• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
65 unsigned int num_processors;
66 
67 unsigned disabled_cpus;
68 
69 /* Processor that is doing the boot up */
70 unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
71 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
72 
73 u8 boot_cpu_apic_version __ro_after_init;
74 
75 /*
76  * The highest APIC ID seen during enumeration.
77  */
78 static unsigned int max_physical_apicid;
79 
80 /*
81  * Bitmask of physically existing CPUs:
82  */
83 physid_mask_t phys_cpu_present_map;
84 
85 /*
86  * Processor to be disabled specified by kernel parameter
87  * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
88  * avoid undefined behaviour caused by sending INIT from AP to BSP.
89  */
90 static unsigned int disabled_cpu_apicid __ro_after_init = BAD_APICID;
91 
92 /*
93  * This variable controls which CPUs receive external NMIs.  By default,
94  * external NMIs are delivered only to the BSP.
95  */
96 static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
97 
98 /*
99  * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
100  */
101 static bool virt_ext_dest_id __ro_after_init;
102 
103 /*
104  * Map cpu index to physical APIC ID
105  */
106 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
107 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
108 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
109 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
110 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
111 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
112 
113 #ifdef CONFIG_X86_32
114 
115 /*
116  * On x86_32, the mapping between cpu and logical apicid may vary
117  * depending on apic in use.  The following early percpu variable is
118  * used for the mapping.  This is where the behaviors of x86_64 and 32
119  * actually diverge.  Let's keep it ugly for now.
120  */
121 DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
122 
123 /* Local APIC was disabled by the BIOS and enabled by the kernel */
124 static int enabled_via_apicbase __ro_after_init;
125 
126 /*
127  * Handle interrupt mode configuration register (IMCR).
128  * This register controls whether the interrupt signals
129  * that reach the BSP come from the master PIC or from the
130  * local APIC. Before entering Symmetric I/O Mode, either
131  * the BIOS or the operating system must switch out of
132  * PIC Mode by changing the IMCR.
133  */
imcr_pic_to_apic(void)134 static inline void imcr_pic_to_apic(void)
135 {
136 	/* NMI and 8259 INTR go through APIC */
137 	pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
138 }
139 
imcr_apic_to_pic(void)140 static inline void imcr_apic_to_pic(void)
141 {
142 	/* NMI and 8259 INTR go directly to BSP */
143 	pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
144 }
145 #endif
146 
147 /*
148  * Knob to control our willingness to enable the local APIC.
149  *
150  * +1=force-enable
151  */
152 static int force_enable_local_apic __initdata;
153 
154 /*
155  * APIC command line parameters
156  */
parse_lapic(char * arg)157 static int __init parse_lapic(char *arg)
158 {
159 	if (IS_ENABLED(CONFIG_X86_32) && !arg)
160 		force_enable_local_apic = 1;
161 	else if (arg && !strncmp(arg, "notscdeadline", 13))
162 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
163 	return 0;
164 }
165 early_param("lapic", parse_lapic);
166 
167 #ifdef CONFIG_X86_64
168 static int apic_calibrate_pmtmr __initdata;
setup_apicpmtimer(char * s)169 static __init int setup_apicpmtimer(char *s)
170 {
171 	apic_calibrate_pmtmr = 1;
172 	notsc_setup(NULL);
173 	return 1;
174 }
175 __setup("apicpmtimer", setup_apicpmtimer);
176 #endif
177 
178 unsigned long mp_lapic_addr __ro_after_init;
179 int disable_apic __ro_after_init;
180 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
181 static int disable_apic_timer __initdata;
182 /* Local APIC timer works in C2 */
183 int local_apic_timer_c2_ok __ro_after_init;
184 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
185 
186 /*
187  * Debug level, exported for io_apic.c
188  */
189 int apic_verbosity __ro_after_init;
190 
191 int pic_mode __ro_after_init;
192 
193 /* Have we found an MP table */
194 int smp_found_config __ro_after_init;
195 
196 static struct resource lapic_resource = {
197 	.name = "Local APIC",
198 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
199 };
200 
201 unsigned int lapic_timer_period = 0;
202 
203 static void apic_pm_activate(void);
204 
205 static unsigned long apic_phys __ro_after_init;
206 
207 /*
208  * Get the LAPIC version
209  */
lapic_get_version(void)210 static inline int lapic_get_version(void)
211 {
212 	return GET_APIC_VERSION(apic_read(APIC_LVR));
213 }
214 
215 /*
216  * Check, if the APIC is integrated or a separate chip
217  */
lapic_is_integrated(void)218 static inline int lapic_is_integrated(void)
219 {
220 	return APIC_INTEGRATED(lapic_get_version());
221 }
222 
223 /*
224  * Check, whether this is a modern or a first generation APIC
225  */
modern_apic(void)226 static int modern_apic(void)
227 {
228 	/* AMD systems use old APIC versions, so check the CPU */
229 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
230 	    boot_cpu_data.x86 >= 0xf)
231 		return 1;
232 
233 	/* Hygon systems use modern APIC */
234 	if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
235 		return 1;
236 
237 	return lapic_get_version() >= 0x14;
238 }
239 
240 /*
241  * right after this call apic become NOOP driven
242  * so apic->write/read doesn't do anything
243  */
apic_disable(void)244 static void __init apic_disable(void)
245 {
246 	pr_info("APIC: switched to apic NOOP\n");
247 	apic = &apic_noop;
248 }
249 
native_apic_wait_icr_idle(void)250 void native_apic_wait_icr_idle(void)
251 {
252 	while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
253 		cpu_relax();
254 }
255 
native_safe_apic_wait_icr_idle(void)256 u32 native_safe_apic_wait_icr_idle(void)
257 {
258 	u32 send_status;
259 	int timeout;
260 
261 	timeout = 0;
262 	do {
263 		send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
264 		if (!send_status)
265 			break;
266 		inc_irq_stat(icr_read_retry_count);
267 		udelay(100);
268 	} while (timeout++ < 1000);
269 
270 	return send_status;
271 }
272 
native_apic_icr_write(u32 low,u32 id)273 void native_apic_icr_write(u32 low, u32 id)
274 {
275 	unsigned long flags;
276 
277 	local_irq_save(flags);
278 	apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
279 	apic_write(APIC_ICR, low);
280 	local_irq_restore(flags);
281 }
282 
native_apic_icr_read(void)283 u64 native_apic_icr_read(void)
284 {
285 	u32 icr1, icr2;
286 
287 	icr2 = apic_read(APIC_ICR2);
288 	icr1 = apic_read(APIC_ICR);
289 
290 	return icr1 | ((u64)icr2 << 32);
291 }
292 
293 #ifdef CONFIG_X86_32
294 /**
295  * get_physical_broadcast - Get number of physical broadcast IDs
296  */
get_physical_broadcast(void)297 int get_physical_broadcast(void)
298 {
299 	return modern_apic() ? 0xff : 0xf;
300 }
301 #endif
302 
303 /**
304  * lapic_get_maxlvt - get the maximum number of local vector table entries
305  */
lapic_get_maxlvt(void)306 int lapic_get_maxlvt(void)
307 {
308 	/*
309 	 * - we always have APIC integrated on 64bit mode
310 	 * - 82489DXs do not report # of LVT entries
311 	 */
312 	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
313 }
314 
315 /*
316  * Local APIC timer
317  */
318 
319 /* Clock divisor */
320 #define APIC_DIVISOR 16
321 #define TSC_DIVISOR  8
322 
323 /*
324  * This function sets up the local APIC timer, with a timeout of
325  * 'clocks' APIC bus clock. During calibration we actually call
326  * this function twice on the boot CPU, once with a bogus timeout
327  * value, second time for real. The other (noncalibrating) CPUs
328  * call this function only once, with the real, calibrated value.
329  *
330  * We do reads before writes even if unnecessary, to get around the
331  * P5 APIC double write bug.
332  */
__setup_APIC_LVTT(unsigned int clocks,int oneshot,int irqen)333 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
334 {
335 	unsigned int lvtt_value, tmp_value;
336 
337 	lvtt_value = LOCAL_TIMER_VECTOR;
338 	if (!oneshot)
339 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
340 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
341 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
342 
343 	if (!lapic_is_integrated())
344 		lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
345 
346 	if (!irqen)
347 		lvtt_value |= APIC_LVT_MASKED;
348 
349 	apic_write(APIC_LVTT, lvtt_value);
350 
351 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
352 		/*
353 		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
354 		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
355 		 * According to Intel, MFENCE can do the serialization here.
356 		 */
357 		asm volatile("mfence" : : : "memory");
358 		return;
359 	}
360 
361 	/*
362 	 * Divide PICLK by 16
363 	 */
364 	tmp_value = apic_read(APIC_TDCR);
365 	apic_write(APIC_TDCR,
366 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
367 		APIC_TDR_DIV_16);
368 
369 	if (!oneshot)
370 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
371 }
372 
373 /*
374  * Setup extended LVT, AMD specific
375  *
376  * Software should use the LVT offsets the BIOS provides.  The offsets
377  * are determined by the subsystems using it like those for MCE
378  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
379  * are supported. Beginning with family 10h at least 4 offsets are
380  * available.
381  *
382  * Since the offsets must be consistent for all cores, we keep track
383  * of the LVT offsets in software and reserve the offset for the same
384  * vector also to be used on other cores. An offset is freed by
385  * setting the entry to APIC_EILVT_MASKED.
386  *
387  * If the BIOS is right, there should be no conflicts. Otherwise a
388  * "[Firmware Bug]: ..." error message is generated. However, if
389  * software does not properly determines the offsets, it is not
390  * necessarily a BIOS bug.
391  */
392 
393 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
394 
eilvt_entry_is_changeable(unsigned int old,unsigned int new)395 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
396 {
397 	return (old & APIC_EILVT_MASKED)
398 		|| (new == APIC_EILVT_MASKED)
399 		|| ((new & ~APIC_EILVT_MASKED) == old);
400 }
401 
reserve_eilvt_offset(int offset,unsigned int new)402 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
403 {
404 	unsigned int rsvd, vector;
405 
406 	if (offset >= APIC_EILVT_NR_MAX)
407 		return ~0;
408 
409 	rsvd = atomic_read(&eilvt_offsets[offset]);
410 	do {
411 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
412 		if (vector && !eilvt_entry_is_changeable(vector, new))
413 			/* may not change if vectors are different */
414 			return rsvd;
415 	} while (!atomic_try_cmpxchg(&eilvt_offsets[offset], &rsvd, new));
416 
417 	rsvd = new & ~APIC_EILVT_MASKED;
418 	if (rsvd && rsvd != vector)
419 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
420 			offset, rsvd);
421 
422 	return new;
423 }
424 
425 /*
426  * If mask=1, the LVT entry does not generate interrupts while mask=0
427  * enables the vector. See also the BKDGs. Must be called with
428  * preemption disabled.
