xref: /linux/arch/x86/kernel/apic/apic.c (revision 87807f77a03d0271211b75f84b2a8b88f4e8e5d4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	Local APIC handling, local APIC timers
4  *
5  *	(c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
6  *
7  *	Fixes
8  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs;
9  *					thanks to Eric Gilmore
10  *					and Rolf G. Tews
11  *					for testing these extensively.
12  *	Maciej W. Rozycki	:	Various updates and fixes.
13  *	Mikael Pettersson	:	Power Management for UP-APIC.
14  *	Pavel Machek and
15  *	Mikael Pettersson	:	PM converted to driver model.
16  */
17 
18 #include <linux/perf_event.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/mc146818rtc.h>
21 #include <linux/acpi_pmtmr.h>
22 #include <linux/clockchips.h>
23 #include <linux/interrupt.h>
24 #include <linux/memblock.h>
25 #include <linux/ftrace.h>
26 #include <linux/ioport.h>
27 #include <linux/export.h>
28 #include <linux/syscore_ops.h>
29 #include <linux/delay.h>
30 #include <linux/timex.h>
31 #include <linux/i8253.h>
32 #include <linux/dmar.h>
33 #include <linux/init.h>
34 #include <linux/cpu.h>
35 #include <linux/dmi.h>
36 #include <linux/smp.h>
37 #include <linux/mm.h>
38 
39 #include <asm/trace/irq_vectors.h>
40 #include <asm/irq_remapping.h>
41 #include <asm/pc-conf-reg.h>
42 #include <asm/perf_event.h>
43 #include <asm/x86_init.h>
44 #include <linux/atomic.h>
45 #include <asm/barrier.h>
46 #include <asm/mpspec.h>
47 #include <asm/i8259.h>
48 #include <asm/proto.h>
49 #include <asm/traps.h>
50 #include <asm/apic.h>
51 #include <asm/acpi.h>
52 #include <asm/io_apic.h>
53 #include <asm/desc.h>
54 #include <asm/hpet.h>
55 #include <asm/mtrr.h>
56 #include <asm/time.h>
57 #include <asm/smp.h>
58 #include <asm/mce.h>
59 #include <asm/tsc.h>
60 #include <asm/hypervisor.h>
61 #include <asm/cpu_device_id.h>
62 #include <asm/intel-family.h>
63 #include <asm/irq_regs.h>
64 #include <asm/cpu.h>
65 
66 unsigned int num_processors;
67 
68 unsigned disabled_cpus;
69 
70 /* Processor that is doing the boot up */
71 unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
72 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
73 
74 u8 boot_cpu_apic_version __ro_after_init;
75 
76 /*
77  * The highest APIC ID seen during enumeration.
78  */
79 static unsigned int max_physical_apicid;
80 
81 /*
82  * Bitmask of physically existing CPUs:
83  */
84 physid_mask_t phys_cpu_present_map;
85 
86 /*
87  * Processor to be disabled specified by kernel parameter
88  * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
89  * avoid undefined behaviour caused by sending INIT from AP to BSP.
90  */
91 static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;
92 
93 /*
94  * This variable controls which CPUs receive external NMIs.  By default,
95  * external NMIs are delivered only to the BSP.
96  */
97 static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
98 
99 /*
100  * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
101  */
102 static bool virt_ext_dest_id __ro_after_init;
103 
104 /*
105  * Map cpu index to physical APIC ID
106  */
107 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
108 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
109 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
110 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
111 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
112 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
113 
114 #ifdef CONFIG_X86_32
115 
116 /*
117  * On x86_32, the mapping between cpu and logical apicid may vary
118  * depending on apic in use.  The following early percpu variable is
119  * used for the mapping.  This is where the behaviors of x86_64 and 32
120  * actually diverge.  Let's keep it ugly for now.
121  */
122 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
123 
124 /* Local APIC was disabled by the BIOS and enabled by the kernel */
125 static int enabled_via_apicbase __ro_after_init;
126 
127 /*
128  * Handle interrupt mode configuration register (IMCR).
129  * This register controls whether the interrupt signals
130  * that reach the BSP come from the master PIC or from the
131  * local APIC. Before entering Symmetric I/O Mode, either
132  * the BIOS or the operating system must switch out of
133  * PIC Mode by changing the IMCR.
134  */
135 static inline void imcr_pic_to_apic(void)
136 {
137 	/* NMI and 8259 INTR go through APIC */
138 	pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
139 }
140 
141 static inline void imcr_apic_to_pic(void)
142 {
143 	/* NMI and 8259 INTR go directly to BSP */
144 	pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
145 }
146 #endif
147 
148 /*
149  * Knob to control our willingness to enable the local APIC.
150  *
151  * +1=force-enable
152  */
153 static int force_enable_local_apic __initdata;
154 
155 /*
156  * APIC command line parameters
157  */
158 static int __init parse_lapic(char *arg)
159 {
160 	if (IS_ENABLED(CONFIG_X86_32) && !arg)
161 		force_enable_local_apic = 1;
162 	else if (arg && !strncmp(arg, "notscdeadline", 13))
163 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
164 	return 0;
165 }
166 early_param("lapic", parse_lapic);
167 
168 #ifdef CONFIG_X86_64
169 static int apic_calibrate_pmtmr __initdata;
170 static __init int setup_apicpmtimer(char *s)
171 {
172 	apic_calibrate_pmtmr = 1;
173 	notsc_setup(NULL);
174 	return 1;
175 }
176 __setup("apicpmtimer", setup_apicpmtimer);
177 #endif
178 
179 unsigned long mp_lapic_addr __ro_after_init;
180 int disable_apic __ro_after_init;
181 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
182 static int disable_apic_timer __initdata;
183 /* Local APIC timer works in C2 */
184 int local_apic_timer_c2_ok __ro_after_init;
185 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
186 
187 /*
188  * Debug level, exported for io_apic.c
189  */
190 int apic_verbosity __ro_after_init;
191 
192 int pic_mode __ro_after_init;
193 
194 /* Have we found an MP table */
195 int smp_found_config __ro_after_init;
196 
197 static struct resource lapic_resource = {
198 	.name = "Local APIC",
199 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
200 };
201 
202 unsigned int lapic_timer_period = 0;
203 
204 static void apic_pm_activate(void);
205 
206 static unsigned long apic_phys __ro_after_init;
207 
208 /*
209  * Get the LAPIC version
210  */
211 static inline int lapic_get_version(void)
212 {
213 	return GET_APIC_VERSION(apic_read(APIC_LVR));
214 }
215 
216 /*
217  * Check, if the APIC is integrated or a separate chip
218  */
219 static inline int lapic_is_integrated(void)
220 {
221 	return APIC_INTEGRATED(lapic_get_version());
222 }
223 
224 /*
225  * Check, whether this is a modern or a first generation APIC
226  */
227 static int modern_apic(void)
228 {
229 	/* AMD systems use old APIC versions, so check the CPU */
230 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
231 	    boot_cpu_data.x86 >= 0xf)
232 		return 1;
233 
234 	/* Hygon systems use modern APIC */
235 	if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
236 		return 1;
237 
238 	return lapic_get_version() >= 0x14;
239 }
240 
241 /*
242  * right after this call apic become NOOP driven
243  * so apic->write/read doesn't do anything
244  */
245 static void __init apic_disable(void)
246 {
247 	pr_info("APIC: switched to apic NOOP\n");
248 	apic = &apic_noop;
249 }
250 
251 void native_apic_wait_icr_idle(void)
252 {
253 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
254 		cpu_relax();
255 }
256 
257 u32 native_safe_apic_wait_icr_idle(void)
258 {
259 	u32 send_status;
260 	int timeout;
261 
262 	timeout = 0;
263 	do {
264 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
265 		if (!send_status)
266 			break;
267 		inc_irq_stat(icr_read_retry_count);
268 		udelay(100);
269 	} while (timeout++ < 1000);
270 
271 	return send_status;
272 }
273 
274 void native_apic_icr_write(u32 low, u32 id)
275 {
276 	unsigned long flags;
277 
278 	local_irq_save(flags);
279 	apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
280 	apic_write(APIC_ICR, low);
281 	local_irq_restore(flags);
282 }
283 
284 u64 native_apic_icr_read(void)
285 {
286 	u32 icr1, icr2;
287 
288 	icr2 = apic_read(APIC_ICR2);
289 	icr1 = apic_read(APIC_ICR);
290 
291 	return icr1 | ((u64)icr2 << 32);
292 }
293 
294 #ifdef CONFIG_X86_32
295 /**
296  * get_physical_broadcast - Get number of physical broadcast IDs
297  */
298 int get_physical_broadcast(void)
299 {
300 	return modern_apic() ? 0xff : 0xf;
301 }
302 #endif
303 
304 /**
305  * lapic_get_maxlvt - get the maximum number of local vector table entries
306  */
307 int lapic_get_maxlvt(void)
308 {
309 	/*
310 	 * - we always have APIC integrated on 64bit mode
311 	 * - 82489DXs do not report # of LVT entries
312 	 */
313 	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
314 }
315 
316 /*
317  * Local APIC timer
318  */
319 
320 /* Clock divisor */
321 #define APIC_DIVISOR 16
322 #define TSC_DIVISOR  8
323 
324 /* i82489DX specific */
325 #define		I82489DX_BASE_DIVIDER		(((0x2) << 18))
326 
327 /*
328  * This function sets up the local APIC timer, with a timeout of
329  * 'clocks' APIC bus clock. During calibration we actually call
330  * this function twice on the boot CPU, once with a bogus timeout
331  * value, second time for real. The other (noncalibrating) CPUs
332  * call this function only once, with the real, calibrated value.
333  *
334  * We do reads before writes even if unnecessary, to get around the
335  * P5 APIC double write bug.
336  */
337 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
338 {
339 	unsigned int lvtt_value, tmp_value;
340 
341 	lvtt_value = LOCAL_TIMER_VECTOR;
342 	if (!oneshot)
343 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
344 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
345 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
346 
347 	/*
348 	 * The i82489DX APIC uses bit 18 and 19 for the base divider.  This
349 	 * overlaps with bit 18 on integrated APICs, but is not documented
350 	 * in the SDM. No problem though. i82489DX equipped systems do not
351 	 * have TSC deadline timer.
352 	 */
353 	if (!lapic_is_integrated())
354 		lvtt_value |= I82489DX_BASE_DIVIDER;
355 
356 	if (!irqen)
357 		lvtt_value |= APIC_LVT_MASKED;
358 
359 	apic_write(APIC_LVTT, lvtt_value);
360 
361 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
362 		/*
363 		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
364 		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
365 		 * According to Intel, MFENCE can do the serialization here.
366 		 */
367 		asm volatile("mfence" : : : "memory");
368 		return;
369 	}
370 
371 	/*
372 	 * Divide PICLK by 16
373 	 */
374 	tmp_value = apic_read(APIC_TDCR);
375 	apic_write(APIC_TDCR,
376 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
377 		APIC_TDR_DIV_16);
378 
379 	if (!oneshot)
380 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
381 }
382 
383 /*
384  * Setup extended LVT, AMD specific
385  *
386  * Software should use the LVT offsets the BIOS provides.  The offsets
387  * are determined by the subsystems using it like those for MCE
388  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
389  * are supported. Beginning with family 10h at least 4 offsets are
390  * available.
