xref: /openbmc/linux/arch/x86/kernel/apic/apic.c (revision 79f08d9e)
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 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  * Map cpu index to physical APIC ID
79  */
80 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
81 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
82 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
83 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
84 
85 #ifdef CONFIG_X86_32
86 
87 /*
88  * On x86_32, the mapping between cpu and logical apicid may vary
89  * depending on apic in use.  The following early percpu variable is
90  * used for the mapping.  This is where the behaviors of x86_64 and 32
91  * actually diverge.  Let's keep it ugly for now.
92  */
93 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
94 
95 /* Local APIC was disabled by the BIOS and enabled by the kernel */
96 static int enabled_via_apicbase;
97 
98 /*
99  * Handle interrupt mode configuration register (IMCR).
100  * This register controls whether the interrupt signals
101  * that reach the BSP come from the master PIC or from the
102  * local APIC. Before entering Symmetric I/O Mode, either
103  * the BIOS or the operating system must switch out of
104  * PIC Mode by changing the IMCR.
105  */
106 static inline void imcr_pic_to_apic(void)
107 {
108 	/* select IMCR register */
109 	outb(0x70, 0x22);
110 	/* NMI and 8259 INTR go through APIC */
111 	outb(0x01, 0x23);
112 }
113 
114 static inline void imcr_apic_to_pic(void)
115 {
116 	/* select IMCR register */
117 	outb(0x70, 0x22);
118 	/* NMI and 8259 INTR go directly to BSP */
119 	outb(0x00, 0x23);
120 }
121 #endif
122 
123 /*
124  * Knob to control our willingness to enable the local APIC.
125  *
126  * +1=force-enable
127  */
128 static int force_enable_local_apic __initdata;
129 /*
130  * APIC command line parameters
131  */
132 static int __init parse_lapic(char *arg)
133 {
134 	if (config_enabled(CONFIG_X86_32) && !arg)
135 		force_enable_local_apic = 1;
136 	else if (arg && !strncmp(arg, "notscdeadline", 13))
137 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
138 	return 0;
139 }
140 early_param("lapic", parse_lapic);
141 
142 #ifdef CONFIG_X86_64
143 static int apic_calibrate_pmtmr __initdata;
144 static __init int setup_apicpmtimer(char *s)
145 {
146 	apic_calibrate_pmtmr = 1;
147 	notsc_setup(NULL);
148 	return 0;
149 }
150 __setup("apicpmtimer", setup_apicpmtimer);
151 #endif
152 
153 int x2apic_mode;
154 #ifdef CONFIG_X86_X2APIC
155 /* x2apic enabled before OS handover */
156 int x2apic_preenabled;
157 static int x2apic_disabled;
158 static int nox2apic;
159 static __init int setup_nox2apic(char *str)
160 {
161 	if (x2apic_enabled()) {
162 		int apicid = native_apic_msr_read(APIC_ID);
163 
164 		if (apicid >= 255) {
165 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
166 				   apicid);
167 			return 0;
168 		}
169 
170 		pr_warning("x2apic already enabled. will disable it\n");
171 	} else
172 		setup_clear_cpu_cap(X86_FEATURE_X2APIC);
173 
174 	nox2apic = 1;
175 
176 	return 0;
177 }
178 early_param("nox2apic", setup_nox2apic);
179 #endif
180 
181 unsigned long mp_lapic_addr;
182 int disable_apic;
183 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
184 static int disable_apic_timer __initdata;
185 /* Local APIC timer works in C2 */
186 int local_apic_timer_c2_ok;
187 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
188 
189 int first_system_vector = 0xfe;
190 
191 /*
192  * Debug level, exported for io_apic.c
193  */
194 unsigned int apic_verbosity;
195 
196 int pic_mode;
197 
198 /* Have we found an MP table */
199 int smp_found_config;
200 
201 static struct resource lapic_resource = {
202 	.name = "Local APIC",
203 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
204 };
205 
206 unsigned int lapic_timer_frequency = 0;
207 
208 static void apic_pm_activate(void);
209 
210 static unsigned long apic_phys;
211 
212 /*
213  * Get the LAPIC version
214  */
215 static inline int lapic_get_version(void)
216 {
217 	return GET_APIC_VERSION(apic_read(APIC_LVR));
218 }
219 
220 /*
221  * Check, if the APIC is integrated or a separate chip
222  */
223 static inline int lapic_is_integrated(void)
224 {
225 #ifdef CONFIG_X86_64
226 	return 1;
227 #else
228 	return APIC_INTEGRATED(lapic_get_version());
229 #endif
230 }
231 
232 /*
233  * Check, whether this is a modern or a first generation APIC
234  */
235 static int modern_apic(void)
236 {
237 	/* AMD systems use old APIC versions, so check the CPU */
238 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
239 	    boot_cpu_data.x86 >= 0xf)
240 		return 1;
241 	return lapic_get_version() >= 0x14;
242 }
243 
244 /*
245  * right after this call apic become NOOP driven
246  * so apic->write/read doesn't do anything
247  */
248 static void __init apic_disable(void)
249 {
250 	pr_info("APIC: switched to apic NOOP\n");
251 	apic = &apic_noop;
252 }
253 
254 void native_apic_wait_icr_idle(void)
255 {
256 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
257 		cpu_relax();
258 }
259 
260 u32 native_safe_apic_wait_icr_idle(void)
261 {
262 	u32 send_status;
263 	int timeout;
264 
265 	timeout = 0;
266 	do {
267 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
268 		if (!send_status)
269 			break;
270 		inc_irq_stat(icr_read_retry_count);
271 		udelay(100);
272 	} while (timeout++ < 1000);
273 
274 	return send_status;
275 }
276 
277 void native_apic_icr_write(u32 low, u32 id)
278 {
279 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
280 	apic_write(APIC_ICR, low);
281 }
282 
283 u64 native_apic_icr_read(void)
284 {
285 	u32 icr1, icr2;
286 
287 	icr2 = apic_read(APIC_ICR2);
288 	icr1 = apic_read(APIC_ICR);
289 
290 	return icr1 | ((u64)icr2 << 32);
291 }
292 
293 #ifdef CONFIG_X86_32
294 /**
295  * get_physical_broadcast - Get number of physical broadcast IDs
296  */
297 int get_physical_broadcast(void)
298 {
299 	return modern_apic() ? 0xff : 0xf;
300 }
301 #endif
302 
303 /**
304  * lapic_get_maxlvt - get the maximum number of local vector table entries
305  */
306 int lapic_get_maxlvt(void)
307 {
308 	unsigned int v;
309 
310 	v = apic_read(APIC_LVR);
311 	/*
312 	 * - we always have APIC integrated on 64bit mode
313 	 * - 82489DXs do not report # of LVT entries
314 	 */
315 	return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
316 }
317 
318 /*
319  * Local APIC timer
320  */
321 
322 /* Clock divisor */
323 #define APIC_DIVISOR 16
324 #define TSC_DIVISOR  32
325 
326 /*
327  * This function sets up the local APIC timer, with a timeout of
328  * 'clocks' APIC bus clock. During calibration we actually call
329  * this function twice on the boot CPU, once with a bogus timeout
330  * value, second time for real. The other (noncalibrating) CPUs
331  * call this function only once, with the real, calibrated value.
332  *
333  * We do reads before writes even if unnecessary, to get around the
334  * P5 APIC double write bug.
335  */
336 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
337 {
338 	unsigned int lvtt_value, tmp_value;
339 
340 	lvtt_value = LOCAL_TIMER_VECTOR;
341 	if (!oneshot)
342 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
343 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
344 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
345 
346 	if (!lapic_is_integrated())
347 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
348 
349 	if (!irqen)
350 		lvtt_value |= APIC_LVT_MASKED;
351 
352 	apic_write(APIC_LVTT, lvtt_value);
353 
354 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
355 		printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
356 		return;
357 	}
358 
359 	/*
360 	 * Divide PICLK by 16
361 	 */
362 	tmp_value = apic_read(APIC_TDCR);
363 	apic_write(APIC_TDCR,
364 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
365 		APIC_TDR_DIV_16);
366 
367 	if (!oneshot)
368 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
369 }
370 
371 /*
372  * Setup extended LVT, AMD specific
373  *
374  * Software should use the LVT offsets the BIOS provides.  The offsets
375  * are determined by the subsystems using it like those for MCE
376  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
377  * are supported. Beginning with family 10h at least 4 offsets are
378  * available.
379  *
380  * Since the offsets must be consistent for all cores, we keep track
381  * of the LVT offsets in software and reserve the offset for the same
382  * vector also to be used on other cores. An offset is freed by
383  * setting the entry to APIC_EILVT_MASKED.
384  *
385  * If the BIOS is right, there should be no conflicts. Otherwise a
386  * "[Firmware Bug]: ..." error message is generated. However, if
387  * software does not properly determines the offsets, it is not
388  * necessarily a BIOS bug.
389  */
390 
391 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
392 
393 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
394 {
395 	return (old & APIC_EILVT_MASKED)
396 		|| (new == APIC_EILVT_MASKED)
397 		|| ((new & ~APIC_EILVT_MASKED) == old);
398 }
399 
400 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
401 {
402 	unsigned int rsvd, vector;
403 
404 	if (offset >= APIC_EILVT_NR_MAX)
405 		return ~0;
406 
407 	rsvd = atomic_read(&eilvt_offsets[offset]);
408 	do {
409 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
410 		if (vector && !eilvt_entry_is_changeable(vector, new))
411 			/* may not change if vectors are different */
412 			return rsvd;
413 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
414 	} while (rsvd != new);
415 
416 	rsvd &= ~APIC_EILVT_MASKED;
417 	if (rsvd && rsvd != vector)
418 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
419 			offset, rsvd);
420 
421 	return new;
422 }
423 
424 /*
425  * If mask=1, the LVT entry does not generate interrupts while mask=0
426  * enables the vector. See also the BKDGs. Must be called with
427  * preemption disabled.
