xref: /openbmc/linux/arch/x86/kernel/apic/apic.c (revision 1504b6f9)
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 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
426 	} while (rsvd != new);
427 
428 	rsvd &= ~APIC_EILVT_MASKED;
429 	if (rsvd && rsvd != vector)
430 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
431 			offset, rsvd);
432 
433 	return new;
434 }
435 
436 /*
437  * If mask=1, the LVT entry does not generate interrupts while mask=0
438  * enables the vector. See also the BKDGs. Must be called with
439  * preemption disabled.
440  */
441 
442 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
443 {
444 	unsigned long reg = APIC_EILVTn(offset);
445 	unsigned int new, old, reserved;
446 
447 	new = (mask << 16) | (msg_type << 8) | vector;
448 	old = apic_read(reg);
449 	reserved = reserve_eilvt_offset(offset, new);
450 
451 	if (reserved != new) {
452 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
453 		       "vector 0x%x, but the register is already in use for "
454 		       "vector 0x%x on another cpu\n",
455 		       smp_processor_id(), reg, offset, new, reserved);
456 		return -EINVAL;
457 	}
458 
459 	if (!eilvt_entry_is_changeable(old, new)) {
460 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
461 		       "vector 0x%x, but the register is already in use for "
462 		       "vector 0x%x on this cpu\n",
463 		       smp_processor_id(), reg, offset, new, old);
464 		return -EBUSY;
465 	}
466 
467 	apic_write(reg, new);
468 
469 	return 0;
470 }
471 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
472 
473 /*
474  * Program the next event, relative to now
475  */
476 static int lapic_next_event(unsigned long delta,
477 			    struct clock_event_device *evt)
478 {
479 	apic_write(APIC_TMICT, delta);
480 	return 0;
481 }
482 
483 static int lapic_next_deadline(unsigned long delta,
484 			       struct clock_event_device *evt)
485 {
486 	u64 tsc;
487 
488 	/* This MSR is special and need a special fence: */
489 	weak_wrmsr_fence();
490 
491 	tsc = rdtsc();
492 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
493 	return 0;
494 }
495 
496 static int lapic_timer_shutdown(struct clock_event_device *evt)
497 {
498 	unsigned int v;
499 
500 	/* Lapic used as dummy for broadcast ? */
501 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
502 		return 0;
503 
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 	return 0;
509 }
510 
511 static inline int
512 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
513 {
514 	/* Lapic used as dummy for broadcast ? */
515 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
516 		return 0;
517 
518 	__setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
519 	return 0;
520 }
521 
522 static int lapic_timer_set_periodic(struct clock_event_device *evt)
523 {
524 	return lapic_timer_set_periodic_oneshot(evt, false);
525 }
526 
527 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
528 {
529 	return lapic_timer_set_periodic_oneshot(evt, true);
530 }
531 
532 /*
533  * Local APIC timer broadcast function
534  */
535 static void lapic_timer_broadcast(const struct cpumask *mask)
536 {
537 #ifdef CONFIG_SMP
538 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
539 #endif
540 }
541 
542 
543 /*
544  * The local apic timer can be used for any function which is CPU local.
545  */
546 static struct clock_event_device lapic_clockevent = {
547 	.name				= "lapic",
548 	.features			= CLOCK_EVT_FEAT_PERIODIC |
549 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
550 					  | CLOCK_EVT_FEAT_DUMMY,
551 	.shift				= 32,
552 	.set_state_shutdown		= lapic_timer_shutdown,
553 	.set_state_periodic		= lapic_timer_set_periodic,
554 	.set_state_oneshot		= lapic_timer_set_oneshot,
555 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
556 	.set_next_event			= lapic_next_event,
557 	.broadcast			= lapic_timer_broadcast,
558 	.rating				= 100,
559 	.irq				= -1,
560 };
561 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
562 
563 static const struct x86_cpu_id deadline_match[] __initconst = {
564 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
565 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
566 
567 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X,	0x0b000020),
568 
569 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
570 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
571 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
572 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
573 
574 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
575 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
576 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
577 
578 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL,		0x22),
579 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L,		0x20),
580 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G,		0x17),
581 
582 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL,		0x25),
583 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G,	0x17),
584 
585 	X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L,		0xb2),
586 	X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE,		0xb2),
587 
588 	X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L,		0x52),
589 	X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE,		0x52),
590 
591 	{},
592 };
593 
594 static __init bool apic_validate_deadline_timer(void)
595 {
596 	const struct x86_cpu_id *m;
597 	u32 rev;
598 
599 	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
600 		return false;
601 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
602 		return true;
603 
604 	m = x86_match_cpu(deadline_match);
605 	if (!m)
606 		return true;
607 
608 	rev = (u32)m->driver_data;
609 
610 	if (boot_cpu_data.microcode >= rev)
611 		return true;
612 
613 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
614 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
615 	       "please update microcode to version: 0x%x (or later)\n", rev);
616 	return false;
617 }
618 
619 /*
620  * Setup the local APIC timer for this CPU. Copy the initialized values
621  * of the boot CPU and register the clock event in the framework.
622  */
623 static void setup_APIC_timer(void)
624 {
625 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
626 
627 	if (this_cpu_has(X86_FEATURE_ARAT)) {
628 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
629 		/* Make LAPIC timer preferable over percpu HPET */
630 		lapic_clockevent.rating = 150;
631 	}
632 
633 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
634 	levt->cpumask = cpumask_of(smp_processor_id());
635 
636 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
637 		levt->name = "lapic-deadline";
638 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
639 				    CLOCK_EVT_FEAT_DUMMY);
640 		levt->set_next_event = lapic_next_deadline;
641 		clockevents_config_and_register(levt,
642 						tsc_khz * (1000 / TSC_DIVISOR),
643 						0xF, ~0UL);
644 	} else
645 		clockevents_register_device(levt);
646 }
647 
648 /*
649  * Install the updated TSC frequency from recalibration at the TSC
650  * deadline clockevent devices.
651  */
652 static void __lapic_update_tsc_freq(void *info)
653 {
654 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
655 
656 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
657 		return;
658 
659 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
660 }
661 
662 void lapic_update_tsc_freq(void)
663 {
664 	/*
665 	 * The clockevent device's ->mult and ->shift can both be
666 	 * changed. In order to avoid races, schedule the frequency
667 	 * update code on each CPU.
668 	 */
669 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
670 }
671 
672 /*
673  * In this functions we calibrate APIC bus clocks to the external timer.
674  *
675  * We want to do the calibration only once since we want to have local timer
676  * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
677  * frequency.
678  *
679  * This was previously done by reading the PIT/HPET and waiting for a wrap
680  * around to find out, that a tick has elapsed. I have a box, where the PIT
681  * readout is broken, so it never gets out of the wait loop again. This was
682  * also reported by others.
683  *
684  * Monitoring the jiffies value is inaccurate and the clockevents
685  * infrastructure allows us to do a simple substitution of the interrupt
686  * handler.
687  *
688  * The calibration routine also uses the pm_timer when possible, as the PIT
689  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
690  * back to normal later in the boot process).
691  */
692 
693 #define LAPIC_CAL_LOOPS		(HZ/10)
694 
695 static __initdata int lapic_cal_loops = -1;
696 static __initdata long lapic_cal_t1, lapic_cal_t2;
697 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
698 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
699 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
700 
701 /*
702  * Temporary interrupt handler and polled calibration function.
703  */
704 static void __init lapic_cal_handler(struct clock_event_device *dev)
705 {
706 	unsigned long long tsc = 0;
707 	long tapic = apic_read(APIC_TMCCT);
708 	unsigned long pm = acpi_pm_read_early();
709 
710 	if (boot_cpu_has(X86_FEATURE_TSC))
711 		tsc = rdtsc();
712 
713 	switch (lapic_cal_loops++) {
714 	case 0:
715 		lapic_cal_t1 = tapic;
716 		lapic_cal_tsc1 = tsc;
717 		lapic_cal_pm1 = pm;
718 		lapic_cal_j1 = jiffies;
719 		break;
720 
721 	case LAPIC_CAL_LOOPS:
722 		lapic_cal_t2 = tapic;
723 		lapic_cal_tsc2 = tsc;
724 		if (pm < lapic_cal_pm1)
725 			pm += ACPI_PM_OVRRUN;
726 		lapic_cal_pm2 = pm;
727 		lapic_cal_j2 = jiffies;
728 		break;
729 	}
730 }
731 
732 static int __init
733 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
734 {
735 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
736 	const long pm_thresh = pm_100ms / 100;
737 	unsigned long mult;
738 	u64 res;
739 
740 #ifndef CONFIG_X86_PM_TIMER
741 	return -1;
742 #endif
743 
744 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
745 
746 	/* Check, if the PM timer is available */
747 	if (!deltapm)
748 		return -1;
749 
750 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
751 
752 	if (deltapm > (pm_100ms - pm_thresh) &&
753 	    deltapm < (pm_100ms + pm_thresh)) {
754 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
755 		return 0;
756 	}
757 
758 	res = (((u64)deltapm) *  mult) >> 22;
759 	do_div(res, 1000000);
760 	pr_warn("APIC calibration not consistent "
761 		"with PM-Timer: %ldms instead of 100ms\n", (long)res);
762 
763 	/* Correct the lapic counter value */
764 	res = (((u64)(*delta)) * pm_100ms);
765 	do_div(res, deltapm);
766 	pr_info("APIC delta adjusted to PM-Timer: "
767 		"%lu (%ld)\n", (unsigned long)res, *delta);
768 	*delta = (long)res;
769 
770 	/* Correct the tsc counter value */
771 	if (boot_cpu_has(X86_FEATURE_TSC)) {
772 		res = (((u64)(*deltatsc)) * pm_100ms);
773 		do_div(res, deltapm);
774 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
775 					  "PM-Timer: %lu (%ld)\n",
776 					(unsigned long)res, *deltatsc);
777 		*deltatsc = (long)res;
778 	}
779 
780 	return 0;
781 }
782 
783 static int __init lapic_init_clockevent(void)
784 {
785 	if (!lapic_timer_period)
786 		return -1;
787 
788 	/* Calculate the scaled math multiplication factor */
789 	lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
790 					TICK_NSEC, lapic_clockevent.shift);
791 	lapic_clockevent.max_delta_ns =
792 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
793 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
794 	lapic_clockevent.min_delta_ns =
795 		clockevent_delta2ns(0xF, &lapic_clockevent);
796 	lapic_clockevent.min_delta_ticks = 0xF;
797 
798 	return 0;
799 }
800 
801 bool __init apic_needs_pit(void)
802 {
803 	/*
804 	 * If the frequencies are not known, PIT is required for both TSC
805 	 * and apic timer calibration.
