xref: /linux/arch/x86/kernel/apic/apic.c (revision 42b16d3ac371a2fac9b6f08fd75f23f34ba3955a)
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/bitmap.h>
23 #include <linux/clockchips.h>
24 #include <linux/interrupt.h>
25 #include <linux/memblock.h>
26 #include <linux/ftrace.h>
27 #include <linux/ioport.h>
28 #include <linux/export.h>
29 #include <linux/syscore_ops.h>
30 #include <linux/delay.h>
31 #include <linux/timex.h>
32 #include <linux/i8253.h>
33 #include <linux/dmar.h>
34 #include <linux/init.h>
35 #include <linux/cpu.h>
36 #include <linux/dmi.h>
37 #include <linux/smp.h>
38 #include <linux/mm.h>
39 
40 #include <xen/xen.h>
41 
42 #include <asm/trace/irq_vectors.h>
43 #include <asm/irq_remapping.h>
44 #include <asm/pc-conf-reg.h>
45 #include <asm/perf_event.h>
46 #include <asm/x86_init.h>
47 #include <linux/atomic.h>
48 #include <asm/barrier.h>
49 #include <asm/mpspec.h>
50 #include <asm/i8259.h>
51 #include <asm/proto.h>
52 #include <asm/traps.h>
53 #include <asm/apic.h>
54 #include <asm/acpi.h>
55 #include <asm/io_apic.h>
56 #include <asm/desc.h>
57 #include <asm/hpet.h>
58 #include <asm/mtrr.h>
59 #include <asm/time.h>
60 #include <asm/smp.h>
61 #include <asm/mce.h>
62 #include <asm/tsc.h>
63 #include <asm/hypervisor.h>
64 #include <asm/cpu_device_id.h>
65 #include <asm/intel-family.h>
66 #include <asm/irq_regs.h>
67 #include <asm/cpu.h>
68 
69 #include "local.h"
70 
71 /* Processor that is doing the boot up */
72 u32 boot_cpu_physical_apicid __ro_after_init = BAD_APICID;
73 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
74 
75 u8 boot_cpu_apic_version __ro_after_init;
76 
77 /*
78  * This variable controls which CPUs receive external NMIs.  By default,
79  * external NMIs are delivered only to the BSP.
80  */
81 static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP;
82 
83 /*
84  * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID
85  */
86 static bool virt_ext_dest_id __ro_after_init;
87 
88 /* For parallel bootup. */
89 unsigned long apic_mmio_base __ro_after_init;
90 
apic_accessible(void)91 static inline bool apic_accessible(void)
92 {
93 	return x2apic_mode || apic_mmio_base;
94 }
95 
96 #ifdef CONFIG_X86_32
97 /* Local APIC was disabled by the BIOS and enabled by the kernel */
98 static int enabled_via_apicbase __ro_after_init;
99 
100 /*
101  * Handle interrupt mode configuration register (IMCR).
102  * This register controls whether the interrupt signals
103  * that reach the BSP come from the master PIC or from the
104  * local APIC. Before entering Symmetric I/O Mode, either
105  * the BIOS or the operating system must switch out of
106  * PIC Mode by changing the IMCR.
107  */
imcr_pic_to_apic(void)108 static inline void imcr_pic_to_apic(void)
109 {
110 	/* NMI and 8259 INTR go through APIC */
111 	pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
112 }
113 
imcr_apic_to_pic(void)114 static inline void imcr_apic_to_pic(void)
115 {
116 	/* NMI and 8259 INTR go directly to BSP */
117 	pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
118 }
119 #endif
120 
121 /*
122  * Knob to control our willingness to enable the local APIC.