428  */
429 
430 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
431 {
432 	unsigned long reg = APIC_EILVTn(offset);
433 	unsigned int new, old, reserved;
434 
435 	new = (mask << 16) | (msg_type << 8) | vector;
436 	old = apic_read(reg);
437 	reserved = reserve_eilvt_offset(offset, new);
438 
439 	if (reserved != new) {
440 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
441 		       "vector 0x%x, but the register is already in use for "
442 		       "vector 0x%x on another cpu\n",
443 		       smp_processor_id(), reg, offset, new, reserved);
444 		return -EINVAL;
445 	}
446 
447 	if (!eilvt_entry_is_changeable(old, new)) {
448 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
449 		       "vector 0x%x, but the register is already in use for "
450 		       "vector 0x%x on this cpu\n",
451 		       smp_processor_id(), reg, offset, new, old);
452 		return -EBUSY;
453 	}
454 
455 	apic_write(reg, new);
456 
457 	return 0;
458 }
459 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
460 
461 /*
462  * Program the next event, relative to now
463  */
464 static int lapic_next_event(unsigned long delta,
465 			    struct clock_event_device *evt)
466 {
467 	apic_write(APIC_TMICT, delta);
468 	return 0;
469 }
470 
471 static int lapic_next_deadline(unsigned long delta,
472 			       struct clock_event_device *evt)
473 {
474 	u64 tsc;
475 
476 	rdtscll(tsc);
477 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
478 	return 0;
479 }
480 
481 /*
482  * Setup the lapic timer in periodic or oneshot mode
483  */
484 static void lapic_timer_setup(enum clock_event_mode mode,
485 			      struct clock_event_device *evt)
486 {
487 	unsigned long flags;
488 	unsigned int v;
489 
490 	/* Lapic used as dummy for broadcast ? */
491 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
492 		return;
493 
494 	local_irq_save(flags);
495 
496 	switch (mode) {
497 	case CLOCK_EVT_MODE_PERIODIC:
498 	case CLOCK_EVT_MODE_ONESHOT:
499 		__setup_APIC_LVTT(lapic_timer_frequency,
500 				  mode != CLOCK_EVT_MODE_PERIODIC, 1);
501 		break;
502 	case CLOCK_EVT_MODE_UNUSED:
503 	case CLOCK_EVT_MODE_SHUTDOWN:
504 		v = apic_read(APIC_LVTT);
505 		v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
506 		apic_write(APIC_LVTT, v);
507 		apic_write(APIC_TMICT, 0);
508 		break;
509 	case CLOCK_EVT_MODE_RESUME:
510 		/* Nothing to do here */
511 		break;
512 	}
513 
514 	local_irq_restore(flags);
515 }
516 
517 /*
518  * Local APIC timer broadcast function
519  */
520 static void lapic_timer_broadcast(const struct cpumask *mask)
521 {
522 #ifdef CONFIG_SMP
523 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
524 #endif
525 }
526 
527 
528 /*
529  * The local apic timer can be used for any function which is CPU local.
530  */
531 static struct clock_event_device lapic_clockevent = {
532 	.name		= "lapic",
533 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
534 			| CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
535 	.shift		= 32,
536 	.set_mode	= lapic_timer_setup,
537 	.set_next_event	= lapic_next_event,
538 	.broadcast	= lapic_timer_broadcast,
539 	.rating		= 100,
540 	.irq		= -1,
541 };
542 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
543 
544 /*
545  * Setup the local APIC timer for this CPU. Copy the initialized values
546  * of the boot CPU and register the clock event in the framework.
547  */
548 static void setup_APIC_timer(void)
549 {
550 	struct clock_event_device *levt = &__get_cpu_var(lapic_events);
551 
552 	if (this_cpu_has(X86_FEATURE_ARAT)) {
553 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
554 		/* Make LAPIC timer preferrable over percpu HPET */
555 		lapic_clockevent.rating = 150;
556 	}
557 
558 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
559 	levt->cpumask = cpumask_of(smp_processor_id());
560 
561 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
562 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
563 				    CLOCK_EVT_FEAT_DUMMY);
564 		levt->set_next_event = lapic_next_deadline;
565 		clockevents_config_and_register(levt,
566 						(tsc_khz / TSC_DIVISOR) * 1000,
567 						0xF, ~0UL);
568 	} else
569 		clockevents_register_device(levt);
570 }
571 
572 /*
573  * In this functions we calibrate APIC bus clocks to the external timer.
574  *
575  * We want to do the calibration only once since we want to have local timer
576  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
577  * frequency.
578  *
579  * This was previously done by reading the PIT/HPET and waiting for a wrap
580  * around to find out, that a tick has elapsed. I have a box, where the PIT
581  * readout is broken, so it never gets out of the wait loop again. This was
582  * also reported by others.
583  *
584  * Monitoring the jiffies value is inaccurate and the clockevents
585  * infrastructure allows us to do a simple substitution of the interrupt
586  * handler.
587  *
588  * The calibration routine also uses the pm_timer when possible, as the PIT
589  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
590  * back to normal later in the boot process).
591  */
592 
593 #define LAPIC_CAL_LOOPS		(HZ/10)
594 
595 static __initdata int lapic_cal_loops = -1;
596 static __initdata long lapic_cal_t1, lapic_cal_t2;
597 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
598 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
599 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
600 
601 /*
602  * Temporary interrupt handler.
603  */
604 static void __init lapic_cal_handler(struct clock_event_device *dev)
605 {
606 	unsigned long long tsc = 0;
607 	long tapic = apic_read(APIC_TMCCT);
608 	unsigned long pm = acpi_pm_read_early();
609 
610 	if (cpu_has_tsc)
611 		rdtscll(tsc);
612 
613 	switch (lapic_cal_loops++) {
614 	case 0:
615 		lapic_cal_t1 = tapic;
616 		lapic_cal_tsc1 = tsc;
617 		lapic_cal_pm1 = pm;
618 		lapic_cal_j1 = jiffies;
619 		break;
620 
621 	case LAPIC_CAL_LOOPS:
622 		lapic_cal_t2 = tapic;
623 		lapic_cal_tsc2 = tsc;
624 		if (pm < lapic_cal_pm1)
625 			pm += ACPI_PM_OVRRUN;
626 		lapic_cal_pm2 = pm;
627 		lapic_cal_j2 = jiffies;
628 		break;
629 	}
630 }
631 
632 static int __init
633 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
634 {
635 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
636 	const long pm_thresh = pm_100ms / 100;
637 	unsigned long mult;
638 	u64 res;
639 
640 #ifndef CONFIG_X86_PM_TIMER
641 	return -1;
642 #endif
643 
644 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
645 
646 	/* Check, if the PM timer is available */
647 	if (!deltapm)
648 		return -1;
649 
650 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
651 
652 	if (deltapm > (pm_100ms - pm_thresh) &&
653 	    deltapm < (pm_100ms + pm_thresh)) {
654 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
655 		return 0;
656 	}
657 
658 	res = (((u64)deltapm) *  mult) >> 22;
659 	do_div(res, 1000000);
660 	pr_warning("APIC calibration not consistent "
661 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
662 
663 	/* Correct the lapic counter value */
664 	res = (((u64)(*delta)) * pm_100ms);
665 	do_div(res, deltapm);
666 	pr_info("APIC delta adjusted to PM-Timer: "
667 		"%lu (%ld)\n", (unsigned long)res, *delta);
668 	*delta = (long)res;
669 
670 	/* Correct the tsc counter value */
671 	if (cpu_has_tsc) {
672 		res = (((u64)(*deltatsc)) * pm_100ms);
673 		do_div(res, deltapm);
674 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
675 					  "PM-Timer: %lu (%ld)\n",
676 					(unsigned long)res, *deltatsc);
677 		*deltatsc = (long)res;
678 	}
679 
680 	return 0;
681 }
682 
683 static int __init calibrate_APIC_clock(void)
684 {
685 	struct clock_event_device *levt = &__get_cpu_var(lapic_events);
686 	void (*real_handler)(struct clock_event_device *dev);
687 	unsigned long deltaj;
688 	long delta, deltatsc;
689 	int pm_referenced = 0;
690 
691 	/**
692 	 * check if lapic timer has already been calibrated by platform
693 	 * specific routine, such as tsc calibration code. if so, we just fill
694 	 * in the clockevent structure and return.