429  */
430 
setup_APIC_eilvt(u8 offset,u8 vector,u8 msg_type,u8 mask)431 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
432 {
433 	unsigned long reg = APIC_EILVTn(offset);
434 	unsigned int new, old, reserved;
435 
436 	new = (mask << 16) | (msg_type << 8) | vector;
437 	old = apic_read(reg);
438 	reserved = reserve_eilvt_offset(offset, new);
439 
440 	if (reserved != new) {
441 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
442 		       "vector 0x%x, but the register is already in use for "
443 		       "vector 0x%x on another cpu\n",
444 		       smp_processor_id(), reg, offset, new, reserved);
445 		return -EINVAL;
446 	}
447 
448 	if (!eilvt_entry_is_changeable(old, new)) {
449 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
450 		       "vector 0x%x, but the register is already in use for "
451 		       "vector 0x%x on this cpu\n",
452 		       smp_processor_id(), reg, offset, new, old);
453 		return -EBUSY;
454 	}
455 
456 	apic_write(reg, new);
457 
458 	return 0;
459 }
460 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
461 
462 /*
463  * Program the next event, relative to now
464  */
lapic_next_event(unsigned long delta,struct clock_event_device * evt)465 static int lapic_next_event(unsigned long delta,
466 			    struct clock_event_device *evt)
467 {
468 	apic_write(APIC_TMICT, delta);
469 	return 0;
470 }
471 
lapic_next_deadline(unsigned long delta,struct clock_event_device * evt)472 static int lapic_next_deadline(unsigned long delta,
473 			       struct clock_event_device *evt)
474 {
475 	u64 tsc;
476 
477 	/* This MSR is special and need a special fence: */
478 	weak_wrmsr_fence();
479 
480 	tsc = rdtsc();
481 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
482 	return 0;
483 }
484 
lapic_timer_shutdown(struct clock_event_device * evt)485 static int lapic_timer_shutdown(struct clock_event_device *evt)
486 {
487 	unsigned int v;
488 
489 	/* Lapic used as dummy for broadcast ? */
490 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
491 		return 0;
492 
493 	v = apic_read(APIC_LVTT);
494 	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
495 	apic_write(APIC_LVTT, v);
496 	apic_write(APIC_TMICT, 0);
497 	return 0;
498 }
499 
500 static inline int
lapic_timer_set_periodic_oneshot(struct clock_event_device * evt,bool oneshot)501 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
502 {
503 	/* Lapic used as dummy for broadcast ? */
504 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
505 		return 0;
506 
507 	__setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
508 	return 0;
509 }
510 
lapic_timer_set_periodic(struct clock_event_device * evt)511 static int lapic_timer_set_periodic(struct clock_event_device *evt)
512 {
513 	return lapic_timer_set_periodic_oneshot(evt, false);
514 }
515 
lapic_timer_set_oneshot(struct clock_event_device * evt)516 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
517 {
518 	return lapic_timer_set_periodic_oneshot(evt, true);
519 }
520 
521 /*
522  * Local APIC timer broadcast function
523  */
lapic_timer_broadcast(const struct cpumask * mask)524 static void lapic_timer_broadcast(const struct cpumask *mask)
525 {
526 #ifdef CONFIG_SMP
527 	apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
528 #endif
529 }
530 
531 
532 /*
533  * The local apic timer can be used for any function which is CPU local.
534  */
535 static struct clock_event_device lapic_clockevent = {
536 	.name				= "lapic",
537 	.features			= CLOCK_EVT_FEAT_PERIODIC |
538 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
539 					  | CLOCK_EVT_FEAT_DUMMY,
540 	.shift				= 32,
541 	.set_state_shutdown		= lapic_timer_shutdown,
542 	.set_state_periodic		= lapic_timer_set_periodic,
543 	.set_state_oneshot		= lapic_timer_set_oneshot,
544 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
545 	.set_next_event			= lapic_next_event,
546 	.broadcast			= lapic_timer_broadcast,
547 	.rating				= 100,
548 	.irq				= -1,
549 };
550 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
551 
552 static const struct x86_cpu_id deadline_match[] __initconst = {
553 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
554 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
555 
556 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X,	0x0b000020),
557 
558 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
559 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
560 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
561 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
562 
563 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
564 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
565 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
566 
567 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL,		0x22),
568 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L,		0x20),
569 	X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G,		0x17),
570 
571 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL,		0x25),
572 	X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G,	0x17),
573 
574 	X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L,		0xb2),
575 	X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE,		0xb2),
576 
577 	X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L,		0x52),
578 	X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE,		0x52),
579 
580 	{},
581 };
582 
apic_validate_deadline_timer(void)583 static __init bool apic_validate_deadline_timer(void)
584 {
585 	const struct x86_cpu_id *m;
586 	u32 rev;
587 
588 	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
589 		return false;
590 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
591 		return true;
592 
593 	m = x86_match_cpu(deadline_match);
594 	if (!m)
595 		return true;
596 
597 	rev = (u32)m->driver_data;
598 
599 	if (boot_cpu_data.microcode >= rev)
600 		return true;
601 
602 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
603 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
604 	       "please update microcode to version: 0x%x (or later)\n", rev);
605 	return false;
606 }
607 
608 /*
609  * Setup the local APIC timer for this CPU. Copy the initialized values
610  * of the boot CPU and register the clock event in the framework.
611  */
setup_APIC_timer(void)612 static void setup_APIC_timer(void)
613 {
614 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
615 
616 	if (this_cpu_has(X86_FEATURE_ARAT)) {
617 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
618 		/* Make LAPIC timer preferable over percpu HPET */
619 		lapic_clockevent.rating = 150;
620 	}
621 
622 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
623 	levt->cpumask = cpumask_of(smp_processor_id());
624 
625 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
626 		levt->name = "lapic-deadline";
627 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
628 				    CLOCK_EVT_FEAT_DUMMY);
629 		levt->set_next_event = lapic_next_deadline;
630 		clockevents_config_and_register(levt,
631 						tsc_khz * (1000 / TSC_DIVISOR),
632 						0xF, ~0UL);
633 	} else
634 		clockevents_register_device(levt);
635 }
636 
637 /*
638  * Install the updated TSC frequency from recalibration at the TSC
639  * deadline clockevent devices.
640  */
__lapic_update_tsc_freq(void * info)641 static void __lapic_update_tsc_freq(void *info)
642 {
643 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
644 
645 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
646 		return;
647 
648 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
649 }
650 
lapic_update_tsc_freq(void)651 void lapic_update_tsc_freq(void)
652 {
653 	/*
654 	 * The clockevent device's ->mult and ->shift can both be
655 	 * changed. In order to avoid races, schedule the frequency
656 	 * update code on each CPU.
657 	 */
658 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
659 }
660 
661 /*
662  * In this functions we calibrate APIC bus clocks to the external timer.
663  *
664  * We want to do the calibration only once since we want to have local timer
665  * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
666  * frequency.
667  *
668  * This was previously done by reading the PIT/HPET and waiting for a wrap
669  * around to find out, that a tick has elapsed. I have a box, where the PIT
670  * readout is broken, so it never gets out of the wait loop again. This was
671  * also reported by others.
672  *
673  * Monitoring the jiffies value is inaccurate and the clockevents
674  * infrastructure allows us to do a simple substitution of the interrupt
675  * handler.
676  *
677  * The calibration routine also uses the pm_timer when possible, as the PIT
678  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
679  * back to normal later in the boot process).
680  */
681 
682 #define LAPIC_CAL_LOOPS		(HZ/10)
683 
684 static __initdata int lapic_cal_loops = -1;
685 static __initdata long lapic_cal_t1, lapic_cal_t2;
686 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
687 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
688 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
689 
690 /*
691  * Temporary interrupt handler and polled calibration function.
692  */
lapic_cal_handler(struct clock_event_device * dev)693 static void __init lapic_cal_handler(struct clock_event_device *dev)
694 {
695 	unsigned long long tsc = 0;
696 	long tapic = apic_read(APIC_TMCCT);
697 	unsigned long pm = acpi_pm_read_early();
698 
699 	if (boot_cpu_has(X86_FEATURE_TSC))
700 		tsc = rdtsc();
701 
702 	switch (lapic_cal_loops++) {
703 	case 0:
704 		lapic_cal_t1 = tapic;
705 		lapic_cal_tsc1 = tsc;
706 		lapic_cal_pm1 = pm;
707 		lapic_cal_j1 = jiffies;
708 		break;
709 
710 	case LAPIC_CAL_LOOPS:
711 		lapic_cal_t2 = tapic;
712 		lapic_cal_tsc2 = tsc;
713 		if (pm < lapic_cal_pm1)
714 			pm += ACPI_PM_OVRRUN;
715 		lapic_cal_pm2 = pm;
716 		lapic_cal_j2 = jiffies;
717 		break;
718 	}
719 }
720 
721 static int __init
calibrate_by_pmtimer(long deltapm,long * delta,long * deltatsc)722 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
723 {
724 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
725 	const long pm_thresh = pm_100ms / 100;
726 	unsigned long mult;
727 	u64 res;
728 
729 #ifndef CONFIG_X86_PM_TIMER
730 	return -1;
731 #endif
732 
733 	apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
734 
735 	/* Check, if the PM timer is available */
736 	if (!deltapm)
737 		return -1;
738 
739 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
740 
741 	if (deltapm > (pm_100ms - pm_thresh) &&
742 	    deltapm < (pm_100ms + pm_thresh)) {
743 		apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
744 		return 0;
745 	}
746 
747 	res = (((u64)deltapm) *  mult) >> 22;
748 	do_div(res, 1000000);
749 	pr_warn("APIC calibration not consistent "
750 		"with PM-Timer: %ldms instead of 100ms\n", (long)res);
751 
752 	/* Correct the lapic counter value */
753 	res = (((u64)(*delta)) * pm_100ms);
754 	do_div(res, deltapm);
755 	pr_info("APIC delta adjusted to PM-Timer: "
756 		"%lu (%ld)\n", (unsigned long)res, *delta);
757 	*delta = (long)res;
758 
759 	/* Correct the tsc counter value */
760 	if (boot_cpu_has(X86_FEATURE_TSC)) {
761 		res = (((u64)(*deltatsc)) * pm_100ms);
762 		do_div(res, deltapm);
763 		apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
764 					  "PM-Timer: %lu (%ld)\n",
765 					(unsigned long)res, *deltatsc);
766 		*deltatsc = (long)res;
767 	}
768 
769 	return 0;
770 }
771 
lapic_init_clockevent(void)772 static int __init lapic_init_clockevent(void)
773 {
774 	if (!lapic_timer_period)
775 		return -1;
776 
777 	/* Calculate the scaled math multiplication factor */
778 	lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
779 					TICK_NSEC, lapic_clockevent.shift);
780 	lapic_clockevent.max_delta_ns =
781 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
782 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
783 	lapic_clockevent.min_delta_ns =
784 		clockevent_delta2ns(0xF, &lapic_clockevent);
785 	lapic_clockevent.min_delta_ticks = 0xF;
786 
787 	return 0;
788 }
789 
apic_needs_pit(void)790 bool __init apic_needs_pit(void)
791 {
792 	/*
793 	 * If the frequencies are not known, PIT is required for both TSC
794 	 * and apic timer calibration.
795 	 */
796 	if (!tsc_khz || !cpu_khz)
797 		return true;
798 
799 	/* Is there an APIC at all or is it disabled? */
800 	if (!boot_cpu_has(X86_FEATURE_APIC) || disable_apic)
801 		return true;
802 
803 	/*
804 	 * If interrupt delivery mode is legacy PIC or virtual wire without
805 	 * configuration, the local APIC timer wont be set up. Make sure
806 	 * that the PIT is initialized.
807 	 */
808 	if (apic_intr_mode == APIC_PIC ||
809 	    apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
810 		return true;
811 
812 	/* Virt guests may lack ARAT, but still have DEADLINE */
813 	if (!boot_cpu_has(X86_FEATURE_ARAT))
814 		return true;
815 
816 	/* Deadline timer is based on TSC so no further PIT action required */
817 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
818 		return false;
819 
820 	/* APIC timer disabled? */
821 	if (disable_apic_timer)
822 		return true;
823 	/*
824 	 * The APIC timer frequency is known already, no PIT calibration
825 	 * required. If unknown, let the PIT be initialized.
