• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *	Local APIC handling, local APIC timers
3  *
4  *	(c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
5  *
6  *	Fixes
7  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs;
8  *					thanks to Eric Gilmore
9  *					and Rolf G. Tews
10  *					for testing these extensively.
11  *	Maciej W. Rozycki	:	Various updates and fixes.
12  *	Mikael Pettersson	:	Power Management for UP-APIC.
13  *	Pavel Machek and
14  *	Mikael Pettersson	:	PM converted to driver model.
15  */
16 
17 #include <linux/perf_event.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/mc146818rtc.h>
20 #include <linux/acpi_pmtmr.h>
21 #include <linux/clockchips.h>
22 #include <linux/interrupt.h>
23 #include <linux/bootmem.h>
24 #include <linux/ftrace.h>
25 #include <linux/ioport.h>
26 #include <linux/export.h>
27 #include <linux/syscore_ops.h>
28 #include <linux/delay.h>
29 #include <linux/timex.h>
30 #include <linux/i8253.h>
31 #include <linux/dmar.h>
32 #include <linux/init.h>
33 #include <linux/cpu.h>
34 #include <linux/dmi.h>
35 #include <linux/smp.h>
36 #include <linux/mm.h>
37 
38 #include <asm/trace/irq_vectors.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/perf_event.h>
41 #include <asm/x86_init.h>
42 #include <asm/pgalloc.h>
43 #include <linux/atomic.h>
44 #include <asm/mpspec.h>
45 #include <asm/i8259.h>
46 #include <asm/proto.h>
47 #include <asm/apic.h>
48 #include <asm/io_apic.h>
49 #include <asm/desc.h>
50 #include <asm/hpet.h>
51 #include <asm/mtrr.h>
52 #include <asm/time.h>
53 #include <asm/smp.h>
54 #include <asm/mce.h>
55 #include <asm/tsc.h>
56 #include <asm/hypervisor.h>
57 #include <asm/cpu_device_id.h>
58 #include <asm/intel-family.h>
59 #include <asm/irq_regs.h>
60 
61 unsigned int num_processors;
62 
63 unsigned disabled_cpus;
64 
65 /* Processor that is doing the boot up */
66 unsigned int boot_cpu_physical_apicid = -1U;
67 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
68 
69 u8 boot_cpu_apic_version;
70 
71 /*
72  * The highest APIC ID seen during enumeration.
73  */
74 static unsigned int max_physical_apicid;
75 
76 /*
77  * Bitmask of physically existing CPUs:
78  */
79 physid_mask_t phys_cpu_present_map;
80 
81 /*
82  * Processor to be disabled specified by kernel parameter
83  * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
84  * avoid undefined behaviour caused by sending INIT from AP to BSP.
85  */
86 static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
87 
88 /*
89  * This variable controls which CPUs receive external NMIs.  By default,
90  * external NMIs are delivered only to the BSP.
91  */
92 static int apic_extnmi = APIC_EXTNMI_BSP;
93 
94 /*
95  * Map cpu index to physical APIC ID
96  */
97 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
98 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
99 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
100 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
101 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
102 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
103 
104 #ifdef CONFIG_X86_32
105 
106 /*
107  * On x86_32, the mapping between cpu and logical apicid may vary
108  * depending on apic in use.  The following early percpu variable is
109  * used for the mapping.  This is where the behaviors of x86_64 and 32
110  * actually diverge.  Let's keep it ugly for now.
111  */
112 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
113 
114 /* Local APIC was disabled by the BIOS and enabled by the kernel */
115 static int enabled_via_apicbase;
116 
117 /*
118  * Handle interrupt mode configuration register (IMCR).
119  * This register controls whether the interrupt signals
120  * that reach the BSP come from the master PIC or from the
121  * local APIC. Before entering Symmetric I/O Mode, either
122  * the BIOS or the operating system must switch out of
123  * PIC Mode by changing the IMCR.
124  */
imcr_pic_to_apic(void)125 static inline void imcr_pic_to_apic(void)
126 {
127 	/* select IMCR register */
128 	outb(0x70, 0x22);
129 	/* NMI and 8259 INTR go through APIC */
130 	outb(0x01, 0x23);
131 }
132 
imcr_apic_to_pic(void)133 static inline void imcr_apic_to_pic(void)
134 {
135 	/* select IMCR register */
136 	outb(0x70, 0x22);
137 	/* NMI and 8259 INTR go directly to BSP */
138 	outb(0x00, 0x23);
139 }
140 #endif
141 
142 /*
143  * Knob to control our willingness to enable the local APIC.
144  *
145  * +1=force-enable
146  */
147 static int force_enable_local_apic __initdata;
148 
149 /*
150  * APIC command line parameters
151  */
parse_lapic(char * arg)152 static int __init parse_lapic(char *arg)
153 {
154 	if (IS_ENABLED(CONFIG_X86_32) && !arg)
155 		force_enable_local_apic = 1;
156 	else if (arg && !strncmp(arg, "notscdeadline", 13))
157 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
158 	return 0;
159 }
160 early_param("lapic", parse_lapic);
161 
162 #ifdef CONFIG_X86_64
163 static int apic_calibrate_pmtmr __initdata;
setup_apicpmtimer(char * s)164 static __init int setup_apicpmtimer(char *s)
165 {
166 	apic_calibrate_pmtmr = 1;
167 	notsc_setup(NULL);
168 	return 0;
169 }
170 __setup("apicpmtimer", setup_apicpmtimer);
171 #endif
172 
173 unsigned long mp_lapic_addr;
174 int disable_apic;
175 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
176 static int disable_apic_timer __initdata;
177 /* Local APIC timer works in C2 */
178 int local_apic_timer_c2_ok;
179 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
180 
181 /*
182  * Debug level, exported for io_apic.c
183  */
184 int apic_verbosity;
185 
186 int pic_mode;
187 
188 /* Have we found an MP table */
189 int smp_found_config;
190 
191 static struct resource lapic_resource = {
192 	.name = "Local APIC",
193 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
194 };
195 
196 unsigned int lapic_timer_frequency = 0;
197 
198 static void apic_pm_activate(void);
199 
200 static unsigned long apic_phys;
201 
202 /*
203  * Get the LAPIC version
204  */
lapic_get_version(void)205 static inline int lapic_get_version(void)
206 {
207 	return GET_APIC_VERSION(apic_read(APIC_LVR));
208 }
209 
210 /*
211  * Check, if the APIC is integrated or a separate chip
212  */
lapic_is_integrated(void)213 static inline int lapic_is_integrated(void)
214 {
215 	return APIC_INTEGRATED(lapic_get_version());
216 }
217 
218 /*
219  * Check, whether this is a modern or a first generation APIC
220  */
modern_apic(void)221 static int modern_apic(void)
222 {
223 	/* AMD systems use old APIC versions, so check the CPU */
224 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
225 	    boot_cpu_data.x86 >= 0xf)
226 		return 1;
227 	return lapic_get_version() >= 0x14;
228 }
229 
230 /*
231  * right after this call apic become NOOP driven
232  * so apic->write/read doesn't do anything
233  */
apic_disable(void)234 static void __init apic_disable(void)
235 {
236 	pr_info("APIC: switched to apic NOOP\n");
237 	apic = &apic_noop;
238 }
239 
native_apic_wait_icr_idle(void)240 void native_apic_wait_icr_idle(void)
241 {
242 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
243 		cpu_relax();
244 }
245 
native_safe_apic_wait_icr_idle(void)246 u32 native_safe_apic_wait_icr_idle(void)
247 {
248 	u32 send_status;
249 	int timeout;
250 
251 	timeout = 0;
252 	do {
253 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
254 		if (!send_status)
255 			break;
256 		inc_irq_stat(icr_read_retry_count);
257 		udelay(100);
258 	} while (timeout++ < 1000);
259 
260 	return send_status;
261 }
262 
native_apic_icr_write(u32 low,u32 id)263 void native_apic_icr_write(u32 low, u32 id)
264 {
265 	unsigned long flags;
266 
267 	local_irq_save(flags);
268 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
269 	apic_write(APIC_ICR, low);
270 	local_irq_restore(flags);
271 }
272 
native_apic_icr_read(void)273 u64 native_apic_icr_read(void)
274 {
275 	u32 icr1, icr2;
276 
277 	icr2 = apic_read(APIC_ICR2);
278 	icr1 = apic_read(APIC_ICR);
279 
280 	return icr1 | ((u64)icr2 << 32);
281 }
282 
283 #ifdef CONFIG_X86_32
284 /**
285  * get_physical_broadcast - Get number of physical broadcast IDs
286  */
get_physical_broadcast(void)287 int get_physical_broadcast(void)
288 {
289 	return modern_apic() ? 0xff : 0xf;
290 }
291 #endif
292 
293 /**
294  * lapic_get_maxlvt - get the maximum number of local vector table entries
295  */
lapic_get_maxlvt(void)296 int lapic_get_maxlvt(void)
297 {
298 	/*
299 	 * - we always have APIC integrated on 64bit mode
300 	 * - 82489DXs do not report # of LVT entries
301 	 */
302 	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
303 }
304 
305 /*
306  * Local APIC timer
307  */
308 
309 /* Clock divisor */
310 #define APIC_DIVISOR 16
311 #define TSC_DIVISOR  8
312 
313 /*
314  * This function sets up the local APIC timer, with a timeout of
315  * 'clocks' APIC bus clock. During calibration we actually call
316  * this function twice on the boot CPU, once with a bogus timeout
317  * value, second time for real. The other (noncalibrating) CPUs
318  * call this function only once, with the real, calibrated value.
319  *
320  * We do reads before writes even if unnecessary, to get around the
321  * P5 APIC double write bug.
322  */
__setup_APIC_LVTT(unsigned int clocks,int oneshot,int irqen)323 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
324 {
325 	unsigned int lvtt_value, tmp_value;
326 
327 	lvtt_value = LOCAL_TIMER_VECTOR;
328 	if (!oneshot)
329 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
330 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
331 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
332 
333 	if (!lapic_is_integrated())
334 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
335 
336 	if (!irqen)
337 		lvtt_value |= APIC_LVT_MASKED;
338 
339 	apic_write(APIC_LVTT, lvtt_value);
340 
341 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
342 		/*
343 		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
344 		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
345 		 * According to Intel, MFENCE can do the serialization here.
346 		 */
347 		asm volatile("mfence" : : : "memory");
348 		return;
349 	}
350 
351 	/*
352 	 * Divide PICLK by 16
353 	 */
354 	tmp_value = apic_read(APIC_TDCR);
355 	apic_write(APIC_TDCR,
356 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
357 		APIC_TDR_DIV_16);
358 
359 	if (!oneshot)
360 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
361 }
362 
363 /*
364  * Setup extended LVT, AMD specific
365  *
366  * Software should use the LVT offsets the BIOS provides.  The offsets
367  * are determined by the subsystems using it like those for MCE
368  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
369  * are supported. Beginning with family 10h at least 4 offsets are
370  * available.
371  *
372  * Since the offsets must be consistent for all cores, we keep track
373  * of the LVT offsets in software and reserve the offset for the same
374  * vector also to be used on other cores. An offset is freed by
375  * setting the entry to APIC_EILVT_MASKED.