695 	 */
696 
697 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
698 		return 0;
699 	} else if (lapic_timer_frequency) {
700 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
701 				lapic_timer_frequency);
702 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
703 					TICK_NSEC, lapic_clockevent.shift);
704 		lapic_clockevent.max_delta_ns =
705 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
706 		lapic_clockevent.min_delta_ns =
707 			clockevent_delta2ns(0xF, &lapic_clockevent);
708 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
709 		return 0;
710 	}
711 
712 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
713 		    "calibrating APIC timer ...\n");
714 
715 	local_irq_disable();
716 
717 	/* Replace the global interrupt handler */
718 	real_handler = global_clock_event->event_handler;
719 	global_clock_event->event_handler = lapic_cal_handler;
720 
721 	/*
722 	 * Setup the APIC counter to maximum. There is no way the lapic
723 	 * can underflow in the 100ms detection time frame
724 	 */
725 	__setup_APIC_LVTT(0xffffffff, 0, 0);
726 
727 	/* Let the interrupts run */
728 	local_irq_enable();
729 
730 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
731 		cpu_relax();
732 
733 	local_irq_disable();
734 
735 	/* Restore the real event handler */
736 	global_clock_event->event_handler = real_handler;
737 
738 	/* Build delta t1-t2 as apic timer counts down */
739 	delta = lapic_cal_t1 - lapic_cal_t2;
740 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
741 
742 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
743 
744 	/* we trust the PM based calibration if possible */
745 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
746 					&delta, &deltatsc);
747 
748 	/* Calculate the scaled math multiplication factor */
749 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
750 				       lapic_clockevent.shift);
751 	lapic_clockevent.max_delta_ns =
752 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
753 	lapic_clockevent.min_delta_ns =
754 		clockevent_delta2ns(0xF, &lapic_clockevent);
755 
756 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
757 
758 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
759 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
760 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
761 		    lapic_timer_frequency);
762 
763 	if (cpu_has_tsc) {
764 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
765 			    "%ld.%04ld MHz.\n",
766 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
767 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
768 	}
769 
770 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
771 		    "%u.%04u MHz.\n",
772 		    lapic_timer_frequency / (1000000 / HZ),
773 		    lapic_timer_frequency % (1000000 / HZ));
774 
775 	/*
776 	 * Do a sanity check on the APIC calibration result
777 	 */
778 	if (lapic_timer_frequency < (1000000 / HZ)) {
779 		local_irq_enable();
780 		pr_warning("APIC frequency too slow, disabling apic timer\n");
781 		return -1;
782 	}
783 
784 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
785 
786 	/*
787 	 * PM timer calibration failed or not turned on
788 	 * so lets try APIC timer based calibration
789 	 */
790 	if (!pm_referenced) {
791 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
792 
793 		/*
794 		 * Setup the apic timer manually
795 		 */
796 		levt->event_handler = lapic_cal_handler;
797 		lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
798 		lapic_cal_loops = -1;
799 
800 		/* Let the interrupts run */
801 		local_irq_enable();
802 
803 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
804 			cpu_relax();
805 
806 		/* Stop the lapic timer */
807 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
808 
809 		/* Jiffies delta */
810 		deltaj = lapic_cal_j2 - lapic_cal_j1;
811 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
812 
813 		/* Check, if the jiffies result is consistent */
814 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
815 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
816 		else
817 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
818 	} else
819 		local_irq_enable();
820 
821 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
822 		pr_warning("APIC timer disabled due to verification failure\n");
823 			return -1;
824 	}
825 
826 	return 0;
827 }
828 
829 /*
830  * Setup the boot APIC
831  *
832  * Calibrate and verify the result.
833  */
834 void __init setup_boot_APIC_clock(void)
835 {
836 	/*
837 	 * The local apic timer can be disabled via the kernel
838 	 * commandline or from the CPU detection code. Register the lapic
839 	 * timer as a dummy clock event source on SMP systems, so the
840 	 * broadcast mechanism is used. On UP systems simply ignore it.
841 	 */
842 	if (disable_apic_timer) {
843 		pr_info("Disabling APIC timer\n");
844 		/* No broadcast on UP ! */
845 		if (num_possible_cpus() > 1) {
846 			lapic_clockevent.mult = 1;
847 			setup_APIC_timer();
848 		}
849 		return;
850 	}
851 
852 	if (calibrate_APIC_clock()) {
853 		/* No broadcast on UP ! */
854 		if (num_possible_cpus() > 1)
855 			setup_APIC_timer();
856 		return;
857 	}
858 
859 	/*
860 	 * If nmi_watchdog is set to IO_APIC, we need the
861 	 * PIT/HPET going.  Otherwise register lapic as a dummy
862 	 * device.
863 	 */
864 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
865 
866 	/* Setup the lapic or request the broadcast */
867 	setup_APIC_timer();
868 }
869 
870 void setup_secondary_APIC_clock(void)
871 {
872 	setup_APIC_timer();
873 }
874 
875 /*
876  * The guts of the apic timer interrupt
877  */
878 static void local_apic_timer_interrupt(void)
879 {
880 	int cpu = smp_processor_id();
881 	struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
882 
883 	/*
884 	 * Normally we should not be here till LAPIC has been initialized but
885 	 * in some cases like kdump, its possible that there is a pending LAPIC
886 	 * timer interrupt from previous kernel's context and is delivered in
887 	 * new kernel the moment interrupts are enabled.
888 	 *
889 	 * Interrupts are enabled early and LAPIC is setup much later, hence
890 	 * its possible that when we get here evt->event_handler is NULL.
891 	 * Check for event_handler being NULL and discard the interrupt as
892 	 * spurious.
893 	 */
894 	if (!evt->event_handler) {
895 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
896 		/* Switch it off */
897 		lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
898 		return;
899 	}
900 
901 	/*
902 	 * the NMI deadlock-detector uses this.
903 	 */
904 	inc_irq_stat(apic_timer_irqs);
905 
906 	evt->event_handler(evt);
907 }
908 
909 /*
910  * Local APIC timer interrupt. This is the most natural way for doing
911  * local interrupts, but local timer interrupts can be emulated by
912  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
913  *
914  * [ if a single-CPU system runs an SMP kernel then we call the local
915  *   interrupt as well. Thus we cannot inline the local irq ... ]
916  */
917 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
918 {
919 	struct pt_regs *old_regs = set_irq_regs(regs);
920 
921 	/*
922 	 * NOTE! We'd better ACK the irq immediately,
923 	 * because timer handling can be slow.
924 	 *
925 	 * update_process_times() expects us to have done irq_enter().
926 	 * Besides, if we don't timer interrupts ignore the global
927 	 * interrupt lock, which is the WrongThing (tm) to do.
928 	 */
929 	entering_ack_irq();
930 	local_apic_timer_interrupt();
931 	exiting_irq();
932 
933 	set_irq_regs(old_regs);
934 }
935 
936 __visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
937 {
938 	struct pt_regs *old_regs = set_irq_regs(regs);
939 
940 	/*
941 	 * NOTE! We'd better ACK the irq immediately,
942 	 * because timer handling can be slow.
943 	 *
944 	 * update_process_times() expects us to have done irq_enter().
945 	 * Besides, if we don't timer interrupts ignore the global
946 	 * interrupt lock, which is the WrongThing (tm) to do.
947 	 */
948 	entering_ack_irq();
949 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
950 	local_apic_timer_interrupt();
951 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
952 	exiting_irq();
953 
954 	set_irq_regs(old_regs);
955 }
956 
957 int setup_profiling_timer(unsigned int multiplier)
958 {
959 	return -EINVAL;
960 }
961 
962 /*
963  * Local APIC start and shutdown
964  */
965 
966 /**
967  * clear_local_APIC - shutdown the local APIC
968  *
969  * This is called, when a CPU is disabled and before rebooting, so the state of
970  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
971  * leftovers during boot.
972  */
973 void clear_local_APIC(void)
974 {
975 	int maxlvt;
976 	u32 v;
977 
978 	/* APIC hasn't been mapped yet */
979 	if (!x2apic_mode && !apic_phys)
980 		return;
981 
982 	maxlvt = lapic_get_maxlvt();
983 	/*
984 	 * Masking an LVT entry can trigger a local APIC error
985 	 * if the vector is zero. Mask LVTERR first to prevent this.
986 	 */
987 	if (maxlvt >= 3) {
988 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
989 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
990 	}
991 	/*
992 	 * Careful: we have to set masks only first to deassert
993 	 * any level-triggered sources.
994 	 */
995 	v = apic_read(APIC_LVTT);
996 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
997 	v = apic_read(APIC_LVT0);
998 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
999 	v = apic_read(APIC_LVT1);
1000 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1001 	if (maxlvt >= 4) {
1002 		v = apic_read(APIC_LVTPC);
1003 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1004 	}
1005 
1006 	/* lets not touch this if we didn't frob it */
1007 #ifdef CONFIG_X86_THERMAL_VECTOR
1008 	if (maxlvt >= 5) {
1009 		v = apic_read(APIC_LVTTHMR);
1010 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1011 	}
1012 #endif
1013 #ifdef CONFIG_X86_MCE_INTEL
1014 	if (maxlvt >= 6) {
1015 		v = apic_read(APIC_LVTCMCI);
1016 		if (!(v & APIC_LVT_MASKED))
1017 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1018 	}
1019 #endif
1020 
1021 	/*
1022 	 * Clean APIC state for other OSs:
1023 	 */
1024 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1025 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1026 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1027 	if (maxlvt >= 3)
1028 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1029 	if (maxlvt >= 4)
1030 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1031 
1032 	/* Integrated APIC (!82489DX) ? */
1033 	if (lapic_is_integrated()) {
1034 		if (maxlvt > 3)
1035 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1036 			apic_write(APIC_ESR, 0);
1037 		apic_read(APIC_ESR);
1038 	}
1039 }
1040 
1041 /**
1042  * disable_local_APIC - clear and disable the local APIC
1043  */
1044 void disable_local_APIC(void)
1045 {
1046 	unsigned int value;
1047 
1048 	/* APIC hasn't been mapped yet */
1049 	if (!x2apic_mode && !apic_phys)
1050 		return;
1051 
1052 	clear_local_APIC();
1053 
1054 	/*
1055 	 * Disable APIC (implies clearing of registers
1056 	 * for 82489DX!).