806 	 */
807 	if (!tsc_khz || !cpu_khz)
808 		return true;
809 
810 	/* Is there an APIC at all or is it disabled? */
811 	if (!boot_cpu_has(X86_FEATURE_APIC) || disable_apic)
812 		return true;
813 
814 	/*
815 	 * If interrupt delivery mode is legacy PIC or virtual wire without
816 	 * configuration, the local APIC timer wont be set up. Make sure
817 	 * that the PIT is initialized.
818 	 */
819 	if (apic_intr_mode == APIC_PIC ||
820 	    apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
821 		return true;
822 
823 	/* Virt guests may lack ARAT, but still have DEADLINE */
824 	if (!boot_cpu_has(X86_FEATURE_ARAT))
825 		return true;
826 
827 	/* Deadline timer is based on TSC so no further PIT action required */
828 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
829 		return false;
830 
831 	/* APIC timer disabled? */
832 	if (disable_apic_timer)
833 		return true;
834 	/*
835 	 * The APIC timer frequency is known already, no PIT calibration
836 	 * required. If unknown, let the PIT be initialized.
837 	 */
838 	return lapic_timer_period == 0;
839 }
840 
841 static int __init calibrate_APIC_clock(void)
842 {
843 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
844 	u64 tsc_perj = 0, tsc_start = 0;
845 	unsigned long jif_start;
846 	unsigned long deltaj;
847 	long delta, deltatsc;
848 	int pm_referenced = 0;
849 
850 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
851 		return 0;
852 
853 	/*
854 	 * Check if lapic timer has already been calibrated by platform
855 	 * specific routine, such as tsc calibration code. If so just fill
856 	 * in the clockevent structure and return.
857 	 */
858 	if (!lapic_init_clockevent()) {
859 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
860 			    lapic_timer_period);
861 		/*
862 		 * Direct calibration methods must have an always running
863 		 * local APIC timer, no need for broadcast timer.
864 		 */
865 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
866 		return 0;
867 	}
868 
869 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
870 		    "calibrating APIC timer ...\n");
871 
872 	/*
873 	 * There are platforms w/o global clockevent devices. Instead of
874 	 * making the calibration conditional on that, use a polling based
875 	 * approach everywhere.
876 	 */
877 	local_irq_disable();
878 
879 	/*
880 	 * Setup the APIC counter to maximum. There is no way the lapic
881 	 * can underflow in the 100ms detection time frame
882 	 */
883 	__setup_APIC_LVTT(0xffffffff, 0, 0);
884 
885 	/*
886 	 * Methods to terminate the calibration loop:
887 	 *  1) Global clockevent if available (jiffies)
888 	 *  2) TSC if available and frequency is known
889 	 */
890 	jif_start = READ_ONCE(jiffies);
891 
892 	if (tsc_khz) {
893 		tsc_start = rdtsc();
894 		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
895 	}
896 
897 	/*
898 	 * Enable interrupts so the tick can fire, if a global
899 	 * clockevent device is available
900 	 */
901 	local_irq_enable();
902 
903 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
904 		/* Wait for a tick to elapse */
905 		while (1) {
906 			if (tsc_khz) {
907 				u64 tsc_now = rdtsc();
908 				if ((tsc_now - tsc_start) >= tsc_perj) {
909 					tsc_start += tsc_perj;
910 					break;
911 				}
912 			} else {
913 				unsigned long jif_now = READ_ONCE(jiffies);
914 
915 				if (time_after(jif_now, jif_start)) {
916 					jif_start = jif_now;
917 					break;
918 				}
919 			}
920 			cpu_relax();
921 		}
922 
923 		/* Invoke the calibration routine */
924 		local_irq_disable();
925 		lapic_cal_handler(NULL);
926 		local_irq_enable();
927 	}
928 
929 	local_irq_disable();
930 
931 	/* Build delta t1-t2 as apic timer counts down */
932 	delta = lapic_cal_t1 - lapic_cal_t2;
933 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
934 
935 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
936 
937 	/* we trust the PM based calibration if possible */
938 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
939 					&delta, &deltatsc);
940 
941 	lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
942 	lapic_init_clockevent();
943 
944 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
945 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
946 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
947 		    lapic_timer_period);
948 
949 	if (boot_cpu_has(X86_FEATURE_TSC)) {
950 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
951 			    "%ld.%04ld MHz.\n",
952 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
953 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
954 	}
955 
956 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
957 		    "%u.%04u MHz.\n",
958 		    lapic_timer_period / (1000000 / HZ),
959 		    lapic_timer_period % (1000000 / HZ));
960 
961 	/*
962 	 * Do a sanity check on the APIC calibration result
963 	 */
964 	if (lapic_timer_period < (1000000 / HZ)) {
965 		local_irq_enable();
966 		pr_warn("APIC frequency too slow, disabling apic timer\n");
967 		return -1;
968 	}
969 
970 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
971 
972 	/*
973 	 * PM timer calibration failed or not turned on so lets try APIC
974 	 * timer based calibration, if a global clockevent device is
975 	 * available.
976 	 */
977 	if (!pm_referenced && global_clock_event) {
978 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
979 
980 		/*
981 		 * Setup the apic timer manually
982 		 */
983 		levt->event_handler = lapic_cal_handler;
984 		lapic_timer_set_periodic(levt);
985 		lapic_cal_loops = -1;
986 
987 		/* Let the interrupts run */
988 		local_irq_enable();
989 
990 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
991 			cpu_relax();
992 
993 		/* Stop the lapic timer */
994 		local_irq_disable();
995 		lapic_timer_shutdown(levt);
996 
997 		/* Jiffies delta */
998 		deltaj = lapic_cal_j2 - lapic_cal_j1;
999 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
1000 
1001 		/* Check, if the jiffies result is consistent */
1002 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
1003 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
1004 		else
1005 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
1006 	}
1007 	local_irq_enable();
1008 
1009 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
1010 		pr_warn("APIC timer disabled due to verification failure\n");
1011 		return -1;
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 /*
1018  * Setup the boot APIC
1019  *
1020  * Calibrate and verify the result.
1021  */
1022 void __init setup_boot_APIC_clock(void)
1023 {
1024 	/*
1025 	 * The local apic timer can be disabled via the kernel
1026 	 * commandline or from the CPU detection code. Register the lapic
1027 	 * timer as a dummy clock event source on SMP systems, so the
1028 	 * broadcast mechanism is used. On UP systems simply ignore it.
1029 	 */
1030 	if (disable_apic_timer) {
1031 		pr_info("Disabling APIC timer\n");
1032 		/* No broadcast on UP ! */
1033 		if (num_possible_cpus() > 1) {
1034 			lapic_clockevent.mult = 1;
1035 			setup_APIC_timer();
1036 		}
1037 		return;
1038 	}
1039 
1040 	if (calibrate_APIC_clock()) {
1041 		/* No broadcast on UP ! */
1042 		if (num_possible_cpus() > 1)
1043 			setup_APIC_timer();
1044 		return;
1045 	}
1046 
1047 	/*
1048 	 * If nmi_watchdog is set to IO_APIC, we need the
1049 	 * PIT/HPET going.  Otherwise register lapic as a dummy
1050 	 * device.
1051 	 */
1052 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1053 
1054 	/* Setup the lapic or request the broadcast */
1055 	setup_APIC_timer();
1056 	amd_e400_c1e_apic_setup();
1057 }
1058 
1059 void setup_secondary_APIC_clock(void)
1060 {
1061 	setup_APIC_timer();
1062 	amd_e400_c1e_apic_setup();
1063 }
1064 
1065 /*
1066  * The guts of the apic timer interrupt
1067  */
1068 static void local_apic_timer_interrupt(void)
1069 {
1070 	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1071 
1072 	/*
1073 	 * Normally we should not be here till LAPIC has been initialized but
1074 	 * in some cases like kdump, its possible that there is a pending LAPIC
1075 	 * timer interrupt from previous kernel's context and is delivered in
1076 	 * new kernel the moment interrupts are enabled.
1077 	 *
1078 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1079 	 * its possible that when we get here evt->event_handler is NULL.
1080 	 * Check for event_handler being NULL and discard the interrupt as
1081 	 * spurious.
1082 	 */
1083 	if (!evt->event_handler) {
1084 		pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1085 			smp_processor_id());
1086 		/* Switch it off */
1087 		lapic_timer_shutdown(evt);
1088 		return;
1089 	}
1090 
1091 	/*
1092 	 * the NMI deadlock-detector uses this.