376  *
377  * If the BIOS is right, there should be no conflicts. Otherwise a
378  * "[Firmware Bug]: ..." error message is generated. However, if
379  * software does not properly determines the offsets, it is not
380  * necessarily a BIOS bug.
381  */
382 
383 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
384 
eilvt_entry_is_changeable(unsigned int old,unsigned int new)385 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
386 {
387 	return (old & APIC_EILVT_MASKED)
388 		|| (new == APIC_EILVT_MASKED)
389 		|| ((new & ~APIC_EILVT_MASKED) == old);
390 }
391 
reserve_eilvt_offset(int offset,unsigned int new)392 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
393 {
394 	unsigned int rsvd, vector;
395 
396 	if (offset >= APIC_EILVT_NR_MAX)
397 		return ~0;
398 
399 	rsvd = atomic_read(&eilvt_offsets[offset]);
400 	do {
401 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
402 		if (vector && !eilvt_entry_is_changeable(vector, new))
403 			/* may not change if vectors are different */
404 			return rsvd;
405 		rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
406 	} while (rsvd != new);
407 
408 	rsvd &= ~APIC_EILVT_MASKED;
409 	if (rsvd && rsvd != vector)
410 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
411 			offset, rsvd);
412 
413 	return new;
414 }
415 
416 /*
417  * If mask=1, the LVT entry does not generate interrupts while mask=0
418  * enables the vector. See also the BKDGs. Must be called with
419  * preemption disabled.
420  */
421 
setup_APIC_eilvt(u8 offset,u8 vector,u8 msg_type,u8 mask)422 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
423 {
424 	unsigned long reg = APIC_EILVTn(offset);
425 	unsigned int new, old, reserved;
426 
427 	new = (mask << 16) | (msg_type << 8) | vector;
428 	old = apic_read(reg);
429 	reserved = reserve_eilvt_offset(offset, new);
430 
431 	if (reserved != new) {
432 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
433 		       "vector 0x%x, but the register is already in use for "
434 		       "vector 0x%x on another cpu\n",
435 		       smp_processor_id(), reg, offset, new, reserved);
436 		return -EINVAL;
437 	}
438 
439 	if (!eilvt_entry_is_changeable(old, new)) {
440 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
441 		       "vector 0x%x, but the register is already in use for "
442 		       "vector 0x%x on this cpu\n",
443 		       smp_processor_id(), reg, offset, new, old);
444 		return -EBUSY;
445 	}
446 
447 	apic_write(reg, new);
448 
449 	return 0;
450 }
451 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
452 
453 /*
454  * Program the next event, relative to now
455  */
lapic_next_event(unsigned long delta,struct clock_event_device * evt)456 static int lapic_next_event(unsigned long delta,
457 			    struct clock_event_device *evt)
458 {
459 	apic_write(APIC_TMICT, delta);
460 	return 0;
461 }
462 
lapic_next_deadline(unsigned long delta,struct clock_event_device * evt)463 static int lapic_next_deadline(unsigned long delta,
464 			       struct clock_event_device *evt)
465 {
466 	u64 tsc;
467 
468 	tsc = rdtsc();
469 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
470 	return 0;
471 }
472 
lapic_timer_shutdown(struct clock_event_device * evt)473 static int lapic_timer_shutdown(struct clock_event_device *evt)
474 {
475 	unsigned int v;
476 
477 	/* Lapic used as dummy for broadcast ? */
478 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
479 		return 0;
480 
481 	v = apic_read(APIC_LVTT);
482 	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
483 	apic_write(APIC_LVTT, v);
484 	apic_write(APIC_TMICT, 0);
485 	return 0;
486 }
487 
488 static inline int
lapic_timer_set_periodic_oneshot(struct clock_event_device * evt,bool oneshot)489 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
490 {
491 	/* Lapic used as dummy for broadcast ? */
492 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
493 		return 0;
494 
495 	__setup_APIC_LVTT(lapic_timer_frequency, oneshot, 1);
496 	return 0;
497 }
498 
lapic_timer_set_periodic(struct clock_event_device * evt)499 static int lapic_timer_set_periodic(struct clock_event_device *evt)
500 {
501 	return lapic_timer_set_periodic_oneshot(evt, false);
502 }
503 
lapic_timer_set_oneshot(struct clock_event_device * evt)504 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
505 {
506 	return lapic_timer_set_periodic_oneshot(evt, true);
507 }
508 
509 /*
510  * Local APIC timer broadcast function
511  */
lapic_timer_broadcast(const struct cpumask * mask)512 static void lapic_timer_broadcast(const struct cpumask *mask)
513 {
514 #ifdef CONFIG_SMP
515 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
516 #endif
517 }
518 
519 
520 /*
521  * The local apic timer can be used for any function which is CPU local.
522  */
523 static struct clock_event_device lapic_clockevent = {
524 	.name				= "lapic",
525 	.features			= CLOCK_EVT_FEAT_PERIODIC |
526 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
527 					  | CLOCK_EVT_FEAT_DUMMY,
528 	.shift				= 32,
529 	.set_state_shutdown		= lapic_timer_shutdown,
530 	.set_state_periodic		= lapic_timer_set_periodic,
531 	.set_state_oneshot		= lapic_timer_set_oneshot,
532 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
533 	.set_next_event			= lapic_next_event,
534 	.broadcast			= lapic_timer_broadcast,
535 	.rating				= 100,
536 	.irq				= -1,
537 };
538 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
539 
540 #define DEADLINE_MODEL_MATCH_FUNC(model, func)	\
541 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&func }
542 
543 #define DEADLINE_MODEL_MATCH_REV(model, rev)	\
544 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)rev }
545 
hsx_deadline_rev(void)546 static __init u32 hsx_deadline_rev(void)
547 {
548 	switch (boot_cpu_data.x86_stepping) {
549 	case 0x02: return 0x3a; /* EP */
550 	case 0x04: return 0x0f; /* EX */
551 	}
552 
553 	return ~0U;
554 }
555 
bdx_deadline_rev(void)556 static __init u32 bdx_deadline_rev(void)
557 {
558 	switch (boot_cpu_data.x86_stepping) {
559 	case 0x02: return 0x00000011;
560 	case 0x03: return 0x0700000e;
561 	case 0x04: return 0x0f00000c;
562 	case 0x05: return 0x0e000003;
563 	}
564 
565 	return ~0U;
566 }
567 
skx_deadline_rev(void)568 static __init u32 skx_deadline_rev(void)
569 {
570 	switch (boot_cpu_data.x86_stepping) {
571 	case 0x03: return 0x01000136;
572 	case 0x04: return 0x02000014;
573 	}
574 
575 	if (boot_cpu_data.x86_stepping > 4)
576 		return 0;
577 
578 	return ~0U;
579 }
580 
581 static const struct x86_cpu_id deadline_match[] __initconst = {
582 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_HASWELL_X,	hsx_deadline_rev),
583 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_X,	0x0b000020),
584 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_BROADWELL_XEON_D,	bdx_deadline_rev),
585 	DEADLINE_MODEL_MATCH_FUNC( INTEL_FAM6_SKYLAKE_X,	skx_deadline_rev),
586 
587 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_CORE,	0x22),
588 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_ULT,	0x20),
589 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_HASWELL_GT3E,	0x17),
590 
591 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_CORE,	0x25),
592 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_BROADWELL_GT3E,	0x17),
593 
594 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE,	0xb2),
595 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP,	0xb2),
596 
597 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE,	0x52),
598 	DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP,	0x52),
599 
600 	{},
601 };
602 
apic_validate_deadline_timer(void)603 static __init bool apic_validate_deadline_timer(void)
604 {
605 	const struct x86_cpu_id *m;
606 	u32 rev;
607 
608 	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
609 		return false;
610 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
611 		return true;
612 
613 	m = x86_match_cpu(deadline_match);
614 	if (!m)
615 		return true;
616 
617 	/*
618 	 * Function pointers will have the MSB set due to address layout,
619 	 * immediate revisions will not.
620 	 */
621 	if ((long)m->driver_data < 0)
622 		rev = ((u32 (*)(void))(m->driver_data))();
623 	else
624 		rev = (u32)m->driver_data;
625 
626 	if (boot_cpu_data.microcode >= rev)
627 		return true;
628 
629 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
630 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
631 	       "please update microcode to version: 0x%x (or later)\n", rev);
632 	return false;
633 }
634 
635 /*
636  * Setup the local APIC timer for this CPU. Copy the initialized values
637  * of the boot CPU and register the clock event in the framework.
638  */
setup_APIC_timer(void)639 static void setup_APIC_timer(void)
640 {
641 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
642 
643 	if (this_cpu_has(X86_FEATURE_ARAT)) {
644 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
645 		/* Make LAPIC timer preferrable over percpu HPET */
646 		lapic_clockevent.rating = 150;
647 	}
648 
649 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
650 	levt->cpumask = cpumask_of(smp_processor_id());
651 
652 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
653 		levt->name = "lapic-deadline";
654 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
655 				    CLOCK_EVT_FEAT_DUMMY);
656 		levt->set_next_event = lapic_next_deadline;
657 		clockevents_config_and_register(levt,
658 						tsc_khz * (1000 / TSC_DIVISOR),
659 						0xF, ~0UL);
660 	} else
661 		clockevents_register_device(levt);
662 }
663 
664 /*
665  * Install the updated TSC frequency from recalibration at the TSC
666  * deadline clockevent devices.
667  */
__lapic_update_tsc_freq(void * info)668 static void __lapic_update_tsc_freq(void *info)
669 {
670 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
671 
672 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
673 		return;
674 
675 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
676 }
677 
lapic_update_tsc_freq(void)678 void lapic_update_tsc_freq(void)
679 {
680 	/*
681 	 * The clockevent device's ->mult and ->shift can both be
682 	 * changed. In order to avoid races, schedule the frequency
683 	 * update code on each CPU.
684 	 */
685 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
686 }
687 
688 /*
689  * In this functions we calibrate APIC bus clocks to the external timer.
690  *
691  * We want to do the calibration only once since we want to have local timer
692  * irqs syncron. CPUs connected by the same APIC bus have the very same bus
693  * frequency.
694  *
695  * This was previously done by reading the PIT/HPET and waiting for a wrap
696  * around to find out, that a tick has elapsed. I have a box, where the PIT
697  * readout is broken, so it never gets out of the wait loop again. This was
698  * also reported by others.
699  *
700  * Monitoring the jiffies value is inaccurate and the clockevents
701  * infrastructure allows us to do a simple substitution of the interrupt
702  * handler.
703  *
704  * The calibration routine also uses the pm_timer when possible, as the PIT
705  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
706  * back to normal later in the boot process).
707  */
708 
709 #define LAPIC_CAL_LOOPS		(HZ/10)
710 
711 static __initdata int lapic_cal_loops = -1;
712 static __initdata long lapic_cal_t1, lapic_cal_t2;
713 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
714 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
715 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
716 
717 /*
718  * Temporary interrupt handler and polled calibration function.