391  *
392  * Since the offsets must be consistent for all cores, we keep track
393  * of the LVT offsets in software and reserve the offset for the same
394  * vector also to be used on other cores. An offset is freed by
395  * setting the entry to APIC_EILVT_MASKED.
396  *
397  * If the BIOS is right, there should be no conflicts. Otherwise a
398  * "[Firmware Bug]: ..." error message is generated. However, if
399  * software does not properly determines the offsets, it is not
400  * necessarily a BIOS bug.
401  */
402 
403 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
404 
405 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
406 {
407 	return (old & APIC_EILVT_MASKED)
408 		|| (new == APIC_EILVT_MASKED)
409 		|| ((new & ~APIC_EILVT_MASKED) == old);
410 }
411 
412 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
413 {
414 	unsigned int rsvd, vector;
415 
416 	if (offset >= APIC_EILVT_NR_MAX)
417 		return ~0;
418 
419 	rsvd = atomic_read(&eilvt_offsets[offset]);
420 	do {
421 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
422 		if (vector && !eilvt_entry_is_changeable(vector, new))
423 			/* may not change if vectors are different */
424 			return rsvd;
425 	} while (!atomic_try_cmpxchg(&eilvt_offsets[offset], &rsvd, new));
426 
427 	rsvd = new & ~APIC_EILVT_MASKED;
428 	if (rsvd && rsvd != vector)
429 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
430 			offset, rsvd);
431 
432 	return new;
433 }
434 
435 /*
436  * If mask=1, the LVT entry does not generate interrupts while mask=0
437  * enables the vector. See also the BKDGs. Must be called with
438  * preemption disabled.
439  */
440 
441 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
442 {
443 	unsigned long reg = APIC_EILVTn(offset);
444 	unsigned int new, old, reserved;
445 
446 	new = (mask << 16) | (msg_type << 8) | vector;
447 	old = apic_read(reg);
448 	reserved = reserve_eilvt_offset(offset, new);
449 
450 	if (reserved != new) {
451 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
452 		       "vector 0x%x, but the register is already in use for "
453 		       "vector 0x%x on another cpu\n",
454 		       smp_processor_id(), reg, offset, new, reserved);
455 		return -EINVAL;
456 	}
457 
458 	if (!eilvt_entry_is_changeable(old, new)) {
459 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
460 		       "vector 0x%x, but the register is already in use for "
461 		       "vector 0x%x on this cpu\n",
462 		       smp_processor_id(), reg, offset, new, old);
463 		return -EBUSY;
464 	}
465 
466 	apic_write(reg, new);
467 
468 	return 0;
469 }
470 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
471 
472 /*
473  * Program the next event, relative to now
474  */
475 static int lapic_next_event(unsigned long delta,
476 			    struct clock_event_device *evt)
477 {
478 	apic_write(APIC_TMICT, delta);
479 	return 0;
480 }
481 
482 static int lapic_next_deadline(unsigned long delta,
483 			       struct clock_event_device *evt)
484 {
485 	u64 tsc;
486 
487 	/* This MSR is special and need a special fence: */
488 	weak_wrmsr_fence();
489 
490 	tsc = rdtsc();
491 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
492 	return 0;
493 }
494 
495 static int lapic_timer_shutdown(struct clock_event_device *evt)
496 {
497 	unsigned int v;
498 
499 	/* Lapic used as dummy for broadcast ? */
500 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
501 		return 0;
502 
503 	v = apic_read(APIC_LVTT);
504 	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
505 	apic_write(APIC_LVTT, v);
506 	apic_write(APIC_TMICT, 0);
507 	return 0;
508 }
509 
510 static inline int
511 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
512 {
513 	/* Lapic used as dummy for broadcast ? */
514 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
515 		return 0;
516 
517 	__setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
518 	return 0;
519 }
520 
521 static int lapic_timer_set_periodic(struct clock_event_device *evt)
522 {
523 	return lapic_timer_set_periodic_oneshot(evt, false);
524 }
525 
526 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
527 {
528 	return lapic_timer_set_periodic_oneshot(evt, true);
529 }
530 
531 /*
532  * Local APIC timer broadcast function
533  */
534 static 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  */
545 static struct clock_event_device lapic_clockevent = {
546 	.name				= "lapic",
547 	.features			= CLOCK_EVT_FEAT_PERIODIC |
548 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
549 					  | CLOCK_EVT_FEAT_DUMMY,
550 	.shift				= 32,
551 	.set_state_shutdown		= lapic_timer_shutdown,
552 	.set_state_periodic		= lapic_timer_set_periodic,
553 	.set_state_oneshot		= lapic_timer_set_oneshot,
554 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
555 	.set_next_event			= lapic_next_event,
556 	.broadcast			= lapic_timer_broadcast,
557 	.rating				= 100,
558 	.irq				= -1,
559 };
560 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
561 
562 static const struct x86_cpu_id deadline_match[] __initconst = {
563 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
564 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
565 
566 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X,	0x0b000020),
567 
568 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
569 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
570 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
571 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
572 
573 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
574 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
575 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
576 
577 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL,		0x22),
578 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L,		0x20),
579 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G,		0x17),
580 
581 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL,		0x25),
582 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G,	0x17),
583 
584 	X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L,		0xb2),
585 	X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE,		0xb2),
586 
587 	X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L,		0x52),
588 	X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE,		0x52),
589 
590 	{},
591 };
592 
593 static __init bool apic_validate_deadline_timer(void)
594 {
595 	const struct x86_cpu_id *m;
596 	u32 rev;
597 
598 	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
599 		return false;
600 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
601 		return true;
602 
603 	m = x86_match_cpu(deadline_match);
604 	if (!m)
605 		return true;
606 
607 	rev = (u32)m->driver_data;
608 
609 	if (boot_cpu_data.microcode >= rev)
610 		return true;
611 
612 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
613 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
614 	       "please update microcode to version: 0x%x (or later)\n", rev);
615 	return false;
616 }
617 
618 /*
619  * Setup the local APIC timer for this CPU. Copy the initialized values
620  * of the boot CPU and register the clock event in the framework.
621  */
622 static void setup_APIC_timer(void)
623 {
624 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
625 
626 	if (this_cpu_has(X86_FEATURE_ARAT)) {
627 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
628 		/* Make LAPIC timer preferable over percpu HPET */
629 		lapic_clockevent.rating = 150;
630 	}
631 
632 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
633 	levt->cpumask = cpumask_of(smp_processor_id());
634 
635 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
636 		levt->name = "lapic-deadline";
637 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
638 				    CLOCK_EVT_FEAT_DUMMY);
639 		levt->set_next_event = lapic_next_deadline;
640 		clockevents_config_and_register(levt,
641 						tsc_khz * (1000 / TSC_DIVISOR),
642 						0xF, ~0UL);
643 	} else
644 		clockevents_register_device(levt);
645 }
646 
647 /*
648  * Install the updated TSC frequency from recalibration at the TSC
649  * deadline clockevent devices.
650  */
651 static void __lapic_update_tsc_freq(void *info)
652 {
653 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
654 
655 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
656 		return;
657 
658 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
659 }
660 
661 void lapic_update_tsc_freq(void)
662 {
663 	/*
664 	 * The clockevent device's ->mult and ->shift can both be
665 	 * changed. In order to avoid races, schedule the frequency
666 	 * update code on each CPU.
667 	 */
668 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
669 }
670 
671 /*
672  * In this functions we calibrate APIC bus clocks to the external timer.
673  *
674  * We want to do the calibration only once since we want to have local timer
675  * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
676  * frequency.
677  *
678  * This was previously done by reading the PIT/HPET and waiting for a wrap
679  * around to find out, that a tick has elapsed. I have a box, where the PIT
680  * readout is broken, so it never gets out of the wait loop again. This was
681  * also reported by others.
682  *
683  * Monitoring the jiffies value is inaccurate and the clockevents
684  * infrastructure allows us to do a simple substitution of the interrupt
685  * handler.
686  *
687  * The calibration routine also uses the pm_timer when possible, as the PIT
688  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
689  * back to normal later in the boot process).
690  */
691 
692 #define LAPIC_CAL_LOOPS		(HZ/10)
693 
694 static __initdata int lapic_cal_loops = -1;
695 static __initdata long lapic_cal_t1, lapic_cal_t2;
696 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
697 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
698 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
699 
700 /*
701  * Temporary interrupt handler and polled calibration function.
702  */
703 static void __init lapic_cal_handler(struct clock_event_device *dev)
704 {
705 	unsigned long long tsc = 0;
706 	long tapic = apic_read(APIC_TMCCT);
707 	unsigned long pm = acpi_pm_read_early();
708 
709 	if (boot_cpu_has(X86_FEATURE_TSC))
710 		tsc = rdtsc();
711 
712 	switch (lapic_cal_loops++) {
713 	case 0:
714 		lapic_cal_t1 = tapic;
715 		lapic_cal_tsc1 = tsc;
716 		lapic_cal_pm1 = pm;
717 		lapic_cal_j1 = jiffies;
718 		break;
719 
720 	case LAPIC_CAL_LOOPS:
721 		lapic_cal_t2 = tapic;
722 		lapic_cal_tsc2 = tsc;
723 		if (pm < lapic_cal_pm1)
724 			pm += ACPI_PM_OVRRUN;
725 		lapic_cal_pm2 = pm;
726 		lapic_cal_j2 = jiffies;
727 		break;
728 	}
729 }
730 
731 static int __init
732 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
733 {
734 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
735 	const long pm_thresh = pm_100ms / 100;
736 	unsigned long mult;
737 	u64 res;
738 
739 #ifndef CONFIG_X86_PM_TIMER
740 	return -1;
741 #endif
742 
743 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
744 
745 	/* Check, if the PM timer is available */
746 	if (!deltapm)
747 		return -1;
748 
749 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
750 
751 	if (deltapm > (pm_100ms - pm_thresh) &&
752 	    deltapm < (pm_100ms + pm_thresh)) {
753 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
754 		return 0;
755 	}
756 
757 	res = (((u64)deltapm) *  mult) >> 22;
758 	do_div(res, 1000000);
759 	pr_warn("APIC calibration not consistent "
760 		"with PM-Timer: %ldms instead of 100ms\n", (long)res);
761 
762 	/* Correct the lapic counter value */
763 	res = (((u64)(*delta)) * pm_100ms);
764 	do_div(res, deltapm);
765 	pr_info("APIC delta adjusted to PM-Timer: "
766 		"%lu (%ld)\n", (unsigned long)res, *delta);
767 	*delta = (long)res;
768 
769 	/* Correct the tsc counter value */
770 	if (boot_cpu_has(X86_FEATURE_TSC)) {
771 		res = (((u64)(*deltatsc)) * pm_100ms);
772 		do_div(res, deltapm);
773 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
774 					  "PM-Timer: %lu (%ld)\n",
775 					(unsigned long)res, *deltatsc);
776 		*deltatsc = (long)res;
777 	}
778 
779 	return 0;
780 }
781 
782 static int __init lapic_init_clockevent(void)
783 {
784 	if (!lapic_timer_period)
785 		return -1;
786 
787 	/* Calculate the scaled math multiplication factor */
788 	lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
789 					TICK_NSEC, lapic_clockevent.shift);
790 	lapic_clockevent.max_delta_ns =
791 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
792 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
793 	lapic_clockevent.min_delta_ns =
794 		clockevent_delta2ns(0xF, &lapic_clockevent);
795 	lapic_clockevent.min_delta_ticks = 0xF;
796 
797 	return 0;
798 }
799 
800 bool __init apic_needs_pit(void)
801 {
802 	/*
803 	 * If the frequencies are not known, PIT is required for both TSC
804 	 * and apic timer calibration.