1093 	 */
1094 	inc_irq_stat(apic_timer_irqs);
1095 
1096 	evt->event_handler(evt);
1097 }
1098 
1099 /*
1100  * Local APIC timer interrupt. This is the most natural way for doing
1101  * local interrupts, but local timer interrupts can be emulated by
1102  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1103  *
1104  * [ if a single-CPU system runs an SMP kernel then we call the local
1105  *   interrupt as well. Thus we cannot inline the local irq ... ]
1106  */
1107 DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1108 {
1109 	struct pt_regs *old_regs = set_irq_regs(regs);
1110 
1111 	ack_APIC_irq();
1112 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1113 	local_apic_timer_interrupt();
1114 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1115 
1116 	set_irq_regs(old_regs);
1117 }
1118 
1119 /*
1120  * Local APIC start and shutdown
1121  */
1122 
1123 /**
1124  * clear_local_APIC - shutdown the local APIC
1125  *
1126  * This is called, when a CPU is disabled and before rebooting, so the state of
1127  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1128  * leftovers during boot.
1129  */
1130 void clear_local_APIC(void)
1131 {
1132 	int maxlvt;
1133 	u32 v;
1134 
1135 	/* APIC hasn't been mapped yet */
1136 	if (!x2apic_mode && !apic_phys)
1137 		return;
1138 
1139 	maxlvt = lapic_get_maxlvt();
1140 	/*
1141 	 * Masking an LVT entry can trigger a local APIC error
1142 	 * if the vector is zero. Mask LVTERR first to prevent this.
1143 	 */
1144 	if (maxlvt >= 3) {
1145 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1146 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1147 	}
1148 	/*
1149 	 * Careful: we have to set masks only first to deassert
1150 	 * any level-triggered sources.
1151 	 */
1152 	v = apic_read(APIC_LVTT);
1153 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1154 	v = apic_read(APIC_LVT0);
1155 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1156 	v = apic_read(APIC_LVT1);
1157 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1158 	if (maxlvt >= 4) {
1159 		v = apic_read(APIC_LVTPC);
1160 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1161 	}
1162 
1163 	/* lets not touch this if we didn't frob it */
1164 #ifdef CONFIG_X86_THERMAL_VECTOR
1165 	if (maxlvt >= 5) {
1166 		v = apic_read(APIC_LVTTHMR);
1167 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1168 	}
1169 #endif
1170 #ifdef CONFIG_X86_MCE_INTEL
1171 	if (maxlvt >= 6) {
1172 		v = apic_read(APIC_LVTCMCI);
1173 		if (!(v & APIC_LVT_MASKED))
1174 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1175 	}
1176 #endif
1177 
1178 	/*
1179 	 * Clean APIC state for other OSs:
1180 	 */
1181 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1182 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1183 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1184 	if (maxlvt >= 3)
1185 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1186 	if (maxlvt >= 4)
1187 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1188 
1189 	/* Integrated APIC (!82489DX) ? */
1190 	if (lapic_is_integrated()) {
1191 		if (maxlvt > 3)
1192 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1193 			apic_write(APIC_ESR, 0);
1194 		apic_read(APIC_ESR);
1195 	}
1196 }
1197 
1198 /**
1199  * apic_soft_disable - Clears and software disables the local APIC on hotplug
1200  *
1201  * Contrary to disable_local_APIC() this does not touch the enable bit in
1202  * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1203  * bus would require a hardware reset as the APIC would lose track of bus
1204  * arbitration. On systems with FSB delivery APICBASE could be disabled,
1205  * but it has to be guaranteed that no interrupt is sent to the APIC while
1206  * in that state and it's not clear from the SDM whether it still responds
1207  * to INIT/SIPI messages. Stay on the safe side and use software disable.
1208  */
1209 void apic_soft_disable(void)
1210 {
1211 	u32 value;
1212 
1213 	clear_local_APIC();
1214 
1215 	/* Soft disable APIC (implies clearing of registers for 82489DX!). */
1216 	value = apic_read(APIC_SPIV);
1217 	value &= ~APIC_SPIV_APIC_ENABLED;
1218 	apic_write(APIC_SPIV, value);
1219 }
1220 
1221 /**
1222  * disable_local_APIC - clear and disable the local APIC
1223  */
1224 void disable_local_APIC(void)
1225 {
1226 	/* APIC hasn't been mapped yet */
1227 	if (!x2apic_mode && !apic_phys)
1228 		return;
1229 
1230 	apic_soft_disable();
1231 
1232 #ifdef CONFIG_X86_32
1233 	/*
1234 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1235 	 * restore the disabled state.
1236 	 */
1237 	if (enabled_via_apicbase) {
1238 		unsigned int l, h;
1239 
1240 		rdmsr(MSR_IA32_APICBASE, l, h);
1241 		l &= ~MSR_IA32_APICBASE_ENABLE;
1242 		wrmsr(MSR_IA32_APICBASE, l, h);
1243 	}
1244 #endif
1245 }
1246 
1247 /*
1248  * If Linux enabled the LAPIC against the BIOS default disable it down before
1249  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1250  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1251  * for the case where Linux didn't enable the LAPIC.
1252  */
1253 void lapic_shutdown(void)
1254 {
1255 	unsigned long flags;
1256 
1257 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1258 		return;
1259 
1260 	local_irq_save(flags);
1261 
1262 #ifdef CONFIG_X86_32
1263 	if (!enabled_via_apicbase)
1264 		clear_local_APIC();
1265 	else
1266 #endif
1267 		disable_local_APIC();
1268 
1269 
1270 	local_irq_restore(flags);
1271 }
1272 
1273 /**
1274  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1275  */
1276 void __init sync_Arb_IDs(void)
1277 {
1278 	/*
1279 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1280 	 * needed on AMD.
1281 	 */
1282 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1283 		return;
1284 
1285 	/*
1286 	 * Wait for idle.
1287 	 */
1288 	apic_wait_icr_idle();
1289 
1290 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1291 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1292 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1293 }
1294 
1295 enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1296 
1297 static int __init __apic_intr_mode_select(void)
1298 {
1299 	/* Check kernel option */
1300 	if (disable_apic) {
1301 		pr_info("APIC disabled via kernel command line\n");
1302 		return APIC_PIC;
1303 	}
1304 
1305 	/* Check BIOS */
1306 #ifdef CONFIG_X86_64
1307 	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1308 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1309 		disable_apic = 1;
1310 		pr_info("APIC disabled by BIOS\n");
1311 		return APIC_PIC;
1312 	}
1313 #else
1314 	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1315 
1316 	/* Neither 82489DX nor integrated APIC ? */
1317 	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1318 		disable_apic = 1;
1319 		return APIC_PIC;
1320 	}
1321 
1322 	/* If the BIOS pretends there is an integrated APIC ? */
1323 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1324 		APIC_INTEGRATED(boot_cpu_apic_version)) {
1325 		disable_apic = 1;
1326 		pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1327 				       boot_cpu_physical_apicid);
1328 		return APIC_PIC;
1329 	}
1330 #endif
1331 
1332 	/* Check MP table or ACPI MADT configuration */
1333 	if (!smp_found_config) {
1334 		disable_ioapic_support();
1335 		if (!acpi_lapic) {
1336 			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1337 			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1338 		}
1339 		return APIC_VIRTUAL_WIRE;
1340 	}
1341 
1342 #ifdef CONFIG_SMP
1343 	/* If SMP should be disabled, then really disable it! */
1344 	if (!setup_max_cpus) {
1345 		pr_info("APIC: SMP mode deactivated\n");
1346 		return APIC_SYMMETRIC_IO_NO_ROUTING;
1347 	}
1348 
1349 	if (read_apic_id() != boot_cpu_physical_apicid) {
1350 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1351 		     read_apic_id(), boot_cpu_physical_apicid);
1352 		/* Or can we switch back to PIC here? */
1353 	}
1354 #endif
1355 
1356 	return APIC_SYMMETRIC_IO;
1357 }
1358 
1359 /* Select the interrupt delivery mode for the BSP */
1360 void __init apic_intr_mode_select(void)
1361 {
1362 	apic_intr_mode = __apic_intr_mode_select();
1363 }
1364 
1365 /*
1366  * An initial setup of the virtual wire mode.
1367  */
1368 void __init init_bsp_APIC(void)
1369 {
1370 	unsigned int value;
1371 
1372 	/*
1373 	 * Don't do the setup now if we have a SMP BIOS as the
1374 	 * through-I/O-APIC virtual wire mode might be active.
1375 	 */
1376 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1377 		return;
1378 
1379 	/*
1380 	 * Do not trust the local APIC being empty at bootup.
1381 	 */
1382 	clear_local_APIC();
1383 
1384 	/*
1385 	 * Enable APIC.
1386 	 */
1387 	value = apic_read(APIC_SPIV);
1388 	value &= ~APIC_VECTOR_MASK;
1389 	value |= APIC_SPIV_APIC_ENABLED;
1390 
1391 #ifdef CONFIG_X86_32
1392 	/* This bit is reserved on P4/Xeon and should be cleared */
1393 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1394 	    (boot_cpu_data.x86 == 15))
1395 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1396 	else
1397 #endif
1398 		value |= APIC_SPIV_FOCUS_DISABLED;
1399 	value |= SPURIOUS_APIC_VECTOR;
1400 	apic_write(APIC_SPIV, value);
1401 
1402 	/*
1403 	 * Set up the virtual wire mode.