719  */
lapic_cal_handler(struct clock_event_device * dev)720 static void __init lapic_cal_handler(struct clock_event_device *dev)
721 {
722 	unsigned long long tsc = 0;
723 	long tapic = apic_read(APIC_TMCCT);
724 	unsigned long pm = acpi_pm_read_early();
725 
726 	if (boot_cpu_has(X86_FEATURE_TSC))
727 		tsc = rdtsc();
728 
729 	switch (lapic_cal_loops++) {
730 	case 0:
731 		lapic_cal_t1 = tapic;
732 		lapic_cal_tsc1 = tsc;
733 		lapic_cal_pm1 = pm;
734 		lapic_cal_j1 = jiffies;
735 		break;
736 
737 	case LAPIC_CAL_LOOPS:
738 		lapic_cal_t2 = tapic;
739 		lapic_cal_tsc2 = tsc;
740 		if (pm < lapic_cal_pm1)
741 			pm += ACPI_PM_OVRRUN;
742 		lapic_cal_pm2 = pm;
743 		lapic_cal_j2 = jiffies;
744 		break;
745 	}
746 }
747 
748 static int __init
calibrate_by_pmtimer(long deltapm,long * delta,long * deltatsc)749 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
750 {
751 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
752 	const long pm_thresh = pm_100ms / 100;
753 	unsigned long mult;
754 	u64 res;
755 
756 #ifndef CONFIG_X86_PM_TIMER
757 	return -1;
758 #endif
759 
760 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
761 
762 	/* Check, if the PM timer is available */
763 	if (!deltapm)
764 		return -1;
765 
766 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
767 
768 	if (deltapm > (pm_100ms - pm_thresh) &&
769 	    deltapm < (pm_100ms + pm_thresh)) {
770 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
771 		return 0;
772 	}
773 
774 	res = (((u64)deltapm) *  mult) >> 22;
775 	do_div(res, 1000000);
776 	pr_warning("APIC calibration not consistent "
777 		   "with PM-Timer: %ldms instead of 100ms\n",(long)res);
778 
779 	/* Correct the lapic counter value */
780 	res = (((u64)(*delta)) * pm_100ms);
781 	do_div(res, deltapm);
782 	pr_info("APIC delta adjusted to PM-Timer: "
783 		"%lu (%ld)\n", (unsigned long)res, *delta);
784 	*delta = (long)res;
785 
786 	/* Correct the tsc counter value */
787 	if (boot_cpu_has(X86_FEATURE_TSC)) {
788 		res = (((u64)(*deltatsc)) * pm_100ms);
789 		do_div(res, deltapm);
790 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
791 					  "PM-Timer: %lu (%ld)\n",
792 					(unsigned long)res, *deltatsc);
793 		*deltatsc = (long)res;
794 	}
795 
796 	return 0;
797 }
798 
calibrate_APIC_clock(void)799 static int __init calibrate_APIC_clock(void)
800 {
801 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
802 	u64 tsc_perj = 0, tsc_start = 0;
803 	unsigned long jif_start;
804 	unsigned long deltaj;
805 	long delta, deltatsc;
806 	int pm_referenced = 0;
807 
808 	/**
809 	 * check if lapic timer has already been calibrated by platform
810 	 * specific routine, such as tsc calibration code. if so, we just fill
811 	 * in the clockevent structure and return.
812 	 */
813 
814 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
815 		return 0;
816 	} else if (lapic_timer_frequency) {
817 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
818 				lapic_timer_frequency);
819 		lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
820 					TICK_NSEC, lapic_clockevent.shift);
821 		lapic_clockevent.max_delta_ns =
822 			clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
823 		lapic_clockevent.max_delta_ticks = 0x7FFFFF;
824 		lapic_clockevent.min_delta_ns =
825 			clockevent_delta2ns(0xF, &lapic_clockevent);
826 		lapic_clockevent.min_delta_ticks = 0xF;
827 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
828 		return 0;
829 	}
830 
831 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
832 		    "calibrating APIC timer ...\n");
833 
834 	/*
835 	 * There are platforms w/o global clockevent devices. Instead of
836 	 * making the calibration conditional on that, use a polling based
837 	 * approach everywhere.
838 	 */
839 	local_irq_disable();
840 
841 	/*
842 	 * Setup the APIC counter to maximum. There is no way the lapic
843 	 * can underflow in the 100ms detection time frame
844 	 */
845 	__setup_APIC_LVTT(0xffffffff, 0, 0);
846 
847 	/*
848 	 * Methods to terminate the calibration loop:
849 	 *  1) Global clockevent if available (jiffies)
850 	 *  2) TSC if available and frequency is known
851 	 */
852 	jif_start = READ_ONCE(jiffies);
853 
854 	if (tsc_khz) {
855 		tsc_start = rdtsc();
856 		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
857 	}
858 
859 	/*
860 	 * Enable interrupts so the tick can fire, if a global
861 	 * clockevent device is available
862 	 */
863 	local_irq_enable();
864 
865 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
866 		/* Wait for a tick to elapse */
867 		while (1) {
868 			if (tsc_khz) {
869 				u64 tsc_now = rdtsc();
870 				if ((tsc_now - tsc_start) >= tsc_perj) {
871 					tsc_start += tsc_perj;
872 					break;
873 				}
874 			} else {
875 				unsigned long jif_now = READ_ONCE(jiffies);
876 
877 				if (time_after(jif_now, jif_start)) {
878 					jif_start = jif_now;
879 					break;
880 				}
881 			}
882 			cpu_relax();
883 		}
884 
885 		/* Invoke the calibration routine */
886 		local_irq_disable();
887 		lapic_cal_handler(NULL);
888 		local_irq_enable();
889 	}
890 
891 	local_irq_disable();
892 
893 	/* Build delta t1-t2 as apic timer counts down */
894 	delta = lapic_cal_t1 - lapic_cal_t2;
895 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
896 
897 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
898 
899 	/* we trust the PM based calibration if possible */
900 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
901 					&delta, &deltatsc);
902 
903 	/* Calculate the scaled math multiplication factor */
904 	lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
905 				       lapic_clockevent.shift);
906 	lapic_clockevent.max_delta_ns =
907 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
908 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
909 	lapic_clockevent.min_delta_ns =
910 		clockevent_delta2ns(0xF, &lapic_clockevent);
911 	lapic_clockevent.min_delta_ticks = 0xF;
912 
913 	lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
914 
915 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
916 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
917 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
918 		    lapic_timer_frequency);
919 
920 	if (boot_cpu_has(X86_FEATURE_TSC)) {
921 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
922 			    "%ld.%04ld MHz.\n",
923 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
924 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
925 	}
926 
927 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
928 		    "%u.%04u MHz.\n",
929 		    lapic_timer_frequency / (1000000 / HZ),
930 		    lapic_timer_frequency % (1000000 / HZ));
931 
932 	/*
933 	 * Do a sanity check on the APIC calibration result
934 	 */
935 	if (lapic_timer_frequency < (1000000 / HZ)) {
936 		local_irq_enable();
937 		pr_warning("APIC frequency too slow, disabling apic timer\n");
938 		return -1;
939 	}
940 
941 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
942 
943 	/*
944 	 * PM timer calibration failed or not turned on so lets try APIC
945 	 * timer based calibration, if a global clockevent device is
946 	 * available.
947 	 */
948 	if (!pm_referenced && global_clock_event) {
949 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
950 
951 		/*
952 		 * Setup the apic timer manually
953 		 */
954 		levt->event_handler = lapic_cal_handler;
955 		lapic_timer_set_periodic(levt);
956 		lapic_cal_loops = -1;
957 
958 		/* Let the interrupts run */
959 		local_irq_enable();
960 
961 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
962 			cpu_relax();
963 
964 		/* Stop the lapic timer */
965 		local_irq_disable();
966 		lapic_timer_shutdown(levt);
967 
968 		/* Jiffies delta */
969 		deltaj = lapic_cal_j2 - lapic_cal_j1;
970 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
971 
972 		/* Check, if the jiffies result is consistent */
973 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
974 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
975 		else
976 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
977 	}
978 	local_irq_enable();
979 
980 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
981 		pr_warning("APIC timer disabled due to verification failure\n");
982 		return -1;
983 	}
984 
985 	return 0;
986 }
987 
988 /*
989  * Setup the boot APIC
990  *
991  * Calibrate and verify the result.
992  */
setup_boot_APIC_clock(void)993 void __init setup_boot_APIC_clock(void)
994 {
995 	/*
996 	 * The local apic timer can be disabled via the kernel
997 	 * commandline or from the CPU detection code. Register the lapic
998 	 * timer as a dummy clock event source on SMP systems, so the
999 	 * broadcast mechanism is used. On UP systems simply ignore it.
1000 	 */
1001 	if (disable_apic_timer) {
1002 		pr_info("Disabling APIC timer\n");
1003 		/* No broadcast on UP ! */
1004 		if (num_possible_cpus() > 1) {
1005 			lapic_clockevent.mult = 1;
1006 			setup_APIC_timer();
1007 		}
1008 		return;
1009 	}
1010 
1011 	if (calibrate_APIC_clock()) {
1012 		/* No broadcast on UP ! */
1013 		if (num_possible_cpus() > 1)
1014 			setup_APIC_timer();
1015 		return;
1016 	}
1017 
1018 	/*
1019 	 * If nmi_watchdog is set to IO_APIC, we need the
1020 	 * PIT/HPET going.  Otherwise register lapic as a dummy
1021 	 * device.
1022 	 */
1023 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1024 
1025 	/* Setup the lapic or request the broadcast */
1026 	setup_APIC_timer();
1027 	amd_e400_c1e_apic_setup();
1028 }
1029 
setup_secondary_APIC_clock(void)1030 void setup_secondary_APIC_clock(void)
1031 {
1032 	setup_APIC_timer();
1033 	amd_e400_c1e_apic_setup();
1034 }
1035 
1036 /*
1037  * The guts of the apic timer interrupt
1038  */
local_apic_timer_interrupt(void)1039 static void local_apic_timer_interrupt(void)
1040 {
1041 	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1042 
1043 	/*
1044 	 * Normally we should not be here till LAPIC has been initialized but
1045 	 * in some cases like kdump, its possible that there is a pending LAPIC
1046 	 * timer interrupt from previous kernel's context and is delivered in
1047 	 * new kernel the moment interrupts are enabled.
1048 	 *
1049 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1050 	 * its possible that when we get here evt->event_handler is NULL.
1051 	 * Check for event_handler being NULL and discard the interrupt as
1052 	 * spurious.
1053 	 */
1054 	if (!evt->event_handler) {
1055 		pr_warning("Spurious LAPIC timer interrupt on cpu %d\n",
1056 			   smp_processor_id());
1057 		/* Switch it off */
1058 		lapic_timer_shutdown(evt);
1059 		return;
1060 	}
1061 
1062 	/*
1063 	 * the NMI deadlock-detector uses this.
1064 	 */
1065 	inc_irq_stat(apic_timer_irqs);
1066 
1067 	evt->event_handler(evt);
1068 }
1069 
1070 /*
1071  * Local APIC timer interrupt. This is the most natural way for doing
1072  * local interrupts, but local timer interrupts can be emulated by
1073  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1074  *
1075  * [ if a single-CPU system runs an SMP kernel then we call the local
1076  *   interrupt as well. Thus we cannot inline the local irq ... ]
1077  */
smp_apic_timer_interrupt(struct pt_regs * regs)1078 __visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
1079 {
1080 	struct pt_regs *old_regs = set_irq_regs(regs);
1081 
1082 	/*
1083 	 * NOTE! We'd better ACK the irq immediately,
1084 	 * because timer handling can be slow.
