)
2029{
2030#ifdef CONFIG_X86_32
2031	if (pic_mode) {
2032		/*
2033		 * Do not trust the local APIC being empty at bootup.
2034		 */
2035		clear_local_APIC();
2036		/*
2037		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2038		 * local APIC to INT and NMI lines.
2039		 */
2040		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2041				"enabling APIC mode.\n");
2042		imcr_pic_to_apic();
2043	}
2044#endif
2045}
2046
2047/**
2048 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2049 * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2050 *
2051 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2052 * APIC is disabled.
2053 */
2054void disconnect_bsp_APIC(int virt_wire_setup)
2055{
2056	unsigned int value;
2057
2058#ifdef CONFIG_X86_32
2059	if (pic_mode) {
2060		/*
2061		 * Put the board back into PIC mode (has an effect only on
2062		 * certain older boards).  Note that APIC interrupts, including
2063		 * IPIs, won't work beyond this point!  The only exception are
2064		 * INIT IPIs.
2065		 */
2066		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2067				"entering PIC mode.\n");
2068		imcr_apic_to_pic();
2069		return;
2070	}
2071#endif
2072
2073	/* Go back to Virtual Wire compatibility mode */
2074
2075	/* For the spurious interrupt use vector F, and enable it */
2076	value = apic_read(APIC_SPIV);
2077	value &= ~APIC_VECTOR_MASK;
2078	value |= APIC_SPIV_APIC_ENABLED;
2079	value |= 0xf;
2080	apic_write(APIC_SPIV, value);
2081
2082	if (!virt_wire_setup) {
2083		/*
2084		 * For LVT0 make it edge triggered, active high,
2085		 * external and enabled
2086		 */
2087		value = apic_read(APIC_LVT0);
2088		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2089			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2090			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2091		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2092		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2093		apic_write(APIC_LVT0, value);
2094	} else {
2095		/* Disable LVT0 */
2096		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2097	}
2098
2099	/*
2100	 * For LVT1 make it edge triggered, active high,
2101	 * nmi and enabled
2102	 */
2103	value = apic_read(APIC_LVT1);
2104	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2105			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2106			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2107	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2108	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2109	apic_write(APIC_LVT1, value);
2110}
2111
2112int generic_processor_info(int apicid, int version)
2113{
2114	int cpu, max = nr_cpu_ids;
2115	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2116				phys_cpu_present_map);
2117
2118	/*
2119	 * boot_cpu_physical_apicid is designed to have the apicid
2120	 * returned by read_apic_id(), i.e, the apicid of the
2121	 * currently booting-up processor. However, on some platforms,
2122	 * it is temporarily modified by the apicid reported as BSP
2123	 * through MP table. Concretely:
2124	 *
2125	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2126	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2127	 *
2128	 * This function is executed with the modified
2129	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2130	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2131	 *
2132	 * Since fixing handling of boot_cpu_physical_apicid requires
2133	 * another discussion and tests on each platform, we leave it
2134	 * for now and here we use read_apic_id() directly in this
2135	 * function, generic_processor_info().
2136	 */
2137	if (disabled_cpu_apicid != BAD_APICID &&
2138	    disabled_cpu_apicid != read_apic_id() &&
2139	    disabled_cpu_apicid == apicid) {
2140		int thiscpu = num_processors + disabled_cpus;
2141
2142		pr_warning("APIC: Disabling requested cpu."
2143			   " Processor %d/0x%x ignored.\n",
2144			   thiscpu, apicid);
2145
2146		disabled_cpus++;
2147		return -ENODEV;
2148	}
2149
2150	/*
2151	 * If boot cpu has not been detected yet, then only allow upto
2152	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2153	 */
2154	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2155	    apicid != boot_cpu_physical_apicid) {
2156		int thiscpu = max + disabled_cpus - 1;
2157
2158		pr_warning(
2159			"ACPI: NR_CPUS/possible_cpus limit of %i almost"
2160			" reached. Keeping one slot for boot cpu."
2161			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2162
2163		disabled_cpus++;
2164		return -ENODEV;
2165	}
2166
2167	if (num_processors >= nr_cpu_ids) {
2168		int thiscpu = max + disabled_cpus;
2169
2170		pr_warning(
2171			"ACPI: NR_CPUS/possible_cpus limit of %i reached."
2172			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2173
2174		disabled_cpus++;
2175		return -EINVAL;
2176	}
2177
2178	num_processors++;
2179	if (apicid == boot_cpu_physical_apicid) {
2180		/*
2181		 * x86_bios_cpu_apicid is required to have processors listed
2182		 * in same order as logical cpu numbers. Hence the first
2183		 * entry is BSP, and so on.
2184		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2185		 * for BSP.