1404 	 */
1405 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1406 	value = APIC_DM_NMI;
1407 	if (!lapic_is_integrated())		/* 82489DX */
1408 		value |= APIC_LVT_LEVEL_TRIGGER;
1409 	if (apic_extnmi == APIC_EXTNMI_NONE)
1410 		value |= APIC_LVT_MASKED;
1411 	apic_write(APIC_LVT1, value);
1412 }
1413 
1414 static void __init apic_bsp_setup(bool upmode);
1415 
1416 /* Init the interrupt delivery mode for the BSP */
1417 void __init apic_intr_mode_init(void)
1418 {
1419 	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1420 
1421 	switch (apic_intr_mode) {
1422 	case APIC_PIC:
1423 		pr_info("APIC: Keep in PIC mode(8259)\n");
1424 		return;
1425 	case APIC_VIRTUAL_WIRE:
1426 		pr_info("APIC: Switch to virtual wire mode setup\n");
1427 		break;
1428 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1429 		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1430 		upmode = true;
1431 		break;
1432 	case APIC_SYMMETRIC_IO:
1433 		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1434 		break;
1435 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1436 		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1437 		break;
1438 	}
1439 
1440 	default_setup_apic_routing();
1441 
1442 	if (x86_platform.apic_post_init)
1443 		x86_platform.apic_post_init();
1444 
1445 	apic_bsp_setup(upmode);
1446 }
1447 
1448 static void lapic_setup_esr(void)
1449 {
1450 	unsigned int oldvalue, value, maxlvt;
1451 
1452 	if (!lapic_is_integrated()) {
1453 		pr_info("No ESR for 82489DX.\n");
1454 		return;
1455 	}
1456 
1457 	if (apic->disable_esr) {
1458 		/*
1459 		 * Something untraceable is creating bad interrupts on
1460 		 * secondary quads ... for the moment, just leave the
1461 		 * ESR disabled - we can't do anything useful with the
1462 		 * errors anyway - mbligh
1463 		 */
1464 		pr_info("Leaving ESR disabled.\n");
1465 		return;
1466 	}
1467 
1468 	maxlvt = lapic_get_maxlvt();
1469 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1470 		apic_write(APIC_ESR, 0);
1471 	oldvalue = apic_read(APIC_ESR);
1472 
1473 	/* enables sending errors */
1474 	value = ERROR_APIC_VECTOR;
1475 	apic_write(APIC_LVTERR, value);
1476 
1477 	/*
1478 	 * spec says clear errors after enabling vector.
1479 	 */
1480 	if (maxlvt > 3)
1481 		apic_write(APIC_ESR, 0);
1482 	value = apic_read(APIC_ESR);
1483 	if (value != oldvalue)
1484 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1485 			"vector: 0x%08x  after: 0x%08x\n",
1486 			oldvalue, value);
1487 }
1488 
1489 #define APIC_IR_REGS		APIC_ISR_NR
1490 #define APIC_IR_BITS		(APIC_IR_REGS * 32)
1491 #define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1492 
1493 union apic_ir {
1494 	unsigned long	map[APIC_IR_MAPSIZE];
1495 	u32		regs[APIC_IR_REGS];
1496 };
1497 
1498 static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1499 {
1500 	int i, bit;
1501 
1502 	/* Read the IRRs */
1503 	for (i = 0; i < APIC_IR_REGS; i++)
1504 		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1505 
1506 	/* Read the ISRs */
1507 	for (i = 0; i < APIC_IR_REGS; i++)
1508 		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1509 
1510 	/*
1511 	 * If the ISR map is not empty. ACK the APIC and run another round
1512 	 * to verify whether a pending IRR has been unblocked and turned
1513 	 * into a ISR.
1514 	 */
1515 	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1516 		/*
1517 		 * There can be multiple ISR bits set when a high priority
1518 		 * interrupt preempted a lower priority one. Issue an ACK
1519 		 * per set bit.
1520 		 */
1521 		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1522 			ack_APIC_irq();
1523 		return true;
1524 	}
1525 
1526 	return !bitmap_empty(irr->map, APIC_IR_BITS);
1527 }
1528 
1529 /*
1530  * After a crash, we no longer service the interrupts and a pending
1531  * interrupt from previous kernel might still have ISR bit set.
1532  *
1533  * Most probably by now the CPU has serviced that pending interrupt and it
1534  * might not have done the ack_APIC_irq() because it thought, interrupt
1535  * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1536  * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1537  * a vector might get locked. It was noticed for timer irq (vector
1538  * 0x31). Issue an extra EOI to clear ISR.
1539  *
1540  * If there are pending IRR bits they turn into ISR bits after a higher
1541  * priority ISR bit has been acked.
1542  */
1543 static void apic_pending_intr_clear(void)
1544 {
1545 	union apic_ir irr, isr;
1546 	unsigned int i;
1547 
1548 	/* 512 loops are way oversized and give the APIC a chance to obey. */
1549 	for (i = 0; i < 512; i++) {
1550 		if (!apic_check_and_ack(&irr, &isr))
1551 			return;
1552 	}
1553 	/* Dump the IRR/ISR content if that failed */
1554 	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1555 }
1556 
1557 /**
1558  * setup_local_APIC - setup the local APIC
1559  *
1560  * Used to setup local APIC while initializing BSP or bringing up APs.
1561  * Always called with preemption disabled.
1562  */
1563 static void setup_local_APIC(void)
1564 {
1565 	int cpu = smp_processor_id();
1566 	unsigned int value;
1567 
1568 	if (disable_apic) {
1569 		disable_ioapic_support();
1570 		return;
1571 	}
1572 
1573 	/*
1574 	 * If this comes from kexec/kcrash the APIC might be enabled in
1575 	 * SPIV. Soft disable it before doing further initialization.
1576 	 */
1577 	value = apic_read(APIC_SPIV);
1578 	value &= ~APIC_SPIV_APIC_ENABLED;
1579 	apic_write(APIC_SPIV, value);
1580 
1581 #ifdef CONFIG_X86_32
1582 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1583 	if (lapic_is_integrated() && apic->disable_esr) {
1584 		apic_write(APIC_ESR, 0);
1585 		apic_write(APIC_ESR, 0);
1586 		apic_write(APIC_ESR, 0);
1587 		apic_write(APIC_ESR, 0);
1588 	}
1589 #endif
1590 	/*
1591 	 * Double-check whether this APIC is really registered.
1592 	 * This is meaningless in clustered apic mode, so we skip it.
1593 	 */
1594 	BUG_ON(!apic->apic_id_registered());
1595 
1596 	/*
1597 	 * Intel recommends to set DFR, LDR and TPR before enabling
1598 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1599 	 * document number 292116).  So here it goes...
1600 	 */
1601 	apic->init_apic_ldr();
1602 
1603 #ifdef CONFIG_X86_32
1604 	if (apic->dest_mode_logical) {
1605 		int logical_apicid, ldr_apicid;
1606 
1607 		/*
1608 		 * APIC LDR is initialized.  If logical_apicid mapping was
1609 		 * initialized during get_smp_config(), make sure it matches
1610 		 * the actual value.
1611 		 */
1612 		logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1613 		ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1614 		if (logical_apicid != BAD_APICID)
1615 			WARN_ON(logical_apicid != ldr_apicid);
1616 		/* Always use the value from LDR. */
1617 		early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1618 	}
1619 #endif
1620 
1621 	/*
1622 	 * Set Task Priority to 'accept all except vectors 0-31'.  An APIC
1623 	 * vector in the 16-31 range could be delivered if TPR == 0, but we
1624 	 * would think it's an exception and terrible things will happen.  We
1625 	 * never change this later on.
1626 	 */
1627 	value = apic_read(APIC_TASKPRI);
1628 	value &= ~APIC_TPRI_MASK;
1629 	value |= 0x10;
1630 	apic_write(APIC_TASKPRI, value);
1631 
1632 	/* Clear eventually stale ISR/IRR bits */
1633 	apic_pending_intr_clear();
1634 
1635 	/*
1636 	 * Now that we are all set up, enable the APIC
1637 	 */
1638 	value = apic_read(APIC_SPIV);
1639 	value &= ~APIC_VECTOR_MASK;
1640 	/*
1641 	 * Enable APIC
1642 	 */
1643 	value |= APIC_SPIV_APIC_ENABLED;
1644 
1645 #ifdef CONFIG_X86_32
1646 	/*
1647 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1648 	 * certain networking cards. If high frequency interrupts are
1649 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1650 	 * entry is masked/unmasked at a high rate as well then sooner or
1651 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1652 	 * from the device. If focus CPU is disabled then the hang goes
1653 	 * away, oh well :-(
1654 	 *
1655 	 * [ This bug can be reproduced easily with a level-triggered
1656 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1657 	 *   BX chipset. ]
1658 	 */
1659 	/*
1660 	 * Actually disabling the focus CPU check just makes the hang less
1661 	 * frequent as it makes the interrupt distribution model be more
1662 	 * like LRU than MRU (the short-term load is more even across CPUs).
1663 	 */
1664 
1665 	/*
1666 	 * - enable focus processor (bit==0)
1667 	 * - 64bit mode always use processor focus
1668 	 *   so no need to set it
1669 	 */
1670 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1671 #endif
1672 
1673 	/*
1674 	 * Set spurious IRQ vector
1675 	 */
1676 	value |= SPURIOUS_APIC_VECTOR;
1677 	apic_write(APIC_SPIV, value);
1678 
1679 	perf_events_lapic_init();
1680 
1681 	/*
1682 	 * Set up LVT0, LVT1:
1683 	 *
1684 	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1685 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1686 	 * we delegate interrupts to the 8259A.
1687 	 */
1688 	/*
1689 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1690 	 */
1691 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1692 	if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1693 		value = APIC_DM_EXTINT;
1694 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1695 	} else {
1696 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1697 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1698 	}
1699 	apic_write(APIC_LVT0, value);
1700 
1701 	/*
1702 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1703 	 * modified by apic_extnmi= boot option.