1085 	 *
1086 	 * update_process_times() expects us to have done irq_enter().
1087 	 * Besides, if we don't timer interrupts ignore the global
1088 	 * interrupt lock, which is the WrongThing (tm) to do.
1089 	 */
1090 	entering_ack_irq();
1091 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1092 	local_apic_timer_interrupt();
1093 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1094 	exiting_irq();
1095 
1096 	set_irq_regs(old_regs);
1097 }
1098 
setup_profiling_timer(unsigned int multiplier)1099 int setup_profiling_timer(unsigned int multiplier)
1100 {
1101 	return -EINVAL;
1102 }
1103 
1104 /*
1105  * Local APIC start and shutdown
1106  */
1107 
1108 /**
1109  * clear_local_APIC - shutdown the local APIC
1110  *
1111  * This is called, when a CPU is disabled and before rebooting, so the state of
1112  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1113  * leftovers during boot.
1114  */
clear_local_APIC(void)1115 void clear_local_APIC(void)
1116 {
1117 	int maxlvt;
1118 	u32 v;
1119 
1120 	/* APIC hasn't been mapped yet */
1121 	if (!x2apic_mode && !apic_phys)
1122 		return;
1123 
1124 	maxlvt = lapic_get_maxlvt();
1125 	/*
1126 	 * Masking an LVT entry can trigger a local APIC error
1127 	 * if the vector is zero. Mask LVTERR first to prevent this.
1128 	 */
1129 	if (maxlvt >= 3) {
1130 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1131 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1132 	}
1133 	/*
1134 	 * Careful: we have to set masks only first to deassert
1135 	 * any level-triggered sources.
1136 	 */
1137 	v = apic_read(APIC_LVTT);
1138 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1139 	v = apic_read(APIC_LVT0);
1140 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1141 	v = apic_read(APIC_LVT1);
1142 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1143 	if (maxlvt >= 4) {
1144 		v = apic_read(APIC_LVTPC);
1145 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1146 	}
1147 
1148 	/* lets not touch this if we didn't frob it */
1149 #ifdef CONFIG_X86_THERMAL_VECTOR
1150 	if (maxlvt >= 5) {
1151 		v = apic_read(APIC_LVTTHMR);
1152 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1153 	}
1154 #endif
1155 #ifdef CONFIG_X86_MCE_INTEL
1156 	if (maxlvt >= 6) {
1157 		v = apic_read(APIC_LVTCMCI);
1158 		if (!(v & APIC_LVT_MASKED))
1159 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1160 	}
1161 #endif
1162 
1163 	/*
1164 	 * Clean APIC state for other OSs:
1165 	 */
1166 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1167 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1168 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1169 	if (maxlvt >= 3)
1170 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1171 	if (maxlvt >= 4)
1172 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1173 
1174 	/* Integrated APIC (!82489DX) ? */
1175 	if (lapic_is_integrated()) {
1176 		if (maxlvt > 3)
1177 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1178 			apic_write(APIC_ESR, 0);
1179 		apic_read(APIC_ESR);
1180 	}
1181 }
1182 
1183 /**
1184  * disable_local_APIC - clear and disable the local APIC
1185  */
disable_local_APIC(void)1186 void disable_local_APIC(void)
1187 {
1188 	unsigned int value;
1189 
1190 	/* APIC hasn't been mapped yet */
1191 	if (!x2apic_mode && !apic_phys)
1192 		return;
1193 
1194 	clear_local_APIC();
1195 
1196 	/*
1197 	 * Disable APIC (implies clearing of registers
1198 	 * for 82489DX!).
1199 	 */
1200 	value = apic_read(APIC_SPIV);
1201 	value &= ~APIC_SPIV_APIC_ENABLED;
1202 	apic_write(APIC_SPIV, value);
1203 
1204 #ifdef CONFIG_X86_32
1205 	/*
1206 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1207 	 * restore the disabled state.
1208 	 */
1209 	if (enabled_via_apicbase) {
1210 		unsigned int l, h;
1211 
1212 		rdmsr(MSR_IA32_APICBASE, l, h);
1213 		l &= ~MSR_IA32_APICBASE_ENABLE;
1214 		wrmsr(MSR_IA32_APICBASE, l, h);
1215 	}
1216 #endif
1217 }
1218 
1219 /*
1220  * If Linux enabled the LAPIC against the BIOS default disable it down before
1221  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1222  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1223  * for the case where Linux didn't enable the LAPIC.
1224  */
lapic_shutdown(void)1225 void lapic_shutdown(void)
1226 {
1227 	unsigned long flags;
1228 
1229 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1230 		return;
1231 
1232 	local_irq_save(flags);
1233 
1234 #ifdef CONFIG_X86_32
1235 	if (!enabled_via_apicbase)
1236 		clear_local_APIC();
1237 	else
1238 #endif
1239 		disable_local_APIC();
1240 
1241 
1242 	local_irq_restore(flags);
1243 }
1244 
1245 /**
1246  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1247  */
sync_Arb_IDs(void)1248 void __init sync_Arb_IDs(void)
1249 {
1250 	/*
1251 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1252 	 * needed on AMD.
1253 	 */
1254 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1255 		return;
1256 
1257 	/*
1258 	 * Wait for idle.
1259 	 */
1260 	apic_wait_icr_idle();
1261 
1262 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1263 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1264 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1265 }
1266 
1267 enum apic_intr_mode_id apic_intr_mode;
1268 
apic_intr_mode_select(void)1269 static int __init apic_intr_mode_select(void)
1270 {
1271 	/* Check kernel option */
1272 	if (disable_apic) {
1273 		pr_info("APIC disabled via kernel command line\n");
1274 		return APIC_PIC;
1275 	}
1276 
1277 	/* Check BIOS */
1278 #ifdef CONFIG_X86_64
1279 	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1280 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1281 		disable_apic = 1;
1282 		pr_info("APIC disabled by BIOS\n");
1283 		return APIC_PIC;
1284 	}
1285 #else
1286 	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1287 
1288 	/* Neither 82489DX nor integrated APIC ? */
1289 	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1290 		disable_apic = 1;
1291 		return APIC_PIC;
1292 	}
1293 
1294 	/* If the BIOS pretends there is an integrated APIC ? */
1295 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1296 		APIC_INTEGRATED(boot_cpu_apic_version)) {
1297 		disable_apic = 1;
1298 		pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1299 				       boot_cpu_physical_apicid);
1300 		return APIC_PIC;
1301 	}
1302 #endif
1303 
1304 	/* Check MP table or ACPI MADT configuration */
1305 	if (!smp_found_config) {
1306 		disable_ioapic_support();
1307 		if (!acpi_lapic) {
1308 			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1309 			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1310 		}
1311 		return APIC_VIRTUAL_WIRE;
1312 	}
1313 
1314 #ifdef CONFIG_SMP
1315 	/* If SMP should be disabled, then really disable it! */
1316 	if (!setup_max_cpus) {
1317 		pr_info("APIC: SMP mode deactivated\n");
1318 		return APIC_SYMMETRIC_IO_NO_ROUTING;
1319 	}
1320 
1321 	if (read_apic_id() != boot_cpu_physical_apicid) {
1322 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1323 		     read_apic_id(), boot_cpu_physical_apicid);
1324 		/* Or can we switch back to PIC here? */
1325 	}
1326 #endif
1327 
1328 	return APIC_SYMMETRIC_IO;
1329 }
1330 
1331 /*
1332  * An initial setup of the virtual wire mode.
1333  */
init_bsp_APIC(void)1334 void __init init_bsp_APIC(void)
1335 {
1336 	unsigned int value;
1337 
1338 	/*
1339 	 * Don't do the setup now if we have a SMP BIOS as the
1340 	 * through-I/O-APIC virtual wire mode might be active.
1341 	 */
1342 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1343 		return;
1344 
1345 	/*
1346 	 * Do not trust the local APIC being empty at bootup.
1347 	 */
1348 	clear_local_APIC();
1349 
1350 	/*
1351 	 * Enable APIC.
1352 	 */
1353 	value = apic_read(APIC_SPIV);
1354 	value &= ~APIC_VECTOR_MASK;
1355 	value |= APIC_SPIV_APIC_ENABLED;
1356 
1357 #ifdef CONFIG_X86_32
1358 	/* This bit is reserved on P4/Xeon and should be cleared */
1359 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1360 	    (boot_cpu_data.x86 == 15))
1361 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1362 	else
1363 #endif
1364 		value |= APIC_SPIV_FOCUS_DISABLED;
1365 	value |= SPURIOUS_APIC_VECTOR;
1366 	apic_write(APIC_SPIV, value);
1367 
1368 	/*
1369 	 * Set up the virtual wire mode.
1370 	 */
1371 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1372 	value = APIC_DM_NMI;
1373 	if (!lapic_is_integrated())		/* 82489DX */
1374 		value |= APIC_LVT_LEVEL_TRIGGER;
1375 	if (apic_extnmi == APIC_EXTNMI_NONE)
1376 		value |= APIC_LVT_MASKED;
1377 	apic_write(APIC_LVT1, value);
1378 }
1379 
1380 /* Init the interrupt delivery mode for the BSP */
apic_intr_mode_init(void)1381 void __init apic_intr_mode_init(void)
1382 {
1383 	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1384 
1385 	apic_intr_mode = apic_intr_mode_select();
1386 
1387 	switch (apic_intr_mode) {
1388 	case APIC_PIC:
1389 		pr_info("APIC: Keep in PIC mode(8259)\n");
1390 		return;
1391 	case APIC_VIRTUAL_WIRE:
1392 		pr_info("APIC: Switch to virtual wire mode setup\n");
1393 		default_setup_apic_routing();
1394 		break;
1395 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1396 		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1397 		upmode = true;
1398 		default_setup_apic_routing();
1399 		break;
1400 	case APIC_SYMMETRIC_IO:
1401 		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1402 		default_setup_apic_routing();
1403 		break;
1404 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1405 		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1406 		break;
1407 	}
1408 
1409 	apic_bsp_setup(upmode);
1410 }
1411 
lapic_setup_esr(void)1412 static void lapic_setup_esr(void)
1413 {
1414 	unsigned int oldvalue, value, maxlvt;
1415 
1416 	if (!lapic_is_integrated()) {
1417 		pr_info("No ESR for 82489DX.\n");
1418 		return;
1419 	}
1420 
1421 	if (apic->disable_esr) {
1422 		/*
1423 		 * Something untraceable is creating bad interrupts on
1424 		 * secondary quads ... for the moment, just leave the
1425 		 * ESR disabled - we can't do anything useful with the
1426 		 * errors anyway - mbligh
1427 		 */
1428 		pr_info("Leaving ESR disabled.\n");
1429 		return;
1430 	}
1431 
1432 	maxlvt = lapic_get_maxlvt();
1433 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1434 		apic_write(APIC_ESR, 0);
1435 	oldvalue = apic_read(APIC_ESR);
1436 
1437 	/* enables sending errors */
1438 	value = ERROR_APIC_VECTOR;
1439 	apic_write(APIC_LVTERR, value);
1440 
1441 	/*
1442 	 * spec says clear errors after enabling vector.