805 	 */
806 	if (!tsc_khz || !cpu_khz)
807 		return true;
808 
809 	/* Is there an APIC at all or is it disabled? */
810 	if (!boot_cpu_has(X86_FEATURE_APIC) || disable_apic)
811 		return true;
812 
813 	/*
814 	 * If interrupt delivery mode is legacy PIC or virtual wire without
815 	 * configuration, the local APIC timer wont be set up. Make sure
816 	 * that the PIT is initialized.
817 	 */
818 	if (apic_intr_mode == APIC_PIC ||
819 	    apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
820 		return true;
821 
822 	/* Virt guests may lack ARAT, but still have DEADLINE */
823 	if (!boot_cpu_has(X86_FEATURE_ARAT))
824 		return true;
825 
826 	/* Deadline timer is based on TSC so no further PIT action required */
827 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
828 		return false;
829 
830 	/* APIC timer disabled? */
831 	if (disable_apic_timer)
832 		return true;
833 	/*
834 	 * The APIC timer frequency is known already, no PIT calibration
835 	 * required. If unknown, let the PIT be initialized.
836 	 */
837 	return lapic_timer_period == 0;
838 }
839 
840 static int __init calibrate_APIC_clock(void)
841 {
842 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
843 	u64 tsc_perj = 0, tsc_start = 0;
844 	unsigned long jif_start;
845 	unsigned long deltaj;
846 	long delta, deltatsc;
847 	int pm_referenced = 0;
848 
849 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
850 		return 0;
851 
852 	/*
853 	 * Check if lapic timer has already been calibrated by platform
854 	 * specific routine, such as tsc calibration code. If so just fill
855 	 * in the clockevent structure and return.
856 	 */
857 	if (!lapic_init_clockevent()) {
858 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
859 			    lapic_timer_period);
860 		/*
861 		 * Direct calibration methods must have an always running
862 		 * local APIC timer, no need for broadcast timer.
863 		 */
864 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
865 		return 0;
866 	}
867 
868 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
869 		    "calibrating APIC timer ...\n");
870 
871 	/*
872 	 * There are platforms w/o global clockevent devices. Instead of
873 	 * making the calibration conditional on that, use a polling based
874 	 * approach everywhere.
875 	 */
876 	local_irq_disable();
877 
878 	/*
879 	 * Setup the APIC counter to maximum. There is no way the lapic
880 	 * can underflow in the 100ms detection time frame
881 	 */
882 	__setup_APIC_LVTT(0xffffffff, 0, 0);
883 
884 	/*
885 	 * Methods to terminate the calibration loop:
886 	 *  1) Global clockevent if available (jiffies)
887 	 *  2) TSC if available and frequency is known
888 	 */
889 	jif_start = READ_ONCE(jiffies);
890 
891 	if (tsc_khz) {
892 		tsc_start = rdtsc();
893 		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
894 	}
895 
896 	/*
897 	 * Enable interrupts so the tick can fire, if a global
898 	 * clockevent device is available
899 	 */
900 	local_irq_enable();
901 
902 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
903 		/* Wait for a tick to elapse */
904 		while (1) {
905 			if (tsc_khz) {
906 				u64 tsc_now = rdtsc();
907 				if ((tsc_now - tsc_start) >= tsc_perj) {
908 					tsc_start += tsc_perj;
909 					break;
910 				}
911 			} else {
912 				unsigned long jif_now = READ_ONCE(jiffies);
913 
914 				if (time_after(jif_now, jif_start)) {
915 					jif_start = jif_now;
916 					break;
917 				}
918 			}
919 			cpu_relax();
920 		}
921 
922 		/* Invoke the calibration routine */
923 		local_irq_disable();
924 		lapic_cal_handler(NULL);
925 		local_irq_enable();
926 	}
927 
928 	local_irq_disable();
929 
930 	/* Build delta t1-t2 as apic timer counts down */
931 	delta = lapic_cal_t1 - lapic_cal_t2;
932 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
933 
934 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
935 
936 	/* we trust the PM based calibration if possible */
937 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
938 					&delta, &deltatsc);
939 
940 	lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
941 	lapic_init_clockevent();
942 
943 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
944 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
945 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
946 		    lapic_timer_period);
947 
948 	if (boot_cpu_has(X86_FEATURE_TSC)) {
949 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
950 			    "%ld.%04ld MHz.\n",
951 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
952 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
953 	}
954 
955 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
956 		    "%u.%04u MHz.\n",
957 		    lapic_timer_period / (1000000 / HZ),
958 		    lapic_timer_period % (1000000 / HZ));
959 
960 	/*
961 	 * Do a sanity check on the APIC calibration result
962 	 */
963 	if (lapic_timer_period < (1000000 / HZ)) {
964 		local_irq_enable();
965 		pr_warn("APIC frequency too slow, disabling apic timer\n");
966 		return -1;
967 	}
968 
969 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
970 
971 	/*
972 	 * PM timer calibration failed or not turned on so lets try APIC
973 	 * timer based calibration, if a global clockevent device is
974 	 * available.
975 	 */
976 	if (!pm_referenced && global_clock_event) {
977 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
978 
979 		/*
980 		 * Setup the apic timer manually
981 		 */
982 		levt->event_handler = lapic_cal_handler;
983 		lapic_timer_set_periodic(levt);
984 		lapic_cal_loops = -1;
985 
986 		/* Let the interrupts run */
987 		local_irq_enable();
988 
989 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
990 			cpu_relax();
991 
992 		/* Stop the lapic timer */
993 		local_irq_disable();
994 		lapic_timer_shutdown(levt);
995 
996 		/* Jiffies delta */
997 		deltaj = lapic_cal_j2 - lapic_cal_j1;
998 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
999 
1000 		/* Check, if the jiffies result is consistent */
1001 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
1002 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
1003 		else
1004 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
1005 	}
1006 	local_irq_enable();
1007 
1008 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
1009 		pr_warn("APIC timer disabled due to verification failure\n");
1010 		return -1;
1011 	}
1012 
1013 	return 0;
1014 }
1015 
1016 /*
1017  * Setup the boot APIC
1018  *
1019  * Calibrate and verify the result.
1020  */
1021 void __init setup_boot_APIC_clock(void)
1022 {
1023 	/*
1024 	 * The local apic timer can be disabled via the kernel
1025 	 * commandline or from the CPU detection code. Register the lapic
1026 	 * timer as a dummy clock event source on SMP systems, so the
1027 	 * broadcast mechanism is used. On UP systems simply ignore it.
1028 	 */
1029 	if (disable_apic_timer) {
1030 		pr_info("Disabling APIC timer\n");
1031 		/* No broadcast on UP ! */
1032 		if (num_possible_cpus() > 1) {
1033 			lapic_clockevent.mult = 1;
1034 			setup_APIC_timer();
1035 		}
1036 		return;
1037 	}
1038 
1039 	if (calibrate_APIC_clock()) {
1040 		/* No broadcast on UP ! */
1041 		if (num_possible_cpus() > 1)
1042 			setup_APIC_timer();
1043 		return;
1044 	}
1045 
1046 	/*
1047 	 * If nmi_watchdog is set to IO_APIC, we need the
1048 	 * PIT/HPET going.  Otherwise register lapic as a dummy
1049 	 * device.
1050 	 */
1051 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1052 
1053 	/* Setup the lapic or request the broadcast */
1054 	setup_APIC_timer();
1055 	amd_e400_c1e_apic_setup();
1056 }
1057 
1058 void setup_secondary_APIC_clock(void)
1059 {
1060 	setup_APIC_timer();
1061 	amd_e400_c1e_apic_setup();
1062 }
1063 
1064 /*
1065  * The guts of the apic timer interrupt
1066  */
1067 static void local_apic_timer_interrupt(void)
1068 {
1069 	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1070 
1071 	/*
1072 	 * Normally we should not be here till LAPIC has been initialized but
1073 	 * in some cases like kdump, its possible that there is a pending LAPIC
1074 	 * timer interrupt from previous kernel's context and is delivered in
1075 	 * new kernel the moment interrupts are enabled.
1076 	 *
1077 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1078 	 * its possible that when we get here evt->event_handler is NULL.
1079 	 * Check for event_handler being NULL and discard the interrupt as
1080 	 * spurious.
1081 	 */
1082 	if (!evt->event_handler) {
1083 		pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1084 			smp_processor_id());
1085 		/* Switch it off */
1086 		lapic_timer_shutdown(evt);
1087 		return;
1088 	}
1089 
1090 	/*
1091 	 * the NMI deadlock-detector uses this.
1092 	 */
1093 	inc_irq_stat(apic_timer_irqs);
1094 
1095 	evt->event_handler(evt);
1096 }
1097 
1098 /*
1099  * Local APIC timer interrupt. This is the most natural way for doing
1100  * local interrupts, but local timer interrupts can be emulated by
1101  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1102  *
1103  * [ if a single-CPU system runs an SMP kernel then we call the local
1104  *   interrupt as well. Thus we cannot inline the local irq ... ]
1105  */
1106 DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1107 {
1108 	struct pt_regs *old_regs = set_irq_regs(regs);
1109 
1110 	ack_APIC_irq();
1111 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1112 	local_apic_timer_interrupt();
1113 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1114 
1115 	set_irq_regs(old_regs);
1116 }
1117 
1118 /*
1119  * Local APIC start and shutdown
1120  */
1121 
1122 /**
1123  * clear_local_APIC - shutdown the local APIC
1124  *
1125  * This is called, when a CPU is disabled and before rebooting, so the state of
1126  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1127  * leftovers during boot.
1128  */
1129 void clear_local_APIC(void)
1130 {
1131 	int maxlvt;
1132 	u32 v;
1133 
1134 	/* APIC hasn't been mapped yet */
1135 	if (!x2apic_mode && !apic_phys)
1136 		return;
1137 
1138 	maxlvt = lapic_get_maxlvt();
1139 	/*
1140 	 * Masking an LVT entry can trigger a local APIC error
1141 	 * if the vector is zero. Mask LVTERR first to prevent this.
1142 	 */
1143 	if (maxlvt >= 3) {
1144 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1145 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1146 	}
1147 	/*
1148 	 * Careful: we have to set masks only first to deassert
1149 	 * any level-triggered sources.