1704 	 */
1705 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1706 	    apic_extnmi == APIC_EXTNMI_ALL)
1707 		value = APIC_DM_NMI;
1708 	else
1709 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1710 
1711 	/* Is 82489DX ? */
1712 	if (!lapic_is_integrated())
1713 		value |= APIC_LVT_LEVEL_TRIGGER;
1714 	apic_write(APIC_LVT1, value);
1715 
1716 #ifdef CONFIG_X86_MCE_INTEL
1717 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1718 	if (!cpu)
1719 		cmci_recheck();
1720 #endif
1721 }
1722 
1723 static void end_local_APIC_setup(void)
1724 {
1725 	lapic_setup_esr();
1726 
1727 #ifdef CONFIG_X86_32
1728 	{
1729 		unsigned int value;
1730 		/* Disable the local apic timer */
1731 		value = apic_read(APIC_LVTT);
1732 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1733 		apic_write(APIC_LVTT, value);
1734 	}
1735 #endif
1736 
1737 	apic_pm_activate();
1738 }
1739 
1740 /*
1741  * APIC setup function for application processors. Called from smpboot.c
1742  */
1743 void apic_ap_setup(void)
1744 {
1745 	setup_local_APIC();
1746 	end_local_APIC_setup();
1747 }
1748 
1749 #ifdef CONFIG_X86_X2APIC
1750 int x2apic_mode;
1751 EXPORT_SYMBOL_GPL(x2apic_mode);
1752 
1753 enum {
1754 	X2APIC_OFF,
1755 	X2APIC_DISABLED,
1756 	/* All states below here have X2APIC enabled */
1757 	X2APIC_ON,
1758 	X2APIC_ON_LOCKED
1759 };
1760 static int x2apic_state;
1761 
1762 static bool x2apic_hw_locked(void)
1763 {
1764 	u64 ia32_cap;
1765 	u64 msr;
1766 
1767 	ia32_cap = x86_read_arch_cap_msr();
1768 	if (ia32_cap & ARCH_CAP_XAPIC_DISABLE) {
1769 		rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1770 		return (msr & LEGACY_XAPIC_DISABLED);
1771 	}
1772 	return false;
1773 }
1774 
1775 static void __x2apic_disable(void)
1776 {
1777 	u64 msr;
1778 
1779 	if (!boot_cpu_has(X86_FEATURE_APIC))
1780 		return;
1781 
1782 	rdmsrl(MSR_IA32_APICBASE, msr);
1783 	if (!(msr & X2APIC_ENABLE))
1784 		return;
1785 	/* Disable xapic and x2apic first and then reenable xapic mode */
1786 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1787 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1788 	printk_once(KERN_INFO "x2apic disabled\n");
1789 }
1790 
1791 static void __x2apic_enable(void)
1792 {
1793 	u64 msr;
1794 
1795 	rdmsrl(MSR_IA32_APICBASE, msr);
1796 	if (msr & X2APIC_ENABLE)
1797 		return;
1798 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1799 	printk_once(KERN_INFO "x2apic enabled\n");
1800 }
1801 
1802 static int __init setup_nox2apic(char *str)
1803 {
1804 	if (x2apic_enabled()) {
1805 		int apicid = native_apic_msr_read(APIC_ID);
1806 
1807 		if (apicid >= 255) {
1808 			pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1809 				apicid);
1810 			return 0;
1811 		}
1812 		if (x2apic_hw_locked()) {
1813 			pr_warn("APIC locked in x2apic mode, can't disable\n");
1814 			return 0;
1815 		}
1816 		pr_warn("x2apic already enabled.\n");
1817 		__x2apic_disable();
1818 	}
1819 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1820 	x2apic_state = X2APIC_DISABLED;
1821 	x2apic_mode = 0;
1822 	return 0;
1823 }
1824 early_param("nox2apic", setup_nox2apic);
1825 
1826 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
1827 void x2apic_setup(void)
1828 {
1829 	/*
1830 	 * Try to make the AP's APIC state match that of the BSP,  but if the
1831 	 * BSP is unlocked and the AP is locked then there is a state mismatch.
1832 	 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1833 	 * trying to be turned off.
1834 	 */
1835 	if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1836 		pr_warn("x2apic lock mismatch between BSP and AP.\n");
1837 	/*
1838 	 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1839 	 * from BIOS.
1840 	 */
1841 	if (x2apic_state < X2APIC_ON) {
1842 		__x2apic_disable();
1843 		return;
1844 	}
1845 	__x2apic_enable();
1846 }
1847 
1848 static __init void x2apic_disable(void)
1849 {
1850 	u32 x2apic_id, state = x2apic_state;
1851 
1852 	x2apic_mode = 0;
1853 	x2apic_state = X2APIC_DISABLED;
1854 
1855 	if (state != X2APIC_ON)
1856 		return;
1857 
1858 	x2apic_id = read_apic_id();
1859 	if (x2apic_id >= 255)
1860 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1861 
1862 	if (x2apic_hw_locked()) {
1863 		pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1864 		return;
1865 	}
1866 
1867 	__x2apic_disable();
1868 	register_lapic_address(mp_lapic_addr);
1869 }
1870 
1871 static __init void x2apic_enable(void)
1872 {
1873 	if (x2apic_state != X2APIC_OFF)
1874 		return;
1875 
1876 	x2apic_mode = 1;
1877 	x2apic_state = X2APIC_ON;
1878 	__x2apic_enable();
1879 }
1880 
1881 static __init void try_to_enable_x2apic(int remap_mode)
1882 {
1883 	if (x2apic_state == X2APIC_DISABLED)
1884 		return;
1885 
1886 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1887 		u32 apic_limit = 255;
1888 
1889 		/*
1890 		 * Using X2APIC without IR is not architecturally supported
1891 		 * on bare metal but may be supported in guests.
1892 		 */
1893 		if (!x86_init.hyper.x2apic_available()) {
1894 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1895 			x2apic_disable();
1896 			return;
1897 		}
1898 
1899 		/*
1900 		 * If the hypervisor supports extended destination ID in
1901 		 * MSI, that increases the maximum APIC ID that can be
1902 		 * used for non-remapped IRQ domains.
1903 		 */
1904 		if (x86_init.hyper.msi_ext_dest_id()) {
1905 			virt_ext_dest_id = 1;
1906 			apic_limit = 32767;
1907 		}
1908 
1909 		/*
1910 		 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1911 		 * in physical mode, and CPUs with an APIC ID that cannot
1912 		 * be addressed must not be brought online.
1913 		 */
1914 		x2apic_set_max_apicid(apic_limit);
1915 		x2apic_phys = 1;
1916 	}
1917 	x2apic_enable();
1918 }
1919 
1920 void __init check_x2apic(void)
1921 {
1922 	if (x2apic_enabled()) {
1923 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1924 		x2apic_mode = 1;
1925 		if (x2apic_hw_locked())
1926 			x2apic_state = X2APIC_ON_LOCKED;
1927 		else
1928 			x2apic_state = X2APIC_ON;
1929 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1930 		x2apic_state = X2APIC_DISABLED;
1931 	}
1932 }
1933 #else /* CONFIG_X86_X2APIC */
1934 static int __init validate_x2apic(void)
1935 {
1936 	if (!apic_is_x2apic_enabled())
1937 		return 0;
1938 	/*
1939 	 * Checkme: Can we simply turn off x2apic here instead of panic?
1940 	 */
1941 	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1942 }
1943 early_initcall(validate_x2apic);
1944 
1945 static inline void try_to_enable_x2apic(int remap_mode) { }
1946 static inline void __x2apic_enable(void) { }
1947 #endif /* !CONFIG_X86_X2APIC */
1948 
1949 void __init enable_IR_x2apic(void)
1950 {
1951 	unsigned long flags;
1952 	int ret, ir_stat;
1953 
1954 	if (skip_ioapic_setup) {
1955 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1956 		return;
1957 	}
1958 
1959 	ir_stat = irq_remapping_prepare();
1960 	if (ir_stat < 0 && !x2apic_supported())
1961 		return;
1962 
1963 	ret = save_ioapic_entries();
1964 	if (ret) {
1965 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1966 		return;
1967 	}
1968 
1969 	local_irq_save(flags);
1970 	legacy_pic->mask_all();
1971 	mask_ioapic_entries();
1972 
1973 	/* If irq_remapping_prepare() succeeded, try to enable it */
1974 	if (ir_stat >= 0)
1975 		ir_stat = irq_remapping_enable();
1976 	/* ir_stat contains the remap mode or an error code */
1977 	try_to_enable_x2apic(ir_stat);
1978 
1979 	if (ir_stat < 0)
1980 		restore_ioapic_entries();
1981 	legacy_pic->restore_mask();
1982 	local_irq_restore(flags);
1983 }
1984 
1985 #ifdef CONFIG_X86_64
1986 /*
1987  * Detect and enable local APICs on non-SMP boards.
1988  * Original code written by Keir Fraser.
1989  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1990  * not correctly set up (usually the APIC timer won't work etc.)
1991  */
1992 static int __init detect_init_APIC(void)
1993 {
1994 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1995 		pr_info("No local APIC present\n");
1996 		return -1;
1997 	}
1998 
1999 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
2000 	return 0;
2001 }
2002 #else
2003 
2004 static int __init apic_verify(void)
2005 {
2006 	u32 features, h, l;
2007 
2008 	/*
2009 	 * The APIC feature bit should now be enabled
2010 	 * in `cpuid'
2011 	 */
2012 	features = cpuid_edx(1);
2013 	if (!(features & (1 << X86_FEATURE_APIC))) {
2014 		pr_warn("Could not enable APIC!\n");
2015 		return -1;
2016 	}
2017 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
2018 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
2019 
2020 	/* The BIOS may have set up the APIC at some other address */
2021 	if (boot_cpu_data.x86 >= 6) {
2022 		rdmsr(MSR_IA32_APICBASE, l, h);
2023 		if (l & MSR_IA32_APICBASE_ENABLE)
2024 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
2025 	}
2026 
2027 	pr_info("Found and enabled local APIC!\n");
2028 	return 0;
2029 }
2030 
2031 int __init apic_force_enable(unsigned long addr)
2032 {
2033 	u32 h, l;
2034 
2035 	if (disable_apic)
2036 		return -1;
2037 
2038 	/*
2039 	 * Some BIOSes disable the local APIC in the APIC_BASE
2040 	 * MSR. This can only be done in software for Intel P6 or later
2041 	 * and AMD K7 (Model > 1) or later.