1443 	 */
1444 	if (maxlvt > 3)
1445 		apic_write(APIC_ESR, 0);
1446 	value = apic_read(APIC_ESR);
1447 	if (value != oldvalue)
1448 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1449 			"vector: 0x%08x  after: 0x%08x\n",
1450 			oldvalue, value);
1451 }
1452 
1453 #define APIC_IR_REGS		APIC_ISR_NR
1454 #define APIC_IR_BITS		(APIC_IR_REGS * 32)
1455 #define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1456 
1457 union apic_ir {
1458 	unsigned long	map[APIC_IR_MAPSIZE];
1459 	u32		regs[APIC_IR_REGS];
1460 };
1461 
apic_check_and_ack(union apic_ir * irr,union apic_ir * isr)1462 static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1463 {
1464 	int i, bit;
1465 
1466 	/* Read the IRRs */
1467 	for (i = 0; i < APIC_IR_REGS; i++)
1468 		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1469 
1470 	/* Read the ISRs */
1471 	for (i = 0; i < APIC_IR_REGS; i++)
1472 		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1473 
1474 	/*
1475 	 * If the ISR map is not empty. ACK the APIC and run another round
1476 	 * to verify whether a pending IRR has been unblocked and turned
1477 	 * into a ISR.
1478 	 */
1479 	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1480 		/*
1481 		 * There can be multiple ISR bits set when a high priority
1482 		 * interrupt preempted a lower priority one. Issue an ACK
1483 		 * per set bit.
1484 		 */
1485 		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1486 			ack_APIC_irq();
1487 		return true;
1488 	}
1489 
1490 	return !bitmap_empty(irr->map, APIC_IR_BITS);
1491 }
1492 
1493 /*
1494  * After a crash, we no longer service the interrupts and a pending
1495  * interrupt from previous kernel might still have ISR bit set.
1496  *
1497  * Most probably by now the CPU has serviced that pending interrupt and it
1498  * might not have done the ack_APIC_irq() because it thought, interrupt
1499  * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1500  * the ISR bit and cpu thinks it has already serivced the interrupt. Hence
1501  * a vector might get locked. It was noticed for timer irq (vector
1502  * 0x31). Issue an extra EOI to clear ISR.
1503  *
1504  * If there are pending IRR bits they turn into ISR bits after a higher
1505  * priority ISR bit has been acked.
1506  */
apic_pending_intr_clear(void)1507 static void apic_pending_intr_clear(void)
1508 {
1509 	union apic_ir irr, isr;
1510 	unsigned int i;
1511 
1512 	/* 512 loops are way oversized and give the APIC a chance to obey. */
1513 	for (i = 0; i < 512; i++) {
1514 		if (!apic_check_and_ack(&irr, &isr))
1515 			return;
1516 	}
1517 	/* Dump the IRR/ISR content if that failed */
1518 	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1519 }
1520 
1521 /**
1522  * setup_local_APIC - setup the local APIC
1523  *
1524  * Used to setup local APIC while initializing BSP or bringing up APs.
1525  * Always called with preemption disabled.
1526  */
setup_local_APIC(void)1527 static void setup_local_APIC(void)
1528 {
1529 	int cpu = smp_processor_id();
1530 	unsigned int value;
1531 
1532 
1533 	if (disable_apic) {
1534 		disable_ioapic_support();
1535 		return;
1536 	}
1537 
1538 	/*
1539 	 * If this comes from kexec/kcrash the APIC might be enabled in
1540 	 * SPIV. Soft disable it before doing further initialization.
1541 	 */
1542 	value = apic_read(APIC_SPIV);
1543 	value &= ~APIC_SPIV_APIC_ENABLED;
1544 	apic_write(APIC_SPIV, value);
1545 
1546 #ifdef CONFIG_X86_32
1547 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1548 	if (lapic_is_integrated() && apic->disable_esr) {
1549 		apic_write(APIC_ESR, 0);
1550 		apic_write(APIC_ESR, 0);
1551 		apic_write(APIC_ESR, 0);
1552 		apic_write(APIC_ESR, 0);
1553 	}
1554 #endif
1555 	perf_events_lapic_init();
1556 
1557 	/*
1558 	 * Double-check whether this APIC is really registered.
1559 	 * This is meaningless in clustered apic mode, so we skip it.
1560 	 */
1561 	BUG_ON(!apic->apic_id_registered());
1562 
1563 	/*
1564 	 * Intel recommends to set DFR, LDR and TPR before enabling
1565 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1566 	 * document number 292116).  So here it goes...
1567 	 */
1568 	apic->init_apic_ldr();
1569 
1570 #ifdef CONFIG_X86_32
1571 	if (apic->dest_logical) {
1572 		int logical_apicid, ldr_apicid;
1573 
1574 		/*
1575 		 * APIC LDR is initialized.  If logical_apicid mapping was
1576 		 * initialized during get_smp_config(), make sure it matches
1577 		 * the actual value.
1578 		 */
1579 		logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1580 		ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1581 		if (logical_apicid != BAD_APICID)
1582 			WARN_ON(logical_apicid != ldr_apicid);
1583 		/* Always use the value from LDR. */
1584 		early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1585 	}
1586 #endif
1587 
1588 	/*
1589 	 * Set Task Priority to 'accept all'. We never change this
1590 	 * later on.
1591 	 */
1592 	value = apic_read(APIC_TASKPRI);
1593 	value &= ~APIC_TPRI_MASK;
1594 	apic_write(APIC_TASKPRI, value);
1595 
1596 	/* Clear eventually stale ISR/IRR bits */
1597 	apic_pending_intr_clear();
1598 
1599 	/*
1600 	 * Now that we are all set up, enable the APIC
1601 	 */
1602 	value = apic_read(APIC_SPIV);
1603 	value &= ~APIC_VECTOR_MASK;
1604 	/*
1605 	 * Enable APIC
1606 	 */
1607 	value |= APIC_SPIV_APIC_ENABLED;
1608 
1609 #ifdef CONFIG_X86_32
1610 	/*
1611 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1612 	 * certain networking cards. If high frequency interrupts are
1613 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1614 	 * entry is masked/unmasked at a high rate as well then sooner or
1615 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1616 	 * from the device. If focus CPU is disabled then the hang goes
1617 	 * away, oh well :-(
1618 	 *
1619 	 * [ This bug can be reproduced easily with a level-triggered
1620 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1621 	 *   BX chipset. ]
1622 	 */
1623 	/*
1624 	 * Actually disabling the focus CPU check just makes the hang less
1625 	 * frequent as it makes the interrupt distributon model be more
1626 	 * like LRU than MRU (the short-term load is more even across CPUs).
1627 	 */
1628 
1629 	/*
1630 	 * - enable focus processor (bit==0)
1631 	 * - 64bit mode always use processor focus
1632 	 *   so no need to set it
1633 	 */
1634 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1635 #endif
1636 
1637 	/*
1638 	 * Set spurious IRQ vector
1639 	 */
1640 	value |= SPURIOUS_APIC_VECTOR;
1641 	apic_write(APIC_SPIV, value);
1642 
1643 	/*
1644 	 * Set up LVT0, LVT1:
1645 	 *
1646 	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1647 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1648 	 * we delegate interrupts to the 8259A.
1649 	 */
1650 	/*
1651 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1652 	 */
1653 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1654 	if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1655 		value = APIC_DM_EXTINT;
1656 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1657 	} else {
1658 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1659 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1660 	}
1661 	apic_write(APIC_LVT0, value);
1662 
1663 	/*
1664 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1665 	 * modified by apic_extnmi= boot option.
1666 	 */
1667 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1668 	    apic_extnmi == APIC_EXTNMI_ALL)
1669 		value = APIC_DM_NMI;
1670 	else
1671 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1672 
1673 	/* Is 82489DX ? */
1674 	if (!lapic_is_integrated())
1675 		value |= APIC_LVT_LEVEL_TRIGGER;
1676 	apic_write(APIC_LVT1, value);
1677 
1678 #ifdef CONFIG_X86_MCE_INTEL
1679 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1680 	if (!cpu)
1681 		cmci_recheck();
1682 #endif
1683 }
1684 
end_local_APIC_setup(void)1685 static void end_local_APIC_setup(void)
1686 {
1687 	lapic_setup_esr();
1688 
1689 #ifdef CONFIG_X86_32
1690 	{
1691 		unsigned int value;
1692 		/* Disable the local apic timer */
1693 		value = apic_read(APIC_LVTT);
1694 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1695 		apic_write(APIC_LVTT, value);
1696 	}
1697 #endif
1698 
1699 	apic_pm_activate();
1700 }
1701 
1702 /*
1703  * APIC setup function for application processors. Called from smpboot.c
1704  */
apic_ap_setup(void)1705 void apic_ap_setup(void)
1706 {
1707 	setup_local_APIC();
1708 	end_local_APIC_setup();
1709 }
1710 
1711 #ifdef CONFIG_X86_X2APIC
1712 int x2apic_mode;
1713 
1714 enum {
1715 	X2APIC_OFF,
1716 	X2APIC_ON,
1717 	X2APIC_DISABLED,
1718 };
1719 static int x2apic_state;
1720 
__x2apic_disable(void)1721 static void __x2apic_disable(void)
1722 {
1723 	u64 msr;
1724 
1725 	if (!boot_cpu_has(X86_FEATURE_APIC))
1726 		return;
1727 
1728 	rdmsrl(MSR_IA32_APICBASE, msr);
1729 	if (!(msr & X2APIC_ENABLE))
1730 		return;
1731 	/* Disable xapic and x2apic first and then reenable xapic mode */
1732 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1733 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1734 	printk_once(KERN_INFO "x2apic disabled\n");
1735 }
1736 
__x2apic_enable(void)1737 static void __x2apic_enable(void)
1738 {
1739 	u64 msr;
1740 
1741 	rdmsrl(MSR_IA32_APICBASE, msr);
1742 	if (msr & X2APIC_ENABLE)
1743 		return;
1744 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1745 	printk_once(KERN_INFO "x2apic enabled\n");
1746 }
1747 
setup_nox2apic(char * str)1748 static int __init setup_nox2apic(char *str)
1749 {
1750 	if (x2apic_enabled()) {
1751 		int apicid = native_apic_msr_read(APIC_ID);
1752 
1753 		if (apicid >= 255) {
1754 			pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
1755 				   apicid);
1756 			return 0;
1757 		}
1758 		pr_warning("x2apic already enabled.\n");
1759 		__x2apic_disable();
1760 	}
1761 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1762 	x2apic_state = X2APIC_DISABLED;
1763 	x2apic_mode = 0;
1764 	return 0;
1765 }
1766 early_param("nox2apic", setup_nox2apic);
1767 
1768 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
x2apic_setup(void)1769 void x2apic_setup(void)
1770 {
1771 	/*
1772 	 * If x2apic is not in ON state, disable it if already enabled
1773 	 * from BIOS.