2186		 */
2187		cpu = 0;
2188	} else
2189		cpu = cpumask_next_zero(-1, cpu_present_mask);
2190
2191	/*
2192	 * Validate version
2193	 */
2194	if (version == 0x0) {
2195		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2196			   cpu, apicid);
2197		version = 0x10;
2198	}
2199	apic_version[apicid] = version;
2200
2201	if (version != apic_version[boot_cpu_physical_apicid]) {
2202		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2203			apic_version[boot_cpu_physical_apicid], cpu, version);
2204	}
2205
2206	physid_set(apicid, phys_cpu_present_map);
2207	if (apicid > max_physical_apicid)
2208		max_physical_apicid = apicid;
2209
2210#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2211	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2212	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2213#endif
2214#ifdef CONFIG_X86_32
2215	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2216		apic->x86_32_early_logical_apicid(cpu);
2217#endif
2218	set_cpu_possible(cpu, true);
2219	set_cpu_present(cpu, true);
2220
2221	return cpu;
2222}
2223
2224int hard_smp_processor_id(void)
2225{
2226	return read_apic_id();
2227}
2228
2229void default_init_apic_ldr(void)
2230{
2231	unsigned long val;
2232
2233	apic_write(APIC_DFR, APIC_DFR_VALUE);
2234	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2235	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2236	apic_write(APIC_LDR, val);
2237}
2238
2239int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2240				   const struct cpumask *andmask,
2241				   unsigned int *apicid)
2242{
2243	unsigned int cpu;
2244
2245	for_each_cpu_and(cpu, cpumask, andmask) {
2246		if (cpumask_test_cpu(cpu, cpu_online_mask))
2247			break;
2248	}
2249
2250	if (likely(cpu < nr_cpu_ids)) {
2251		*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2252		return 0;
2253	}
2254
2255	return -EINVAL;
2256}
2257
2258/*
2259 * Override the generic EOI implementation with an optimized version.
2260 * Only called during early boot when only one CPU is active and with
2261 * interrupts disabled, so we know this does not race with actual APIC driver
2262 * use.
2263 */
2264void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2265{
2266	struct apic **drv;
2267
2268	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2269		/* Should happen once for each apic */
2270		WARN_ON((*drv)->eoi_write == eoi_write);
2271		(*drv)->eoi_write = eoi_write;
2272	}
2273}
2274
2275/*
2276 * Power management
2277 */
2278#ifdef CONFIG_PM
2279
2280static struct {
2281	/*
2282	 * 'active' is true if the local APIC was enabled by us and
2283	 * not the BIOS; this signifies that we are also responsible
2284	 * for disabling it before entering apm/acpi suspend
2285	 */
2286	int active;
2287	/* r/w apic fields */
2288	unsigned int apic_id;
2289	unsigned int apic_taskpri;
2290	unsigned int apic_ldr;
2291	unsigned int apic_dfr;
2292	unsigned int apic_spiv;
2293	unsigned int apic_lvtt;
2294	unsigned int apic_lvtpc;
2295	unsigned int apic_lvt0;
2296	unsigned int apic_lvt1;
2297	unsigned int apic_lvterr;
2298	unsigned int apic_tmict;
2299	unsigned int apic_tdcr;
2300	unsigned int apic_thmr;
2301} apic_pm_state;
2302
2303static int lapic_suspend(void)
2304{
2305	unsigned long flags;
2306	int maxlvt;
2307
2308	if (!apic_pm_state.active)
2309		return 0;
2310
2311	maxlvt = lapic_get_maxlvt();
2312
2313	apic_pm_state.apic_id = apic_read(APIC_ID);
2314	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2315	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2316	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2317	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2318	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2319	if (maxlvt >= 4)
2320		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2321	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2322	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2323	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2324	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2325	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2326#ifdef CONFIG_X86_THERMAL_VECTOR
2327	if (maxlvt >= 5)
2328		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2329#endif
2330
2331	local_irq_save(flags);
2332	disable_local_APIC();
2333
2334	irq_remapping_disable();
2335
2336	local_irq_restore(flags);
2337	return 0;
2338}
2339
2340static void lapic_resume(void)
2341{
2342	unsigned int l, h;
2343	unsigned long flags;
2344	int maxlvt;
2345
2346	if (!apic_pm_state.active)
2347		return;
2348
2349	local_irq_save(flags);
2350
2351	/*
2352	 * IO-APIC and PIC have their own resume routines.
2353	 * We just mask them here to make sure the interrupt
2354	 * subsystem is completely quiet while we enable x2apic
2355	 * and interrupt-remapping.
2356	 */
2357	mask_ioapic_entries();
2358	legacy_pic->mask_all();
2359
2360	if (x2apic_mode)
2361		enable_x2apic();
2362	else {
2363		/*
2364		 * Make sure the APICBASE points to the right address
2365		 *
2366		 * FIXME! This will be wrong if we ever support suspend on
2367		 * SMP! We'll need to do this as part of the CPU restore!
2368		 */
2369		if (boot_cpu_data.x86 >= 6) {
2370			rdmsr(MSR_IA32_APICBASE, l, h);
2371			l &= ~MSR_IA32_APICBASE_BASE;
2372			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2373			wrmsr(MSR_IA32_APICBASE, l, h);
2374		}
2375	}
2376
2377	maxlvt = lapic_get_maxlvt();
2378	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2379	apic_write(APIC_ID, apic_pm_state.apic_id);
2380	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2381	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2382	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2383	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2384	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2385	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2386#if defined(CONFIG_X86_MCE_INTEL)
2387	if (maxlvt >= 5)
2388		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2389#endif
2390	if (maxlvt >= 4)
2391		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2392	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2393	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2394	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2395	apic_write(APIC_ESR, 0);
2396	apic_read(APIC_ESR);
2397	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2398	apic_write(APIC_ESR, 0);
2399	apic_read(APIC_ESR);
2400
2401	irq_remapping_reenable(x2apic_mode);
2402
2403	local_irq_restore(flags);
2404}
2405
2406/*
2407 * This device has no shutdown method - fully functioning local APICs
2408 * are needed on every CPU up until machine_halt/restart/poweroff.