826 	 */
827 	return lapic_timer_period == 0;
828 }
829 
calibrate_APIC_clock(void)830 static int __init calibrate_APIC_clock(void)
831 {
832 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
833 	u64 tsc_perj = 0, tsc_start = 0;
834 	unsigned long jif_start;
835 	unsigned long deltaj;
836 	long delta, deltatsc;
837 	int pm_referenced = 0;
838 
839 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
840 		return 0;
841 
842 	/*
843 	 * Check if lapic timer has already been calibrated by platform
844 	 * specific routine, such as tsc calibration code. If so just fill
845 	 * in the clockevent structure and return.
846 	 */
847 	if (!lapic_init_clockevent()) {
848 		apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
849 			    lapic_timer_period);
850 		/*
851 		 * Direct calibration methods must have an always running
852 		 * local APIC timer, no need for broadcast timer.
853 		 */
854 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
855 		return 0;
856 	}
857 
858 	apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
859 		    "calibrating APIC timer ...\n");
860 
861 	/*
862 	 * There are platforms w/o global clockevent devices. Instead of
863 	 * making the calibration conditional on that, use a polling based
864 	 * approach everywhere.
865 	 */
866 	local_irq_disable();
867 
868 	/*
869 	 * Setup the APIC counter to maximum. There is no way the lapic
870 	 * can underflow in the 100ms detection time frame
871 	 */
872 	__setup_APIC_LVTT(0xffffffff, 0, 0);
873 
874 	/*
875 	 * Methods to terminate the calibration loop:
876 	 *  1) Global clockevent if available (jiffies)
877 	 *  2) TSC if available and frequency is known
878 	 */
879 	jif_start = READ_ONCE(jiffies);
880 
881 	if (tsc_khz) {
882 		tsc_start = rdtsc();
883 		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
884 	}
885 
886 	/*
887 	 * Enable interrupts so the tick can fire, if a global
888 	 * clockevent device is available
889 	 */
890 	local_irq_enable();
891 
892 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
893 		/* Wait for a tick to elapse */
894 		while (1) {
895 			if (tsc_khz) {
896 				u64 tsc_now = rdtsc();
897 				if ((tsc_now - tsc_start) >= tsc_perj) {
898 					tsc_start += tsc_perj;
899 					break;
900 				}
901 			} else {
902 				unsigned long jif_now = READ_ONCE(jiffies);
903 
904 				if (time_after(jif_now, jif_start)) {
905 					jif_start = jif_now;
906 					break;
907 				}
908 			}
909 			cpu_relax();
910 		}
911 
912 		/* Invoke the calibration routine */
913 		local_irq_disable();
914 		lapic_cal_handler(NULL);
915 		local_irq_enable();
916 	}
917 
918 	local_irq_disable();
919 
920 	/* Build delta t1-t2 as apic timer counts down */
921 	delta = lapic_cal_t1 - lapic_cal_t2;
922 	apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
923 
924 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
925 
926 	/* we trust the PM based calibration if possible */
927 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
928 					&delta, &deltatsc);
929 
930 	lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
931 	lapic_init_clockevent();
932 
933 	apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
934 	apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
935 	apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
936 		    lapic_timer_period);
937 
938 	if (boot_cpu_has(X86_FEATURE_TSC)) {
939 		apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
940 			    "%ld.%04ld MHz.\n",
941 			    (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
942 			    (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
943 	}
944 
945 	apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
946 		    "%u.%04u MHz.\n",
947 		    lapic_timer_period / (1000000 / HZ),
948 		    lapic_timer_period % (1000000 / HZ));
949 
950 	/*
951 	 * Do a sanity check on the APIC calibration result
952 	 */
953 	if (lapic_timer_period < (1000000 / HZ)) {
954 		local_irq_enable();
955 		pr_warn("APIC frequency too slow, disabling apic timer\n");
956 		return -1;
957 	}
958 
959 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
960 
961 	/*
962 	 * PM timer calibration failed or not turned on so lets try APIC
963 	 * timer based calibration, if a global clockevent device is
964 	 * available.
965 	 */
966 	if (!pm_referenced && global_clock_event) {
967 		apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
968 
969 		/*
970 		 * Setup the apic timer manually
971 		 */
972 		levt->event_handler = lapic_cal_handler;
973 		lapic_timer_set_periodic(levt);
974 		lapic_cal_loops = -1;
975 
976 		/* Let the interrupts run */
977 		local_irq_enable();
978 
979 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
980 			cpu_relax();
981 
982 		/* Stop the lapic timer */
983 		local_irq_disable();
984 		lapic_timer_shutdown(levt);
985 
986 		/* Jiffies delta */
987 		deltaj = lapic_cal_j2 - lapic_cal_j1;
988 		apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
989 
990 		/* Check, if the jiffies result is consistent */
991 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
992 			apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
993 		else
994 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
995 	}
996 	local_irq_enable();
997 
998 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
999 		pr_warn("APIC timer disabled due to verification failure\n");
1000 		return -1;
1001 	}
1002 
1003 	return 0;
1004 }
1005 
1006 /*
1007  * Setup the boot APIC
1008  *
1009  * Calibrate and verify the result.
1010  */
setup_boot_APIC_clock(void)1011 void __init setup_boot_APIC_clock(void)
1012 {
1013 	/*
1014 	 * The local apic timer can be disabled via the kernel
1015 	 * commandline or from the CPU detection code. Register the lapic
1016 	 * timer as a dummy clock event source on SMP systems, so the
1017 	 * broadcast mechanism is used. On UP systems simply ignore it.
1018 	 */
1019 	if (disable_apic_timer) {
1020 		pr_info("Disabling APIC timer\n");
1021 		/* No broadcast on UP ! */
1022 		if (num_possible_cpus() > 1) {
1023 			lapic_clockevent.mult = 1;
1024 			setup_APIC_timer();
1025 		}
1026 		return;
1027 	}
1028 
1029 	if (calibrate_APIC_clock()) {
1030 		/* No broadcast on UP ! */
1031 		if (num_possible_cpus() > 1)
1032 			setup_APIC_timer();
1033 		return;
1034 	}
1035 
1036 	/*
1037 	 * If nmi_watchdog is set to IO_APIC, we need the
1038 	 * PIT/HPET going.  Otherwise register lapic as a dummy
1039 	 * device.
1040 	 */
1041 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1042 
1043 	/* Setup the lapic or request the broadcast */
1044 	setup_APIC_timer();
1045 	amd_e400_c1e_apic_setup();
1046 }
1047 
setup_secondary_APIC_clock(void)1048 void setup_secondary_APIC_clock(void)
1049 {
1050 	setup_APIC_timer();
1051 	amd_e400_c1e_apic_setup();
1052 }
1053 
1054 /*
1055  * The guts of the apic timer interrupt
1056  */
local_apic_timer_interrupt(void)1057 static void local_apic_timer_interrupt(void)
1058 {
1059 	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1060 
1061 	/*
1062 	 * Normally we should not be here till LAPIC has been initialized but
1063 	 * in some cases like kdump, its possible that there is a pending LAPIC
1064 	 * timer interrupt from previous kernel's context and is delivered in
1065 	 * new kernel the moment interrupts are enabled.
1066 	 *
1067 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1068 	 * its possible that when we get here evt->event_handler is NULL.
1069 	 * Check for event_handler being NULL and discard the interrupt as
1070 	 * spurious.
1071 	 */
1072 	if (!evt->event_handler) {
1073 		pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1074 			smp_processor_id());
1075 		/* Switch it off */
1076 		lapic_timer_shutdown(evt);
1077 		return;
1078 	}
1079 
1080 	/*
1081 	 * the NMI deadlock-detector uses this.
1082 	 */
1083 	inc_irq_stat(apic_timer_irqs);
1084 
1085 	evt->event_handler(evt);
1086 }
1087 
1088 /*
1089  * Local APIC timer interrupt. This is the most natural way for doing
1090  * local interrupts, but local timer interrupts can be emulated by
1091  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1092  *
1093  * [ if a single-CPU system runs an SMP kernel then we call the local
1094  *   interrupt as well. Thus we cannot inline the local irq ... ]
1095  */
DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)1096 DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1097 {
1098 	struct pt_regs *old_regs = set_irq_regs(regs);
1099 
1100 	ack_APIC_irq();
1101 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1102 	local_apic_timer_interrupt();
1103 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1104 
1105 	set_irq_regs(old_regs);
1106 }
1107 
setup_profiling_timer(unsigned int multiplier)1108 int setup_profiling_timer(unsigned int multiplier)
1109 {
1110 	return -EINVAL;
1111 }
1112 
1113 /*
1114  * Local APIC start and shutdown
1115  */
1116 
1117 /**
1118  * clear_local_APIC - shutdown the local APIC
1119  *
1120  * This is called, when a CPU is disabled and before rebooting, so the state of
1121  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1122  * leftovers during boot.
1123  */
clear_local_APIC(void)1124 void clear_local_APIC(void)
1125 {
1126 	int maxlvt;
1127 	u32 v;
1128 
1129 	/* APIC hasn't been mapped yet */
1130 	if (!x2apic_mode && !apic_phys)
1131 		return;
1132 
1133 	maxlvt = lapic_get_maxlvt();
1134 	/*
1135 	 * Masking an LVT entry can trigger a local APIC error
1136 	 * if the vector is zero. Mask LVTERR first to prevent this.
1137 	 */
1138 	if (maxlvt >= 3) {
1139 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1140 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1141 	}
1142 	/*
1143 	 * Careful: we have to set masks only first to deassert
1144 	 * any level-triggered sources.
1145 	 */
1146 	v = apic_read(APIC_LVTT);
1147 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1148 	v = apic_read(APIC_LVT0);
1149 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1150 	v = apic_read(APIC_LVT1);
1151 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1152 	if (maxlvt >= 4) {
1153 		v = apic_read(APIC_LVTPC);
1154 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1155 	}
1156 
1157 	/* lets not touch this if we didn't frob it */
1158 #ifdef CONFIG_X86_THERMAL_VECTOR
1159 	if (maxlvt >= 5) {
1160 		v = apic_read(APIC_LVTTHMR);
1161 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1162 	}
1163 #endif
1164 #ifdef CONFIG_X86_MCE_INTEL
1165 	if (maxlvt >= 6) {
1166 		v = apic_read(APIC_LVTCMCI);
1167 		if (!(v & APIC_LVT_MASKED))
1168 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1169 	}
1170 #endif
1171 
1172 	/*
1173 	 * Clean APIC state for other OSs:
1174 	 */
1175 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1176 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1177 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1178 	if (maxlvt >= 3)
1179 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1180 	if (maxlvt >= 4)
1181 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1182 
1183 	/* Integrated APIC (!82489DX) ? */
1184 	if (lapic_is_integrated()) {
1185 		if (maxlvt > 3)
1186 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1187 			apic_write(APIC_ESR, 0);
1188 		apic_read(APIC_ESR);
1189 	}
1190 }
1191 
1192 /**
1193  * apic_soft_disable - Clears and software disables the local APIC on hotplug
1194  *
1195  * Contrary to disable_local_APIC() this does not touch the enable bit in
1196  * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1197  * bus would require a hardware reset as the APIC would lose track of bus
1198  * arbitration. On systems with FSB delivery APICBASE could be disabled,
1199  * but it has to be guaranteed that no interrupt is sent to the APIC while
1200  * in that state and it's not clear from the SDM whether it still responds
1201  * to INIT/SIPI messages. Stay on the safe side and use software disable.