1150 	 */
1151 	v = apic_read(APIC_LVTT);
1152 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1153 	v = apic_read(APIC_LVT0);
1154 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1155 	v = apic_read(APIC_LVT1);
1156 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1157 	if (maxlvt >= 4) {
1158 		v = apic_read(APIC_LVTPC);
1159 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1160 	}
1161 
1162 	/* lets not touch this if we didn't frob it */
1163 #ifdef CONFIG_X86_THERMAL_VECTOR
1164 	if (maxlvt >= 5) {
1165 		v = apic_read(APIC_LVTTHMR);
1166 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1167 	}
1168 #endif
1169 #ifdef CONFIG_X86_MCE_INTEL
1170 	if (maxlvt >= 6) {
1171 		v = apic_read(APIC_LVTCMCI);
1172 		if (!(v & APIC_LVT_MASKED))
1173 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1174 	}
1175 #endif
1176 
1177 	/*
1178 	 * Clean APIC state for other OSs:
1179 	 */
1180 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1181 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1182 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1183 	if (maxlvt >= 3)
1184 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1185 	if (maxlvt >= 4)
1186 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1187 
1188 	/* Integrated APIC (!82489DX) ? */
1189 	if (lapic_is_integrated()) {
1190 		if (maxlvt > 3)
1191 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1192 			apic_write(APIC_ESR, 0);
1193 		apic_read(APIC_ESR);
1194 	}
1195 }
1196 
1197 /**
1198  * apic_soft_disable - Clears and software disables the local APIC on hotplug
1199  *
1200  * Contrary to disable_local_APIC() this does not touch the enable bit in
1201  * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1202  * bus would require a hardware reset as the APIC would lose track of bus
1203  * arbitration. On systems with FSB delivery APICBASE could be disabled,
1204  * but it has to be guaranteed that no interrupt is sent to the APIC while
1205  * in that state and it's not clear from the SDM whether it still responds
1206  * to INIT/SIPI messages. Stay on the safe side and use software disable.
1207  */
1208 void apic_soft_disable(void)
1209 {
1210 	u32 value;
1211 
1212 	clear_local_APIC();
1213 
1214 	/* Soft disable APIC (implies clearing of registers for 82489DX!). */
1215 	value = apic_read(APIC_SPIV);
1216 	value &= ~APIC_SPIV_APIC_ENABLED;
1217 	apic_write(APIC_SPIV, value);
1218 }
1219 
1220 /**
1221  * disable_local_APIC - clear and disable the local APIC
1222  */
1223 void disable_local_APIC(void)
1224 {
1225 	/* APIC hasn't been mapped yet */
1226 	if (!x2apic_mode && !apic_phys)
1227 		return;
1228 
1229 	apic_soft_disable();
1230 
1231 #ifdef CONFIG_X86_32
1232 	/*
1233 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1234 	 * restore the disabled state.
1235 	 */
1236 	if (enabled_via_apicbase) {
1237 		unsigned int l, h;
1238 
1239 		rdmsr(MSR_IA32_APICBASE, l, h);
1240 		l &= ~MSR_IA32_APICBASE_ENABLE;
1241 		wrmsr(MSR_IA32_APICBASE, l, h);
1242 	}
1243 #endif
1244 }
1245 
1246 /*
1247  * If Linux enabled the LAPIC against the BIOS default disable it down before
1248  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1249  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1250  * for the case where Linux didn't enable the LAPIC.
1251  */
1252 void lapic_shutdown(void)
1253 {
1254 	unsigned long flags;
1255 
1256 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1257 		return;
1258 
1259 	local_irq_save(flags);
1260 
1261 #ifdef CONFIG_X86_32
1262 	if (!enabled_via_apicbase)
1263 		clear_local_APIC();
1264 	else
1265 #endif
1266 		disable_local_APIC();
1267 
1268 
1269 	local_irq_restore(flags);
1270 }
1271 
1272 /**
1273  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1274  */
1275 void __init sync_Arb_IDs(void)
1276 {
1277 	/*
1278 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1279 	 * needed on AMD.
1280 	 */
1281 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1282 		return;
1283 
1284 	/*
1285 	 * Wait for idle.
1286 	 */
1287 	apic_wait_icr_idle();
1288 
1289 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1290 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1291 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1292 }
1293 
1294 enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1295 
1296 static int __init __apic_intr_mode_select(void)
1297 {
1298 	/* Check kernel option */
1299 	if (disable_apic) {
1300 		pr_info("APIC disabled via kernel command line\n");
1301 		return APIC_PIC;
1302 	}
1303 
1304 	/* Check BIOS */
1305 #ifdef CONFIG_X86_64
1306 	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1307 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1308 		disable_apic = 1;
1309 		pr_info("APIC disabled by BIOS\n");
1310 		return APIC_PIC;
1311 	}
1312 #else
1313 	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1314 
1315 	/* Neither 82489DX nor integrated APIC ? */
1316 	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1317 		disable_apic = 1;
1318 		return APIC_PIC;
1319 	}
1320 
1321 	/* If the BIOS pretends there is an integrated APIC ? */
1322 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1323 		APIC_INTEGRATED(boot_cpu_apic_version)) {
1324 		disable_apic = 1;
1325 		pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1326 				       boot_cpu_physical_apicid);
1327 		return APIC_PIC;
1328 	}
1329 #endif
1330 
1331 	/* Check MP table or ACPI MADT configuration */
1332 	if (!smp_found_config) {
1333 		disable_ioapic_support();
1334 		if (!acpi_lapic) {
1335 			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1336 			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1337 		}
1338 		return APIC_VIRTUAL_WIRE;
1339 	}
1340 
1341 #ifdef CONFIG_SMP
1342 	/* If SMP should be disabled, then really disable it! */
1343 	if (!setup_max_cpus) {
1344 		pr_info("APIC: SMP mode deactivated\n");
1345 		return APIC_SYMMETRIC_IO_NO_ROUTING;
1346 	}
1347 
1348 	if (read_apic_id() != boot_cpu_physical_apicid) {
1349 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1350 		     read_apic_id(), boot_cpu_physical_apicid);
1351 		/* Or can we switch back to PIC here? */
1352 	}
1353 #endif
1354 
1355 	return APIC_SYMMETRIC_IO;
1356 }
1357 
1358 /* Select the interrupt delivery mode for the BSP */
1359 void __init apic_intr_mode_select(void)
1360 {
1361 	apic_intr_mode = __apic_intr_mode_select();
1362 }
1363 
1364 /*
1365  * An initial setup of the virtual wire mode.
1366  */
1367 void __init init_bsp_APIC(void)
1368 {
1369 	unsigned int value;
1370 
1371 	/*
1372 	 * Don't do the setup now if we have a SMP BIOS as the
1373 	 * through-I/O-APIC virtual wire mode might be active.
1374 	 */
1375 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1376 		return;
1377 
1378 	/*
1379 	 * Do not trust the local APIC being empty at bootup.
1380 	 */
1381 	clear_local_APIC();
1382 
1383 	/*
1384 	 * Enable APIC.
1385 	 */
1386 	value = apic_read(APIC_SPIV);
1387 	value &= ~APIC_VECTOR_MASK;
1388 	value |= APIC_SPIV_APIC_ENABLED;
1389 
1390 #ifdef CONFIG_X86_32
1391 	/* This bit is reserved on P4/Xeon and should be cleared */
1392 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1393 	    (boot_cpu_data.x86 == 15))
1394 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1395 	else
1396 #endif
1397 		value |= APIC_SPIV_FOCUS_DISABLED;
1398 	value |= SPURIOUS_APIC_VECTOR;
1399 	apic_write(APIC_SPIV, value);
1400 
1401 	/*
1402 	 * Set up the virtual wire mode.
1403 	 */
1404 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1405 	value = APIC_DM_NMI;
1406 	if (!lapic_is_integrated())		/* 82489DX */
1407 		value |= APIC_LVT_LEVEL_TRIGGER;
1408 	if (apic_extnmi == APIC_EXTNMI_NONE)
1409 		value |= APIC_LVT_MASKED;
1410 	apic_write(APIC_LVT1, value);
1411 }
1412 
1413 static void __init apic_bsp_setup(bool upmode);
1414 
1415 /* Init the interrupt delivery mode for the BSP */
1416 void __init apic_intr_mode_init(void)
1417 {
1418 	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1419 
1420 	switch (apic_intr_mode) {
1421 	case APIC_PIC:
1422 		pr_info("APIC: Keep in PIC mode(8259)\n");
1423 		return;
1424 	case APIC_VIRTUAL_WIRE:
1425 		pr_info("APIC: Switch to virtual wire mode setup\n");
1426 		break;
1427 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1428 		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1429 		upmode = true;
1430 		break;
1431 	case APIC_SYMMETRIC_IO:
1432 		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1433 		break;
1434 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1435 		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1436 		break;
1437 	}
1438 
1439 	default_setup_apic_routing();
1440 
1441 	if (x86_platform.apic_post_init)
1442 		x86_platform.apic_post_init();
1443 
1444 	apic_bsp_setup(upmode);
1445 }
1446 
1447 static void lapic_setup_esr(void)
1448 {
1449 	unsigned int oldvalue, value, maxlvt;
1450 
1451 	if (!lapic_is_integrated()) {
1452 		pr_info("No ESR for 82489DX.\n");
1453 		return;
1454 	}
1455 
1456 	if (apic->disable_esr) {
1457 		/*
1458 		 * Something untraceable is creating bad interrupts on
1459 		 * secondary quads ... for the moment, just leave the
1460 		 * ESR disabled - we can't do anything useful with the
1461 		 * errors anyway - mbligh
1462 		 */
1463 		pr_info("Leaving ESR disabled.\n");
1464 		return;
1465 	}
1466 
1467 	maxlvt = lapic_get_maxlvt();
1468 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1469 		apic_write(APIC_ESR, 0);
1470 	oldvalue = apic_read(APIC_ESR);
1471 
1472 	/* enables sending errors */
1473 	value = ERROR_APIC_VECTOR;
1474 	apic_write(APIC_LVTERR, value);
1475 
1476 	/*
1477 	 * spec says clear errors after enabling vector.
1478 	 */
1479 	if (maxlvt > 3)
1480 		apic_write(APIC_ESR, 0);
1481 	value = apic_read(APIC_ESR);
1482 	if (value != oldvalue)
1483 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1484 			"vector: 0x%08x  after: 0x%08x\n",
1485 			oldvalue, value);
1486 }
1487 
1488 #define APIC_IR_REGS		APIC_ISR_NR
1489 #define APIC_IR_BITS		(APIC_IR_REGS * 32)
1490 #define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1491 
1492 union apic_ir {
1493 	unsigned long	map[APIC_IR_MAPSIZE];
1494 	u32		regs[APIC_IR_REGS];
1495 };
1496 
1497 static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1498 {
1499 	int i, bit;
1500 
1501 	/* Read the IRRs */
1502 	for (i = 0; i < APIC_IR_REGS; i++)
1503 		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1504 
1505 	/* Read the ISRs */
1506 	for (i = 0; i < APIC_IR_REGS; i++)
1507 		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1508 
1509 	/*
1510 	 * If the ISR map is not empty. ACK the APIC and run another round
1511 	 * to verify whether a pending IRR has been unblocked and turned
1512 	 * into a ISR.
1513 	 */
1514 	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1515 		/*
1516 		 * There can be multiple ISR bits set when a high priority
1517 		 * interrupt preempted a lower priority one. Issue an ACK
1518 		 * per set bit.
1519 		 */
1520 		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1521 			ack_APIC_irq();
1522 		return true;
1523 	}
1524 
1525 	return !bitmap_empty(irr->map, APIC_IR_BITS);
1526 }
1527 
1528 /*
1529  * After a crash, we no longer service the interrupts and a pending
1530  * interrupt from previous kernel might still have ISR bit set.