1774 	 */
1775 	if (x2apic_state != X2APIC_ON) {
1776 		__x2apic_disable();
1777 		return;
1778 	}
1779 	__x2apic_enable();
1780 }
1781 
x2apic_disable(void)1782 static __init void x2apic_disable(void)
1783 {
1784 	u32 x2apic_id, state = x2apic_state;
1785 
1786 	x2apic_mode = 0;
1787 	x2apic_state = X2APIC_DISABLED;
1788 
1789 	if (state != X2APIC_ON)
1790 		return;
1791 
1792 	x2apic_id = read_apic_id();
1793 	if (x2apic_id >= 255)
1794 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1795 
1796 	__x2apic_disable();
1797 	register_lapic_address(mp_lapic_addr);
1798 }
1799 
x2apic_enable(void)1800 static __init void x2apic_enable(void)
1801 {
1802 	if (x2apic_state != X2APIC_OFF)
1803 		return;
1804 
1805 	x2apic_mode = 1;
1806 	x2apic_state = X2APIC_ON;
1807 	__x2apic_enable();
1808 }
1809 
try_to_enable_x2apic(int remap_mode)1810 static __init void try_to_enable_x2apic(int remap_mode)
1811 {
1812 	if (x2apic_state == X2APIC_DISABLED)
1813 		return;
1814 
1815 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1816 		/* IR is required if there is APIC ID > 255 even when running
1817 		 * under KVM
1818 		 */
1819 		if (max_physical_apicid > 255 ||
1820 		    !x86_init.hyper.x2apic_available()) {
1821 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1822 			x2apic_disable();
1823 			return;
1824 		}
1825 
1826 		/*
1827 		 * without IR all CPUs can be addressed by IOAPIC/MSI
1828 		 * only in physical mode
1829 		 */
1830 		x2apic_phys = 1;
1831 	}
1832 	x2apic_enable();
1833 }
1834 
check_x2apic(void)1835 void __init check_x2apic(void)
1836 {
1837 	if (x2apic_enabled()) {
1838 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1839 		x2apic_mode = 1;
1840 		x2apic_state = X2APIC_ON;
1841 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1842 		x2apic_state = X2APIC_DISABLED;
1843 	}
1844 }
1845 #else /* CONFIG_X86_X2APIC */
validate_x2apic(void)1846 static int __init validate_x2apic(void)
1847 {
1848 	if (!apic_is_x2apic_enabled())
1849 		return 0;
1850 	/*
1851 	 * Checkme: Can we simply turn off x2apic here instead of panic?
1852 	 */
1853 	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1854 }
1855 early_initcall(validate_x2apic);
1856 
try_to_enable_x2apic(int remap_mode)1857 static inline void try_to_enable_x2apic(int remap_mode) { }
__x2apic_enable(void)1858 static inline void __x2apic_enable(void) { }
1859 #endif /* !CONFIG_X86_X2APIC */
1860 
enable_IR_x2apic(void)1861 void __init enable_IR_x2apic(void)
1862 {
1863 	unsigned long flags;
1864 	int ret, ir_stat;
1865 
1866 	if (skip_ioapic_setup) {
1867 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1868 		return;
1869 	}
1870 
1871 	ir_stat = irq_remapping_prepare();
1872 	if (ir_stat < 0 && !x2apic_supported())
1873 		return;
1874 
1875 	ret = save_ioapic_entries();
1876 	if (ret) {
1877 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1878 		return;
1879 	}
1880 
1881 	local_irq_save(flags);
1882 	legacy_pic->mask_all();
1883 	mask_ioapic_entries();
1884 
1885 	/* If irq_remapping_prepare() succeeded, try to enable it */
1886 	if (ir_stat >= 0)
1887 		ir_stat = irq_remapping_enable();
1888 	/* ir_stat contains the remap mode or an error code */
1889 	try_to_enable_x2apic(ir_stat);
1890 
1891 	if (ir_stat < 0)
1892 		restore_ioapic_entries();
1893 	legacy_pic->restore_mask();
1894 	local_irq_restore(flags);
1895 }
1896 
1897 #ifdef CONFIG_X86_64
1898 /*
1899  * Detect and enable local APICs on non-SMP boards.
1900  * Original code written by Keir Fraser.
1901  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1902  * not correctly set up (usually the APIC timer won't work etc.)
1903  */
detect_init_APIC(void)1904 static int __init detect_init_APIC(void)
1905 {
1906 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1907 		pr_info("No local APIC present\n");
1908 		return -1;
1909 	}
1910 
1911 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1912 	return 0;
1913 }
1914 #else
1915 
apic_verify(void)1916 static int __init apic_verify(void)
1917 {
1918 	u32 features, h, l;
1919 
1920 	/*
1921 	 * The APIC feature bit should now be enabled
1922 	 * in `cpuid'
1923 	 */
1924 	features = cpuid_edx(1);
1925 	if (!(features & (1 << X86_FEATURE_APIC))) {
1926 		pr_warning("Could not enable APIC!\n");
1927 		return -1;
1928 	}
1929 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1930 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1931 
1932 	/* The BIOS may have set up the APIC at some other address */
1933 	if (boot_cpu_data.x86 >= 6) {
1934 		rdmsr(MSR_IA32_APICBASE, l, h);
1935 		if (l & MSR_IA32_APICBASE_ENABLE)
1936 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1937 	}
1938 
1939 	pr_info("Found and enabled local APIC!\n");
1940 	return 0;
1941 }
1942 
apic_force_enable(unsigned long addr)1943 int __init apic_force_enable(unsigned long addr)
1944 {
1945 	u32 h, l;
1946 
1947 	if (disable_apic)
1948 		return -1;
1949 
1950 	/*
1951 	 * Some BIOSes disable the local APIC in the APIC_BASE
1952 	 * MSR. This can only be done in software for Intel P6 or later
1953 	 * and AMD K7 (Model > 1) or later.
1954 	 */
1955 	if (boot_cpu_data.x86 >= 6) {
1956 		rdmsr(MSR_IA32_APICBASE, l, h);
1957 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1958 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1959 			l &= ~MSR_IA32_APICBASE_BASE;
1960 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1961 			wrmsr(MSR_IA32_APICBASE, l, h);
1962 			enabled_via_apicbase = 1;
1963 		}
1964 	}
1965 	return apic_verify();
1966 }
1967 
1968 /*
1969  * Detect and initialize APIC
1970  */
detect_init_APIC(void)1971 static int __init detect_init_APIC(void)
1972 {
1973 	/* Disabled by kernel option? */
1974 	if (disable_apic)
1975 		return -1;
1976 
1977 	switch (boot_cpu_data.x86_vendor) {
1978 	case X86_VENDOR_AMD:
1979 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1980 		    (boot_cpu_data.x86 >= 15))
1981 			break;
1982 		goto no_apic;
1983 	case X86_VENDOR_INTEL:
1984 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1985 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
1986 			break;
1987 		goto no_apic;
1988 	default:
1989 		goto no_apic;
1990 	}
1991 
1992 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1993 		/*
1994 		 * Over-ride BIOS and try to enable the local APIC only if
1995 		 * "lapic" specified.
1996 		 */
1997 		if (!force_enable_local_apic) {
1998 			pr_info("Local APIC disabled by BIOS -- "
1999 				"you can enable it with \"lapic\"\n");
2000 			return -1;
2001 		}
2002 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2003 			return -1;
2004 	} else {
2005 		if (apic_verify())
2006 			return -1;
2007 	}
2008 
2009 	apic_pm_activate();
2010 
2011 	return 0;
2012 
2013 no_apic:
2014 	pr_info("No local APIC present or hardware disabled\n");
2015 	return -1;
2016 }
2017 #endif
2018 
2019 /**
2020  * init_apic_mappings - initialize APIC mappings
2021  */
init_apic_mappings(void)2022 void __init init_apic_mappings(void)
2023 {
2024 	unsigned int new_apicid;
2025 
2026 	if (apic_validate_deadline_timer())
2027 		pr_info("TSC deadline timer available\n");
2028 
2029 	if (x2apic_mode) {
2030 		boot_cpu_physical_apicid = read_apic_id();
2031 		return;
2032 	}
2033 
2034 	/* If no local APIC can be found return early */
2035 	if (!smp_found_config && detect_init_APIC()) {
2036 		/* lets NOP'ify apic operations */
2037 		pr_info("APIC: disable apic facility\n");
2038 		apic_disable();
2039 	} else {
2040 		apic_phys = mp_lapic_addr;
2041 
2042 		/*
2043 		 * If the system has ACPI MADT tables or MP info, the LAPIC
2044 		 * address is already registered.
2045 		 */
2046 		if (!acpi_lapic && !smp_found_config)
2047 			register_lapic_address(apic_phys);
2048 	}
2049 
2050 	/*
2051 	 * Fetch the APIC ID of the BSP in case we have a
2052 	 * default configuration (or the MP table is broken).
2053 	 */
2054 	new_apicid = read_apic_id();
2055 	if (boot_cpu_physical_apicid != new_apicid) {
2056 		boot_cpu_physical_apicid = new_apicid;
2057 		/*
2058 		 * yeah -- we lie about apic_version
2059 		 * in case if apic was disabled via boot option
2060 		 * but it's not a problem for SMP compiled kernel
2061 		 * since apic_intr_mode_select is prepared for such
2062 		 * a case and disable smp mode
2063 		 */
2064 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2065 	}
2066 }
2067 
register_lapic_address(unsigned long address)2068 void __init register_lapic_address(unsigned long address)
2069 {
2070 	mp_lapic_addr = address;
2071 
2072 	if (!x2apic_mode) {
2073 		set_fixmap_nocache(FIX_APIC_BASE, address);
2074 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2075 			    APIC_BASE, address);
2076 	}
2077 	if (boot_cpu_physical_apicid == -1U) {
2078 		boot_cpu_physical_apicid  = read_apic_id();
2079 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2080 	}
2081 }
2082 
2083 /*
2084  * Local APIC interrupts
2085  */
2086 
2087 /*
2088  * This interrupt should _never_ happen with our APIC/SMP architecture
2089  */
smp_spurious_interrupt(struct pt_regs * regs)2090 __visible void __irq_entry smp_spurious_interrupt(struct pt_regs *regs)
2091 {
2092 	u8 vector = ~regs->orig_ax;
2093 	u32 v;
2094 
2095 	entering_irq();
2096 	trace_spurious_apic_entry(vector);
2097 
2098 	inc_irq_stat(irq_spurious_count);
2099 
2100 	/*
2101 	 * If this is a spurious interrupt then do not acknowledge
2102 	 */
2103 	if (vector == SPURIOUS_APIC_VECTOR) {
2104 		/* See SDM vol 3 */
2105 		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2106 			smp_processor_id());
2107 		goto out;
2108 	}
2109 
2110 	/*
2111 	 * If it is a vectored one, verify it's set in the ISR. If set,
2112 	 * acknowledge it.