2042 	 */
2043 	if (boot_cpu_data.x86 >= 6) {
2044 		rdmsr(MSR_IA32_APICBASE, l, h);
2045 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2046 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2047 			l &= ~MSR_IA32_APICBASE_BASE;
2048 			l |= MSR_IA32_APICBASE_ENABLE | addr;
2049 			wrmsr(MSR_IA32_APICBASE, l, h);
2050 			enabled_via_apicbase = 1;
2051 		}
2052 	}
2053 	return apic_verify();
2054 }
2055 
2056 /*
2057  * Detect and initialize APIC
2058  */
2059 static int __init detect_init_APIC(void)
2060 {
2061 	/* Disabled by kernel option? */
2062 	if (disable_apic)
2063 		return -1;
2064 
2065 	switch (boot_cpu_data.x86_vendor) {
2066 	case X86_VENDOR_AMD:
2067 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2068 		    (boot_cpu_data.x86 >= 15))
2069 			break;
2070 		goto no_apic;
2071 	case X86_VENDOR_HYGON:
2072 		break;
2073 	case X86_VENDOR_INTEL:
2074 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2075 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2076 			break;
2077 		goto no_apic;
2078 	default:
2079 		goto no_apic;
2080 	}
2081 
2082 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2083 		/*
2084 		 * Over-ride BIOS and try to enable the local APIC only if
2085 		 * "lapic" specified.
2086 		 */
2087 		if (!force_enable_local_apic) {
2088 			pr_info("Local APIC disabled by BIOS -- "
2089 				"you can enable it with \"lapic\"\n");
2090 			return -1;
2091 		}
2092 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2093 			return -1;
2094 	} else {
2095 		if (apic_verify())
2096 			return -1;
2097 	}
2098 
2099 	apic_pm_activate();
2100 
2101 	return 0;
2102 
2103 no_apic:
2104 	pr_info("No local APIC present or hardware disabled\n");
2105 	return -1;
2106 }
2107 #endif
2108 
2109 /**
2110  * init_apic_mappings - initialize APIC mappings
2111  */
2112 void __init init_apic_mappings(void)
2113 {
2114 	unsigned int new_apicid;
2115 
2116 	if (apic_validate_deadline_timer())
2117 		pr_info("TSC deadline timer available\n");
2118 
2119 	if (x2apic_mode) {
2120 		boot_cpu_physical_apicid = read_apic_id();
2121 		return;
2122 	}
2123 
2124 	/* If no local APIC can be found return early */
2125 	if (!smp_found_config && detect_init_APIC()) {
2126 		/* lets NOP'ify apic operations */
2127 		pr_info("APIC: disable apic facility\n");
2128 		apic_disable();
2129 	} else {
2130 		apic_phys = mp_lapic_addr;
2131 
2132 		/*
2133 		 * If the system has ACPI MADT tables or MP info, the LAPIC
2134 		 * address is already registered.
2135 		 */
2136 		if (!acpi_lapic && !smp_found_config)
2137 			register_lapic_address(apic_phys);
2138 	}
2139 
2140 	/*
2141 	 * Fetch the APIC ID of the BSP in case we have a
2142 	 * default configuration (or the MP table is broken).
2143 	 */
2144 	new_apicid = read_apic_id();
2145 	if (boot_cpu_physical_apicid != new_apicid) {
2146 		boot_cpu_physical_apicid = new_apicid;
2147 		/*
2148 		 * yeah -- we lie about apic_version
2149 		 * in case if apic was disabled via boot option
2150 		 * but it's not a problem for SMP compiled kernel
2151 		 * since apic_intr_mode_select is prepared for such
2152 		 * a case and disable smp mode
2153 		 */
2154 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2155 	}
2156 }
2157 
2158 void __init register_lapic_address(unsigned long address)
2159 {
2160 	mp_lapic_addr = address;
2161 
2162 	if (!x2apic_mode) {
2163 		set_fixmap_nocache(FIX_APIC_BASE, address);
2164 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2165 			    APIC_BASE, address);
2166 	}
2167 	if (boot_cpu_physical_apicid == -1U) {
2168 		boot_cpu_physical_apicid  = read_apic_id();
2169 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2170 	}
2171 }
2172 
2173 /*
2174  * Local APIC interrupts
2175  */
2176 
2177 /*
2178  * Common handling code for spurious_interrupt and spurious_vector entry
2179  * points below. No point in allowing the compiler to inline it twice.
2180  */
2181 static noinline void handle_spurious_interrupt(u8 vector)
2182 {
2183 	u32 v;
2184 
2185 	trace_spurious_apic_entry(vector);
2186 
2187 	inc_irq_stat(irq_spurious_count);
2188 
2189 	/*
2190 	 * If this is a spurious interrupt then do not acknowledge
2191 	 */
2192 	if (vector == SPURIOUS_APIC_VECTOR) {
2193 		/* See SDM vol 3 */
2194 		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2195 			smp_processor_id());
2196 		goto out;
2197 	}
2198 
2199 	/*
2200 	 * If it is a vectored one, verify it's set in the ISR. If set,
2201 	 * acknowledge it.
2202 	 */
2203 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2204 	if (v & (1 << (vector & 0x1f))) {
2205 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2206 			vector, smp_processor_id());
2207 		ack_APIC_irq();
2208 	} else {
2209 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2210 			vector, smp_processor_id());
2211 	}
2212 out:
2213 	trace_spurious_apic_exit(vector);
2214 }
2215 
2216 /**
2217  * spurious_interrupt - Catch all for interrupts raised on unused vectors
2218  * @regs:	Pointer to pt_regs on stack
2219  * @vector:	The vector number
2220  *
2221  * This is invoked from ASM entry code to catch all interrupts which
2222  * trigger on an entry which is routed to the common_spurious idtentry
2223  * point.
2224  */
2225 DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2226 {
2227 	handle_spurious_interrupt(vector);
2228 }
2229 
2230 DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2231 {
2232 	handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2233 }
2234 
2235 /*
2236  * This interrupt should never happen with our APIC/SMP architecture
2237  */
2238 DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2239 {
2240 	static const char * const error_interrupt_reason[] = {
2241 		"Send CS error",		/* APIC Error Bit 0 */
2242 		"Receive CS error",		/* APIC Error Bit 1 */
2243 		"Send accept error",		/* APIC Error Bit 2 */
2244 		"Receive accept error",		/* APIC Error Bit 3 */
2245 		"Redirectable IPI",		/* APIC Error Bit 4 */
2246 		"Send illegal vector",		/* APIC Error Bit 5 */
2247 		"Received illegal vector",	/* APIC Error Bit 6 */
2248 		"Illegal register address",	/* APIC Error Bit 7 */
2249 	};
2250 	u32 v, i = 0;
2251 
2252 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2253 
2254 	/* First tickle the hardware, only then report what went on. -- REW */
2255 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2256 		apic_write(APIC_ESR, 0);
2257 	v = apic_read(APIC_ESR);
2258 	ack_APIC_irq();
2259 	atomic_inc(&irq_err_count);
2260 
2261 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2262 		    smp_processor_id(), v);
2263 
2264 	v &= 0xff;
2265 	while (v) {
2266 		if (v & 0x1)
2267 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2268 		i++;
2269 		v >>= 1;
2270 	}
2271 
2272 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2273 
2274 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2275 }
2276 
2277 /**
2278  * connect_bsp_APIC - attach the APIC to the interrupt system
2279  */
2280 static void __init connect_bsp_APIC(void)
2281 {
2282 #ifdef CONFIG_X86_32
2283 	if (pic_mode) {
2284 		/*
2285 		 * Do not trust the local APIC being empty at bootup.
2286 		 */
2287 		clear_local_APIC();
2288 		/*
2289 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2290 		 * local APIC to INT and NMI lines.
2291 		 */
2292 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2293 				"enabling APIC mode.\n");
2294 		imcr_pic_to_apic();
2295 	}
2296 #endif
2297 }
2298 
2299 /**
2300  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2301  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2302  *
2303  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2304  * APIC is disabled.
2305  */
2306 void disconnect_bsp_APIC(int virt_wire_setup)
2307 {
2308 	unsigned int value;
2309 
2310 #ifdef CONFIG_X86_32
2311 	if (pic_mode) {
2312 		/*
2313 		 * Put the board back into PIC mode (has an effect only on
2314 		 * certain older boards).  Note that APIC interrupts, including
2315 		 * IPIs, won't work beyond this point!  The only exception are
2316 		 * INIT IPIs.
2317 		 */
2318 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2319 				"entering PIC mode.\n");
2320 		imcr_apic_to_pic();
2321 		return;
2322 	}
2323 #endif
2324 
2325 	/* Go back to Virtual Wire compatibility mode */
2326 
2327 	/* For the spurious interrupt use vector F, and enable it */
2328 	value = apic_read(APIC_SPIV);
2329 	value &= ~APIC_VECTOR_MASK;
2330 	value |= APIC_SPIV_APIC_ENABLED;
2331 	value |= 0xf;
2332 	apic_write(APIC_SPIV, value);
2333 
2334 	if (!virt_wire_setup) {
2335 		/*
2336 		 * For LVT0 make it edge triggered, active high,
2337 		 * external and enabled
2338 		 */
2339 		value = apic_read(APIC_LVT0);
2340 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2341 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2342 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2343 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2344 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2345 		apic_write(APIC_LVT0, value);
2346 	} else {
2347 		/* Disable LVT0 */
2348 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2349 	}
2350 
2351 	/*
2352 	 * For LVT1 make it edge triggered, active high,
2353 	 * nmi and enabled
2354 	 */
2355 	value = apic_read(APIC_LVT1);
2356 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2357 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2358 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2359 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2360 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2361 	apic_write(APIC_LVT1, value);
2362 }
2363 
2364 /*
2365  * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2366  * contiguously, it equals to current allocated max logical CPU ID plus 1.