123  *
124  * +1=force-enable
125  */
126 static int force_enable_local_apic __initdata;
127 
128 /*
129  * APIC command line parameters
130  */
parse_lapic(char * arg)131 static int __init parse_lapic(char *arg)
132 {
133 	if (IS_ENABLED(CONFIG_X86_32) && !arg)
134 		force_enable_local_apic = 1;
135 	else if (arg && !strncmp(arg, "notscdeadline", 13))
136 		setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
137 	return 0;
138 }
139 early_param("lapic", parse_lapic);
140 
141 #ifdef CONFIG_X86_64
142 static int apic_calibrate_pmtmr __initdata;
setup_apicpmtimer(char * s)143 static __init int setup_apicpmtimer(char *s)
144 {
145 	apic_calibrate_pmtmr = 1;
146 	notsc_setup(NULL);
147 	return 1;
148 }
149 __setup("apicpmtimer", setup_apicpmtimer);
150 #endif
151 
152 static unsigned long mp_lapic_addr __ro_after_init;
153 bool apic_is_disabled __ro_after_init;
154 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
155 static int disable_apic_timer __initdata;
156 /* Local APIC timer works in C2 */
157 int local_apic_timer_c2_ok __ro_after_init;
158 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
159 
160 /*
161  * Debug level, exported for io_apic.c
162  */
163 int apic_verbosity __ro_after_init;
164 
165 int pic_mode __ro_after_init;
166 
167 /* Have we found an MP table */
168 int smp_found_config __ro_after_init;
169 
170 static struct resource lapic_resource = {
171 	.name = "Local APIC",
172 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
173 };
174 
175 unsigned int lapic_timer_period = 0;
176 
177 static void apic_pm_activate(void);
178 
179 /*
180  * Get the LAPIC version
181  */
lapic_get_version(void)182 static inline int lapic_get_version(void)
183 {
184 	return GET_APIC_VERSION(apic_read(APIC_LVR));
185 }
186 
187 /*
188  * Check, if the APIC is integrated or a separate chip
189  */
lapic_is_integrated(void)190 static inline int lapic_is_integrated(void)
191 {
192 	return APIC_INTEGRATED(lapic_get_version());
193 }
194 
195 /*
196  * Check, whether this is a modern or a first generation APIC
197  */
modern_apic(void)198 static int modern_apic(void)
199 {
200 	/* AMD systems use old APIC versions, so check the CPU */
201 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
202 	    boot_cpu_data.x86 >= 0xf)
203 		return 1;
204 
205 	/* Hygon systems use modern APIC */
206 	if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
207 		return 1;
208 
209 	return lapic_get_version() >= 0x14;
210 }
211 
212 /*
213  * right after this call apic become NOOP driven
214  * so apic->write/read doesn't do anything
215  */
apic_disable(void)216 static void __init apic_disable(void)
217 {
218 	apic_install_driver(&apic_noop);
219 }
220 
native_apic_icr_write(u32 low,u32 id)221 void native_apic_icr_write(u32 low, u32 id)
222 {
223 	unsigned long flags;
224 
225 	local_irq_save(flags);
226 	apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
227 	apic_write(APIC_ICR, low);
228 	local_irq_restore(flags);
229 }
230 
native_apic_icr_read(void)231 u64 native_apic_icr_read(void)
232 {
233 	u32 icr1, icr2;
234 
235 	icr2 = apic_read(APIC_ICR2);
236 	icr1 = apic_read(APIC_ICR);
237 
238 	return icr1 | ((u64)icr2 << 32);
239 }
240 
241 /**
242  * lapic_get_maxlvt - get the maximum number of local vector table entries
243  */
lapic_get_maxlvt(void)244 int lapic_get_maxlvt(void)
245 {
246 	/*
247 	 * - we always have APIC integrated on 64bit mode
248 	 * - 82489DXs do not report # of LVT entries
249 	 */
250 	return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
251 }
252 
253 /*
254  * Local APIC timer
255  */
256 
257 /* Clock divisor */
258 #define APIC_DIVISOR 16
259 #define TSC_DIVISOR  8
260 
261 /* i82489DX specific */
262 #define		I82489DX_BASE_DIVIDER		(((0x2) << 18))
263 
264 /*
265  * This function sets up the local APIC timer, with a timeout of
266  * 'clocks' APIC bus clock. During calibration we actually call
267  * this function twice on the boot CPU, once with a bogus timeout
268  * value, second time for real. The other (noncalibrating) CPUs
269  * call this function only once, with the real, calibrated value.
270  *
271  * We do reads before writes even if unnecessary, to get around the
272  * P5 APIC double write bug.
273  */
__setup_APIC_LVTT(unsigned int clocks,int oneshot,int irqen)274 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
275 {
276 	unsigned int lvtt_value, tmp_value;
277 
278 	lvtt_value = LOCAL_TIMER_VECTOR;
279 	if (!oneshot)
280 		lvtt_value |= APIC_LVT_TIMER_PERIODIC;
281 	else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
282 		lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
283 
284 	/*
285 	 * The i82489DX APIC uses bit 18 and 19 for the base divider.  This
286 	 * overlaps with bit 18 on integrated APICs, but is not documented
287 	 * in the SDM. No problem though. i82489DX equipped systems do not
288 	 * have TSC deadline timer.
289 	 */
290 	if (!lapic_is_integrated())
291 		lvtt_value |= I82489DX_BASE_DIVIDER;
292 
293 	if (!irqen)
294 		lvtt_value |= APIC_LVT_MASKED;
295 
296 	apic_write(APIC_LVTT, lvtt_value);
297 
298 	if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
299 		/*
300 		 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
301 		 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
302 		 * According to Intel, MFENCE can do the serialization here.
303 		 */
304 		asm volatile("mfence" : : : "memory");
305 		return;
306 	}
307 
308 	/*
309 	 * Divide PICLK by 16
310 	 */
311 	tmp_value = apic_read(APIC_TDCR);
312 	apic_write(APIC_TDCR,
313 		(tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
314 		APIC_TDR_DIV_16);
315 
316 	if (!oneshot)
317 		apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
318 }
319 
320 /*
321  * Setup extended LVT, AMD specific
322  *
323  * Software should use the LVT offsets the BIOS provides.  The offsets
324  * are determined by the subsystems using it like those for MCE
325  * threshold or IBS.  On K8 only offset 0 (APIC500) and MCE interrupts
326  * are supported. Beginning with family 10h at least 4 offsets are
327  * available.