1057 	 */
1058 	value = apic_read(APIC_SPIV);
1059 	value &= ~APIC_SPIV_APIC_ENABLED;
1060 	apic_write(APIC_SPIV, value);
1061 
1062 #ifdef CONFIG_X86_32
1063 	/*
1064 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1065 	 * restore the disabled state.
1066 	 */
1067 	if (enabled_via_apicbase) {
1068 		unsigned int l, h;
1069 
1070 		rdmsr(MSR_IA32_APICBASE, l, h);
1071 		l &= ~MSR_IA32_APICBASE_ENABLE;
1072 		wrmsr(MSR_IA32_APICBASE, l, h);
1073 	}
1074 #endif
1075 }
1076 
1077 /*
1078  * If Linux enabled the LAPIC against the BIOS default disable it down before
1079  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1080  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1081  * for the case where Linux didn't enable the LAPIC.
1082  */
1083 void lapic_shutdown(void)
1084 {
1085 	unsigned long flags;
1086 
1087 	if (!cpu_has_apic && !apic_from_smp_config())
1088 		return;
1089 
1090 	local_irq_save(flags);
1091 
1092 #ifdef CONFIG_X86_32
1093 	if (!enabled_via_apicbase)
1094 		clear_local_APIC();
1095 	else
1096 #endif
1097 		disable_local_APIC();
1098 
1099 
1100 	local_irq_restore(flags);
1101 }
1102 
1103 /*
1104  * This is to verify that we're looking at a real local APIC.
1105  * Check these against your board if the CPUs aren't getting
1106  * started for no apparent reason.
1107  */
1108 int __init verify_local_APIC(void)
1109 {
1110 	unsigned int reg0, reg1;
1111 
1112 	/*
1113 	 * The version register is read-only in a real APIC.
1114 	 */
1115 	reg0 = apic_read(APIC_LVR);
1116 	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1117 	apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1118 	reg1 = apic_read(APIC_LVR);
1119 	apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1120 
1121 	/*
1122 	 * The two version reads above should print the same
1123 	 * numbers.  If the second one is different, then we
1124 	 * poke at a non-APIC.
1125 	 */
1126 	if (reg1 != reg0)
1127 		return 0;
1128 
1129 	/*
1130 	 * Check if the version looks reasonably.
1131 	 */
1132 	reg1 = GET_APIC_VERSION(reg0);
1133 	if (reg1 == 0x00 || reg1 == 0xff)
1134 		return 0;
1135 	reg1 = lapic_get_maxlvt();
1136 	if (reg1 < 0x02 || reg1 == 0xff)
1137 		return 0;
1138 
1139 	/*
1140 	 * The ID register is read/write in a real APIC.
1141 	 */
1142 	reg0 = apic_read(APIC_ID);
1143 	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1144 	apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1145 	reg1 = apic_read(APIC_ID);
1146 	apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1147 	apic_write(APIC_ID, reg0);
1148 	if (reg1 != (reg0 ^ apic->apic_id_mask))
1149 		return 0;
1150 
1151 	/*
1152 	 * The next two are just to see if we have sane values.
1153 	 * They're only really relevant if we're in Virtual Wire
1154 	 * compatibility mode, but most boxes are anymore.
1155 	 */
1156 	reg0 = apic_read(APIC_LVT0);
1157 	apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1158 	reg1 = apic_read(APIC_LVT1);
1159 	apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1160 
1161 	return 1;
1162 }
1163 
1164 /**
1165  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1166  */
1167 void __init sync_Arb_IDs(void)
1168 {
1169 	/*
1170 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1171 	 * needed on AMD.
1172 	 */
1173 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1174 		return;
1175 
1176 	/*
1177 	 * Wait for idle.
1178 	 */
1179 	apic_wait_icr_idle();
1180 
1181 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1182 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1183 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1184 }
1185 
1186 /*
1187  * An initial setup of the virtual wire mode.
1188  */
1189 void __init init_bsp_APIC(void)
1190 {
1191 	unsigned int value;
1192 
1193 	/*
1194 	 * Don't do the setup now if we have a SMP BIOS as the
1195 	 * through-I/O-APIC virtual wire mode might be active.
1196 	 */
1197 	if (smp_found_config || !cpu_has_apic)
1198 		return;
1199 
1200 	/*
1201 	 * Do not trust the local APIC being empty at bootup.
1202 	 */
1203 	clear_local_APIC();
1204 
1205 	/*
1206 	 * Enable APIC.
1207 	 */
1208 	value = apic_read(APIC_SPIV);
1209 	value &= ~APIC_VECTOR_MASK;
1210 	value |= APIC_SPIV_APIC_ENABLED;
1211 
1212 #ifdef CONFIG_X86_32
1213 	/* This bit is reserved on P4/Xeon and should be cleared */
1214 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1215 	    (boot_cpu_data.x86 == 15))
1216 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1217 	else
1218 #endif
1219 		value |= APIC_SPIV_FOCUS_DISABLED;
1220 	value |= SPURIOUS_APIC_VECTOR;
1221 	apic_write(APIC_SPIV, value);
1222 
1223 	/*
1224 	 * Set up the virtual wire mode.
1225 	 */
1226 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1227 	value = APIC_DM_NMI;
1228 	if (!lapic_is_integrated())		/* 82489DX */
1229 		value |= APIC_LVT_LEVEL_TRIGGER;
1230 	apic_write(APIC_LVT1, value);
1231 }
1232 
1233 static void lapic_setup_esr(void)
1234 {
1235 	unsigned int oldvalue, value, maxlvt;
1236 
1237 	if (!lapic_is_integrated()) {
1238 		pr_info("No ESR for 82489DX.\n");
1239 		return;
1240 	}
1241 
1242 	if (apic->disable_esr) {
1243 		/*
1244 		 * Something untraceable is creating bad interrupts on
1245 		 * secondary quads ... for the moment, just leave the
1246 		 * ESR disabled - we can't do anything useful with the
1247 		 * errors anyway - mbligh
1248 		 */
1249 		pr_info("Leaving ESR disabled.\n");
1250 		return;
1251 	}
1252 
1253 	maxlvt = lapic_get_maxlvt();
1254 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1255 		apic_write(APIC_ESR, 0);
1256 	oldvalue = apic_read(APIC_ESR);
1257 
1258 	/* enables sending errors */
1259 	value = ERROR_APIC_VECTOR;
1260 	apic_write(APIC_LVTERR, value);
1261 
1262 	/*
1263 	 * spec says clear errors after enabling vector.
1264 	 */
1265 	if (maxlvt > 3)
1266 		apic_write(APIC_ESR, 0);
1267 	value = apic_read(APIC_ESR);
1268 	if (value != oldvalue)
1269 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1270 			"vector: 0x%08x  after: 0x%08x\n",
1271 			oldvalue, value);
1272 }
1273 
1274 /**
1275  * setup_local_APIC - setup the local APIC
1276  *
1277  * Used to setup local APIC while initializing BSP or bringin up APs.
1278  * Always called with preemption disabled.
1279  */
1280 void setup_local_APIC(void)
1281 {
1282 	int cpu = smp_processor_id();
1283 	unsigned int value, queued;
1284 	int i, j, acked = 0;
1285 	unsigned long long tsc = 0, ntsc;
1286 	long long max_loops = cpu_khz;
1287 
1288 	if (cpu_has_tsc)
1289 		rdtscll(tsc);
1290 
1291 	if (disable_apic) {
1292 		disable_ioapic_support();
1293 		return;
1294 	}
1295 
1296 #ifdef CONFIG_X86_32
1297 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1298 	if (lapic_is_integrated() && apic->disable_esr) {
1299 		apic_write(APIC_ESR, 0);
1300 		apic_write(APIC_ESR, 0);
1301 		apic_write(APIC_ESR, 0);
1302 		apic_write(APIC_ESR, 0);
1303 	}
1304 #endif
1305 	perf_events_lapic_init();
1306 
1307 	/*
1308 	 * Double-check whether this APIC is really registered.
1309 	 * This is meaningless in clustered apic mode, so we skip it.
1310 	 */
1311 	BUG_ON(!apic->apic_id_registered());
1312 
1313 	/*
1314 	 * Intel recommends to set DFR, LDR and TPR before enabling
1315 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1316 	 * document number 292116).  So here it goes...
1317 	 */
1318 	apic->init_apic_ldr();
1319 
1320 #ifdef CONFIG_X86_32
1321 	/*
1322 	 * APIC LDR is initialized.  If logical_apicid mapping was
1323 	 * initialized during get_smp_config(), make sure it matches the
1324 	 * actual value.
1325 	 */
1326 	i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1327 	WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1328 	/* always use the value from LDR */
1329 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1330 		logical_smp_processor_id();
1331 
1332 	/*
1333 	 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1334 	 * node mapping during NUMA init.  Now that logical apicid is
1335 	 * guaranteed to be known, give it another chance.  This is already
1336 	 * a bit too late - percpu allocation has already happened without
1337 	 * proper NUMA affinity.
1338 	 */
1339 	if (apic->x86_32_numa_cpu_node)
1340 		set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1341 				   apic->x86_32_numa_cpu_node(cpu));
1342 #endif
1343 
1344 	/*
1345 	 * Set Task Priority to 'accept all'. We never change this
1346 	 * later on.
1347 	 */
1348 	value = apic_read(APIC_TASKPRI);
1349 	value &= ~APIC_TPRI_MASK;
1350 	apic_write(APIC_TASKPRI, value);
1351 
1352 	/*
1353 	 * After a crash, we no longer service the interrupts and a pending
1354 	 * interrupt from previous kernel might still have ISR bit set.