1531  *
1532  * Most probably by now the CPU has serviced that pending interrupt and it
1533  * might not have done the ack_APIC_irq() because it thought, interrupt
1534  * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1535  * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1536  * a vector might get locked. It was noticed for timer irq (vector
1537  * 0x31). Issue an extra EOI to clear ISR.
1538  *
1539  * If there are pending IRR bits they turn into ISR bits after a higher
1540  * priority ISR bit has been acked.
1541  */
1542 static void apic_pending_intr_clear(void)
1543 {
1544 	union apic_ir irr, isr;
1545 	unsigned int i;
1546 
1547 	/* 512 loops are way oversized and give the APIC a chance to obey. */
1548 	for (i = 0; i < 512; i++) {
1549 		if (!apic_check_and_ack(&irr, &isr))
1550 			return;
1551 	}
1552 	/* Dump the IRR/ISR content if that failed */
1553 	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1554 }
1555 
1556 /**
1557  * setup_local_APIC - setup the local APIC
1558  *
1559  * Used to setup local APIC while initializing BSP or bringing up APs.
1560  * Always called with preemption disabled.
1561  */
1562 static void setup_local_APIC(void)
1563 {
1564 	int cpu = smp_processor_id();
1565 	unsigned int value;
1566 
1567 	if (disable_apic) {
1568 		disable_ioapic_support();
1569 		return;
1570 	}
1571 
1572 	/*
1573 	 * If this comes from kexec/kcrash the APIC might be enabled in
1574 	 * SPIV. Soft disable it before doing further initialization.
1575 	 */
1576 	value = apic_read(APIC_SPIV);
1577 	value &= ~APIC_SPIV_APIC_ENABLED;
1578 	apic_write(APIC_SPIV, value);
1579 
1580 #ifdef CONFIG_X86_32
1581 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1582 	if (lapic_is_integrated() && apic->disable_esr) {
1583 		apic_write(APIC_ESR, 0);
1584 		apic_write(APIC_ESR, 0);
1585 		apic_write(APIC_ESR, 0);
1586 		apic_write(APIC_ESR, 0);
1587 	}
1588 #endif
1589 	/*
1590 	 * Double-check whether this APIC is really registered.
1591 	 * This is meaningless in clustered apic mode, so we skip it.
1592 	 */
1593 	BUG_ON(!apic->apic_id_registered());
1594 
1595 	/*
1596 	 * Intel recommends to set DFR, LDR and TPR before enabling
1597 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1598 	 * document number 292116).  So here it goes...
1599 	 */
1600 	apic->init_apic_ldr();
1601 
1602 #ifdef CONFIG_X86_32
1603 	if (apic->dest_mode_logical) {
1604 		int logical_apicid, ldr_apicid;
1605 
1606 		/*
1607 		 * APIC LDR is initialized.  If logical_apicid mapping was
1608 		 * initialized during get_smp_config(), make sure it matches
1609 		 * the actual value.
1610 		 */
1611 		logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1612 		ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1613 		if (logical_apicid != BAD_APICID)
1614 			WARN_ON(logical_apicid != ldr_apicid);
1615 		/* Always use the value from LDR. */
1616 		early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1617 	}
1618 #endif
1619 
1620 	/*
1621 	 * Set Task Priority to 'accept all except vectors 0-31'.  An APIC
1622 	 * vector in the 16-31 range could be delivered if TPR == 0, but we
1623 	 * would think it's an exception and terrible things will happen.  We
1624 	 * never change this later on.
1625 	 */
1626 	value = apic_read(APIC_TASKPRI);
1627 	value &= ~APIC_TPRI_MASK;
1628 	value |= 0x10;
1629 	apic_write(APIC_TASKPRI, value);
1630 
1631 	/* Clear eventually stale ISR/IRR bits */
1632 	apic_pending_intr_clear();
1633 
1634 	/*
1635 	 * Now that we are all set up, enable the APIC
1636 	 */
1637 	value = apic_read(APIC_SPIV);
1638 	value &= ~APIC_VECTOR_MASK;
1639 	/*
1640 	 * Enable APIC
1641 	 */
1642 	value |= APIC_SPIV_APIC_ENABLED;
1643 
1644 #ifdef CONFIG_X86_32
1645 	/*
1646 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1647 	 * certain networking cards. If high frequency interrupts are
1648 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1649 	 * entry is masked/unmasked at a high rate as well then sooner or
1650 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1651 	 * from the device. If focus CPU is disabled then the hang goes
1652 	 * away, oh well :-(
1653 	 *
1654 	 * [ This bug can be reproduced easily with a level-triggered
1655 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1656 	 *   BX chipset. ]
1657 	 */
1658 	/*
1659 	 * Actually disabling the focus CPU check just makes the hang less
1660 	 * frequent as it makes the interrupt distribution model be more
1661 	 * like LRU than MRU (the short-term load is more even across CPUs).
1662 	 */
1663 
1664 	/*
1665 	 * - enable focus processor (bit==0)
1666 	 * - 64bit mode always use processor focus
1667 	 *   so no need to set it
1668 	 */
1669 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1670 #endif
1671 
1672 	/*
1673 	 * Set spurious IRQ vector
1674 	 */
1675 	value |= SPURIOUS_APIC_VECTOR;
1676 	apic_write(APIC_SPIV, value);
1677 
1678 	perf_events_lapic_init();
1679 
1680 	/*
1681 	 * Set up LVT0, LVT1:
1682 	 *
1683 	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1684 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1685 	 * we delegate interrupts to the 8259A.
1686 	 */
1687 	/*
1688 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1689 	 */
1690 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1691 	if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1692 		value = APIC_DM_EXTINT;
1693 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1694 	} else {
1695 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1696 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1697 	}
1698 	apic_write(APIC_LVT0, value);
1699 
1700 	/*
1701 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1702 	 * modified by apic_extnmi= boot option.
1703 	 */
1704 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1705 	    apic_extnmi == APIC_EXTNMI_ALL)
1706 		value = APIC_DM_NMI;
1707 	else
1708 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1709 
1710 	/* Is 82489DX ? */
1711 	if (!lapic_is_integrated())
1712 		value |= APIC_LVT_LEVEL_TRIGGER;
1713 	apic_write(APIC_LVT1, value);
1714 
1715 #ifdef CONFIG_X86_MCE_INTEL
1716 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1717 	if (!cpu)
1718 		cmci_recheck();
1719 #endif
1720 }
1721 
1722 static void end_local_APIC_setup(void)
1723 {
1724 	lapic_setup_esr();
1725 
1726 #ifdef CONFIG_X86_32
1727 	{
1728 		unsigned int value;
1729 		/* Disable the local apic timer */
1730 		value = apic_read(APIC_LVTT);
1731 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1732 		apic_write(APIC_LVTT, value);
1733 	}
1734 #endif
1735 
1736 	apic_pm_activate();
1737 }
1738 
1739 /*
1740  * APIC setup function for application processors. Called from smpboot.c
1741  */
1742 void apic_ap_setup(void)
1743 {
1744 	setup_local_APIC();
1745 	end_local_APIC_setup();
1746 }
1747 
1748 #ifdef CONFIG_X86_X2APIC
1749 int x2apic_mode;
1750 EXPORT_SYMBOL_GPL(x2apic_mode);
1751 
1752 enum {
1753 	X2APIC_OFF,
1754 	X2APIC_DISABLED,
1755 	/* All states below here have X2APIC enabled */
1756 	X2APIC_ON,
1757 	X2APIC_ON_LOCKED
1758 };
1759 static int x2apic_state;
1760 
1761 static bool x2apic_hw_locked(void)
1762 {
1763 	u64 ia32_cap;
1764 	u64 msr;
1765 
1766 	ia32_cap = x86_read_arch_cap_msr();
1767 	if (ia32_cap & ARCH_CAP_XAPIC_DISABLE) {
1768 		rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1769 		return (msr & LEGACY_XAPIC_DISABLED);
1770 	}
1771 	return false;
1772 }
1773 
1774 static void __x2apic_disable(void)
1775 {
1776 	u64 msr;
1777 
1778 	if (!boot_cpu_has(X86_FEATURE_APIC))
1779 		return;
1780 
1781 	rdmsrl(MSR_IA32_APICBASE, msr);
1782 	if (!(msr & X2APIC_ENABLE))
1783 		return;
1784 	/* Disable xapic and x2apic first and then reenable xapic mode */
1785 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1786 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1787 	printk_once(KERN_INFO "x2apic disabled\n");
1788 }
1789 
1790 static void __x2apic_enable(void)
1791 {
1792 	u64 msr;
1793 
1794 	rdmsrl(MSR_IA32_APICBASE, msr);
1795 	if (msr & X2APIC_ENABLE)
1796 		return;
1797 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1798 	printk_once(KERN_INFO "x2apic enabled\n");
1799 }
1800 
1801 static int __init setup_nox2apic(char *str)
1802 {
1803 	if (x2apic_enabled()) {
1804 		int apicid = native_apic_msr_read(APIC_ID);
1805 
1806 		if (apicid >= 255) {
1807 			pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1808 				apicid);
1809 			return 0;
1810 		}
1811 		if (x2apic_hw_locked()) {
1812 			pr_warn("APIC locked in x2apic mode, can't disable\n");
1813 			return 0;
1814 		}
1815 		pr_warn("x2apic already enabled.\n");
1816 		__x2apic_disable();
1817 	}
1818 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1819 	x2apic_state = X2APIC_DISABLED;
1820 	x2apic_mode = 0;
1821 	return 0;
1822 }
1823 early_param("nox2apic", setup_nox2apic);
1824 
1825 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
1826 void x2apic_setup(void)
1827 {
1828 	/*
1829 	 * Try to make the AP's APIC state match that of the BSP,  but if the
1830 	 * BSP is unlocked and the AP is locked then there is a state mismatch.
1831 	 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1832 	 * trying to be turned off.
1833 	 */
1834 	if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1835 		pr_warn("x2apic lock mismatch between BSP and AP.\n");
1836 	/*
1837 	 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1838 	 * from BIOS.
1839 	 */
1840 	if (x2apic_state < X2APIC_ON) {
1841 		__x2apic_disable();
1842 		return;
1843 	}
1844 	__x2apic_enable();
1845 }
1846 
1847 static __init void x2apic_disable(void)
1848 {
1849 	u32 x2apic_id, state = x2apic_state;
1850 
1851 	x2apic_mode = 0;
1852 	x2apic_state = X2APIC_DISABLED;
1853 
1854 	if (state != X2APIC_ON)
1855 		return;
1856 
1857 	x2apic_id = read_apic_id();
1858 	if (x2apic_id >= 255)
1859 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1860 
1861 	if (x2apic_hw_locked()) {
1862 		pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1863 		return;
1864 	}
1865 
1866 	__x2apic_disable();
1867 	register_lapic_address(mp_lapic_addr);
1868 }
1869 
1870 static __init void x2apic_enable(void)
1871 {
1872 	if (x2apic_state != X2APIC_OFF)
1873 		return;
1874 
1875 	x2apic_mode = 1;
1876 	x2apic_state = X2APIC_ON;
1877 	__x2apic_enable();
1878 }
1879 
1880 static __init void try_to_enable_x2apic(int remap_mode)
1881 {
1882 	if (x2apic_state == X2APIC_DISABLED)
1883 		return;
1884 
1885 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1886 		u32 apic_limit = 255;
1887 
1888 		/*
1889 		 * Using X2APIC without IR is not architecturally supported
1890 		 * on bare metal but may be supported in guests.