1202  */
apic_soft_disable(void)1203 void apic_soft_disable(void)
1204 {
1205 	u32 value;
1206 
1207 	clear_local_APIC();
1208 
1209 	/* Soft disable APIC (implies clearing of registers for 82489DX!). */
1210 	value = apic_read(APIC_SPIV);
1211 	value &= ~APIC_SPIV_APIC_ENABLED;
1212 	apic_write(APIC_SPIV, value);
1213 }
1214 
1215 /**
1216  * disable_local_APIC - clear and disable the local APIC
1217  */
disable_local_APIC(void)1218 void disable_local_APIC(void)
1219 {
1220 	/* APIC hasn't been mapped yet */
1221 	if (!x2apic_mode && !apic_phys)
1222 		return;
1223 
1224 	apic_soft_disable();
1225 
1226 #ifdef CONFIG_X86_32
1227 	/*
1228 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1229 	 * restore the disabled state.
1230 	 */
1231 	if (enabled_via_apicbase) {
1232 		unsigned int l, h;
1233 
1234 		rdmsr(MSR_IA32_APICBASE, l, h);
1235 		l &= ~MSR_IA32_APICBASE_ENABLE;
1236 		wrmsr(MSR_IA32_APICBASE, l, h);
1237 	}
1238 #endif
1239 }
1240 
1241 /*
1242  * If Linux enabled the LAPIC against the BIOS default disable it down before
1243  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1244  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1245  * for the case where Linux didn't enable the LAPIC.
1246  */
lapic_shutdown(void)1247 void lapic_shutdown(void)
1248 {
1249 	unsigned long flags;
1250 
1251 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1252 		return;
1253 
1254 	local_irq_save(flags);
1255 
1256 #ifdef CONFIG_X86_32
1257 	if (!enabled_via_apicbase)
1258 		clear_local_APIC();
1259 	else
1260 #endif
1261 		disable_local_APIC();
1262 
1263 
1264 	local_irq_restore(flags);
1265 }
1266 
1267 /**
1268  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1269  */
sync_Arb_IDs(void)1270 void __init sync_Arb_IDs(void)
1271 {
1272 	/*
1273 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1274 	 * needed on AMD.
1275 	 */
1276 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1277 		return;
1278 
1279 	/*
1280 	 * Wait for idle.
1281 	 */
1282 	apic_wait_icr_idle();
1283 
1284 	apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1285 	apic_write(APIC_ICR, APIC_DEST_ALLINC |
1286 			APIC_INT_LEVELTRIG | APIC_DM_INIT);
1287 }
1288 
1289 enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1290 
__apic_intr_mode_select(void)1291 static int __init __apic_intr_mode_select(void)
1292 {
1293 	/* Check kernel option */
1294 	if (disable_apic) {
1295 		pr_info("APIC disabled via kernel command line\n");
1296 		return APIC_PIC;
1297 	}
1298 
1299 	/* Check BIOS */
1300 #ifdef CONFIG_X86_64
1301 	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1302 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1303 		disable_apic = 1;
1304 		pr_info("APIC disabled by BIOS\n");
1305 		return APIC_PIC;
1306 	}
1307 #else
1308 	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1309 
1310 	/* Neither 82489DX nor integrated APIC ? */
1311 	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1312 		disable_apic = 1;
1313 		return APIC_PIC;
1314 	}
1315 
1316 	/* If the BIOS pretends there is an integrated APIC ? */
1317 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1318 		APIC_INTEGRATED(boot_cpu_apic_version)) {
1319 		disable_apic = 1;
1320 		pr_err(FW_BUG "Local APIC %d not detected, force emulation\n",
1321 				       boot_cpu_physical_apicid);
1322 		return APIC_PIC;
1323 	}
1324 #endif
1325 
1326 	/* Check MP table or ACPI MADT configuration */
1327 	if (!smp_found_config) {
1328 		disable_ioapic_support();
1329 		if (!acpi_lapic) {
1330 			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1331 			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1332 		}
1333 		return APIC_VIRTUAL_WIRE;
1334 	}
1335 
1336 #ifdef CONFIG_SMP
1337 	/* If SMP should be disabled, then really disable it! */
1338 	if (!setup_max_cpus) {
1339 		pr_info("APIC: SMP mode deactivated\n");
1340 		return APIC_SYMMETRIC_IO_NO_ROUTING;
1341 	}
1342 
1343 	if (read_apic_id() != boot_cpu_physical_apicid) {
1344 		panic("Boot APIC ID in local APIC unexpected (%d vs %d)",
1345 		     read_apic_id(), boot_cpu_physical_apicid);
1346 		/* Or can we switch back to PIC here? */
1347 	}
1348 #endif
1349 
1350 	return APIC_SYMMETRIC_IO;
1351 }
1352 
1353 /* Select the interrupt delivery mode for the BSP */
apic_intr_mode_select(void)1354 void __init apic_intr_mode_select(void)
1355 {
1356 	apic_intr_mode = __apic_intr_mode_select();
1357 }
1358 
1359 /*
1360  * An initial setup of the virtual wire mode.
1361  */
init_bsp_APIC(void)1362 void __init init_bsp_APIC(void)
1363 {
1364 	unsigned int value;
1365 
1366 	/*
1367 	 * Don't do the setup now if we have a SMP BIOS as the
1368 	 * through-I/O-APIC virtual wire mode might be active.
1369 	 */
1370 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1371 		return;
1372 
1373 	/*
1374 	 * Do not trust the local APIC being empty at bootup.
1375 	 */
1376 	clear_local_APIC();
1377 
1378 	/*
1379 	 * Enable APIC.
1380 	 */
1381 	value = apic_read(APIC_SPIV);
1382 	value &= ~APIC_VECTOR_MASK;
1383 	value |= APIC_SPIV_APIC_ENABLED;
1384 
1385 #ifdef CONFIG_X86_32
1386 	/* This bit is reserved on P4/Xeon and should be cleared */
1387 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1388 	    (boot_cpu_data.x86 == 15))
1389 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1390 	else
1391 #endif
1392 		value |= APIC_SPIV_FOCUS_DISABLED;
1393 	value |= SPURIOUS_APIC_VECTOR;
1394 	apic_write(APIC_SPIV, value);
1395 
1396 	/*
1397 	 * Set up the virtual wire mode.
1398 	 */
1399 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1400 	value = APIC_DM_NMI;
1401 	if (!lapic_is_integrated())		/* 82489DX */
1402 		value |= APIC_LVT_LEVEL_TRIGGER;
1403 	if (apic_extnmi == APIC_EXTNMI_NONE)
1404 		value |= APIC_LVT_MASKED;
1405 	apic_write(APIC_LVT1, value);
1406 }
1407 
1408 static void __init apic_bsp_setup(bool upmode);
1409 
1410 /* Init the interrupt delivery mode for the BSP */
apic_intr_mode_init(void)1411 void __init apic_intr_mode_init(void)
1412 {
1413 	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1414 
1415 	switch (apic_intr_mode) {
1416 	case APIC_PIC:
1417 		pr_info("APIC: Keep in PIC mode(8259)\n");
1418 		return;
1419 	case APIC_VIRTUAL_WIRE:
1420 		pr_info("APIC: Switch to virtual wire mode setup\n");
1421 		default_setup_apic_routing();
1422 		break;
1423 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1424 		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1425 		upmode = true;
1426 		default_setup_apic_routing();
1427 		break;
1428 	case APIC_SYMMETRIC_IO:
1429 		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1430 		default_setup_apic_routing();
1431 		break;
1432 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1433 		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1434 		break;
1435 	}
1436 
1437 	if (x86_platform.apic_post_init)
1438 		x86_platform.apic_post_init();
1439 
1440 	apic_bsp_setup(upmode);
1441 }
1442 
lapic_setup_esr(void)1443 static void lapic_setup_esr(void)
1444 {
1445 	unsigned int oldvalue, value, maxlvt;
1446 
1447 	if (!lapic_is_integrated()) {
1448 		pr_info("No ESR for 82489DX.\n");
1449 		return;
1450 	}
1451 
1452 	if (apic->disable_esr) {
1453 		/*
1454 		 * Something untraceable is creating bad interrupts on
1455 		 * secondary quads ... for the moment, just leave the
1456 		 * ESR disabled - we can't do anything useful with the
1457 		 * errors anyway - mbligh
1458 		 */
1459 		pr_info("Leaving ESR disabled.\n");
1460 		return;
1461 	}
1462 
1463 	maxlvt = lapic_get_maxlvt();
1464 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1465 		apic_write(APIC_ESR, 0);
1466 	oldvalue = apic_read(APIC_ESR);
1467 
1468 	/* enables sending errors */
1469 	value = ERROR_APIC_VECTOR;
1470 	apic_write(APIC_LVTERR, value);
1471 
1472 	/*
1473 	 * spec says clear errors after enabling vector.
1474 	 */
1475 	if (maxlvt > 3)
1476 		apic_write(APIC_ESR, 0);
1477 	value = apic_read(APIC_ESR);
1478 	if (value != oldvalue)
1479 		apic_printk(APIC_VERBOSE, "ESR value before enabling "
1480 			"vector: 0x%08x  after: 0x%08x\n",
1481 			oldvalue, value);
1482 }
1483 
1484 #define APIC_IR_REGS		APIC_ISR_NR
1485 #define APIC_IR_BITS		(APIC_IR_REGS * 32)
1486 #define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1487 
1488 union apic_ir {
1489 	unsigned long	map[APIC_IR_MAPSIZE];
1490 	u32		regs[APIC_IR_REGS];
1491 };
1492 
apic_check_and_ack(union apic_ir * irr,union apic_ir * isr)1493 static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1494 {
1495 	int i, bit;
1496 
1497 	/* Read the IRRs */
1498 	for (i = 0; i < APIC_IR_REGS; i++)
1499 		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1500 
1501 	/* Read the ISRs */
1502 	for (i = 0; i < APIC_IR_REGS; i++)
1503 		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1504 
1505 	/*
1506 	 * If the ISR map is not empty. ACK the APIC and run another round
1507 	 * to verify whether a pending IRR has been unblocked and turned
1508 	 * into a ISR.
1509 	 */
1510 	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1511 		/*
1512 		 * There can be multiple ISR bits set when a high priority
1513 		 * interrupt preempted a lower priority one. Issue an ACK
1514 		 * per set bit.
1515 		 */
1516 		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1517 			ack_APIC_irq();
1518 		return true;
1519 	}
1520 
1521 	return !bitmap_empty(irr->map, APIC_IR_BITS);
1522 }
1523 
1524 /*
1525  * After a crash, we no longer service the interrupts and a pending
1526  * interrupt from previous kernel might still have ISR bit set.
1527  *
1528  * Most probably by now the CPU has serviced that pending interrupt and it
1529  * might not have done the ack_APIC_irq() because it thought, interrupt
1530  * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1531  * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1532  * a vector might get locked. It was noticed for timer irq (vector
1533  * 0x31). Issue an extra EOI to clear ISR.
1534  *
1535  * If there are pending IRR bits they turn into ISR bits after a higher
1536  * priority ISR bit has been acked.