1355 	 *
1356 	 * Most probably by now CPU has serviced that pending interrupt and
1357 	 * it might not have done the ack_APIC_irq() because it thought,
1358 	 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1359 	 * does not clear the ISR bit and cpu thinks it has already serivced
1360 	 * the interrupt. Hence a vector might get locked. It was noticed
1361 	 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1362 	 */
1363 	do {
1364 		queued = 0;
1365 		for (i = APIC_ISR_NR - 1; i >= 0; i--)
1366 			queued |= apic_read(APIC_IRR + i*0x10);
1367 
1368 		for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1369 			value = apic_read(APIC_ISR + i*0x10);
1370 			for (j = 31; j >= 0; j--) {
1371 				if (value & (1<<j)) {
1372 					ack_APIC_irq();
1373 					acked++;
1374 				}
1375 			}
1376 		}
1377 		if (acked > 256) {
1378 			printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1379 			       acked);
1380 			break;
1381 		}
1382 		if (queued) {
1383 			if (cpu_has_tsc) {
1384 				rdtscll(ntsc);
1385 				max_loops = (cpu_khz << 10) - (ntsc - tsc);
1386 			} else
1387 				max_loops--;
1388 		}
1389 	} while (queued && max_loops > 0);
1390 	WARN_ON(max_loops <= 0);
1391 
1392 	/*
1393 	 * Now that we are all set up, enable the APIC
1394 	 */
1395 	value = apic_read(APIC_SPIV);
1396 	value &= ~APIC_VECTOR_MASK;
1397 	/*
1398 	 * Enable APIC
1399 	 */
1400 	value |= APIC_SPIV_APIC_ENABLED;
1401 
1402 #ifdef CONFIG_X86_32
1403 	/*
1404 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1405 	 * certain networking cards. If high frequency interrupts are
1406 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1407 	 * entry is masked/unmasked at a high rate as well then sooner or
1408 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1409 	 * from the device. If focus CPU is disabled then the hang goes
1410 	 * away, oh well :-(
1411 	 *
1412 	 * [ This bug can be reproduced easily with a level-triggered
1413 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1414 	 *   BX chipset. ]
1415 	 */
1416 	/*
1417 	 * Actually disabling the focus CPU check just makes the hang less
1418 	 * frequent as it makes the interrupt distributon model be more
1419 	 * like LRU than MRU (the short-term load is more even across CPUs).
1420 	 * See also the comment in end_level_ioapic_irq().  --macro
1421 	 */
1422 
1423 	/*
1424 	 * - enable focus processor (bit==0)
1425 	 * - 64bit mode always use processor focus
1426 	 *   so no need to set it
1427 	 */
1428 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1429 #endif
1430 
1431 	/*
1432 	 * Set spurious IRQ vector
1433 	 */
1434 	value |= SPURIOUS_APIC_VECTOR;
1435 	apic_write(APIC_SPIV, value);
1436 
1437 	/*
1438 	 * Set up LVT0, LVT1:
1439 	 *
1440 	 * set up through-local-APIC on the BP's LINT0. This is not
1441 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1442 	 * we delegate interrupts to the 8259A.
1443 	 */
1444 	/*
1445 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1446 	 */
1447 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1448 	if (!cpu && (pic_mode || !value)) {
1449 		value = APIC_DM_EXTINT;
1450 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1451 	} else {
1452 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1453 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1454 	}
1455 	apic_write(APIC_LVT0, value);
1456 
1457 	/*
1458 	 * only the BP should see the LINT1 NMI signal, obviously.
1459 	 */
1460 	if (!cpu)
1461 		value = APIC_DM_NMI;
1462 	else
1463 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1464 	if (!lapic_is_integrated())		/* 82489DX */
1465 		value |= APIC_LVT_LEVEL_TRIGGER;
1466 	apic_write(APIC_LVT1, value);
1467 
1468 #ifdef CONFIG_X86_MCE_INTEL
1469 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1470 	if (!cpu)
1471 		cmci_recheck();
1472 #endif
1473 }
1474 
1475 void end_local_APIC_setup(void)
1476 {
1477 	lapic_setup_esr();
1478 
1479 #ifdef CONFIG_X86_32
1480 	{
1481 		unsigned int value;
1482 		/* Disable the local apic timer */
1483 		value = apic_read(APIC_LVTT);
1484 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1485 		apic_write(APIC_LVTT, value);
1486 	}
1487 #endif
1488 
1489 	apic_pm_activate();
1490 }
1491 
1492 void __init bsp_end_local_APIC_setup(void)
1493 {
1494 	end_local_APIC_setup();
1495 
1496 	/*
1497 	 * Now that local APIC setup is completed for BP, configure the fault
1498 	 * handling for interrupt remapping.
1499 	 */
1500 	irq_remap_enable_fault_handling();
1501 
1502 }
1503 
1504 #ifdef CONFIG_X86_X2APIC
1505 /*
1506  * Need to disable xapic and x2apic at the same time and then enable xapic mode
1507  */
1508 static inline void __disable_x2apic(u64 msr)
1509 {
1510 	wrmsrl(MSR_IA32_APICBASE,
1511 	       msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1512 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1513 }
1514 
1515 static __init void disable_x2apic(void)
1516 {
1517 	u64 msr;
1518 
1519 	if (!cpu_has_x2apic)
1520 		return;
1521 
1522 	rdmsrl(MSR_IA32_APICBASE, msr);
1523 	if (msr & X2APIC_ENABLE) {
1524 		u32 x2apic_id = read_apic_id();
1525 
1526 		if (x2apic_id >= 255)
1527 			panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1528 
1529 		pr_info("Disabling x2apic\n");
1530 		__disable_x2apic(msr);
1531 
1532 		if (nox2apic) {
1533 			clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1534 			setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1535 		}
1536 
1537 		x2apic_disabled = 1;
1538 		x2apic_mode = 0;
1539 
1540 		register_lapic_address(mp_lapic_addr);
1541 	}
1542 }
1543 
1544 void check_x2apic(void)
1545 {
1546 	if (x2apic_enabled()) {
1547 		pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1548 		x2apic_preenabled = x2apic_mode = 1;
1549 	}
1550 }
1551 
1552 void enable_x2apic(void)
1553 {
1554 	u64 msr;
1555 
1556 	rdmsrl(MSR_IA32_APICBASE, msr);
1557 	if (x2apic_disabled) {
1558 		__disable_x2apic(msr);
1559 		return;
1560 	}
1561 
1562 	if (!x2apic_mode)
1563 		return;
1564 
1565 	if (!(msr & X2APIC_ENABLE)) {
1566 		printk_once(KERN_INFO "Enabling x2apic\n");
1567 		wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1568 	}
1569 }
1570 #endif /* CONFIG_X86_X2APIC */
1571 
1572 int __init enable_IR(void)
1573 {
1574 #ifdef CONFIG_IRQ_REMAP
1575 	if (!irq_remapping_supported()) {
1576 		pr_debug("intr-remapping not supported\n");
1577 		return -1;
1578 	}
1579 
1580 	if (!x2apic_preenabled && skip_ioapic_setup) {
1581 		pr_info("Skipped enabling intr-remap because of skipping "
1582 			"io-apic setup\n");
1583 		return -1;
1584 	}
1585 
1586 	return irq_remapping_enable();
1587 #endif
1588 	return -1;
1589 }
1590 
1591 void __init enable_IR_x2apic(void)
1592 {
1593 	unsigned long flags;
1594 	int ret, x2apic_enabled = 0;
1595 	int hardware_init_ret;
1596 
1597 	/* Make sure irq_remap_ops are initialized */
1598 	setup_irq_remapping_ops();
1599 
1600 	hardware_init_ret = irq_remapping_prepare();
1601 	if (hardware_init_ret && !x2apic_supported())
1602 		return;
1603 
1604 	ret = save_ioapic_entries();
1605 	if (ret) {
1606 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1607 		return;
1608 	}
1609 
1610 	local_irq_save(flags);
1611 	legacy_pic->mask_all();
1612 	mask_ioapic_entries();
1613 
1614 	if (x2apic_preenabled && nox2apic)
1615 		disable_x2apic();
1616 
1617 	if (hardware_init_ret)
1618 		ret = -1;
1619 	else
1620 		ret = enable_IR();
1621 
1622 	if (!x2apic_supported())
1623 		goto skip_x2apic;
1624 
1625 	if (ret < 0) {
1626 		/* IR is required if there is APIC ID > 255 even when running
1627 		 * under KVM
1628 		 */
1629 		if (max_physical_apicid > 255 ||
1630 		    !hypervisor_x2apic_available()) {
1631 			if (x2apic_preenabled)
1632 				disable_x2apic();
1633 			goto skip_x2apic;
1634 		}
1635 		/*
1636 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1637 		 * only in physical mode
1638 		 */
1639 		x2apic_force_phys();
1640 	}
1641 
1642 	if (ret == IRQ_REMAP_XAPIC_MODE) {
1643 		pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1644 		goto skip_x2apic;
1645 	}
1646 
1647 	x2apic_enabled = 1;
1648 
1649 	if (x2apic_supported() && !x2apic_mode) {
1650 		x2apic_mode = 1;
1651 		enable_x2apic();
1652 		pr_info("Enabled x2apic\n");
1653 	}
1654 
1655 skip_x2apic:
1656 	if (ret < 0) /* IR enabling failed */
1657 		restore_ioapic_entries();
1658 	legacy_pic->restore_mask();
1659 	local_irq_restore(flags);
1660 }
1661 
1662 #ifdef CONFIG_X86_64
1663 /*
1664  * Detect and enable local APICs on non-SMP boards.