1891 		 */
1892 		if (!x86_init.hyper.x2apic_available()) {
1893 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1894 			x2apic_disable();
1895 			return;
1896 		}
1897 
1898 		/*
1899 		 * If the hypervisor supports extended destination ID in
1900 		 * MSI, that increases the maximum APIC ID that can be
1901 		 * used for non-remapped IRQ domains.
1902 		 */
1903 		if (x86_init.hyper.msi_ext_dest_id()) {
1904 			virt_ext_dest_id = 1;
1905 			apic_limit = 32767;
1906 		}
1907 
1908 		/*
1909 		 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1910 		 * in physical mode, and CPUs with an APIC ID that cannot
1911 		 * be addressed must not be brought online.
1912 		 */
1913 		x2apic_set_max_apicid(apic_limit);
1914 		x2apic_phys = 1;
1915 	}
1916 	x2apic_enable();
1917 }
1918 
1919 void __init check_x2apic(void)
1920 {
1921 	if (x2apic_enabled()) {
1922 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1923 		x2apic_mode = 1;
1924 		if (x2apic_hw_locked())
1925 			x2apic_state = X2APIC_ON_LOCKED;
1926 		else
1927 			x2apic_state = X2APIC_ON;
1928 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1929 		x2apic_state = X2APIC_DISABLED;
1930 	}
1931 }
1932 #else /* CONFIG_X86_X2APIC */
1933 void __init check_x2apic(void)
1934 {
1935 	if (!apic_is_x2apic_enabled())
1936 		return;
1937 	/*
1938 	 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1939 	 */
1940 	pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1941 	pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1942 
1943 	disable_apic = 1;
1944 	setup_clear_cpu_cap(X86_FEATURE_APIC);
1945 }
1946 
1947 static inline void try_to_enable_x2apic(int remap_mode) { }
1948 static inline void __x2apic_enable(void) { }
1949 #endif /* !CONFIG_X86_X2APIC */
1950 
1951 void __init enable_IR_x2apic(void)
1952 {
1953 	unsigned long flags;
1954 	int ret, ir_stat;
1955 
1956 	if (skip_ioapic_setup) {
1957 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1958 		return;
1959 	}
1960 
1961 	ir_stat = irq_remapping_prepare();
1962 	if (ir_stat < 0 && !x2apic_supported())
1963 		return;
1964 
1965 	ret = save_ioapic_entries();
1966 	if (ret) {
1967 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1968 		return;
1969 	}
1970 
1971 	local_irq_save(flags);
1972 	legacy_pic->mask_all();
1973 	mask_ioapic_entries();
1974 
1975 	/* If irq_remapping_prepare() succeeded, try to enable it */
1976 	if (ir_stat >= 0)
1977 		ir_stat = irq_remapping_enable();
1978 	/* ir_stat contains the remap mode or an error code */
1979 	try_to_enable_x2apic(ir_stat);
1980 
1981 	if (ir_stat < 0)
1982 		restore_ioapic_entries();
1983 	legacy_pic->restore_mask();
1984 	local_irq_restore(flags);
1985 }
1986 
1987 #ifdef CONFIG_X86_64
1988 /*
1989  * Detect and enable local APICs on non-SMP boards.
1990  * Original code written by Keir Fraser.
1991  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1992  * not correctly set up (usually the APIC timer won't work etc.)
1993  */
1994 static int __init detect_init_APIC(void)
1995 {
1996 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1997 		pr_info("No local APIC present\n");
1998 		return -1;
1999 	}
2000 
2001 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
2002 	return 0;
2003 }
2004 #else
2005 
2006 static int __init apic_verify(void)
2007 {
2008 	u32 features, h, l;
2009 
2010 	/*
2011 	 * The APIC feature bit should now be enabled
2012 	 * in `cpuid'
2013 	 */
2014 	features = cpuid_edx(1);
2015 	if (!(features & (1 << X86_FEATURE_APIC))) {
2016 		pr_warn("Could not enable APIC!\n");
2017 		return -1;
2018 	}
2019 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
2020 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
2021 
2022 	/* The BIOS may have set up the APIC at some other address */
2023 	if (boot_cpu_data.x86 >= 6) {
2024 		rdmsr(MSR_IA32_APICBASE, l, h);
2025 		if (l & MSR_IA32_APICBASE_ENABLE)
2026 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
2027 	}
2028 
2029 	pr_info("Found and enabled local APIC!\n");
2030 	return 0;
2031 }
2032 
2033 int __init apic_force_enable(unsigned long addr)
2034 {
2035 	u32 h, l;
2036 
2037 	if (disable_apic)
2038 		return -1;
2039 
2040 	/*
2041 	 * Some BIOSes disable the local APIC in the APIC_BASE
2042 	 * MSR. This can only be done in software for Intel P6 or later
2043 	 * and AMD K7 (Model > 1) or later.
2044 	 */
2045 	if (boot_cpu_data.x86 >= 6) {
2046 		rdmsr(MSR_IA32_APICBASE, l, h);
2047 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2048 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2049 			l &= ~MSR_IA32_APICBASE_BASE;
2050 			l |= MSR_IA32_APICBASE_ENABLE | addr;
2051 			wrmsr(MSR_IA32_APICBASE, l, h);
2052 			enabled_via_apicbase = 1;
2053 		}
2054 	}
2055 	return apic_verify();
2056 }
2057 
2058 /*
2059  * Detect and initialize APIC
2060  */
2061 static int __init detect_init_APIC(void)
2062 {
2063 	/* Disabled by kernel option? */
2064 	if (disable_apic)
2065 		return -1;
2066 
2067 	switch (boot_cpu_data.x86_vendor) {
2068 	case X86_VENDOR_AMD:
2069 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2070 		    (boot_cpu_data.x86 >= 15))
2071 			break;
2072 		goto no_apic;
2073 	case X86_VENDOR_HYGON:
2074 		break;
2075 	case X86_VENDOR_INTEL:
2076 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2077 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2078 			break;
2079 		goto no_apic;
2080 	default:
2081 		goto no_apic;
2082 	}
2083 
2084 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2085 		/*
2086 		 * Over-ride BIOS and try to enable the local APIC only if
2087 		 * "lapic" specified.
2088 		 */
2089 		if (!force_enable_local_apic) {
2090 			pr_info("Local APIC disabled by BIOS -- "
2091 				"you can enable it with \"lapic\"\n");
2092 			return -1;
2093 		}
2094 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2095 			return -1;
2096 	} else {
2097 		if (apic_verify())
2098 			return -1;
2099 	}
2100 
2101 	apic_pm_activate();
2102 
2103 	return 0;
2104 
2105 no_apic:
2106 	pr_info("No local APIC present or hardware disabled\n");
2107 	return -1;
2108 }
2109 #endif
2110 
2111 /**
2112  * init_apic_mappings - initialize APIC mappings
2113  */
2114 void __init init_apic_mappings(void)
2115 {
2116 	unsigned int new_apicid;
2117 
2118 	if (apic_validate_deadline_timer())
2119 		pr_info("TSC deadline timer available\n");
2120 
2121 	if (x2apic_mode) {
2122 		boot_cpu_physical_apicid = read_apic_id();
2123 		return;
2124 	}
2125 
2126 	/* If no local APIC can be found return early */
2127 	if (!smp_found_config && detect_init_APIC()) {
2128 		/* lets NOP'ify apic operations */
2129 		pr_info("APIC: disable apic facility\n");
2130 		apic_disable();
2131 	} else {
2132 		apic_phys = mp_lapic_addr;
2133 
2134 		/*
2135 		 * If the system has ACPI MADT tables or MP info, the LAPIC
2136 		 * address is already registered.
2137 		 */
2138 		if (!acpi_lapic && !smp_found_config)
2139 			register_lapic_address(apic_phys);
2140 	}
2141 
2142 	/*
2143 	 * Fetch the APIC ID of the BSP in case we have a
2144 	 * default configuration (or the MP table is broken).
2145 	 */
2146 	new_apicid = read_apic_id();
2147 	if (boot_cpu_physical_apicid != new_apicid) {
2148 		boot_cpu_physical_apicid = new_apicid;
2149 		/*
2150 		 * yeah -- we lie about apic_version
2151 		 * in case if apic was disabled via boot option
2152 		 * but it's not a problem for SMP compiled kernel
2153 		 * since apic_intr_mode_select is prepared for such
2154 		 * a case and disable smp mode
2155 		 */
2156 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2157 	}
2158 }
2159 
2160 void __init register_lapic_address(unsigned long address)
2161 {
2162 	mp_lapic_addr = address;
2163 
2164 	if (!x2apic_mode) {
2165 		set_fixmap_nocache(FIX_APIC_BASE, address);
2166 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2167 			    APIC_BASE, address);
2168 	}
2169 	if (boot_cpu_physical_apicid == -1U) {
2170 		boot_cpu_physical_apicid  = read_apic_id();
2171 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2172 	}
2173 }
2174 
2175 /*
2176  * Local APIC interrupts
2177  */
2178 
2179 /*
2180  * Common handling code for spurious_interrupt and spurious_vector entry
2181  * points below. No point in allowing the compiler to inline it twice.
2182  */
2183 static noinline void handle_spurious_interrupt(u8 vector)
2184 {
2185 	u32 v;
2186 
2187 	trace_spurious_apic_entry(vector);
2188 
2189 	inc_irq_stat(irq_spurious_count);
2190 
2191 	/*
2192 	 * If this is a spurious interrupt then do not acknowledge
2193 	 */
2194 	if (vector == SPURIOUS_APIC_VECTOR) {
2195 		/* See SDM vol 3 */
2196 		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2197 			smp_processor_id());
2198 		goto out;
2199 	}
2200 
2201 	/*
2202 	 * If it is a vectored one, verify it's set in the ISR. If set,
2203 	 * acknowledge it.
2204 	 */
2205 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2206 	if (v & (1 << (vector & 0x1f))) {
2207 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2208 			vector, smp_processor_id());
2209 		ack_APIC_irq();
2210 	} else {
2211 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2212 			vector, smp_processor_id());
2213 	}
2214 out:
2215 	trace_spurious_apic_exit(vector);
2216 }
2217 
2218 /**
2219  * spurious_interrupt - Catch all for interrupts raised on unused vectors
2220  * @regs:	Pointer to pt_regs on stack
2221  * @vector:	The vector number
2222  *
2223  * This is invoked from ASM entry code to catch all interrupts which
2224  * trigger on an entry which is routed to the common_spurious idtentry
2225  * point.