1537  */
apic_pending_intr_clear(void)1538 static void apic_pending_intr_clear(void)
1539 {
1540 	union apic_ir irr, isr;
1541 	unsigned int i;
1542 
1543 	/* 512 loops are way oversized and give the APIC a chance to obey. */
1544 	for (i = 0; i < 512; i++) {
1545 		if (!apic_check_and_ack(&irr, &isr))
1546 			return;
1547 	}
1548 	/* Dump the IRR/ISR content if that failed */
1549 	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1550 }
1551 
1552 /**
1553  * setup_local_APIC - setup the local APIC
1554  *
1555  * Used to setup local APIC while initializing BSP or bringing up APs.
1556  * Always called with preemption disabled.
1557  */
setup_local_APIC(void)1558 static void setup_local_APIC(void)
1559 {
1560 	int cpu = smp_processor_id();
1561 	unsigned int value;
1562 
1563 	if (disable_apic) {
1564 		disable_ioapic_support();
1565 		return;
1566 	}
1567 
1568 	/*
1569 	 * If this comes from kexec/kcrash the APIC might be enabled in
1570 	 * SPIV. Soft disable it before doing further initialization.
1571 	 */
1572 	value = apic_read(APIC_SPIV);
1573 	value &= ~APIC_SPIV_APIC_ENABLED;
1574 	apic_write(APIC_SPIV, value);
1575 
1576 #ifdef CONFIG_X86_32
1577 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1578 	if (lapic_is_integrated() && apic->disable_esr) {
1579 		apic_write(APIC_ESR, 0);
1580 		apic_write(APIC_ESR, 0);
1581 		apic_write(APIC_ESR, 0);
1582 		apic_write(APIC_ESR, 0);
1583 	}
1584 #endif
1585 	/*
1586 	 * Double-check whether this APIC is really registered.
1587 	 * This is meaningless in clustered apic mode, so we skip it.
1588 	 */
1589 	BUG_ON(!apic->apic_id_registered());
1590 
1591 	/*
1592 	 * Intel recommends to set DFR, LDR and TPR before enabling
1593 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1594 	 * document number 292116).  So here it goes...
1595 	 */
1596 	apic->init_apic_ldr();
1597 
1598 #ifdef CONFIG_X86_32
1599 	if (apic->dest_mode_logical) {
1600 		int logical_apicid, ldr_apicid;
1601 
1602 		/*
1603 		 * APIC LDR is initialized.  If logical_apicid mapping was
1604 		 * initialized during get_smp_config(), make sure it matches
1605 		 * the actual value.
1606 		 */
1607 		logical_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1608 		ldr_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
1609 		if (logical_apicid != BAD_APICID)
1610 			WARN_ON(logical_apicid != ldr_apicid);
1611 		/* Always use the value from LDR. */
1612 		early_per_cpu(x86_cpu_to_logical_apicid, cpu) = ldr_apicid;
1613 	}
1614 #endif
1615 
1616 	/*
1617 	 * Set Task Priority to 'accept all except vectors 0-31'.  An APIC
1618 	 * vector in the 16-31 range could be delivered if TPR == 0, but we
1619 	 * would think it's an exception and terrible things will happen.  We
1620 	 * never change this later on.
1621 	 */
1622 	value = apic_read(APIC_TASKPRI);
1623 	value &= ~APIC_TPRI_MASK;
1624 	value |= 0x10;
1625 	apic_write(APIC_TASKPRI, value);
1626 
1627 	/* Clear eventually stale ISR/IRR bits */
1628 	apic_pending_intr_clear();
1629 
1630 	/*
1631 	 * Now that we are all set up, enable the APIC
1632 	 */
1633 	value = apic_read(APIC_SPIV);
1634 	value &= ~APIC_VECTOR_MASK;
1635 	/*
1636 	 * Enable APIC
1637 	 */
1638 	value |= APIC_SPIV_APIC_ENABLED;
1639 
1640 #ifdef CONFIG_X86_32
1641 	/*
1642 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1643 	 * certain networking cards. If high frequency interrupts are
1644 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1645 	 * entry is masked/unmasked at a high rate as well then sooner or
1646 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1647 	 * from the device. If focus CPU is disabled then the hang goes
1648 	 * away, oh well :-(
1649 	 *
1650 	 * [ This bug can be reproduced easily with a level-triggered
1651 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1652 	 *   BX chipset. ]
1653 	 */
1654 	/*
1655 	 * Actually disabling the focus CPU check just makes the hang less
1656 	 * frequent as it makes the interrupt distribution model be more
1657 	 * like LRU than MRU (the short-term load is more even across CPUs).
1658 	 */
1659 
1660 	/*
1661 	 * - enable focus processor (bit==0)
1662 	 * - 64bit mode always use processor focus
1663 	 *   so no need to set it
1664 	 */
1665 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1666 #endif
1667 
1668 	/*
1669 	 * Set spurious IRQ vector
1670 	 */
1671 	value |= SPURIOUS_APIC_VECTOR;
1672 	apic_write(APIC_SPIV, value);
1673 
1674 	perf_events_lapic_init();
1675 
1676 	/*
1677 	 * Set up LVT0, LVT1:
1678 	 *
1679 	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1680 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1681 	 * we delegate interrupts to the 8259A.
1682 	 */
1683 	/*
1684 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1685 	 */
1686 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1687 	if (!cpu && (pic_mode || !value || skip_ioapic_setup)) {
1688 		value = APIC_DM_EXTINT;
1689 		apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1690 	} else {
1691 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1692 		apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1693 	}
1694 	apic_write(APIC_LVT0, value);
1695 
1696 	/*
1697 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1698 	 * modified by apic_extnmi= boot option.
1699 	 */
1700 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1701 	    apic_extnmi == APIC_EXTNMI_ALL)
1702 		value = APIC_DM_NMI;
1703 	else
1704 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1705 
1706 	/* Is 82489DX ? */
1707 	if (!lapic_is_integrated())
1708 		value |= APIC_LVT_LEVEL_TRIGGER;
1709 	apic_write(APIC_LVT1, value);
1710 
1711 #ifdef CONFIG_X86_MCE_INTEL
1712 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1713 	if (!cpu)
1714 		cmci_recheck();
1715 #endif
1716 }
1717 
end_local_APIC_setup(void)1718 static void end_local_APIC_setup(void)
1719 {
1720 	lapic_setup_esr();
1721 
1722 #ifdef CONFIG_X86_32
1723 	{
1724 		unsigned int value;
1725 		/* Disable the local apic timer */
1726 		value = apic_read(APIC_LVTT);
1727 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1728 		apic_write(APIC_LVTT, value);
1729 	}
1730 #endif
1731 
1732 	apic_pm_activate();
1733 }
1734 
1735 /*
1736  * APIC setup function for application processors. Called from smpboot.c
1737  */
apic_ap_setup(void)1738 void apic_ap_setup(void)
1739 {
1740 	setup_local_APIC();
1741 	end_local_APIC_setup();
1742 }
1743 
1744 #ifdef CONFIG_X86_X2APIC
1745 int x2apic_mode;
1746 EXPORT_SYMBOL_GPL(x2apic_mode);
1747 
1748 enum {
1749 	X2APIC_OFF,
1750 	X2APIC_ON,
1751 	X2APIC_DISABLED,
1752 };
1753 static int x2apic_state;
1754 
__x2apic_disable(void)1755 static void __x2apic_disable(void)
1756 {
1757 	u64 msr;
1758 
1759 	if (!boot_cpu_has(X86_FEATURE_APIC))
1760 		return;
1761 
1762 	rdmsrl(MSR_IA32_APICBASE, msr);
1763 	if (!(msr & X2APIC_ENABLE))
1764 		return;
1765 	/* Disable xapic and x2apic first and then reenable xapic mode */
1766 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1767 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1768 	printk_once(KERN_INFO "x2apic disabled\n");
1769 }
1770 
__x2apic_enable(void)1771 static void __x2apic_enable(void)
1772 {
1773 	u64 msr;
1774 
1775 	rdmsrl(MSR_IA32_APICBASE, msr);
1776 	if (msr & X2APIC_ENABLE)
1777 		return;
1778 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1779 	printk_once(KERN_INFO "x2apic enabled\n");
1780 }
1781 
setup_nox2apic(char * str)1782 static int __init setup_nox2apic(char *str)
1783 {
1784 	if (x2apic_enabled()) {
1785 		int apicid = native_apic_msr_read(APIC_ID);
1786 
1787 		if (apicid >= 255) {
1788 			pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1789 				apicid);
1790 			return 0;
1791 		}
1792 		pr_warn("x2apic already enabled.\n");
1793 		__x2apic_disable();
1794 	}
1795 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1796 	x2apic_state = X2APIC_DISABLED;
1797 	x2apic_mode = 0;
1798 	return 0;
1799 }
1800 early_param("nox2apic", setup_nox2apic);
1801 
1802 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
x2apic_setup(void)1803 void x2apic_setup(void)
1804 {
1805 	/*
1806 	 * If x2apic is not in ON state, disable it if already enabled
1807 	 * from BIOS.
1808 	 */
1809 	if (x2apic_state != X2APIC_ON) {
1810 		__x2apic_disable();
1811 		return;
1812 	}
1813 	__x2apic_enable();
1814 }
1815 
x2apic_disable(void)1816 static __init void x2apic_disable(void)
1817 {
1818 	u32 x2apic_id, state = x2apic_state;
1819 
1820 	x2apic_mode = 0;
1821 	x2apic_state = X2APIC_DISABLED;
1822 
1823 	if (state != X2APIC_ON)
1824 		return;
1825 
1826 	x2apic_id = read_apic_id();
1827 	if (x2apic_id >= 255)
1828 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1829 
1830 	__x2apic_disable();
1831 	register_lapic_address(mp_lapic_addr);
1832 }
1833 
x2apic_enable(void)1834 static __init void x2apic_enable(void)
1835 {
1836 	if (x2apic_state != X2APIC_OFF)
1837 		return;
1838 
1839 	x2apic_mode = 1;
1840 	x2apic_state = X2APIC_ON;
1841 	__x2apic_enable();
1842 }
1843 
try_to_enable_x2apic(int remap_mode)1844 static __init void try_to_enable_x2apic(int remap_mode)
1845 {
1846 	if (x2apic_state == X2APIC_DISABLED)
1847 		return;
1848 
1849 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1850 		u32 apic_limit = 255;
1851 
1852 		/*
1853 		 * Using X2APIC without IR is not architecturally supported
1854 		 * on bare metal but may be supported in guests.
1855 		 */
1856 		if (!x86_init.hyper.x2apic_available()) {
1857 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1858 			x2apic_disable();
1859 			return;
1860 		}
1861 
1862 		/*
1863 		 * If the hypervisor supports extended destination ID in
1864 		 * MSI, that increases the maximum APIC ID that can be
1865 		 * used for non-remapped IRQ domains.
1866 		 */
1867 		if (x86_init.hyper.msi_ext_dest_id()) {
1868 			virt_ext_dest_id = 1;
1869 			apic_limit = 32767;
1870 		}
1871 
1872 		/*
1873 		 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1874 		 * in physical mode, and CPUs with an APIC ID that cannot
1875 		 * be addressed must not be brought online.