1665  * Original code written by Keir Fraser.
1666  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1667  * not correctly set up (usually the APIC timer won't work etc.)
1668  */
1669 static int __init detect_init_APIC(void)
1670 {
1671 	if (!cpu_has_apic) {
1672 		pr_info("No local APIC present\n");
1673 		return -1;
1674 	}
1675 
1676 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1677 	return 0;
1678 }
1679 #else
1680 
1681 static int __init apic_verify(void)
1682 {
1683 	u32 features, h, l;
1684 
1685 	/*
1686 	 * The APIC feature bit should now be enabled
1687 	 * in `cpuid'
1688 	 */
1689 	features = cpuid_edx(1);
1690 	if (!(features & (1 << X86_FEATURE_APIC))) {
1691 		pr_warning("Could not enable APIC!\n");
1692 		return -1;
1693 	}
1694 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1695 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1696 
1697 	/* The BIOS may have set up the APIC at some other address */
1698 	if (boot_cpu_data.x86 >= 6) {
1699 		rdmsr(MSR_IA32_APICBASE, l, h);
1700 		if (l & MSR_IA32_APICBASE_ENABLE)
1701 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1702 	}
1703 
1704 	pr_info("Found and enabled local APIC!\n");
1705 	return 0;
1706 }
1707 
1708 int __init apic_force_enable(unsigned long addr)
1709 {
1710 	u32 h, l;
1711 
1712 	if (disable_apic)
1713 		return -1;
1714 
1715 	/*
1716 	 * Some BIOSes disable the local APIC in the APIC_BASE
1717 	 * MSR. This can only be done in software for Intel P6 or later
1718 	 * and AMD K7 (Model > 1) or later.
1719 	 */
1720 	if (boot_cpu_data.x86 >= 6) {
1721 		rdmsr(MSR_IA32_APICBASE, l, h);
1722 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1723 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1724 			l &= ~MSR_IA32_APICBASE_BASE;
1725 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1726 			wrmsr(MSR_IA32_APICBASE, l, h);
1727 			enabled_via_apicbase = 1;
1728 		}
1729 	}
1730 	return apic_verify();
1731 }
1732 
1733 /*
1734  * Detect and initialize APIC
1735  */
1736 static int __init detect_init_APIC(void)
1737 {
1738 	/* Disabled by kernel option? */
1739 	if (disable_apic)
1740 		return -1;
1741 
1742 	switch (boot_cpu_data.x86_vendor) {
1743 	case X86_VENDOR_AMD:
1744 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1745 		    (boot_cpu_data.x86 >= 15))
1746 			break;
1747 		goto no_apic;
1748 	case X86_VENDOR_INTEL:
1749 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1750 		    (boot_cpu_data.x86 == 5 && cpu_has_apic))
1751 			break;
1752 		goto no_apic;
1753 	default:
1754 		goto no_apic;
1755 	}
1756 
1757 	if (!cpu_has_apic) {
1758 		/*
1759 		 * Over-ride BIOS and try to enable the local APIC only if
1760 		 * "lapic" specified.
1761 		 */
1762 		if (!force_enable_local_apic) {
1763 			pr_info("Local APIC disabled by BIOS -- "
1764 				"you can enable it with \"lapic\"\n");
1765 			return -1;
1766 		}
1767 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1768 			return -1;
1769 	} else {
1770 		if (apic_verify())
1771 			return -1;
1772 	}
1773 
1774 	apic_pm_activate();
1775 
1776 	return 0;
1777 
1778 no_apic:
1779 	pr_info("No local APIC present or hardware disabled\n");
1780 	return -1;
1781 }
1782 #endif
1783 
1784 /**
1785  * init_apic_mappings - initialize APIC mappings
1786  */
1787 void __init init_apic_mappings(void)
1788 {
1789 	unsigned int new_apicid;
1790 
1791 	if (x2apic_mode) {
1792 		boot_cpu_physical_apicid = read_apic_id();
1793 		return;
1794 	}
1795 
1796 	/* If no local APIC can be found return early */
1797 	if (!smp_found_config && detect_init_APIC()) {
1798 		/* lets NOP'ify apic operations */
1799 		pr_info("APIC: disable apic facility\n");
1800 		apic_disable();
1801 	} else {
1802 		apic_phys = mp_lapic_addr;
1803 
1804 		/*
1805 		 * acpi lapic path already maps that address in
1806 		 * acpi_register_lapic_address()
1807 		 */
1808 		if (!acpi_lapic && !smp_found_config)
1809 			register_lapic_address(apic_phys);
1810 	}
1811 
1812 	/*
1813 	 * Fetch the APIC ID of the BSP in case we have a
1814 	 * default configuration (or the MP table is broken).
1815 	 */
1816 	new_apicid = read_apic_id();
1817 	if (boot_cpu_physical_apicid != new_apicid) {
1818 		boot_cpu_physical_apicid = new_apicid;
1819 		/*
1820 		 * yeah -- we lie about apic_version
1821 		 * in case if apic was disabled via boot option
1822 		 * but it's not a problem for SMP compiled kernel
1823 		 * since smp_sanity_check is prepared for such a case
1824 		 * and disable smp mode
1825 		 */
1826 		apic_version[new_apicid] =
1827 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1828 	}
1829 }
1830 
1831 void __init register_lapic_address(unsigned long address)
1832 {
1833 	mp_lapic_addr = address;
1834 
1835 	if (!x2apic_mode) {
1836 		set_fixmap_nocache(FIX_APIC_BASE, address);
1837 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1838 			    APIC_BASE, mp_lapic_addr);
1839 	}
1840 	if (boot_cpu_physical_apicid == -1U) {
1841 		boot_cpu_physical_apicid  = read_apic_id();
1842 		apic_version[boot_cpu_physical_apicid] =
1843 			 GET_APIC_VERSION(apic_read(APIC_LVR));
1844 	}
1845 }
1846 
1847 /*
1848  * This initializes the IO-APIC and APIC hardware if this is
1849  * a UP kernel.
1850  */
1851 int apic_version[MAX_LOCAL_APIC];
1852 
1853 int __init APIC_init_uniprocessor(void)
1854 {
1855 	if (disable_apic) {
1856 		pr_info("Apic disabled\n");
1857 		return -1;
1858 	}
1859 #ifdef CONFIG_X86_64
1860 	if (!cpu_has_apic) {
1861 		disable_apic = 1;
1862 		pr_info("Apic disabled by BIOS\n");
1863 		return -1;
1864 	}
1865 #else
1866 	if (!smp_found_config && !cpu_has_apic)
1867 		return -1;
1868 
1869 	/*
1870 	 * Complain if the BIOS pretends there is one.
1871 	 */
1872 	if (!cpu_has_apic &&
1873 	    APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1874 		pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1875 			boot_cpu_physical_apicid);
1876 		return -1;
1877 	}
1878 #endif
1879 
1880 	default_setup_apic_routing();
1881 
1882 	verify_local_APIC();
1883 	connect_bsp_APIC();
1884 
1885 #ifdef CONFIG_X86_64
1886 	apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1887 #else
1888 	/*
1889 	 * Hack: In case of kdump, after a crash, kernel might be booting
1890 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1891 	 * might be zero if read from MP tables. Get it from LAPIC.
1892 	 */
1893 # ifdef CONFIG_CRASH_DUMP
1894 	boot_cpu_physical_apicid = read_apic_id();
1895 # endif
1896 #endif
1897 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1898 	setup_local_APIC();
1899 
1900 #ifdef CONFIG_X86_IO_APIC
1901 	/*
1902 	 * Now enable IO-APICs, actually call clear_IO_APIC
1903 	 * We need clear_IO_APIC before enabling error vector
1904 	 */
1905 	if (!skip_ioapic_setup && nr_ioapics)
1906 		enable_IO_APIC();
1907 #endif
1908 
1909 	bsp_end_local_APIC_setup();
1910 
1911 #ifdef CONFIG_X86_IO_APIC
1912 	if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1913 		setup_IO_APIC();
1914 	else {
1915 		nr_ioapics = 0;
1916 	}
1917 #endif
1918 
1919 	x86_init.timers.setup_percpu_clockev();
1920 	return 0;
1921 }
1922 
1923 /*
1924  * Local APIC interrupts
1925  */
1926 
1927 /*
1928  * This interrupt should _never_ happen with our APIC/SMP architecture
1929  */
1930 static inline void __smp_spurious_interrupt(void)
1931 {
1932 	u32 v;
1933 
1934 	/*
1935 	 * Check if this really is a spurious interrupt and ACK it
1936 	 * if it is a vectored one.  Just in case...
1937 	 * Spurious interrupts should not be ACKed.