2226  */
2227 DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2228 {
2229 	handle_spurious_interrupt(vector);
2230 }
2231 
2232 DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2233 {
2234 	handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2235 }
2236 
2237 /*
2238  * This interrupt should never happen with our APIC/SMP architecture
2239  */
2240 DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2241 {
2242 	static const char * const error_interrupt_reason[] = {
2243 		"Send CS error",		/* APIC Error Bit 0 */
2244 		"Receive CS error",		/* APIC Error Bit 1 */
2245 		"Send accept error",		/* APIC Error Bit 2 */
2246 		"Receive accept error",		/* APIC Error Bit 3 */
2247 		"Redirectable IPI",		/* APIC Error Bit 4 */
2248 		"Send illegal vector",		/* APIC Error Bit 5 */
2249 		"Received illegal vector",	/* APIC Error Bit 6 */
2250 		"Illegal register address",	/* APIC Error Bit 7 */
2251 	};
2252 	u32 v, i = 0;
2253 
2254 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2255 
2256 	/* First tickle the hardware, only then report what went on. -- REW */
2257 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2258 		apic_write(APIC_ESR, 0);
2259 	v = apic_read(APIC_ESR);
2260 	ack_APIC_irq();
2261 	atomic_inc(&irq_err_count);
2262 
2263 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2264 		    smp_processor_id(), v);
2265 
2266 	v &= 0xff;
2267 	while (v) {
2268 		if (v & 0x1)
2269 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2270 		i++;
2271 		v >>= 1;
2272 	}
2273 
2274 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2275 
2276 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2277 }
2278 
2279 /**
2280  * connect_bsp_APIC - attach the APIC to the interrupt system
2281  */
2282 static void __init connect_bsp_APIC(void)
2283 {
2284 #ifdef CONFIG_X86_32
2285 	if (pic_mode) {
2286 		/*
2287 		 * Do not trust the local APIC being empty at bootup.
2288 		 */
2289 		clear_local_APIC();
2290 		/*
2291 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2292 		 * local APIC to INT and NMI lines.
2293 		 */
2294 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2295 				"enabling APIC mode.\n");
2296 		imcr_pic_to_apic();
2297 	}
2298 #endif
2299 }
2300 
2301 /**
2302  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2303  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2304  *
2305  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2306  * APIC is disabled.
2307  */
2308 void disconnect_bsp_APIC(int virt_wire_setup)
2309 {
2310 	unsigned int value;
2311 
2312 #ifdef CONFIG_X86_32
2313 	if (pic_mode) {
2314 		/*
2315 		 * Put the board back into PIC mode (has an effect only on
2316 		 * certain older boards).  Note that APIC interrupts, including
2317 		 * IPIs, won't work beyond this point!  The only exception are
2318 		 * INIT IPIs.
2319 		 */
2320 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2321 				"entering PIC mode.\n");
2322 		imcr_apic_to_pic();
2323 		return;
2324 	}
2325 #endif
2326 
2327 	/* Go back to Virtual Wire compatibility mode */
2328 
2329 	/* For the spurious interrupt use vector F, and enable it */
2330 	value = apic_read(APIC_SPIV);
2331 	value &= ~APIC_VECTOR_MASK;
2332 	value |= APIC_SPIV_APIC_ENABLED;
2333 	value |= 0xf;
2334 	apic_write(APIC_SPIV, value);
2335 
2336 	if (!virt_wire_setup) {
2337 		/*
2338 		 * For LVT0 make it edge triggered, active high,
2339 		 * external and enabled
2340 		 */
2341 		value = apic_read(APIC_LVT0);
2342 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2343 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2344 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2345 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2346 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2347 		apic_write(APIC_LVT0, value);
2348 	} else {
2349 		/* Disable LVT0 */
2350 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2351 	}
2352 
2353 	/*
2354 	 * For LVT1 make it edge triggered, active high,
2355 	 * nmi and enabled
2356 	 */
2357 	value = apic_read(APIC_LVT1);
2358 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2359 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2360 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2361 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2362 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2363 	apic_write(APIC_LVT1, value);
2364 }
2365 
2366 /*
2367  * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2368  * contiguously, it equals to current allocated max logical CPU ID plus 1.
2369  * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2370  * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2371  *
2372  * NOTE: Reserve 0 for BSP.
2373  */
2374 static int nr_logical_cpuids = 1;
2375 
2376 /*
2377  * Used to store mapping between logical CPU IDs and APIC IDs.
2378  */
2379 static int cpuid_to_apicid[] = {
2380 	[0 ... NR_CPUS - 1] = -1,
2381 };
2382 
2383 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
2384 {
2385 	return phys_id == cpuid_to_apicid[cpu];
2386 }
2387 
2388 #ifdef CONFIG_SMP
2389 /**
2390  * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2391  * @apicid: APIC ID to check
2392  */
2393 bool apic_id_is_primary_thread(unsigned int apicid)
2394 {
2395 	u32 mask;
2396 
2397 	if (smp_num_siblings == 1)
2398 		return true;
2399 	/* Isolate the SMT bit(s) in the APICID and check for 0 */
2400 	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2401 	return !(apicid & mask);
2402 }
2403 #endif
2404 
2405 /*
2406  * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2407  * and cpuid_to_apicid[] synchronized.
2408  */
2409 static int allocate_logical_cpuid(int apicid)
2410 {
2411 	int i;
2412 
2413 	/*
2414 	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2415 	 * check if the kernel has allocated a cpuid for it.
2416 	 */
2417 	for (i = 0; i < nr_logical_cpuids; i++) {
2418 		if (cpuid_to_apicid[i] == apicid)
2419 			return i;
2420 	}
2421 
2422 	/* Allocate a new cpuid. */
2423 	if (nr_logical_cpuids >= nr_cpu_ids) {
2424 		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2425 			     "Processor %d/0x%x and the rest are ignored.\n",
2426 			     nr_cpu_ids, nr_logical_cpuids, apicid);
2427 		return -EINVAL;
2428 	}
2429 
2430 	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2431 	return nr_logical_cpuids++;
2432 }
2433 
2434 int generic_processor_info(int apicid, int version)
2435 {
2436 	int cpu, max = nr_cpu_ids;
2437 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2438 				phys_cpu_present_map);
2439 
2440 	/*
2441 	 * boot_cpu_physical_apicid is designed to have the apicid
2442 	 * returned by read_apic_id(), i.e, the apicid of the
2443 	 * currently booting-up processor. However, on some platforms,
2444 	 * it is temporarily modified by the apicid reported as BSP
2445 	 * through MP table. Concretely:
2446 	 *
2447 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2448 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2449 	 *
2450 	 * This function is executed with the modified
2451 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2452 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2453 	 *
2454 	 * Since fixing handling of boot_cpu_physical_apicid requires
2455 	 * another discussion and tests on each platform, we leave it
2456 	 * for now and here we use read_apic_id() directly in this
2457 	 * function, generic_processor_info().
2458 	 */
2459 	if (disabled_cpu_apicid != BAD_APICID &&
2460 	    disabled_cpu_apicid != read_apic_id() &&
2461 	    disabled_cpu_apicid == apicid) {
2462 		int thiscpu = num_processors + disabled_cpus;
2463 
2464 		pr_warn("APIC: Disabling requested cpu."
2465 			" Processor %d/0x%x ignored.\n", thiscpu, apicid);
2466 
2467 		disabled_cpus++;
2468 		return -ENODEV;
2469 	}
2470 
2471 	/*
2472 	 * If boot cpu has not been detected yet, then only allow upto
2473 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2474 	 */
2475 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2476 	    apicid != boot_cpu_physical_apicid) {
2477 		int thiscpu = max + disabled_cpus - 1;
2478 
2479 		pr_warn("APIC: NR_CPUS/possible_cpus limit of %i almost"
2480 			" reached. Keeping one slot for boot cpu."
2481 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2482 
2483 		disabled_cpus++;
2484 		return -ENODEV;
2485 	}
2486 
2487 	if (num_processors >= nr_cpu_ids) {
2488 		int thiscpu = max + disabled_cpus;
2489 
2490 		pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2491 			"Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2492 
2493 		disabled_cpus++;
2494 		return -EINVAL;
2495 	}
2496 
2497 	if (apicid == boot_cpu_physical_apicid) {
2498 		/*
2499 		 * x86_bios_cpu_apicid is required to have processors listed
2500 		 * in same order as logical cpu numbers. Hence the first
2501 		 * entry is BSP, and so on.
2502 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2503 		 * for BSP.
2504 		 */
2505 		cpu = 0;
2506 
2507 		/* Logical cpuid 0 is reserved for BSP. */
2508 		cpuid_to_apicid[0] = apicid;
2509 	} else {
2510 		cpu = allocate_logical_cpuid(apicid);
2511 		if (cpu < 0) {
2512 			disabled_cpus++;
2513 			return -EINVAL;
2514 		}
2515 	}
2516 
2517 	/*
2518 	 * Validate version
2519 	 */
2520 	if (version == 0x0) {
2521 		pr_warn("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2522 			cpu, apicid);
2523 		version = 0x10;
2524 	}
2525 
2526 	if (version != boot_cpu_apic_version) {
2527 		pr_warn("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2528 			boot_cpu_apic_version, cpu, version);
2529 	}
2530 
2531 	if (apicid > max_physical_apicid)
2532 		max_physical_apicid = apicid;
2533 
2534 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2535 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2536 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2537 #endif
2538 #ifdef CONFIG_X86_32
2539 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2540 		apic->x86_32_early_logical_apicid(cpu);
2541 #endif
2542 	set_cpu_possible(cpu, true);
2543 	physid_set(apicid, phys_cpu_present_map);
2544 	set_cpu_present(cpu, true);
2545 	num_processors++;
2546 
2547 	return cpu;
2548 }
2549 
2550 int hard_smp_processor_id(void)
2551 {
2552 	return read_apic_id();
2553 }
2554 
2555 void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2556 			   bool dmar)
2557 {
2558 	memset(msg, 0, sizeof(*msg));
2559 
2560 	msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2561 	msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2562 	msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2563 
2564 	msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2565 	msg->arch_data.vector = cfg->vector;
2566 
2567 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2568 	/*
2569 	 * Only the IOMMU itself can use the trick of putting destination
2570 	 * APIC ID into the high bits of the address. Anything else would
2571 	 * just be writing to memory if it tried that, and needs IR to
2572 	 * address APICs which can't be addressed in the normal 32-bit
2573 	 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2574 	 * some hypervisors allow the extended destination ID field in bits
2575 	 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2576 	 */
2577 	if (dmar)
2578 		msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2579 	else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2580 		msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2581 	else
2582 		WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2583 }
2584 
2585 u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2586 {
2587 	u32 dest = msg->arch_addr_lo.destid_0_7;
2588 
2589 	if (extid)
2590 		dest |= msg->arch_addr_hi.destid_8_31 << 8;
2591 	return dest;
2592 }
2593 EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2594 
2595 #ifdef CONFIG_X86_64
2596 void __init acpi_wake_cpu_handler_update(wakeup_cpu_handler handler)
2597 {
2598 	struct apic **drv;
2599 
2600 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++)
2601 		(*drv)->wakeup_secondary_cpu_64 = handler;
2602 }
2603 #endif
2604 
2605 /*
2606  * Override the generic EOI implementation with an optimized version.
2607  * Only called during early boot when only one CPU is active and with
2608  * interrupts disabled, so we know this does not race with actual APIC driver
2609  * use.
2610  */
2611 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2612 {
2613 	struct apic **drv;
2614 
2615 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2616 		/* Should happen once for each apic */
2617 		WARN_ON((*drv)->eoi_write == eoi_write);
2618 		(*drv)->native_eoi_write = (*drv)->eoi_write;
2619 		(*drv)->eoi_write = eoi_write;
2620 	}
2621 }
2622 
2623 static void __init apic_bsp_up_setup(void)
2624 {
2625 #ifdef CONFIG_X86_64
2626 	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2627 #else
2628 	/*
2629 	 * Hack: In case of kdump, after a crash, kernel might be booting
2630 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2631 	 * might be zero if read from MP tables. Get it from LAPIC.