2367  * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2368  * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2369  *
2370  * NOTE: Reserve 0 for BSP.
2371  */
2372 static int nr_logical_cpuids = 1;
2373 
2374 /*
2375  * Used to store mapping between logical CPU IDs and APIC IDs.
2376  */
2377 static int cpuid_to_apicid[] = {
2378 	[0 ... NR_CPUS - 1] = -1,
2379 };
2380 
2381 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
2382 {
2383 	return phys_id == cpuid_to_apicid[cpu];
2384 }
2385 
2386 #ifdef CONFIG_SMP
2387 /**
2388  * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2389  * @apicid: APIC ID to check
2390  */
2391 bool apic_id_is_primary_thread(unsigned int apicid)
2392 {
2393 	u32 mask;
2394 
2395 	if (smp_num_siblings == 1)
2396 		return true;
2397 	/* Isolate the SMT bit(s) in the APICID and check for 0 */
2398 	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2399 	return !(apicid & mask);
2400 }
2401 #endif
2402 
2403 /*
2404  * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2405  * and cpuid_to_apicid[] synchronized.
2406  */
2407 static int allocate_logical_cpuid(int apicid)
2408 {
2409 	int i;
2410 
2411 	/*
2412 	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2413 	 * check if the kernel has allocated a cpuid for it.
2414 	 */
2415 	for (i = 0; i < nr_logical_cpuids; i++) {
2416 		if (cpuid_to_apicid[i] == apicid)
2417 			return i;
2418 	}
2419 
2420 	/* Allocate a new cpuid. */
2421 	if (nr_logical_cpuids >= nr_cpu_ids) {
2422 		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2423 			     "Processor %d/0x%x and the rest are ignored.\n",
2424 			     nr_cpu_ids, nr_logical_cpuids, apicid);
2425 		return -EINVAL;
2426 	}
2427 
2428 	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2429 	return nr_logical_cpuids++;
2430 }
2431 
2432 int generic_processor_info(int apicid, int version)
2433 {
2434 	int cpu, max = nr_cpu_ids;
2435 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2436 				phys_cpu_present_map);
2437 
2438 	/*
2439 	 * boot_cpu_physical_apicid is designed to have the apicid
2440 	 * returned by read_apic_id(), i.e, the apicid of the
2441 	 * currently booting-up processor. However, on some platforms,
2442 	 * it is temporarily modified by the apicid reported as BSP
2443 	 * through MP table. Concretely:
2444 	 *
2445 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2446 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2447 	 *
2448 	 * This function is executed with the modified
2449 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2450 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2451 	 *
2452 	 * Since fixing handling of boot_cpu_physical_apicid requires
2453 	 * another discussion and tests on each platform, we leave it
2454 	 * for now and here we use read_apic_id() directly in this
2455 	 * function, generic_processor_info().
2456 	 */
2457 	if (disabled_cpu_apicid != BAD_APICID &&
2458 	    disabled_cpu_apicid != read_apic_id() &&
2459 	    disabled_cpu_apicid == apicid) {
2460 		int thiscpu = num_processors + disabled_cpus;
2461 
2462 		pr_warn("APIC: Disabling requested cpu."
2463 			" Processor %d/0x%x ignored.\n", thiscpu, apicid);
2464 
2465 		disabled_cpus++;
2466 		return -ENODEV;
2467 	}
2468 
2469 	/*
2470 	 * If boot cpu has not been detected yet, then only allow upto
2471 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2472 	 */
2473 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2474 	    apicid != boot_cpu_physical_apicid) {
2475 		int thiscpu = max + disabled_cpus - 1;
2476 
2477 		pr_warn("APIC: NR_CPUS/possible_cpus limit of %i almost"
2478 			" reached. Keeping one slot for boot cpu."
2479 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2480 
2481 		disabled_cpus++;
2482 		return -ENODEV;
2483 	}
2484 
2485 	if (num_processors >= nr_cpu_ids) {
2486 		int thiscpu = max + disabled_cpus;
2487 
2488 		pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2489 			"Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2490 
2491 		disabled_cpus++;
2492 		return -EINVAL;
2493 	}
2494 
2495 	if (apicid == boot_cpu_physical_apicid) {
2496 		/*
2497 		 * x86_bios_cpu_apicid is required to have processors listed
2498 		 * in same order as logical cpu numbers. Hence the first
2499 		 * entry is BSP, and so on.
2500 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2501 		 * for BSP.
2502 		 */
2503 		cpu = 0;
2504 
2505 		/* Logical cpuid 0 is reserved for BSP. */
2506 		cpuid_to_apicid[0] = apicid;
2507 	} else {
2508 		cpu = allocate_logical_cpuid(apicid);
2509 		if (cpu < 0) {
2510 			disabled_cpus++;
2511 			return -EINVAL;
2512 		}
2513 	}
2514 
2515 	/*
2516 	 * Validate version
2517 	 */
2518 	if (version == 0x0) {
2519 		pr_warn("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2520 			cpu, apicid);
2521 		version = 0x10;
2522 	}
2523 
2524 	if (version != boot_cpu_apic_version) {
2525 		pr_warn("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2526 			boot_cpu_apic_version, cpu, version);
2527 	}
2528 
2529 	if (apicid > max_physical_apicid)
2530 		max_physical_apicid = apicid;
2531 
2532 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2533 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2534 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2535 #endif
2536 #ifdef CONFIG_X86_32
2537 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2538 		apic->x86_32_early_logical_apicid(cpu);
2539 #endif
2540 	set_cpu_possible(cpu, true);
2541 	physid_set(apicid, phys_cpu_present_map);
2542 	set_cpu_present(cpu, true);
2543 	num_processors++;
2544 
2545 	return cpu;
2546 }
2547 
2548 int hard_smp_processor_id(void)
2549 {
2550 	return read_apic_id();
2551 }
2552 
2553 void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2554 			   bool dmar)
2555 {
2556 	memset(msg, 0, sizeof(*msg));
2557 
2558 	msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2559 	msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2560 	msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2561 
2562 	msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2563 	msg->arch_data.vector = cfg->vector;
2564 
2565 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2566 	/*
2567 	 * Only the IOMMU itself can use the trick of putting destination
2568 	 * APIC ID into the high bits of the address. Anything else would
2569 	 * just be writing to memory if it tried that, and needs IR to
2570 	 * address APICs which can't be addressed in the normal 32-bit
2571 	 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2572 	 * some hypervisors allow the extended destination ID field in bits
2573 	 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2574 	 */
2575 	if (dmar)
2576 		msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2577 	else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2578 		msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2579 	else
2580 		WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2581 }
2582 
2583 u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2584 {
2585 	u32 dest = msg->arch_addr_lo.destid_0_7;
2586 
2587 	if (extid)
2588 		dest |= msg->arch_addr_hi.destid_8_31 << 8;
2589 	return dest;
2590 }
2591 EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2592 
2593 #ifdef CONFIG_X86_64
2594 void __init acpi_wake_cpu_handler_update(wakeup_cpu_handler handler)
2595 {
2596 	struct apic **drv;
2597 
2598 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++)
2599 		(*drv)->wakeup_secondary_cpu_64 = handler;
2600 }
2601 #endif
2602 
2603 /*
2604  * Override the generic EOI implementation with an optimized version.
2605  * Only called during early boot when only one CPU is active and with
2606  * interrupts disabled, so we know this does not race with actual APIC driver
2607  * use.
2608  */
2609 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2610 {
2611 	struct apic **drv;
2612 
2613 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2614 		/* Should happen once for each apic */
2615 		WARN_ON((*drv)->eoi_write == eoi_write);
2616 		(*drv)->native_eoi_write = (*drv)->eoi_write;
2617 		(*drv)->eoi_write = eoi_write;
2618 	}
2619 }
2620 
2621 static void __init apic_bsp_up_setup(void)
2622 {
2623 #ifdef CONFIG_X86_64
2624 	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2625 #else
2626 	/*
2627 	 * Hack: In case of kdump, after a crash, kernel might be booting
2628 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2629 	 * might be zero if read from MP tables. Get it from LAPIC.