2113 	 */
2114 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2115 	if (v & (1 << (vector & 0x1f))) {
2116 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2117 			vector, smp_processor_id());
2118 		ack_APIC_irq();
2119 	} else {
2120 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2121 			vector, smp_processor_id());
2122 	}
2123 out:
2124 	trace_spurious_apic_exit(vector);
2125 	exiting_irq();
2126 }
2127 
2128 /*
2129  * This interrupt should never happen with our APIC/SMP architecture
2130  */
smp_error_interrupt(struct pt_regs * regs)2131 __visible void __irq_entry smp_error_interrupt(struct pt_regs *regs)
2132 {
2133 	static const char * const error_interrupt_reason[] = {
2134 		"Send CS error",		/* APIC Error Bit 0 */
2135 		"Receive CS error",		/* APIC Error Bit 1 */
2136 		"Send accept error",		/* APIC Error Bit 2 */
2137 		"Receive accept error",		/* APIC Error Bit 3 */
2138 		"Redirectable IPI",		/* APIC Error Bit 4 */
2139 		"Send illegal vector",		/* APIC Error Bit 5 */
2140 		"Received illegal vector",	/* APIC Error Bit 6 */
2141 		"Illegal register address",	/* APIC Error Bit 7 */
2142 	};
2143 	u32 v, i = 0;
2144 
2145 	entering_irq();
2146 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2147 
2148 	/* First tickle the hardware, only then report what went on. -- REW */
2149 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2150 		apic_write(APIC_ESR, 0);
2151 	v = apic_read(APIC_ESR);
2152 	ack_APIC_irq();
2153 	atomic_inc(&irq_err_count);
2154 
2155 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2156 		    smp_processor_id(), v);
2157 
2158 	v &= 0xff;
2159 	while (v) {
2160 		if (v & 0x1)
2161 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2162 		i++;
2163 		v >>= 1;
2164 	}
2165 
2166 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2167 
2168 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2169 	exiting_irq();
2170 }
2171 
2172 /**
2173  * connect_bsp_APIC - attach the APIC to the interrupt system
2174  */
connect_bsp_APIC(void)2175 static void __init connect_bsp_APIC(void)
2176 {
2177 #ifdef CONFIG_X86_32
2178 	if (pic_mode) {
2179 		/*
2180 		 * Do not trust the local APIC being empty at bootup.
2181 		 */
2182 		clear_local_APIC();
2183 		/*
2184 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2185 		 * local APIC to INT and NMI lines.
2186 		 */
2187 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2188 				"enabling APIC mode.\n");
2189 		imcr_pic_to_apic();
2190 	}
2191 #endif
2192 }
2193 
2194 /**
2195  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2196  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2197  *
2198  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2199  * APIC is disabled.
2200  */
disconnect_bsp_APIC(int virt_wire_setup)2201 void disconnect_bsp_APIC(int virt_wire_setup)
2202 {
2203 	unsigned int value;
2204 
2205 #ifdef CONFIG_X86_32
2206 	if (pic_mode) {
2207 		/*
2208 		 * Put the board back into PIC mode (has an effect only on
2209 		 * certain older boards).  Note that APIC interrupts, including
2210 		 * IPIs, won't work beyond this point!  The only exception are
2211 		 * INIT IPIs.
2212 		 */
2213 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2214 				"entering PIC mode.\n");
2215 		imcr_apic_to_pic();
2216 		return;
2217 	}
2218 #endif
2219 
2220 	/* Go back to Virtual Wire compatibility mode */
2221 
2222 	/* For the spurious interrupt use vector F, and enable it */
2223 	value = apic_read(APIC_SPIV);
2224 	value &= ~APIC_VECTOR_MASK;
2225 	value |= APIC_SPIV_APIC_ENABLED;
2226 	value |= 0xf;
2227 	apic_write(APIC_SPIV, value);
2228 
2229 	if (!virt_wire_setup) {
2230 		/*
2231 		 * For LVT0 make it edge triggered, active high,
2232 		 * external and enabled
2233 		 */
2234 		value = apic_read(APIC_LVT0);
2235 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2236 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2237 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2238 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2239 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2240 		apic_write(APIC_LVT0, value);
2241 	} else {
2242 		/* Disable LVT0 */
2243 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2244 	}
2245 
2246 	/*
2247 	 * For LVT1 make it edge triggered, active high,
2248 	 * nmi and enabled
2249 	 */
2250 	value = apic_read(APIC_LVT1);
2251 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2252 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2253 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2254 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2255 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2256 	apic_write(APIC_LVT1, value);
2257 }
2258 
2259 /*
2260  * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2261  * contiguously, it equals to current allocated max logical CPU ID plus 1.
2262  * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2263  * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2264  *
2265  * NOTE: Reserve 0 for BSP.
2266  */
2267 static int nr_logical_cpuids = 1;
2268 
2269 /*
2270  * Used to store mapping between logical CPU IDs and APIC IDs.
2271  */
2272 static int cpuid_to_apicid[] = {
2273 	[0 ... NR_CPUS - 1] = -1,
2274 };
2275 
2276 #ifdef CONFIG_SMP
2277 /**
2278  * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2279  * @id:	APIC ID to check
2280  */
apic_id_is_primary_thread(unsigned int apicid)2281 bool apic_id_is_primary_thread(unsigned int apicid)
2282 {
2283 	u32 mask;
2284 
2285 	if (smp_num_siblings == 1)
2286 		return true;
2287 	/* Isolate the SMT bit(s) in the APICID and check for 0 */
2288 	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2289 	return !(apicid & mask);
2290 }
2291 #endif
2292 
2293 /*
2294  * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2295  * and cpuid_to_apicid[] synchronized.
2296  */
allocate_logical_cpuid(int apicid)2297 static int allocate_logical_cpuid(int apicid)
2298 {
2299 	int i;
2300 
2301 	/*
2302 	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2303 	 * check if the kernel has allocated a cpuid for it.
2304 	 */
2305 	for (i = 0; i < nr_logical_cpuids; i++) {
2306 		if (cpuid_to_apicid[i] == apicid)
2307 			return i;
2308 	}
2309 
2310 	/* Allocate a new cpuid. */
2311 	if (nr_logical_cpuids >= nr_cpu_ids) {
2312 		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2313 			     "Processor %d/0x%x and the rest are ignored.\n",
2314 			     nr_cpu_ids, nr_logical_cpuids, apicid);
2315 		return -EINVAL;
2316 	}
2317 
2318 	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2319 	return nr_logical_cpuids++;
2320 }
2321 
generic_processor_info(int apicid,int version)2322 int generic_processor_info(int apicid, int version)
2323 {
2324 	int cpu, max = nr_cpu_ids;
2325 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2326 				phys_cpu_present_map);
2327 
2328 	/*
2329 	 * boot_cpu_physical_apicid is designed to have the apicid
2330 	 * returned by read_apic_id(), i.e, the apicid of the
2331 	 * currently booting-up processor. However, on some platforms,
2332 	 * it is temporarily modified by the apicid reported as BSP
2333 	 * through MP table. Concretely:
2334 	 *
2335 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2336 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2337 	 *
2338 	 * This function is executed with the modified
2339 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2340 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2341 	 *
2342 	 * Since fixing handling of boot_cpu_physical_apicid requires
2343 	 * another discussion and tests on each platform, we leave it
2344 	 * for now and here we use read_apic_id() directly in this
2345 	 * function, generic_processor_info().
2346 	 */
2347 	if (disabled_cpu_apicid != BAD_APICID &&
2348 	    disabled_cpu_apicid != read_apic_id() &&
2349 	    disabled_cpu_apicid == apicid) {
2350 		int thiscpu = num_processors + disabled_cpus;
2351 
2352 		pr_warning("APIC: Disabling requested cpu."
2353 			   " Processor %d/0x%x ignored.\n",
2354 			   thiscpu, apicid);
2355 
2356 		disabled_cpus++;
2357 		return -ENODEV;
2358 	}
2359 
2360 	/*
2361 	 * If boot cpu has not been detected yet, then only allow upto
2362 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2363 	 */
2364 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2365 	    apicid != boot_cpu_physical_apicid) {
2366 		int thiscpu = max + disabled_cpus - 1;
2367 
2368 		pr_warning(
2369 			"APIC: NR_CPUS/possible_cpus limit of %i almost"
2370 			" reached. Keeping one slot for boot cpu."
2371 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2372 
2373 		disabled_cpus++;
2374 		return -ENODEV;
2375 	}
2376 
2377 	if (num_processors >= nr_cpu_ids) {
2378 		int thiscpu = max + disabled_cpus;
2379 
2380 		pr_warning("APIC: NR_CPUS/possible_cpus limit of %i "
2381 			   "reached. Processor %d/0x%x ignored.\n",
2382 			   max, thiscpu, apicid);
2383 
2384 		disabled_cpus++;
2385 		return -EINVAL;
2386 	}
2387 
2388 	if (apicid == boot_cpu_physical_apicid) {
2389 		/*
2390 		 * x86_bios_cpu_apicid is required to have processors listed
2391 		 * in same order as logical cpu numbers. Hence the first
2392 		 * entry is BSP, and so on.
2393 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2394 		 * for BSP.
2395 		 */
2396 		cpu = 0;
2397 
2398 		/* Logical cpuid 0 is reserved for BSP. */
2399 		cpuid_to_apicid[0] = apicid;
2400 	} else {
2401 		cpu = allocate_logical_cpuid(apicid);
2402 		if (cpu < 0) {
2403 			disabled_cpus++;
2404 			return -EINVAL;
2405 		}
2406 	}
2407 
2408 	/*
2409 	 * Validate version
2410 	 */
2411 	if (version == 0x0) {
2412 		pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2413 			   cpu, apicid);
2414 		version = 0x10;
2415 	}
2416 
2417 	if (version != boot_cpu_apic_version) {
2418 		pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2419 			boot_cpu_apic_version, cpu, version);
2420 	}
2421 
2422 	if (apicid > max_physical_apicid)
2423 		max_physical_apicid = apicid;
2424 
2425 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2426 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2427 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2428 #endif
2429 #ifdef CONFIG_X86_32
2430 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2431 		apic->x86_32_early_logical_apicid(cpu);
2432 #endif
2433 	set_cpu_possible(cpu, true);
2434 	physid_set(apicid, phys_cpu_present_map);
2435 	set_cpu_present(cpu, true);
2436 	num_processors++;
2437 
2438 	return cpu;
2439 }
2440 
hard_smp_processor_id(void)2441 int hard_smp_processor_id(void)
2442 {
2443 	return read_apic_id();
2444 }
2445 
2446 /*
2447  * Override the generic EOI implementation with an optimized version.
2448  * Only called during early boot when only one CPU is active and with
2449  * interrupts disabled, so we know this does not race with actual APIC driver
2450  * use.
2451  */
apic_set_eoi_write(void (* eoi_write)(u32 reg,u32 v))2452 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2453 {
2454 	struct apic **drv;
2455 
2456 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2457 		/* Should happen once for each apic */
2458 		WARN_ON((*drv)->eoi_write == eoi_write);
2459 		(*drv)->native_eoi_write = (*drv)->eoi_write;
2460 		(*drv)->eoi_write = eoi_write;
2461 	}
2462 }
2463 
apic_bsp_up_setup(void)2464 static void __init apic_bsp_up_setup(void)
2465 {
2466 #ifdef CONFIG_X86_64
2467 	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2468 #else
2469 	/*
2470 	 * Hack: In case of kdump, after a crash, kernel might be booting
2471 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2472 	 * might be zero if read from MP tables. Get it from LAPIC.