1876 		 */
1877 		x2apic_set_max_apicid(apic_limit);
1878 		x2apic_phys = 1;
1879 	}
1880 	x2apic_enable();
1881 }
1882 
check_x2apic(void)1883 void __init check_x2apic(void)
1884 {
1885 	if (x2apic_enabled()) {
1886 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1887 		x2apic_mode = 1;
1888 		x2apic_state = X2APIC_ON;
1889 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1890 		x2apic_state = X2APIC_DISABLED;
1891 	}
1892 }
1893 #else /* CONFIG_X86_X2APIC */
validate_x2apic(void)1894 static int __init validate_x2apic(void)
1895 {
1896 	if (!apic_is_x2apic_enabled())
1897 		return 0;
1898 	/*
1899 	 * Checkme: Can we simply turn off x2apic here instead of panic?
1900 	 */
1901 	panic("BIOS has enabled x2apic but kernel doesn't support x2apic, please disable x2apic in BIOS.\n");
1902 }
1903 early_initcall(validate_x2apic);
1904 
try_to_enable_x2apic(int remap_mode)1905 static inline void try_to_enable_x2apic(int remap_mode) { }
__x2apic_enable(void)1906 static inline void __x2apic_enable(void) { }
1907 #endif /* !CONFIG_X86_X2APIC */
1908 
enable_IR_x2apic(void)1909 void __init enable_IR_x2apic(void)
1910 {
1911 	unsigned long flags;
1912 	int ret, ir_stat;
1913 
1914 	if (skip_ioapic_setup) {
1915 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1916 		return;
1917 	}
1918 
1919 	ir_stat = irq_remapping_prepare();
1920 	if (ir_stat < 0 && !x2apic_supported())
1921 		return;
1922 
1923 	ret = save_ioapic_entries();
1924 	if (ret) {
1925 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1926 		return;
1927 	}
1928 
1929 	local_irq_save(flags);
1930 	legacy_pic->mask_all();
1931 	mask_ioapic_entries();
1932 
1933 	/* If irq_remapping_prepare() succeeded, try to enable it */
1934 	if (ir_stat >= 0)
1935 		ir_stat = irq_remapping_enable();
1936 	/* ir_stat contains the remap mode or an error code */
1937 	try_to_enable_x2apic(ir_stat);
1938 
1939 	if (ir_stat < 0)
1940 		restore_ioapic_entries();
1941 	legacy_pic->restore_mask();
1942 	local_irq_restore(flags);
1943 }
1944 
1945 #ifdef CONFIG_X86_64
1946 /*
1947  * Detect and enable local APICs on non-SMP boards.
1948  * Original code written by Keir Fraser.
1949  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1950  * not correctly set up (usually the APIC timer won't work etc.)
1951  */
detect_init_APIC(void)1952 static int __init detect_init_APIC(void)
1953 {
1954 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1955 		pr_info("No local APIC present\n");
1956 		return -1;
1957 	}
1958 
1959 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1960 	return 0;
1961 }
1962 #else
1963 
apic_verify(void)1964 static int __init apic_verify(void)
1965 {
1966 	u32 features, h, l;
1967 
1968 	/*
1969 	 * The APIC feature bit should now be enabled
1970 	 * in `cpuid'
1971 	 */
1972 	features = cpuid_edx(1);
1973 	if (!(features & (1 << X86_FEATURE_APIC))) {
1974 		pr_warn("Could not enable APIC!\n");
1975 		return -1;
1976 	}
1977 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1978 	mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1979 
1980 	/* The BIOS may have set up the APIC at some other address */
1981 	if (boot_cpu_data.x86 >= 6) {
1982 		rdmsr(MSR_IA32_APICBASE, l, h);
1983 		if (l & MSR_IA32_APICBASE_ENABLE)
1984 			mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1985 	}
1986 
1987 	pr_info("Found and enabled local APIC!\n");
1988 	return 0;
1989 }
1990 
apic_force_enable(unsigned long addr)1991 int __init apic_force_enable(unsigned long addr)
1992 {
1993 	u32 h, l;
1994 
1995 	if (disable_apic)
1996 		return -1;
1997 
1998 	/*
1999 	 * Some BIOSes disable the local APIC in the APIC_BASE
2000 	 * MSR. This can only be done in software for Intel P6 or later
2001 	 * and AMD K7 (Model > 1) or later.
2002 	 */
2003 	if (boot_cpu_data.x86 >= 6) {
2004 		rdmsr(MSR_IA32_APICBASE, l, h);
2005 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2006 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2007 			l &= ~MSR_IA32_APICBASE_BASE;
2008 			l |= MSR_IA32_APICBASE_ENABLE | addr;
2009 			wrmsr(MSR_IA32_APICBASE, l, h);
2010 			enabled_via_apicbase = 1;
2011 		}
2012 	}
2013 	return apic_verify();
2014 }
2015 
2016 /*
2017  * Detect and initialize APIC
2018  */
detect_init_APIC(void)2019 static int __init detect_init_APIC(void)
2020 {
2021 	/* Disabled by kernel option? */
2022 	if (disable_apic)
2023 		return -1;
2024 
2025 	switch (boot_cpu_data.x86_vendor) {
2026 	case X86_VENDOR_AMD:
2027 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2028 		    (boot_cpu_data.x86 >= 15))
2029 			break;
2030 		goto no_apic;
2031 	case X86_VENDOR_HYGON:
2032 		break;
2033 	case X86_VENDOR_INTEL:
2034 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2035 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2036 			break;
2037 		goto no_apic;
2038 	default:
2039 		goto no_apic;
2040 	}
2041 
2042 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2043 		/*
2044 		 * Over-ride BIOS and try to enable the local APIC only if
2045 		 * "lapic" specified.
2046 		 */
2047 		if (!force_enable_local_apic) {
2048 			pr_info("Local APIC disabled by BIOS -- "
2049 				"you can enable it with \"lapic\"\n");
2050 			return -1;
2051 		}
2052 		if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2053 			return -1;
2054 	} else {
2055 		if (apic_verify())
2056 			return -1;
2057 	}
2058 
2059 	apic_pm_activate();
2060 
2061 	return 0;
2062 
2063 no_apic:
2064 	pr_info("No local APIC present or hardware disabled\n");
2065 	return -1;
2066 }
2067 #endif
2068 
2069 /**
2070  * init_apic_mappings - initialize APIC mappings
2071  */
init_apic_mappings(void)2072 void __init init_apic_mappings(void)
2073 {
2074 	unsigned int new_apicid;
2075 
2076 	if (apic_validate_deadline_timer())
2077 		pr_info("TSC deadline timer available\n");
2078 
2079 	if (x2apic_mode) {
2080 		boot_cpu_physical_apicid = read_apic_id();
2081 		return;
2082 	}
2083 
2084 	/* If no local APIC can be found return early */
2085 	if (!smp_found_config && detect_init_APIC()) {
2086 		/* lets NOP'ify apic operations */
2087 		pr_info("APIC: disable apic facility\n");
2088 		apic_disable();
2089 	} else {
2090 		apic_phys = mp_lapic_addr;
2091 
2092 		/*
2093 		 * If the system has ACPI MADT tables or MP info, the LAPIC
2094 		 * address is already registered.
2095 		 */
2096 		if (!acpi_lapic && !smp_found_config)
2097 			register_lapic_address(apic_phys);
2098 	}
2099 
2100 	/*
2101 	 * Fetch the APIC ID of the BSP in case we have a
2102 	 * default configuration (or the MP table is broken).
2103 	 */
2104 	new_apicid = read_apic_id();
2105 	if (boot_cpu_physical_apicid != new_apicid) {
2106 		boot_cpu_physical_apicid = new_apicid;
2107 		/*
2108 		 * yeah -- we lie about apic_version
2109 		 * in case if apic was disabled via boot option
2110 		 * but it's not a problem for SMP compiled kernel
2111 		 * since apic_intr_mode_select is prepared for such
2112 		 * a case and disable smp mode
2113 		 */
2114 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2115 	}
2116 }
2117 
register_lapic_address(unsigned long address)2118 void __init register_lapic_address(unsigned long address)
2119 {
2120 	mp_lapic_addr = address;
2121 
2122 	if (!x2apic_mode) {
2123 		set_fixmap_nocache(FIX_APIC_BASE, address);
2124 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2125 			    APIC_BASE, address);
2126 	}
2127 	if (boot_cpu_physical_apicid == -1U) {
2128 		boot_cpu_physical_apicid  = read_apic_id();
2129 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
2130 	}
2131 }
2132 
2133 /*
2134  * Local APIC interrupts
2135  */
2136 
2137 /*
2138  * Common handling code for spurious_interrupt and spurious_vector entry
2139  * points below. No point in allowing the compiler to inline it twice.
2140  */
handle_spurious_interrupt(u8 vector)2141 static noinline void handle_spurious_interrupt(u8 vector)
2142 {
2143 	u32 v;
2144 
2145 	trace_spurious_apic_entry(vector);
2146 
2147 	inc_irq_stat(irq_spurious_count);
2148 
2149 	/*
2150 	 * If this is a spurious interrupt then do not acknowledge
2151 	 */
2152 	if (vector == SPURIOUS_APIC_VECTOR) {
2153 		/* See SDM vol 3 */
2154 		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2155 			smp_processor_id());
2156 		goto out;
2157 	}
2158 
2159 	/*
2160 	 * If it is a vectored one, verify it's set in the ISR. If set,
2161 	 * acknowledge it.
2162 	 */
2163 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2164 	if (v & (1 << (vector & 0x1f))) {
2165 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2166 			vector, smp_processor_id());
2167 		ack_APIC_irq();
2168 	} else {
2169 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2170 			vector, smp_processor_id());
2171 	}
2172 out:
2173 	trace_spurious_apic_exit(vector);
2174 }
2175 
2176 /**
2177  * spurious_interrupt - Catch all for interrupts raised on unused vectors
2178  * @regs:	Pointer to pt_regs on stack
2179  * @vector:	The vector number
2180  *
2181  * This is invoked from ASM entry code to catch all interrupts which
2182  * trigger on an entry which is routed to the common_spurious idtentry
2183  * point.
2184  */
DEFINE_IDTENTRY_IRQ(spurious_interrupt)2185 DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2186 {
2187 	handle_spurious_interrupt(vector);
2188 }
2189 
DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)2190 DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2191 {
2192 	handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2193 }
2194 
2195 /*
2196  * This interrupt should never happen with our APIC/SMP architecture
2197  */
DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)2198 DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2199 {
2200 	static const char * const error_interrupt_reason[] = {
2201 		"Send CS error",		/* APIC Error Bit 0 */
2202 		"Receive CS error",		/* APIC Error Bit 1 */
2203 		"Send accept error",		/* APIC Error Bit 2 */
2204 		"Receive accept error",		/* APIC Error Bit 3 */
2205 		"Redirectable IPI",		/* APIC Error Bit 4 */
2206 		"Send illegal vector",		/* APIC Error Bit 5 */
2207 		"Received illegal vector",	/* APIC Error Bit 6 */
2208 		"Illegal register address",	/* APIC Error Bit 7 */
2209 	};
2210 	u32 v, i = 0;
2211 
2212 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2213 
2214 	/* First tickle the hardware, only then report what went on. -- REW */
2215 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2216 		apic_write(APIC_ESR, 0);
2217 	v = apic_read(APIC_ESR);
2218 	ack_APIC_irq();
2219 	atomic_inc(&irq_err_count);
2220 
2221 	apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2222 		    smp_processor_id(), v);
2223 
2224 	v &= 0xff;
2225 	while (v) {
2226 		if (v & 0x1)
2227 			apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2228 		i++;
2229 		v >>= 1;
2230 	}
2231 
2232 	apic_printk(APIC_DEBUG, KERN_CONT "\n");
2233 
2234 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2235 }
2236 
2237 /**
2238  * connect_bsp_APIC - attach the APIC to the interrupt system
2239  */
connect_bsp_APIC(void)2240 static void __init connect_bsp_APIC(void)
2241 {
2242 #ifdef CONFIG_X86_32
2243 	if (pic_mode) {
2244 		/*
2245 		 * Do not trust the local APIC being empty at bootup.