2630 	 */
2631 # ifdef CONFIG_CRASH_DUMP
2632 	boot_cpu_physical_apicid = read_apic_id();
2633 # endif
2634 #endif
2635 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2636 }
2637 
2638 /**
2639  * apic_bsp_setup - Setup function for local apic and io-apic
2640  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2641  */
2642 static void __init apic_bsp_setup(bool upmode)
2643 {
2644 	connect_bsp_APIC();
2645 	if (upmode)
2646 		apic_bsp_up_setup();
2647 	setup_local_APIC();
2648 
2649 	enable_IO_APIC();
2650 	end_local_APIC_setup();
2651 	irq_remap_enable_fault_handling();
2652 	setup_IO_APIC();
2653 	lapic_update_legacy_vectors();
2654 }
2655 
2656 #ifdef CONFIG_UP_LATE_INIT
2657 void __init up_late_init(void)
2658 {
2659 	if (apic_intr_mode == APIC_PIC)
2660 		return;
2661 
2662 	/* Setup local timer */
2663 	x86_init.timers.setup_percpu_clockev();
2664 }
2665 #endif
2666 
2667 /*
2668  * Power management
2669  */
2670 #ifdef CONFIG_PM
2671 
2672 static struct {
2673 	/*
2674 	 * 'active' is true if the local APIC was enabled by us and
2675 	 * not the BIOS; this signifies that we are also responsible
2676 	 * for disabling it before entering apm/acpi suspend
2677 	 */
2678 	int active;
2679 	/* r/w apic fields */
2680 	unsigned int apic_id;
2681 	unsigned int apic_taskpri;
2682 	unsigned int apic_ldr;
2683 	unsigned int apic_dfr;
2684 	unsigned int apic_spiv;
2685 	unsigned int apic_lvtt;
2686 	unsigned int apic_lvtpc;
2687 	unsigned int apic_lvt0;
2688 	unsigned int apic_lvt1;
2689 	unsigned int apic_lvterr;
2690 	unsigned int apic_tmict;
2691 	unsigned int apic_tdcr;
2692 	unsigned int apic_thmr;
2693 	unsigned int apic_cmci;
2694 } apic_pm_state;
2695 
2696 static int lapic_suspend(void)
2697 {
2698 	unsigned long flags;
2699 	int maxlvt;
2700 
2701 	if (!apic_pm_state.active)
2702 		return 0;
2703 
2704 	maxlvt = lapic_get_maxlvt();
2705 
2706 	apic_pm_state.apic_id = apic_read(APIC_ID);
2707 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2708 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2709 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2710 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2711 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2712 	if (maxlvt >= 4)
2713 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2714 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2715 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2716 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2717 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2718 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2719 #ifdef CONFIG_X86_THERMAL_VECTOR
2720 	if (maxlvt >= 5)
2721 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2722 #endif
2723 #ifdef CONFIG_X86_MCE_INTEL
2724 	if (maxlvt >= 6)
2725 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2726 #endif
2727 
2728 	local_irq_save(flags);
2729 
2730 	/*
2731 	 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2732 	 * entries on some implementations.
2733 	 */
2734 	mask_ioapic_entries();
2735 
2736 	disable_local_APIC();
2737 
2738 	irq_remapping_disable();
2739 
2740 	local_irq_restore(flags);
2741 	return 0;
2742 }
2743 
2744 static void lapic_resume(void)
2745 {
2746 	unsigned int l, h;
2747 	unsigned long flags;
2748 	int maxlvt;
2749 
2750 	if (!apic_pm_state.active)
2751 		return;
2752 
2753 	local_irq_save(flags);
2754 
2755 	/*
2756 	 * IO-APIC and PIC have their own resume routines.
2757 	 * We just mask them here to make sure the interrupt
2758 	 * subsystem is completely quiet while we enable x2apic
2759 	 * and interrupt-remapping.
2760 	 */
2761 	mask_ioapic_entries();
2762 	legacy_pic->mask_all();
2763 
2764 	if (x2apic_mode) {
2765 		__x2apic_enable();
2766 	} else {
2767 		/*
2768 		 * Make sure the APICBASE points to the right address
2769 		 *
2770 		 * FIXME! This will be wrong if we ever support suspend on
2771 		 * SMP! We'll need to do this as part of the CPU restore!
2772 		 */
2773 		if (boot_cpu_data.x86 >= 6) {
2774 			rdmsr(MSR_IA32_APICBASE, l, h);
2775 			l &= ~MSR_IA32_APICBASE_BASE;
2776 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2777 			wrmsr(MSR_IA32_APICBASE, l, h);
2778 		}
2779 	}
2780 
2781 	maxlvt = lapic_get_maxlvt();
2782 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2783 	apic_write(APIC_ID, apic_pm_state.apic_id);
2784 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2785 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2786 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2787 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2788 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2789 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2790 #ifdef CONFIG_X86_THERMAL_VECTOR
2791 	if (maxlvt >= 5)
2792 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2793 #endif
2794 #ifdef CONFIG_X86_MCE_INTEL
2795 	if (maxlvt >= 6)
2796 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2797 #endif
2798 	if (maxlvt >= 4)
2799 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2800 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2801 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2802 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2803 	apic_write(APIC_ESR, 0);
2804 	apic_read(APIC_ESR);
2805 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2806 	apic_write(APIC_ESR, 0);
2807 	apic_read(APIC_ESR);
2808 
2809 	irq_remapping_reenable(x2apic_mode);
2810 
2811 	local_irq_restore(flags);
2812 }
2813 
2814 /*
2815  * This device has no shutdown method - fully functioning local APICs
2816  * are needed on every CPU up until machine_halt/restart/poweroff.
2817  */
2818 
2819 static struct syscore_ops lapic_syscore_ops = {
2820 	.resume		= lapic_resume,
2821 	.suspend	= lapic_suspend,
2822 };
2823 
2824 static void apic_pm_activate(void)
2825 {
2826 	apic_pm_state.active = 1;
2827 }
2828 
2829 static int __init init_lapic_sysfs(void)
2830 {
2831 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2832 	if (boot_cpu_has(X86_FEATURE_APIC))
2833 		register_syscore_ops(&lapic_syscore_ops);
2834 
2835 	return 0;
2836 }
2837 
2838 /* local apic needs to resume before other devices access its registers. */
2839 core_initcall(init_lapic_sysfs);
2840 
2841 #else	/* CONFIG_PM */
2842 
2843 static void apic_pm_activate(void) { }
2844 
2845 #endif	/* CONFIG_PM */
2846 
2847 #ifdef CONFIG_X86_64
2848 
2849 static int multi_checked;
2850 static int multi;
2851 
2852 static int set_multi(const struct dmi_system_id *d)
2853 {
2854 	if (multi)
2855 		return 0;
2856 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2857 	multi = 1;
2858 	return 0;
2859 }
2860 
2861 static const struct dmi_system_id multi_dmi_table[] = {
2862 	{
2863 		.callback = set_multi,
2864 		.ident = "IBM System Summit2",
2865 		.matches = {
2866 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2867 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2868 		},
2869 	},
2870 	{}
2871 };
2872 
2873 static void dmi_check_multi(void)
2874 {
2875 	if (multi_checked)
2876 		return;
2877 
2878 	dmi_check_system(multi_dmi_table);
2879 	multi_checked = 1;
2880 }
2881 
2882 /*
2883  * apic_is_clustered_box() -- Check if we can expect good TSC
2884  *
2885  * Thus far, the major user of this is IBM's Summit2 series:
2886  * Clustered boxes may have unsynced TSC problems if they are
2887  * multi-chassis.
2888  * Use DMI to check them
2889  */
2890 int apic_is_clustered_box(void)
2891 {
2892 	dmi_check_multi();
2893 	return multi;
2894 }
2895 #endif
2896 
2897 /*
2898  * APIC command line parameters
2899  */
2900 static int __init setup_disableapic(char *arg)
2901 {
2902 	disable_apic = 1;
2903 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2904 	return 0;
2905 }
2906 early_param("disableapic", setup_disableapic);
2907 
2908 /* same as disableapic, for compatibility */
2909 static int __init setup_nolapic(char *arg)
2910 {
2911 	return setup_disableapic(arg);
2912 }
2913 early_param("nolapic", setup_nolapic);
2914 
2915 static int __init parse_lapic_timer_c2_ok(char *arg)
2916 {
2917 	local_apic_timer_c2_ok = 1;
2918 	return 0;
2919 }
2920 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2921 
2922 static int __init parse_disable_apic_timer(char *arg)
2923 {
2924 	disable_apic_timer = 1;
2925 	return 0;
2926 }
2927 early_param("noapictimer", parse_disable_apic_timer);
2928 
2929 static int __init parse_nolapic_timer(char *arg)
2930 {
2931 	disable_apic_timer = 1;
2932 	return 0;
2933 }
2934 early_param("nolapic_timer", parse_nolapic_timer);
2935 
2936 static int __init apic_set_verbosity(char *arg)
2937 {
2938 	if (!arg)  {
2939 #ifdef CONFIG_X86_64
2940 		skip_ioapic_setup = 0;
2941 		return 0;
2942 #endif
2943 		return -EINVAL;
2944 	}
2945 
2946 	if (strcmp("debug", arg) == 0)
2947 		apic_verbosity = APIC_DEBUG;
2948 	else if (strcmp("verbose", arg) == 0)
2949 		apic_verbosity = APIC_VERBOSE;
2950 #ifdef CONFIG_X86_64
2951 	else {
2952 		pr_warn("APIC Verbosity level %s not recognised"
2953 			" use apic=verbose or apic=debug\n", arg);
2954 		return -EINVAL;
2955 	}
2956 #endif
2957 
2958 	return 0;
2959 }
2960 early_param("apic", apic_set_verbosity);
2961 
2962 static int __init lapic_insert_resource(void)
2963 {
2964 	if (!apic_phys)
2965 		return -1;
2966 
2967 	/* Put local APIC into the resource map. */
2968 	lapic_resource.start = apic_phys;
2969 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2970 	insert_resource(&iomem_resource, &lapic_resource);
2971 
2972 	return 0;
2973 }
2974 
2975 /*
2976  * need call insert after e820__reserve_resources()
2977  * that is using request_resource
2978  */
2979 late_initcall(lapic_insert_resource);
2980 
2981 static int __init apic_set_disabled_cpu_apicid(char *arg)
2982 {
2983 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2984 		return -EINVAL;
2985 
2986 	return 0;
2987 }
2988 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2989 
2990 static int __init apic_set_extnmi(char *arg)
2991 {
2992 	if (!arg)
2993 		return -EINVAL;
2994 
2995 	if (!strncmp("all", arg, 3))
2996 		apic_extnmi = APIC_EXTNMI_ALL;
2997 	else if (!strncmp("none", arg, 4))
2998 		apic_extnmi = APIC_EXTNMI_NONE;
2999 	else if (!strncmp("bsp", arg, 3))
3000 		apic_extnmi = APIC_EXTNMI_BSP;
3001 	else {
3002 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
3003 		return -EINVAL;
3004 	}
3005 
3006 	return 0;
3007 }
3008 early_param("apic_extnmi", apic_set_extnmi);
3009