2473 	 */
2474 # ifdef CONFIG_CRASH_DUMP
2475 	boot_cpu_physical_apicid = read_apic_id();
2476 # endif
2477 #endif
2478 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2479 }
2480 
2481 /**
2482  * apic_bsp_setup - Setup function for local apic and io-apic
2483  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2484  *
2485  * Returns:
2486  * apic_id of BSP APIC
2487  */
apic_bsp_setup(bool upmode)2488 void __init apic_bsp_setup(bool upmode)
2489 {
2490 	connect_bsp_APIC();
2491 	if (upmode)
2492 		apic_bsp_up_setup();
2493 	setup_local_APIC();
2494 
2495 	enable_IO_APIC();
2496 	end_local_APIC_setup();
2497 	irq_remap_enable_fault_handling();
2498 	setup_IO_APIC();
2499 }
2500 
2501 #ifdef CONFIG_UP_LATE_INIT
up_late_init(void)2502 void __init up_late_init(void)
2503 {
2504 	if (apic_intr_mode == APIC_PIC)
2505 		return;
2506 
2507 	/* Setup local timer */
2508 	x86_init.timers.setup_percpu_clockev();
2509 }
2510 #endif
2511 
2512 /*
2513  * Power management
2514  */
2515 #ifdef CONFIG_PM
2516 
2517 static struct {
2518 	/*
2519 	 * 'active' is true if the local APIC was enabled by us and
2520 	 * not the BIOS; this signifies that we are also responsible
2521 	 * for disabling it before entering apm/acpi suspend
2522 	 */
2523 	int active;
2524 	/* r/w apic fields */
2525 	unsigned int apic_id;
2526 	unsigned int apic_taskpri;
2527 	unsigned int apic_ldr;
2528 	unsigned int apic_dfr;
2529 	unsigned int apic_spiv;
2530 	unsigned int apic_lvtt;
2531 	unsigned int apic_lvtpc;
2532 	unsigned int apic_lvt0;
2533 	unsigned int apic_lvt1;
2534 	unsigned int apic_lvterr;
2535 	unsigned int apic_tmict;
2536 	unsigned int apic_tdcr;
2537 	unsigned int apic_thmr;
2538 	unsigned int apic_cmci;
2539 } apic_pm_state;
2540 
lapic_suspend(void)2541 static int lapic_suspend(void)
2542 {
2543 	unsigned long flags;
2544 	int maxlvt;
2545 
2546 	if (!apic_pm_state.active)
2547 		return 0;
2548 
2549 	maxlvt = lapic_get_maxlvt();
2550 
2551 	apic_pm_state.apic_id = apic_read(APIC_ID);
2552 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2553 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2554 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2555 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2556 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2557 	if (maxlvt >= 4)
2558 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2559 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2560 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2561 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2562 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2563 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2564 #ifdef CONFIG_X86_THERMAL_VECTOR
2565 	if (maxlvt >= 5)
2566 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2567 #endif
2568 #ifdef CONFIG_X86_MCE_INTEL
2569 	if (maxlvt >= 6)
2570 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2571 #endif
2572 
2573 	local_irq_save(flags);
2574 	disable_local_APIC();
2575 
2576 	irq_remapping_disable();
2577 
2578 	local_irq_restore(flags);
2579 	return 0;
2580 }
2581 
lapic_resume(void)2582 static void lapic_resume(void)
2583 {
2584 	unsigned int l, h;
2585 	unsigned long flags;
2586 	int maxlvt;
2587 
2588 	if (!apic_pm_state.active)
2589 		return;
2590 
2591 	local_irq_save(flags);
2592 
2593 	/*
2594 	 * IO-APIC and PIC have their own resume routines.
2595 	 * We just mask them here to make sure the interrupt
2596 	 * subsystem is completely quiet while we enable x2apic
2597 	 * and interrupt-remapping.
2598 	 */
2599 	mask_ioapic_entries();
2600 	legacy_pic->mask_all();
2601 
2602 	if (x2apic_mode) {
2603 		__x2apic_enable();
2604 	} else {
2605 		/*
2606 		 * Make sure the APICBASE points to the right address
2607 		 *
2608 		 * FIXME! This will be wrong if we ever support suspend on
2609 		 * SMP! We'll need to do this as part of the CPU restore!
2610 		 */
2611 		if (boot_cpu_data.x86 >= 6) {
2612 			rdmsr(MSR_IA32_APICBASE, l, h);
2613 			l &= ~MSR_IA32_APICBASE_BASE;
2614 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2615 			wrmsr(MSR_IA32_APICBASE, l, h);
2616 		}
2617 	}
2618 
2619 	maxlvt = lapic_get_maxlvt();
2620 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2621 	apic_write(APIC_ID, apic_pm_state.apic_id);
2622 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2623 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2624 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2625 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2626 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2627 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2628 #ifdef CONFIG_X86_THERMAL_VECTOR
2629 	if (maxlvt >= 5)
2630 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2631 #endif
2632 #ifdef CONFIG_X86_MCE_INTEL
2633 	if (maxlvt >= 6)
2634 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2635 #endif
2636 	if (maxlvt >= 4)
2637 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2638 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2639 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2640 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2641 	apic_write(APIC_ESR, 0);
2642 	apic_read(APIC_ESR);
2643 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2644 	apic_write(APIC_ESR, 0);
2645 	apic_read(APIC_ESR);
2646 
2647 	irq_remapping_reenable(x2apic_mode);
2648 
2649 	local_irq_restore(flags);
2650 }
2651 
2652 /*
2653  * This device has no shutdown method - fully functioning local APICs
2654  * are needed on every CPU up until machine_halt/restart/poweroff.
2655  */
2656 
2657 static struct syscore_ops lapic_syscore_ops = {
2658 	.resume		= lapic_resume,
2659 	.suspend	= lapic_suspend,
2660 };
2661 
apic_pm_activate(void)2662 static void apic_pm_activate(void)
2663 {
2664 	apic_pm_state.active = 1;
2665 }
2666 
init_lapic_sysfs(void)2667 static int __init init_lapic_sysfs(void)
2668 {
2669 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2670 	if (boot_cpu_has(X86_FEATURE_APIC))
2671 		register_syscore_ops(&lapic_syscore_ops);
2672 
2673 	return 0;
2674 }
2675 
2676 /* local apic needs to resume before other devices access its registers. */
2677 core_initcall(init_lapic_sysfs);
2678 
2679 #else	/* CONFIG_PM */
2680 
apic_pm_activate(void)2681 static void apic_pm_activate(void) { }
2682 
2683 #endif	/* CONFIG_PM */
2684 
2685 #ifdef CONFIG_X86_64
2686 
2687 static int multi_checked;
2688 static int multi;
2689 
set_multi(const struct dmi_system_id * d)2690 static int set_multi(const struct dmi_system_id *d)
2691 {
2692 	if (multi)
2693 		return 0;
2694 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2695 	multi = 1;
2696 	return 0;
2697 }
2698 
2699 static const struct dmi_system_id multi_dmi_table[] = {
2700 	{
2701 		.callback = set_multi,
2702 		.ident = "IBM System Summit2",
2703 		.matches = {
2704 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2705 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2706 		},
2707 	},
2708 	{}
2709 };
2710 
dmi_check_multi(void)2711 static void dmi_check_multi(void)
2712 {
2713 	if (multi_checked)
2714 		return;
2715 
2716 	dmi_check_system(multi_dmi_table);
2717 	multi_checked = 1;
2718 }
2719 
2720 /*
2721  * apic_is_clustered_box() -- Check if we can expect good TSC
2722  *
2723  * Thus far, the major user of this is IBM's Summit2 series:
2724  * Clustered boxes may have unsynced TSC problems if they are
2725  * multi-chassis.
2726  * Use DMI to check them
2727  */
apic_is_clustered_box(void)2728 int apic_is_clustered_box(void)
2729 {
2730 	dmi_check_multi();
2731 	return multi;
2732 }
2733 #endif
2734 
2735 /*
2736  * APIC command line parameters
2737  */
setup_disableapic(char * arg)2738 static int __init setup_disableapic(char *arg)
2739 {
2740 	disable_apic = 1;
2741 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2742 	return 0;
2743 }
2744 early_param("disableapic", setup_disableapic);
2745 
2746 /* same as disableapic, for compatibility */
setup_nolapic(char * arg)2747 static int __init setup_nolapic(char *arg)
2748 {
2749 	return setup_disableapic(arg);
2750 }
2751 early_param("nolapic", setup_nolapic);
2752 
parse_lapic_timer_c2_ok(char * arg)2753 static int __init parse_lapic_timer_c2_ok(char *arg)
2754 {
2755 	local_apic_timer_c2_ok = 1;
2756 	return 0;
2757 }
2758 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2759 
parse_disable_apic_timer(char * arg)2760 static int __init parse_disable_apic_timer(char *arg)
2761 {
2762 	disable_apic_timer = 1;
2763 	return 0;
2764 }
2765 early_param("noapictimer", parse_disable_apic_timer);
2766 
parse_nolapic_timer(char * arg)2767 static int __init parse_nolapic_timer(char *arg)
2768 {
2769 	disable_apic_timer = 1;
2770 	return 0;
2771 }
2772 early_param("nolapic_timer", parse_nolapic_timer);
2773 
apic_set_verbosity(char * arg)2774 static int __init apic_set_verbosity(char *arg)
2775 {
2776 	if (!arg)  {
2777 #ifdef CONFIG_X86_64
2778 		skip_ioapic_setup = 0;
2779 		return 0;
2780 #endif
2781 		return -EINVAL;
2782 	}
2783 
2784 	if (strcmp("debug", arg) == 0)
2785 		apic_verbosity = APIC_DEBUG;
2786 	else if (strcmp("verbose", arg) == 0)
2787 		apic_verbosity = APIC_VERBOSE;
2788 #ifdef CONFIG_X86_64
2789 	else {
2790 		pr_warning("APIC Verbosity level %s not recognised"
2791 			" use apic=verbose or apic=debug\n", arg);
2792 		return -EINVAL;
2793 	}
2794 #endif
2795 
2796 	return 0;
2797 }
2798 early_param("apic", apic_set_verbosity);
2799 
lapic_insert_resource(void)2800 static int __init lapic_insert_resource(void)
2801 {
2802 	if (!apic_phys)
2803 		return -1;
2804 
2805 	/* Put local APIC into the resource map. */
2806 	lapic_resource.start = apic_phys;
2807 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2808 	insert_resource(&iomem_resource, &lapic_resource);
2809 
2810 	return 0;
2811 }
2812 
2813 /*
2814  * need call insert after e820__reserve_resources()
2815  * that is using request_resource
2816  */
2817 late_initcall(lapic_insert_resource);
2818 
apic_set_disabled_cpu_apicid(char * arg)2819 static int __init apic_set_disabled_cpu_apicid(char *arg)
2820 {
2821 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2822 		return -EINVAL;
2823 
2824 	return 0;
2825 }
2826 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2827 
apic_set_extnmi(char * arg)2828 static int __init apic_set_extnmi(char *arg)
2829 {
2830 	if (!arg)
2831 		return -EINVAL;
2832 
2833 	if (!strncmp("all", arg, 3))
2834 		apic_extnmi = APIC_EXTNMI_ALL;
2835 	else if (!strncmp("none", arg, 4))
2836 		apic_extnmi = APIC_EXTNMI_NONE;
2837 	else if (!strncmp("bsp", arg, 3))
2838 		apic_extnmi = APIC_EXTNMI_BSP;
2839 	else {
2840 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2841 		return -EINVAL;
2842 	}
2843 
2844 	return 0;
2845 }
2846 early_param("apic_extnmi", apic_set_extnmi);
2847