2246 		 */
2247 		clear_local_APIC();
2248 		/*
2249 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2250 		 * local APIC to INT and NMI lines.
2251 		 */
2252 		apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2253 				"enabling APIC mode.\n");
2254 		imcr_pic_to_apic();
2255 	}
2256 #endif
2257 }
2258 
2259 /**
2260  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2261  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2262  *
2263  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2264  * APIC is disabled.
2265  */
disconnect_bsp_APIC(int virt_wire_setup)2266 void disconnect_bsp_APIC(int virt_wire_setup)
2267 {
2268 	unsigned int value;
2269 
2270 #ifdef CONFIG_X86_32
2271 	if (pic_mode) {
2272 		/*
2273 		 * Put the board back into PIC mode (has an effect only on
2274 		 * certain older boards).  Note that APIC interrupts, including
2275 		 * IPIs, won't work beyond this point!  The only exception are
2276 		 * INIT IPIs.
2277 		 */
2278 		apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2279 				"entering PIC mode.\n");
2280 		imcr_apic_to_pic();
2281 		return;
2282 	}
2283 #endif
2284 
2285 	/* Go back to Virtual Wire compatibility mode */
2286 
2287 	/* For the spurious interrupt use vector F, and enable it */
2288 	value = apic_read(APIC_SPIV);
2289 	value &= ~APIC_VECTOR_MASK;
2290 	value |= APIC_SPIV_APIC_ENABLED;
2291 	value |= 0xf;
2292 	apic_write(APIC_SPIV, value);
2293 
2294 	if (!virt_wire_setup) {
2295 		/*
2296 		 * For LVT0 make it edge triggered, active high,
2297 		 * external and enabled
2298 		 */
2299 		value = apic_read(APIC_LVT0);
2300 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2301 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2302 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2303 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2304 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2305 		apic_write(APIC_LVT0, value);
2306 	} else {
2307 		/* Disable LVT0 */
2308 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2309 	}
2310 
2311 	/*
2312 	 * For LVT1 make it edge triggered, active high,
2313 	 * nmi and enabled
2314 	 */
2315 	value = apic_read(APIC_LVT1);
2316 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2317 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2318 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2319 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2320 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2321 	apic_write(APIC_LVT1, value);
2322 }
2323 
2324 /*
2325  * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2326  * contiguously, it equals to current allocated max logical CPU ID plus 1.
2327  * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2328  * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2329  *
2330  * NOTE: Reserve 0 for BSP.
2331  */
2332 static int nr_logical_cpuids = 1;
2333 
2334 /*
2335  * Used to store mapping between logical CPU IDs and APIC IDs.
2336  */
2337 static int cpuid_to_apicid[] = {
2338 	[0 ... NR_CPUS - 1] = -1,
2339 };
2340 
arch_match_cpu_phys_id(int cpu,u64 phys_id)2341 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
2342 {
2343 	return phys_id == cpuid_to_apicid[cpu];
2344 }
2345 
2346 #ifdef CONFIG_SMP
2347 /**
2348  * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
2349  * @apicid: APIC ID to check
2350  */
apic_id_is_primary_thread(unsigned int apicid)2351 bool apic_id_is_primary_thread(unsigned int apicid)
2352 {
2353 	u32 mask;
2354 
2355 	if (smp_num_siblings == 1)
2356 		return true;
2357 	/* Isolate the SMT bit(s) in the APICID and check for 0 */
2358 	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2359 	return !(apicid & mask);
2360 }
2361 #endif
2362 
2363 /*
2364  * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2365  * and cpuid_to_apicid[] synchronized.
2366  */
allocate_logical_cpuid(int apicid)2367 static int allocate_logical_cpuid(int apicid)
2368 {
2369 	int i;
2370 
2371 	/*
2372 	 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2373 	 * check if the kernel has allocated a cpuid for it.
2374 	 */
2375 	for (i = 0; i < nr_logical_cpuids; i++) {
2376 		if (cpuid_to_apicid[i] == apicid)
2377 			return i;
2378 	}
2379 
2380 	/* Allocate a new cpuid. */
2381 	if (nr_logical_cpuids >= nr_cpu_ids) {
2382 		WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2383 			     "Processor %d/0x%x and the rest are ignored.\n",
2384 			     nr_cpu_ids, nr_logical_cpuids, apicid);
2385 		return -EINVAL;
2386 	}
2387 
2388 	cpuid_to_apicid[nr_logical_cpuids] = apicid;
2389 	return nr_logical_cpuids++;
2390 }
2391 
generic_processor_info(int apicid,int version)2392 int generic_processor_info(int apicid, int version)
2393 {
2394 	int cpu, max = nr_cpu_ids;
2395 	bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2396 				phys_cpu_present_map);
2397 
2398 	/*
2399 	 * boot_cpu_physical_apicid is designed to have the apicid
2400 	 * returned by read_apic_id(), i.e, the apicid of the
2401 	 * currently booting-up processor. However, on some platforms,
2402 	 * it is temporarily modified by the apicid reported as BSP
2403 	 * through MP table. Concretely:
2404 	 *
2405 	 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2406 	 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2407 	 *
2408 	 * This function is executed with the modified
2409 	 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2410 	 * parameter doesn't work to disable APs on kdump 2nd kernel.
2411 	 *
2412 	 * Since fixing handling of boot_cpu_physical_apicid requires
2413 	 * another discussion and tests on each platform, we leave it
2414 	 * for now and here we use read_apic_id() directly in this
2415 	 * function, generic_processor_info().
2416 	 */
2417 	if (disabled_cpu_apicid != BAD_APICID &&
2418 	    disabled_cpu_apicid != read_apic_id() &&
2419 	    disabled_cpu_apicid == apicid) {
2420 		int thiscpu = num_processors + disabled_cpus;
2421 
2422 		pr_warn("APIC: Disabling requested cpu."
2423 			" Processor %d/0x%x ignored.\n", thiscpu, apicid);
2424 
2425 		disabled_cpus++;
2426 		return -ENODEV;
2427 	}
2428 
2429 	/*
2430 	 * If boot cpu has not been detected yet, then only allow upto
2431 	 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2432 	 */
2433 	if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2434 	    apicid != boot_cpu_physical_apicid) {
2435 		int thiscpu = max + disabled_cpus - 1;
2436 
2437 		pr_warn("APIC: NR_CPUS/possible_cpus limit of %i almost"
2438 			" reached. Keeping one slot for boot cpu."
2439 			"  Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2440 
2441 		disabled_cpus++;
2442 		return -ENODEV;
2443 	}
2444 
2445 	if (num_processors >= nr_cpu_ids) {
2446 		int thiscpu = max + disabled_cpus;
2447 
2448 		pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2449 			"Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2450 
2451 		disabled_cpus++;
2452 		return -EINVAL;
2453 	}
2454 
2455 	if (apicid == boot_cpu_physical_apicid) {
2456 		/*
2457 		 * x86_bios_cpu_apicid is required to have processors listed
2458 		 * in same order as logical cpu numbers. Hence the first
2459 		 * entry is BSP, and so on.
2460 		 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2461 		 * for BSP.
2462 		 */
2463 		cpu = 0;
2464 
2465 		/* Logical cpuid 0 is reserved for BSP. */
2466 		cpuid_to_apicid[0] = apicid;
2467 	} else {
2468 		cpu = allocate_logical_cpuid(apicid);
2469 		if (cpu < 0) {
2470 			disabled_cpus++;
2471 			return -EINVAL;
2472 		}
2473 	}
2474 
2475 	/*
2476 	 * Validate version
2477 	 */
2478 	if (version == 0x0) {
2479 		pr_warn("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2480 			cpu, apicid);
2481 		version = 0x10;
2482 	}
2483 
2484 	if (version != boot_cpu_apic_version) {
2485 		pr_warn("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2486 			boot_cpu_apic_version, cpu, version);
2487 	}
2488 
2489 	if (apicid > max_physical_apicid)
2490 		max_physical_apicid = apicid;
2491 
2492 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2493 	early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2494 	early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2495 #endif
2496 #ifdef CONFIG_X86_32
2497 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2498 		apic->x86_32_early_logical_apicid(cpu);
2499 #endif
2500 	set_cpu_possible(cpu, true);
2501 	physid_set(apicid, phys_cpu_present_map);
2502 	set_cpu_present(cpu, true);
2503 	num_processors++;
2504 
2505 	return cpu;
2506 }
2507 
hard_smp_processor_id(void)2508 int hard_smp_processor_id(void)
2509 {
2510 	return read_apic_id();
2511 }
2512 
__irq_msi_compose_msg(struct irq_cfg * cfg,struct msi_msg * msg,bool dmar)2513 void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2514 			   bool dmar)
2515 {
2516 	memset(msg, 0, sizeof(*msg));
2517 
2518 	msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2519 	msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2520 	msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2521 
2522 	msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2523 	msg->arch_data.vector = cfg->vector;
2524 
2525 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2526 	/*
2527 	 * Only the IOMMU itself can use the trick of putting destination
2528 	 * APIC ID into the high bits of the address. Anything else would
2529 	 * just be writing to memory if it tried that, and needs IR to
2530 	 * address APICs which can't be addressed in the normal 32-bit
2531 	 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2532 	 * some hypervisors allow the extended destination ID field in bits
2533 	 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2534 	 */
2535 	if (dmar)
2536 		msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2537 	else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2538 		msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2539 	else
2540 		WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2541 }
2542 
x86_msi_msg_get_destid(struct msi_msg * msg,bool extid)2543 u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2544 {
2545 	u32 dest = msg->arch_addr_lo.destid_0_7;
2546 
2547 	if (extid)
2548 		dest |= msg->arch_addr_hi.destid_8_31 << 8;
2549 	return dest;
2550 }
2551 EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2552 
2553 /*
2554  * Override the generic EOI implementation with an optimized version.
2555  * Only called during early boot when only one CPU is active and with
2556  * interrupts disabled, so we know this does not race with actual APIC driver
2557  * use.
2558  */
apic_set_eoi_write(void (* eoi_write)(u32 reg,u32 v))2559 void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2560 {
2561 	struct apic **drv;
2562 
2563 	for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2564 		/* Should happen once for each apic */
2565 		WARN_ON((*drv)->eoi_write == eoi_write);
2566 		(*drv)->native_eoi_write = (*drv)->eoi_write;
2567 		(*drv)->eoi_write = eoi_write;
2568 	}
2569 }
2570 
apic_bsp_up_setup(void)2571 static void __init apic_bsp_up_setup(void)
2572 {
2573 #ifdef CONFIG_X86_64
2574 	apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2575 #else
2576 	/*
2577 	 * Hack: In case of kdump, after a crash, kernel might be booting
2578 	 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
2579 	 * might be zero if read from MP tables. Get it from LAPIC.