1938 	 */
1939 	v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1940 	if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1941 		ack_APIC_irq();
1942 
1943 	inc_irq_stat(irq_spurious_count);
1944 
1945 	/* see sw-dev-man vol 3, chapter 7.4.13.5 */
1946 	pr_info("spurious APIC interrupt on CPU#%d, "
1947 		"should never happen.\n", smp_processor_id());
1948 }
1949 
1950 __visible void smp_spurious_interrupt(struct pt_regs *regs)
1951 {
1952 	entering_irq();
1953 	__smp_spurious_interrupt();
1954 	exiting_irq();
1955 }
1956 
1957 __visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1958 {
1959 	entering_irq();
1960 	trace_spurious_apic_entry(SPURIOUS_APIC_VECTOR);
1961 	__smp_spurious_interrupt();
1962 	trace_spurious_apic_exit(SPURIOUS_APIC_VECTOR);
1963 	exiting_irq();
1964 }
1965 
1966 /*
1967  * This interrupt should never happen with our APIC/SMP architecture
1968  */
1969 static inline void __smp_error_interrupt(struct pt_regs *regs)
1970 {
1971 	u32 v0, v1;
1972 	u32 i = 0;
1973 	static const char * const error_interrupt_reason[] = {
1974 		"Send CS error",		/* APIC Error Bit 0 */
1975 		"Receive CS error",		/* APIC Error Bit 1 */
1976 		"Send accept error",		/* APIC Error Bit 2 */
1977 		"Receive accept error",		/* APIC Error Bit 3 */
1978 		"Redirectable IPI",		/* APIC Error Bit 4 */
1979 		"Send illegal vector",		/* APIC Error Bit 5 */
1980 		"Received illegal vector",	/* APIC Error Bit 6 */
1981 		"Illegal register address",	/* APIC Error Bit 7 */
1982 	};
1983 
1984 	/* First tickle the hardware, only then report what went on. -- REW */
1985 	v0 = apic_read(APIC_ESR);
1986 	apic_write(APIC_ESR, 0);
1987 	v1 = apic_read(APIC_ESR);
1988 	ack_APIC_irq();
1989 	atomic_inc(&irq_err_count);
1990 
1991 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x(%02x)",
1992 		    smp_processor_id(), v0 , v1);
1993 
1994 	v1 = v1 & 0xff;
1995 	while (v1) {
1996 		if (v1 & 0x1)
1997 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
1998 		i++;
1999 		v1 >>= 1;
2000 	}
2001 
2002 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2003 
2004 }
2005 
2006 __visible void smp_error_interrupt(struct pt_regs *regs)
2007 {
2008 	entering_irq();
2009 	__smp_error_interrupt(regs);
2010 	exiting_irq();
2011 }
2012 
2013 __visible void smp_trace_error_interrupt(struct pt_regs *regs)
2014 {
2015 	entering_irq();
2016 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2017 	__smp_error_interrupt(regs);
2018 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2019 	exiting_irq();
2020 }
2021 
2022 /**
2023  * connect_bsp_APIC - attach the APIC to the interrupt system
2024  */
2025 void __init connect_bsp_APIC(void)
2026 {
2027 #ifdef CONFIG_X86_32
2028 	if (pic_mode) {
2029 		/*
2030 		 * Do not trust the local APIC being empty at bootup.
2031 		 */
2032 		clear_local_APIC();
2033 		/*
2034 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2035 		 * local APIC to INT and NMI lines.
2036 		 */
2037 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2038 				"enabling APIC mode.\n");
2039 		imcr_pic_to_apic();
2040 	}
2041 #endif
2042 	if (apic->enable_apic_mode)
2043 		apic->enable_apic_mode();
2044 }
2045 
2046 /**
2047  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2048  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2049  *
2050  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2051  * APIC is disabled.
2052  */
2053 void disconnect_bsp_APIC(int virt_wire_setup)
2054 {
2055 	unsigned int value;
2056 
2057 #ifdef CONFIG_X86_32
2058 	if (pic_mode) {
2059 		/*
2060 		 * Put the board back into PIC mode (has an effect only on
2061 		 * certain older boards).  Note that APIC interrupts, including
2062 		 * IPIs, won't work beyond this point!  The only exception are
2063 		 * INIT IPIs.
2064 		 */
2065 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2066 				"entering PIC mode.\n");
2067 		imcr_apic_to_pic();
2068 		return;
2069 	}
2070 #endif
2071 
2072 	/* Go back to Virtual Wire compatibility mode */
2073 
2074 	/* For the spurious interrupt use vector F, and enable it */
2075 	value = apic_read(APIC_SPIV);
2076 	value &= ~APIC_VECTOR_MASK;
2077 	value |= APIC_SPIV_APIC_ENABLED;
2078 	value |= 0xf;
2079 	apic_write(APIC_SPIV, value);
2080 
2081 	if (!virt_wire_setup) {
2082 		/*
2083 		 * For LVT0 make it edge triggered, active high,
2084 		 * external and enabled
2085 		 */
2086 		value = apic_read(APIC_LVT0);
2087 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2088 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2089 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2090 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2091 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2092 		apic_write(APIC_LVT0, value);
2093 	} else {
2094 		/* Disable LVT0 */
2095 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2096 	}
2097 
2098 	/*
2099 	 * For LVT1 make it edge triggered, active high,
2100 	 * nmi and enabled
2101 	 */
2102 	value = apic_read(APIC_LVT1);
2103 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2104 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2105 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2106 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2107 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2108 	apic_write(APIC_LVT1, value);
2109 }
2110 
2111 int generic_processor_info(int apicid, int version)
2112 {
2113 	int cpu, max = nr_cpu_ids;
2114 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2115 				phys_cpu_present_map);
2116 
2117 	/*
2118 	 * If boot cpu has not been detected yet, then only allow upto
2119 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2120 	 */
2121 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2122 	    apicid != boot_cpu_physical_apicid) {
2123 		int thiscpu = max + disabled_cpus - 1;
2124 
2125 		pr_warning(
2126 			"ACPI: NR_CPUS/possible_cpus limit of %i almost"
2127 			" reached. Keeping one slot for boot cpu."
2128 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2129 
2130 		disabled_cpus++;
2131 		return -ENODEV;
2132 	}
2133 
2134 	if (num_processors >= nr_cpu_ids) {
2135 		int thiscpu = max + disabled_cpus;
2136 
2137 		pr_warning(
2138 			"ACPI: NR_CPUS/possible_cpus limit of %i reached."
2139 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2140 
2141 		disabled_cpus++;
2142 		return -EINVAL;
2143 	}
2144 
2145 	num_processors++;
2146 	if (apicid == boot_cpu_physical_apicid) {
2147 		/*
2148 		 * x86_bios_cpu_apicid is required to have processors listed
2149 		 * in same order as logical cpu numbers. Hence the first
2150 		 * entry is BSP, and so on.
2151 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2152 		 * for BSP.
2153 		 */
2154 		cpu = 0;
2155 	} else
2156 		cpu = cpumask_next_zero(-1, cpu_present_mask);
2157 
2158 	/*
2159 	 * Validate version
2160 	 */
2161 	if (version == 0x0) {
2162 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2163 			   cpu, apicid);
2164 		version = 0x10;
2165 	}
2166 	apic_version[apicid] = version;
2167 
2168 	if (version != apic_version[boot_cpu_physical_apicid]) {
2169 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2170 			apic_version[boot_cpu_physical_apicid], cpu, version);
2171 	}
2172 
2173 	physid_set(apicid, phys_cpu_present_map);
2174 	if (apicid > max_physical_apicid)
2175 		max_physical_apicid = apicid;
2176 
2177 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2178 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2179 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2180 #endif
2181 #ifdef CONFIG_X86_32
2182 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2183 		apic->x86_32_early_logical_apicid(cpu);
2184 #endif
2185 	set_cpu_possible(cpu, true);
2186 	set_cpu_present(cpu, true);
2187 
2188 	return cpu;
2189 }
2190 
2191 int hard_smp_processor_id(void)
2192 {
2193 	return read_apic_id();
2194 }
2195 
2196 void default_init_apic_ldr(void)
2197 {
2198 	unsigned long val;
2199 
2200 	apic_write(APIC_DFR, APIC_DFR_VALUE);
2201 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2202 	val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2203 	apic_write(APIC_LDR, val);
2204 }
2205 
2206 int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2207 				   const struct cpumask *andmask,
2208 				   unsigned int *apicid)
2209 {
2210 	unsigned int cpu;
2211 
2212 	for_each_cpu_and(cpu, cpumask, andmask) {
2213 		if (cpumask_test_cpu(cpu, cpu_online_mask))
2214 			break;
2215 	}
2216 
2217 	if (likely(cpu < nr_cpu_ids)) {
2218 		*apicid = per_cpu(x86_cpu_to_apicid, cpu);
2219 		return 0;
2220 	}
2221 
2222 	return -EINVAL;
2223 }
2224 
2225 /*
2226  * Override the generic EOI implementation with an optimized version.
2227  * Only called during early boot when only one CPU is active and with
2228  * interrupts disabled, so we know this does not race with actual APIC driver
2229  * use.