2632 	 */
2633 # ifdef CONFIG_CRASH_DUMP
2634 	boot_cpu_physical_apicid = read_apic_id();
2635 # endif
2636 #endif
2637 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2638 }
2639 
2640 /**
2641  * apic_bsp_setup - Setup function for local apic and io-apic
2642  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2643  */
2644 static void __init apic_bsp_setup(bool upmode)
2645 {
2646 	connect_bsp_APIC();
2647 	if (upmode)
2648 		apic_bsp_up_setup();
2649 	setup_local_APIC();
2650 
2651 	enable_IO_APIC();
2652 	end_local_APIC_setup();
2653 	irq_remap_enable_fault_handling();
2654 	setup_IO_APIC();
2655 	lapic_update_legacy_vectors();
2656 }
2657 
2658 #ifdef CONFIG_UP_LATE_INIT
2659 void __init up_late_init(void)
2660 {
2661 	if (apic_intr_mode == APIC_PIC)
2662 		return;
2663 
2664 	/* Setup local timer */
2665 	x86_init.timers.setup_percpu_clockev();
2666 }
2667 #endif
2668 
2669 /*
2670  * Power management
2671  */
2672 #ifdef CONFIG_PM
2673 
2674 static struct {
2675 	/*
2676 	 * 'active' is true if the local APIC was enabled by us and
2677 	 * not the BIOS; this signifies that we are also responsible
2678 	 * for disabling it before entering apm/acpi suspend
2679 	 */
2680 	int active;
2681 	/* r/w apic fields */
2682 	unsigned int apic_id;
2683 	unsigned int apic_taskpri;
2684 	unsigned int apic_ldr;
2685 	unsigned int apic_dfr;
2686 	unsigned int apic_spiv;
2687 	unsigned int apic_lvtt;
2688 	unsigned int apic_lvtpc;
2689 	unsigned int apic_lvt0;
2690 	unsigned int apic_lvt1;
2691 	unsigned int apic_lvterr;
2692 	unsigned int apic_tmict;
2693 	unsigned int apic_tdcr;
2694 	unsigned int apic_thmr;
2695 	unsigned int apic_cmci;
2696 } apic_pm_state;
2697 
2698 static int lapic_suspend(void)
2699 {
2700 	unsigned long flags;
2701 	int maxlvt;
2702 
2703 	if (!apic_pm_state.active)
2704 		return 0;
2705 
2706 	maxlvt = lapic_get_maxlvt();
2707 
2708 	apic_pm_state.apic_id = apic_read(APIC_ID);
2709 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2710 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2711 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2712 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2713 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2714 	if (maxlvt >= 4)
2715 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2716 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2717 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2718 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2719 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2720 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2721 #ifdef CONFIG_X86_THERMAL_VECTOR
2722 	if (maxlvt >= 5)
2723 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2724 #endif
2725 #ifdef CONFIG_X86_MCE_INTEL
2726 	if (maxlvt >= 6)
2727 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2728 #endif
2729 
2730 	local_irq_save(flags);
2731 
2732 	/*
2733 	 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2734 	 * entries on some implementations.
2735 	 */
2736 	mask_ioapic_entries();
2737 
2738 	disable_local_APIC();
2739 
2740 	irq_remapping_disable();
2741 
2742 	local_irq_restore(flags);
2743 	return 0;
2744 }
2745 
2746 static void lapic_resume(void)
2747 {
2748 	unsigned int l, h;
2749 	unsigned long flags;
2750 	int maxlvt;
2751 
2752 	if (!apic_pm_state.active)
2753 		return;
2754 
2755 	local_irq_save(flags);
2756 
2757 	/*
2758 	 * IO-APIC and PIC have their own resume routines.
2759 	 * We just mask them here to make sure the interrupt
2760 	 * subsystem is completely quiet while we enable x2apic
2761 	 * and interrupt-remapping.
2762 	 */
2763 	mask_ioapic_entries();
2764 	legacy_pic->mask_all();
2765 
2766 	if (x2apic_mode) {
2767 		__x2apic_enable();
2768 	} else {
2769 		/*
2770 		 * Make sure the APICBASE points to the right address
2771 		 *
2772 		 * FIXME! This will be wrong if we ever support suspend on
2773 		 * SMP! We'll need to do this as part of the CPU restore!
2774 		 */
2775 		if (boot_cpu_data.x86 >= 6) {
2776 			rdmsr(MSR_IA32_APICBASE, l, h);
2777 			l &= ~MSR_IA32_APICBASE_BASE;
2778 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2779 			wrmsr(MSR_IA32_APICBASE, l, h);
2780 		}
2781 	}
2782 
2783 	maxlvt = lapic_get_maxlvt();
2784 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2785 	apic_write(APIC_ID, apic_pm_state.apic_id);
2786 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2787 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2788 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2789 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2790 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2791 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2792 #ifdef CONFIG_X86_THERMAL_VECTOR
2793 	if (maxlvt >= 5)
2794 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2795 #endif
2796 #ifdef CONFIG_X86_MCE_INTEL
2797 	if (maxlvt >= 6)
2798 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2799 #endif
2800 	if (maxlvt >= 4)
2801 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2802 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2803 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2804 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2805 	apic_write(APIC_ESR, 0);
2806 	apic_read(APIC_ESR);
2807 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2808 	apic_write(APIC_ESR, 0);
2809 	apic_read(APIC_ESR);
2810 
2811 	irq_remapping_reenable(x2apic_mode);
2812 
2813 	local_irq_restore(flags);
2814 }
2815 
2816 /*
2817  * This device has no shutdown method - fully functioning local APICs
2818  * are needed on every CPU up until machine_halt/restart/poweroff.
2819  */
2820 
2821 static struct syscore_ops lapic_syscore_ops = {
2822 	.resume		= lapic_resume,
2823 	.suspend	= lapic_suspend,
2824 };
2825 
2826 static void apic_pm_activate(void)
2827 {
2828 	apic_pm_state.active = 1;
2829 }
2830 
2831 static int __init init_lapic_sysfs(void)
2832 {
2833 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2834 	if (boot_cpu_has(X86_FEATURE_APIC))
2835 		register_syscore_ops(&lapic_syscore_ops);
2836 
2837 	return 0;
2838 }
2839 
2840 /* local apic needs to resume before other devices access its registers. */
2841 core_initcall(init_lapic_sysfs);
2842 
2843 #else	/* CONFIG_PM */
2844 
2845 static void apic_pm_activate(void) { }
2846 
2847 #endif	/* CONFIG_PM */
2848 
2849 #ifdef CONFIG_X86_64
2850 
2851 static int multi_checked;
2852 static int multi;
2853 
2854 static int set_multi(const struct dmi_system_id *d)
2855 {
2856 	if (multi)
2857 		return 0;
2858 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2859 	multi = 1;
2860 	return 0;
2861 }
2862 
2863 static const struct dmi_system_id multi_dmi_table[] = {
2864 	{
2865 		.callback = set_multi,
2866 		.ident = "IBM System Summit2",
2867 		.matches = {
2868 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2869 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2870 		},
2871 	},
2872 	{}
2873 };
2874 
2875 static void dmi_check_multi(void)
2876 {
2877 	if (multi_checked)
2878 		return;
2879 
2880 	dmi_check_system(multi_dmi_table);
2881 	multi_checked = 1;
2882 }
2883 
2884 /*
2885  * apic_is_clustered_box() -- Check if we can expect good TSC
2886  *
2887  * Thus far, the major user of this is IBM's Summit2 series:
2888  * Clustered boxes may have unsynced TSC problems if they are
2889  * multi-chassis.
2890  * Use DMI to check them
2891  */
2892 int apic_is_clustered_box(void)
2893 {
2894 	dmi_check_multi();
2895 	return multi;
2896 }
2897 #endif
2898 
2899 /*
2900  * APIC command line parameters
2901  */
2902 static int __init setup_disableapic(char *arg)
2903 {
2904 	disable_apic = 1;
2905 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2906 	return 0;
2907 }
2908 early_param("disableapic", setup_disableapic);
2909 
2910 /* same as disableapic, for compatibility */
2911 static int __init setup_nolapic(char *arg)
2912 {
2913 	return setup_disableapic(arg);
2914 }
2915 early_param("nolapic", setup_nolapic);
2916 
2917 static int __init parse_lapic_timer_c2_ok(char *arg)
2918 {
2919 	local_apic_timer_c2_ok = 1;
2920 	return 0;
2921 }
2922 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2923 
2924 static int __init parse_disable_apic_timer(char *arg)
2925 {
2926 	disable_apic_timer = 1;
2927 	return 0;
2928 }
2929 early_param("noapictimer", parse_disable_apic_timer);
2930 
2931 static int __init parse_nolapic_timer(char *arg)
2932 {
2933 	disable_apic_timer = 1;
2934 	return 0;
2935 }
2936 early_param("nolapic_timer", parse_nolapic_timer);
2937 
2938 static int __init apic_set_verbosity(char *arg)
2939 {
2940 	if (!arg)  {
2941 #ifdef CONFIG_X86_64
2942 		skip_ioapic_setup = 0;
2943 		return 0;
2944 #endif
2945 		return -EINVAL;
2946 	}
2947 
2948 	if (strcmp("debug", arg) == 0)
2949 		apic_verbosity = APIC_DEBUG;
2950 	else if (strcmp("verbose", arg) == 0)
2951 		apic_verbosity = APIC_VERBOSE;
2952 #ifdef CONFIG_X86_64
2953 	else {
2954 		pr_warn("APIC Verbosity level %s not recognised"
2955 			" use apic=verbose or apic=debug\n", arg);
2956 		return -EINVAL;
2957 	}
2958 #endif
2959 
2960 	return 0;
2961 }
2962 early_param("apic", apic_set_verbosity);
2963 
2964 static int __init lapic_insert_resource(void)
2965 {
2966 	if (!apic_phys)
2967 		return -1;
2968 
2969 	/* Put local APIC into the resource map. */
2970 	lapic_resource.start = apic_phys;
2971 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2972 	insert_resource(&iomem_resource, &lapic_resource);
2973 
2974 	return 0;
2975 }
2976 
2977 /*
2978  * need call insert after e820__reserve_resources()
2979  * that is using request_resource
2980  */
2981 late_initcall(lapic_insert_resource);
2982 
2983 static int __init apic_set_disabled_cpu_apicid(char *arg)
2984 {
2985 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2986 		return -EINVAL;
2987 
2988 	return 0;
2989 }
2990 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2991 
2992 static int __init apic_set_extnmi(char *arg)
2993 {
2994 	if (!arg)
2995 		return -EINVAL;
2996 
2997 	if (!strncmp("all", arg, 3))
2998 		apic_extnmi = APIC_EXTNMI_ALL;
2999 	else if (!strncmp("none", arg, 4))
3000 		apic_extnmi = APIC_EXTNMI_NONE;
3001 	else if (!strncmp("bsp", arg, 3))
3002 		apic_extnmi = APIC_EXTNMI_BSP;
3003 	else {
3004 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
3005 		return -EINVAL;
3006 	}
3007 
3008 	return 0;
3009 }
3010 early_param("apic_extnmi", apic_set_extnmi);
3011