2580 	 */
2581 # ifdef CONFIG_CRASH_DUMP
2582 	boot_cpu_physical_apicid = read_apic_id();
2583 # endif
2584 #endif
2585 	physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2586 }
2587 
2588 /**
2589  * apic_bsp_setup - Setup function for local apic and io-apic
2590  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2591  */
apic_bsp_setup(bool upmode)2592 static void __init apic_bsp_setup(bool upmode)
2593 {
2594 	connect_bsp_APIC();
2595 	if (upmode)
2596 		apic_bsp_up_setup();
2597 	setup_local_APIC();
2598 
2599 	enable_IO_APIC();
2600 	end_local_APIC_setup();
2601 	irq_remap_enable_fault_handling();
2602 	setup_IO_APIC();
2603 	lapic_update_legacy_vectors();
2604 }
2605 
2606 #ifdef CONFIG_UP_LATE_INIT
up_late_init(void)2607 void __init up_late_init(void)
2608 {
2609 	if (apic_intr_mode == APIC_PIC)
2610 		return;
2611 
2612 	/* Setup local timer */
2613 	x86_init.timers.setup_percpu_clockev();
2614 }
2615 #endif
2616 
2617 /*
2618  * Power management
2619  */
2620 #ifdef CONFIG_PM
2621 
2622 static struct {
2623 	/*
2624 	 * 'active' is true if the local APIC was enabled by us and
2625 	 * not the BIOS; this signifies that we are also responsible
2626 	 * for disabling it before entering apm/acpi suspend
2627 	 */
2628 	int active;
2629 	/* r/w apic fields */
2630 	unsigned int apic_id;
2631 	unsigned int apic_taskpri;
2632 	unsigned int apic_ldr;
2633 	unsigned int apic_dfr;
2634 	unsigned int apic_spiv;
2635 	unsigned int apic_lvtt;
2636 	unsigned int apic_lvtpc;
2637 	unsigned int apic_lvt0;
2638 	unsigned int apic_lvt1;
2639 	unsigned int apic_lvterr;
2640 	unsigned int apic_tmict;
2641 	unsigned int apic_tdcr;
2642 	unsigned int apic_thmr;
2643 	unsigned int apic_cmci;
2644 } apic_pm_state;
2645 
lapic_suspend(void)2646 static int lapic_suspend(void)
2647 {
2648 	unsigned long flags;
2649 	int maxlvt;
2650 
2651 	if (!apic_pm_state.active)
2652 		return 0;
2653 
2654 	maxlvt = lapic_get_maxlvt();
2655 
2656 	apic_pm_state.apic_id = apic_read(APIC_ID);
2657 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2658 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2659 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2660 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2661 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2662 	if (maxlvt >= 4)
2663 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2664 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2665 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2666 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2667 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2668 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2669 #ifdef CONFIG_X86_THERMAL_VECTOR
2670 	if (maxlvt >= 5)
2671 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2672 #endif
2673 #ifdef CONFIG_X86_MCE_INTEL
2674 	if (maxlvt >= 6)
2675 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2676 #endif
2677 
2678 	local_irq_save(flags);
2679 
2680 	/*
2681 	 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2682 	 * entries on some implementations.
2683 	 */
2684 	mask_ioapic_entries();
2685 
2686 	disable_local_APIC();
2687 
2688 	irq_remapping_disable();
2689 
2690 	local_irq_restore(flags);
2691 	return 0;
2692 }
2693 
lapic_resume(void)2694 static void lapic_resume(void)
2695 {
2696 	unsigned int l, h;
2697 	unsigned long flags;
2698 	int maxlvt;
2699 
2700 	if (!apic_pm_state.active)
2701 		return;
2702 
2703 	local_irq_save(flags);
2704 
2705 	/*
2706 	 * IO-APIC and PIC have their own resume routines.
2707 	 * We just mask them here to make sure the interrupt
2708 	 * subsystem is completely quiet while we enable x2apic
2709 	 * and interrupt-remapping.
2710 	 */
2711 	mask_ioapic_entries();
2712 	legacy_pic->mask_all();
2713 
2714 	if (x2apic_mode) {
2715 		__x2apic_enable();
2716 	} else {
2717 		/*
2718 		 * Make sure the APICBASE points to the right address
2719 		 *
2720 		 * FIXME! This will be wrong if we ever support suspend on
2721 		 * SMP! We'll need to do this as part of the CPU restore!
2722 		 */
2723 		if (boot_cpu_data.x86 >= 6) {
2724 			rdmsr(MSR_IA32_APICBASE, l, h);
2725 			l &= ~MSR_IA32_APICBASE_BASE;
2726 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2727 			wrmsr(MSR_IA32_APICBASE, l, h);
2728 		}
2729 	}
2730 
2731 	maxlvt = lapic_get_maxlvt();
2732 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2733 	apic_write(APIC_ID, apic_pm_state.apic_id);
2734 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2735 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2736 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2737 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2738 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2739 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2740 #ifdef CONFIG_X86_THERMAL_VECTOR
2741 	if (maxlvt >= 5)
2742 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2743 #endif
2744 #ifdef CONFIG_X86_MCE_INTEL
2745 	if (maxlvt >= 6)
2746 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2747 #endif
2748 	if (maxlvt >= 4)
2749 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2750 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2751 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2752 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2753 	apic_write(APIC_ESR, 0);
2754 	apic_read(APIC_ESR);
2755 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2756 	apic_write(APIC_ESR, 0);
2757 	apic_read(APIC_ESR);
2758 
2759 	irq_remapping_reenable(x2apic_mode);
2760 
2761 	local_irq_restore(flags);
2762 }
2763 
2764 /*
2765  * This device has no shutdown method - fully functioning local APICs
2766  * are needed on every CPU up until machine_halt/restart/poweroff.
2767  */
2768 
2769 static struct syscore_ops lapic_syscore_ops = {
2770 	.resume		= lapic_resume,
2771 	.suspend	= lapic_suspend,
2772 };
2773 
apic_pm_activate(void)2774 static void apic_pm_activate(void)
2775 {
2776 	apic_pm_state.active = 1;
2777 }
2778 
init_lapic_sysfs(void)2779 static int __init init_lapic_sysfs(void)
2780 {
2781 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2782 	if (boot_cpu_has(X86_FEATURE_APIC))
2783 		register_syscore_ops(&lapic_syscore_ops);
2784 
2785 	return 0;
2786 }
2787 
2788 /* local apic needs to resume before other devices access its registers. */
2789 core_initcall(init_lapic_sysfs);
2790 
2791 #else	/* CONFIG_PM */
2792 
apic_pm_activate(void)2793 static void apic_pm_activate(void) { }
2794 
2795 #endif	/* CONFIG_PM */
2796 
2797 #ifdef CONFIG_X86_64
2798 
2799 static int multi_checked;
2800 static int multi;
2801 
set_multi(const struct dmi_system_id * d)2802 static int set_multi(const struct dmi_system_id *d)
2803 {
2804 	if (multi)
2805 		return 0;
2806 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2807 	multi = 1;
2808 	return 0;
2809 }
2810 
2811 static const struct dmi_system_id multi_dmi_table[] = {
2812 	{
2813 		.callback = set_multi,
2814 		.ident = "IBM System Summit2",
2815 		.matches = {
2816 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2817 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2818 		},
2819 	},
2820 	{}
2821 };
2822 
dmi_check_multi(void)2823 static void dmi_check_multi(void)
2824 {
2825 	if (multi_checked)
2826 		return;
2827 
2828 	dmi_check_system(multi_dmi_table);
2829 	multi_checked = 1;
2830 }
2831 
2832 /*
2833  * apic_is_clustered_box() -- Check if we can expect good TSC
2834  *
2835  * Thus far, the major user of this is IBM's Summit2 series:
2836  * Clustered boxes may have unsynced TSC problems if they are
2837  * multi-chassis.
2838  * Use DMI to check them
2839  */
apic_is_clustered_box(void)2840 int apic_is_clustered_box(void)
2841 {
2842 	dmi_check_multi();
2843 	return multi;
2844 }
2845 #endif
2846 
2847 /*
2848  * APIC command line parameters
2849  */
setup_disableapic(char * arg)2850 static int __init setup_disableapic(char *arg)
2851 {
2852 	disable_apic = 1;
2853 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2854 	return 0;
2855 }
2856 early_param("disableapic", setup_disableapic);
2857 
2858 /* same as disableapic, for compatibility */
setup_nolapic(char * arg)2859 static int __init setup_nolapic(char *arg)
2860 {
2861 	return setup_disableapic(arg);
2862 }
2863 early_param("nolapic", setup_nolapic);
2864 
parse_lapic_timer_c2_ok(char * arg)2865 static int __init parse_lapic_timer_c2_ok(char *arg)
2866 {
2867 	local_apic_timer_c2_ok = 1;
2868 	return 0;
2869 }
2870 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2871 
parse_disable_apic_timer(char * arg)2872 static int __init parse_disable_apic_timer(char *arg)
2873 {
2874 	disable_apic_timer = 1;
2875 	return 0;
2876 }
2877 early_param("noapictimer", parse_disable_apic_timer);
2878 
parse_nolapic_timer(char * arg)2879 static int __init parse_nolapic_timer(char *arg)
2880 {
2881 	disable_apic_timer = 1;
2882 	return 0;
2883 }
2884 early_param("nolapic_timer", parse_nolapic_timer);
2885 
apic_set_verbosity(char * arg)2886 static int __init apic_set_verbosity(char *arg)
2887 {
2888 	if (!arg)  {
2889 #ifdef CONFIG_X86_64
2890 		skip_ioapic_setup = 0;
2891 		return 0;
2892 #endif
2893 		return -EINVAL;
2894 	}
2895 
2896 	if (strcmp("debug", arg) == 0)
2897 		apic_verbosity = APIC_DEBUG;
2898 	else if (strcmp("verbose", arg) == 0)
2899 		apic_verbosity = APIC_VERBOSE;
2900 #ifdef CONFIG_X86_64
2901 	else {
2902 		pr_warn("APIC Verbosity level %s not recognised"
2903 			" use apic=verbose or apic=debug\n", arg);
2904 		return -EINVAL;
2905 	}
2906 #endif
2907 
2908 	return 0;
2909 }
2910 early_param("apic", apic_set_verbosity);
2911 
lapic_insert_resource(void)2912 static int __init lapic_insert_resource(void)
2913 {
2914 	if (!apic_phys)
2915 		return -1;
2916 
2917 	/* Put local APIC into the resource map. */
2918 	lapic_resource.start = apic_phys;
2919 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2920 	insert_resource(&iomem_resource, &lapic_resource);
2921 
2922 	return 0;
2923 }
2924 
2925 /*
2926  * need call insert after e820__reserve_resources()
2927  * that is using request_resource
2928  */
2929 late_initcall(lapic_insert_resource);
2930 
apic_set_disabled_cpu_apicid(char * arg)2931 static int __init apic_set_disabled_cpu_apicid(char *arg)
2932 {
2933 	if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2934 		return -EINVAL;
2935 
2936 	return 0;
2937 }
2938 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2939 
apic_set_extnmi(char * arg)2940 static int __init apic_set_extnmi(char *arg)
2941 {
2942 	if (!arg)
2943 		return -EINVAL;
2944 
2945 	if (!strncmp("all", arg, 3))
2946 		apic_extnmi = APIC_EXTNMI_ALL;
2947 	else if (!strncmp("none", arg, 4))
2948 		apic_extnmi = APIC_EXTNMI_NONE;
2949 	else if (!strncmp("bsp", arg, 3))
2950 		apic_extnmi = APIC_EXTNMI_BSP;
2951 	else {
2952 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2953 		return -EINVAL;
2954 	}
2955 
2956 	return 0;
2957 }
2958 early_param("apic_extnmi", apic_set_extnmi);
2959