2230  */
2231 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2232 {
2233 	struct apic **drv;
2234 
2235 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2236 		/* Should happen once for each apic */
2237 		WARN_ON((*drv)->eoi_write == eoi_write);
2238 		(*drv)->eoi_write = eoi_write;
2239 	}
2240 }
2241 
2242 /*
2243  * Power management
2244  */
2245 #ifdef CONFIG_PM
2246 
2247 static struct {
2248 	/*
2249 	 * 'active' is true if the local APIC was enabled by us and
2250 	 * not the BIOS; this signifies that we are also responsible
2251 	 * for disabling it before entering apm/acpi suspend
2252 	 */
2253 	int active;
2254 	/* r/w apic fields */
2255 	unsigned int apic_id;
2256 	unsigned int apic_taskpri;
2257 	unsigned int apic_ldr;
2258 	unsigned int apic_dfr;
2259 	unsigned int apic_spiv;
2260 	unsigned int apic_lvtt;
2261 	unsigned int apic_lvtpc;
2262 	unsigned int apic_lvt0;
2263 	unsigned int apic_lvt1;
2264 	unsigned int apic_lvterr;
2265 	unsigned int apic_tmict;
2266 	unsigned int apic_tdcr;
2267 	unsigned int apic_thmr;
2268 } apic_pm_state;
2269 
2270 static int lapic_suspend(void)
2271 {
2272 	unsigned long flags;
2273 	int maxlvt;
2274 
2275 	if (!apic_pm_state.active)
2276 		return 0;
2277 
2278 	maxlvt = lapic_get_maxlvt();
2279 
2280 	apic_pm_state.apic_id = apic_read(APIC_ID);
2281 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2282 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2283 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2284 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2285 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2286 	if (maxlvt >= 4)
2287 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2288 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2289 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2290 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2291 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2292 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2293 #ifdef CONFIG_X86_THERMAL_VECTOR
2294 	if (maxlvt >= 5)
2295 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2296 #endif
2297 
2298 	local_irq_save(flags);
2299 	disable_local_APIC();
2300 
2301 	irq_remapping_disable();
2302 
2303 	local_irq_restore(flags);
2304 	return 0;
2305 }
2306 
2307 static void lapic_resume(void)
2308 {
2309 	unsigned int l, h;
2310 	unsigned long flags;
2311 	int maxlvt;
2312 
2313 	if (!apic_pm_state.active)
2314 		return;
2315 
2316 	local_irq_save(flags);
2317 
2318 	/*
2319 	 * IO-APIC and PIC have their own resume routines.
2320 	 * We just mask them here to make sure the interrupt
2321 	 * subsystem is completely quiet while we enable x2apic
2322 	 * and interrupt-remapping.
2323 	 */
2324 	mask_ioapic_entries();
2325 	legacy_pic->mask_all();
2326 
2327 	if (x2apic_mode)
2328 		enable_x2apic();
2329 	else {
2330 		/*
2331 		 * Make sure the APICBASE points to the right address
2332 		 *
2333 		 * FIXME! This will be wrong if we ever support suspend on
2334 		 * SMP! We'll need to do this as part of the CPU restore!
2335 		 */
2336 		if (boot_cpu_data.x86 >= 6) {
2337 			rdmsr(MSR_IA32_APICBASE, l, h);
2338 			l &= ~MSR_IA32_APICBASE_BASE;
2339 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2340 			wrmsr(MSR_IA32_APICBASE, l, h);
2341 		}
2342 	}
2343 
2344 	maxlvt = lapic_get_maxlvt();
2345 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2346 	apic_write(APIC_ID, apic_pm_state.apic_id);
2347 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2348 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2349 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2350 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2351 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2352 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2353 #if defined(CONFIG_X86_MCE_INTEL)
2354 	if (maxlvt >= 5)
2355 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2356 #endif
2357 	if (maxlvt >= 4)
2358 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2359 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2360 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2361 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2362 	apic_write(APIC_ESR, 0);
2363 	apic_read(APIC_ESR);
2364 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2365 	apic_write(APIC_ESR, 0);
2366 	apic_read(APIC_ESR);
2367 
2368 	irq_remapping_reenable(x2apic_mode);
2369 
2370 	local_irq_restore(flags);
2371 }
2372 
2373 /*
2374  * This device has no shutdown method - fully functioning local APICs
2375  * are needed on every CPU up until machine_halt/restart/poweroff.
2376  */
2377 
2378 static struct syscore_ops lapic_syscore_ops = {
2379 	.resume		= lapic_resume,
2380 	.suspend	= lapic_suspend,
2381 };
2382 
2383 static void apic_pm_activate(void)
2384 {
2385 	apic_pm_state.active = 1;
2386 }
2387 
2388 static int __init init_lapic_sysfs(void)
2389 {
2390 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2391 	if (cpu_has_apic)
2392 		register_syscore_ops(&lapic_syscore_ops);
2393 
2394 	return 0;
2395 }
2396 
2397 /* local apic needs to resume before other devices access its registers. */
2398 core_initcall(init_lapic_sysfs);
2399 
2400 #else	/* CONFIG_PM */
2401 
2402 static void apic_pm_activate(void) { }
2403 
2404 #endif	/* CONFIG_PM */
2405 
2406 #ifdef CONFIG_X86_64
2407 
2408 static int apic_cluster_num(void)
2409 {
2410 	int i, clusters, zeros;
2411 	unsigned id;
2412 	u16 *bios_cpu_apicid;
2413 	DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2414 
2415 	bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2416 	bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2417 
2418 	for (i = 0; i < nr_cpu_ids; i++) {
2419 		/* are we being called early in kernel startup? */
2420 		if (bios_cpu_apicid) {
2421 			id = bios_cpu_apicid[i];
2422 		} else if (i < nr_cpu_ids) {
2423 			if (cpu_present(i))
2424 				id = per_cpu(x86_bios_cpu_apicid, i);
2425 			else
2426 				continue;
2427 		} else
2428 			break;
2429 
2430 		if (id != BAD_APICID)
2431 			__set_bit(APIC_CLUSTERID(id), clustermap);
2432 	}
2433 
2434 	/* Problem:  Partially populated chassis may not have CPUs in some of
2435 	 * the APIC clusters they have been allocated.  Only present CPUs have
2436 	 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2437 	 * Since clusters are allocated sequentially, count zeros only if
2438 	 * they are bounded by ones.
2439 	 */
2440 	clusters = 0;
2441 	zeros = 0;
2442 	for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2443 		if (test_bit(i, clustermap)) {
2444 			clusters += 1 + zeros;
2445 			zeros = 0;
2446 		} else
2447 			++zeros;
2448 	}
2449 
2450 	return clusters;
2451 }
2452 
2453 static int multi_checked;
2454 static int multi;
2455 
2456 static int set_multi(const struct dmi_system_id *d)
2457 {
2458 	if (multi)
2459 		return 0;
2460 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2461 	multi = 1;
2462 	return 0;
2463 }
2464 
2465 static const struct dmi_system_id multi_dmi_table[] = {
2466 	{
2467 		.callback = set_multi,
2468 		.ident = "IBM System Summit2",
2469 		.matches = {
2470 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2471 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2472 		},
2473 	},
2474 	{}
2475 };
2476 
2477 static void dmi_check_multi(void)
2478 {
2479 	if (multi_checked)
2480 		return;
2481 
2482 	dmi_check_system(multi_dmi_table);
2483 	multi_checked = 1;
2484 }
2485 
2486 /*
2487  * apic_is_clustered_box() -- Check if we can expect good TSC
2488  *
2489  * Thus far, the major user of this is IBM's Summit2 series:
2490  * Clustered boxes may have unsynced TSC problems if they are
2491  * multi-chassis.
2492  * Use DMI to check them
2493  */
2494 int apic_is_clustered_box(void)
2495 {
2496 	dmi_check_multi();
2497 	if (multi)
2498 		return 1;
2499 
2500 	if (!is_vsmp_box())
2501 		return 0;
2502 
2503 	/*
2504 	 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2505 	 * not guaranteed to be synced between boards
2506 	 */
2507 	if (apic_cluster_num() > 1)
2508 		return 1;
2509 
2510 	return 0;
2511 }
2512 #endif
2513 
2514 /*
2515  * APIC command line parameters
2516  */
2517 static int __init setup_disableapic(char *arg)
2518 {
2519 	disable_apic = 1;
2520 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2521 	return 0;
2522 }
2523 early_param("disableapic", setup_disableapic);
2524 
2525 /* same as disableapic, for compatibility */
2526 static int __init setup_nolapic(char *arg)
2527 {
2528 	return setup_disableapic(arg);
2529 }
2530 early_param("nolapic", setup_nolapic);
2531 
2532 static int __init parse_lapic_timer_c2_ok(char *arg)
2533 {
2534 	local_apic_timer_c2_ok = 1;
2535 	return 0;
2536 }
2537 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2538 
2539 static int __init parse_disable_apic_timer(char *arg)
2540 {
2541 	disable_apic_timer = 1;
2542 	return 0;
2543 }
2544 early_param("noapictimer", parse_disable_apic_timer);
2545 
2546 static int __init parse_nolapic_timer(char *arg)
2547 {
2548 	disable_apic_timer = 1;
2549 	return 0;
2550 }
2551 early_param("nolapic_timer", parse_nolapic_timer);
2552 
2553 static int __init apic_set_verbosity(char *arg)
2554 {
2555 	if (!arg)  {
2556 #ifdef CONFIG_X86_64
2557 		skip_ioapic_setup = 0;
2558 		return 0;
2559 #endif
2560 		return -EINVAL;
2561 	}
2562 
2563 	if (strcmp("debug", arg) == 0)
2564 		apic_verbosity = APIC_DEBUG;
2565 	else if (strcmp("verbose", arg) == 0)
2566 		apic_verbosity = APIC_VERBOSE;
2567 	else {
2568 		pr_warning("APIC Verbosity level %s not recognised"
2569 			" use apic=verbose or apic=debug\n", arg);
2570 		return -EINVAL;
2571 	}
2572 
2573 	return 0;
2574 }
2575 early_param("apic", apic_set_verbosity);
2576 
2577 static int __init lapic_insert_resource(void)
2578 {
2579 	if (!apic_phys)
2580 		return -1;
2581 
2582 	/* Put local APIC into the resource map. */
2583 	lapic_resource.start = apic_phys;
2584 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2585 	insert_resource(&iomem_resource, &lapic_resource);
2586 
2587 	return 0;
2588 }
2589 
2590 /*
2591  * need call insert after e820_reserve_resources()
2592  * that is using request_resource
2593  */
2594 late_initcall(lapic_insert_resource);
2595