2409 */
2410
2411static struct syscore_ops lapic_syscore_ops = {
2412	.resume		= lapic_resume,
2413	.suspend	= lapic_suspend,
2414};
2415
2416static void apic_pm_activate(void)
2417{
2418	apic_pm_state.active = 1;
2419}
2420
2421static int __init init_lapic_sysfs(void)
2422{
2423	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2424	if (cpu_has_apic)
2425		register_syscore_ops(&lapic_syscore_ops);
2426
2427	return 0;
2428}
2429
2430/* local apic needs to resume before other devices access its registers. */
2431core_initcall(init_lapic_sysfs);
2432
2433#else	/* CONFIG_PM */
2434
2435static void apic_pm_activate(void) { }
2436
2437#endif	/* CONFIG_PM */
2438
2439#ifdef CONFIG_X86_64
2440
2441static int multi_checked;
2442static int multi;
2443
2444static int set_multi(const struct dmi_system_id *d)
2445{
2446	if (multi)
2447		return 0;
2448	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2449	multi = 1;
2450	return 0;
2451}
2452
2453static const struct dmi_system_id multi_dmi_table[] = {
2454	{
2455		.callback = set_multi,
2456		.ident = "IBM System Summit2",
2457		.matches = {
2458			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2459			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2460		},
2461	},
2462	{}
2463};
2464
2465static void dmi_check_multi(void)
2466{
2467	if (multi_checked)
2468		return;
2469
2470	dmi_check_system(multi_dmi_table);
2471	multi_checked = 1;
2472}
2473
2474/*
2475 * apic_is_clustered_box() -- Check if we can expect good TSC
2476 *
2477 * Thus far, the major user of this is IBM's Summit2 series:
2478 * Clustered boxes may have unsynced TSC problems if they are
2479 * multi-chassis.
2480 * Use DMI to check them
2481 */
2482int apic_is_clustered_box(void)
2483{
2484	dmi_check_multi();
2485	return multi;
2486}
2487#endif
2488
2489/*
2490 * APIC command line parameters
2491 */
2492static int __init setup_disableapic(char *arg)
2493{
2494	disable_apic = 1;
2495	setup_clear_cpu_cap(X86_FEATURE_APIC);
2496	return 0;
2497}
2498early_param("disableapic", setup_disableapic);
2499
2500/* same as disableapic, for compatibility */
2501static int __init setup_nolapic(char *arg)
2502{
2503	return setup_disableapic(arg);
2504}
2505early_param("nolapic", setup_nolapic);
2506
2507static int __init parse_lapic_timer_c2_ok(char *arg)
2508{
2509	local_apic_timer_c2_ok = 1;
2510	return 0;
2511}
2512early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2513
2514static int __init parse_disable_apic_timer(char *arg)
2515{
2516	disable_apic_timer = 1;
2517	return 0;
2518}
2519early_param("noapictimer", parse_disable_apic_timer);
2520
2521static int __init parse_nolapic_timer(char *arg)
2522{
2523	disable_apic_timer = 1;
2524	return 0;
2525}
2526early_param("nolapic_timer", parse_nolapic_timer);
2527
2528static int __init apic_set_verbosity(char *arg)
2529{
2530	if (!arg)  {
2531#ifdef CONFIG_X86_64
2532		skip_ioapic_setup = 0;
2533		return 0;
2534#endif
2535		return -EINVAL;
2536	}
2537
2538	if (strcmp("debug", arg) == 0)
2539		apic_verbosity = APIC_DEBUG;
2540	else if (strcmp("verbose", arg) == 0)
2541		apic_verbosity = APIC_VERBOSE;
2542	else {
2543		pr_warning("APIC Verbosity level %s not recognised"
2544			" use apic=verbose or apic=debug\n", arg);
2545		return -EINVAL;
2546	}
2547
2548	return 0;
2549}
2550early_param("apic", apic_set_verbosity);
2551
2552static int __init lapic_insert_resource(void)
2553{
2554	if (!apic_phys)
2555		return -1;
2556
2557	/* Put local APIC into the resource map. */
2558	lapic_resource.start = apic_phys;
2559	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2560	insert_resource(&iomem_resource, &lapic_resource);
2561
2562	return 0;
2563}
2564
2565/*
2566 * need call insert after e820_reserve_resources()
2567 * that is using request_resource
2568 */
2569late_initcall(lapic_insert_resource);
2570
2571static int __init apic_set_disabled_cpu_apicid(char *arg)
2572{
2573	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2574		return -EINVAL;
2575
2576	return 0;
2577}
2578early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2579