328  *
329  * Since the offsets must be consistent for all cores, we keep track
330  * of the LVT offsets in software and reserve the offset for the same
331  * vector also to be used on other cores. An offset is freed by
332  * setting the entry to APIC_EILVT_MASKED.
333  *
334  * If the BIOS is right, there should be no conflicts. Otherwise a
335  * "[Firmware Bug]: ..." error message is generated. However, if
336  * software does not properly determines the offsets, it is not
337  * necessarily a BIOS bug.
338  */
339 
340 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
341 
eilvt_entry_is_changeable(unsigned int old,unsigned int new)342 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
343 {
344 	return (old & APIC_EILVT_MASKED)
345 		|| (new == APIC_EILVT_MASKED)
346 		|| ((new & ~APIC_EILVT_MASKED) == old);
347 }
348 
reserve_eilvt_offset(int offset,unsigned int new)349 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
350 {
351 	unsigned int rsvd, vector;
352 
353 	if (offset >= APIC_EILVT_NR_MAX)
354 		return ~0;
355 
356 	rsvd = atomic_read(&eilvt_offsets[offset]);
357 	do {
358 		vector = rsvd & ~APIC_EILVT_MASKED;	/* 0: unassigned */
359 		if (vector && !eilvt_entry_is_changeable(vector, new))
360 			/* may not change if vectors are different */
361 			return rsvd;
362 	} while (!atomic_try_cmpxchg(&eilvt_offsets[offset], &rsvd, new));
363 
364 	rsvd = new & ~APIC_EILVT_MASKED;
365 	if (rsvd && rsvd != vector)
366 		pr_info("LVT offset %d assigned for vector 0x%02x\n",
367 			offset, rsvd);
368 
369 	return new;
370 }
371 
372 /*
373  * If mask=1, the LVT entry does not generate interrupts while mask=0
374  * enables the vector. See also the BKDGs. Must be called with
375  * preemption disabled.
376  */
377 
setup_APIC_eilvt(u8 offset,u8 vector,u8 msg_type,u8 mask)378 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
379 {
380 	unsigned long reg = APIC_EILVTn(offset);
381 	unsigned int new, old, reserved;
382 
383 	new = (mask << 16) | (msg_type << 8) | vector;
384 	old = apic_read(reg);
385 	reserved = reserve_eilvt_offset(offset, new);
386 
387 	if (reserved != new) {
388 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
389 		       "vector 0x%x, but the register is already in use for "
390 		       "vector 0x%x on another cpu\n",
391 		       smp_processor_id(), reg, offset, new, reserved);
392 		return -EINVAL;
393 	}
394 
395 	if (!eilvt_entry_is_changeable(old, new)) {
396 		pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
397 		       "vector 0x%x, but the register is already in use for "
398 		       "vector 0x%x on this cpu\n",
399 		       smp_processor_id(), reg, offset, new, old);
400 		return -EBUSY;
401 	}
402 
403 	apic_write(reg, new);
404 
405 	return 0;
406 }
407 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
408 
409 /*
410  * Program the next event, relative to now
411  */
lapic_next_event(unsigned long delta,struct clock_event_device * evt)412 static int lapic_next_event(unsigned long delta,
413 			    struct clock_event_device *evt)
414 {
415 	apic_write(APIC_TMICT, delta);
416 	return 0;
417 }
418 
lapic_next_deadline(unsigned long delta,struct clock_event_device * evt)419 static int lapic_next_deadline(unsigned long delta,
420 			       struct clock_event_device *evt)
421 {
422 	u64 tsc;
423 
424 	/* This MSR is special and need a special fence: */
425 	weak_wrmsr_fence();
426 
427 	tsc = rdtsc();
428 	wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
429 	return 0;
430 }
431 
lapic_timer_shutdown(struct clock_event_device * evt)432 static int lapic_timer_shutdown(struct clock_event_device *evt)
433 {
434 	unsigned int v;
435 
436 	/* Lapic used as dummy for broadcast ? */
437 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
438 		return 0;
439 
440 	v = apic_read(APIC_LVTT);
441 	v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
442 	apic_write(APIC_LVTT, v);
443 	apic_write(APIC_TMICT, 0);
444 	return 0;
445 }
446 
447 static inline int
lapic_timer_set_periodic_oneshot(struct clock_event_device * evt,bool oneshot)448 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
449 {
450 	/* Lapic used as dummy for broadcast ? */
451 	if (evt->features & CLOCK_EVT_FEAT_DUMMY)
452 		return 0;
453 
454 	__setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
455 	return 0;
456 }
457 
lapic_timer_set_periodic(struct clock_event_device * evt)458 static int lapic_timer_set_periodic(struct clock_event_device *evt)
459 {
460 	return lapic_timer_set_periodic_oneshot(evt, false);
461 }
462 
lapic_timer_set_oneshot(struct clock_event_device * evt)463 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
464 {
465 	return lapic_timer_set_periodic_oneshot(evt, true);
466 }
467 
468 /*
469  * Local APIC timer broadcast function
470  */
lapic_timer_broadcast(const struct cpumask * mask)471 static void lapic_timer_broadcast(const struct cpumask *mask)
472 {
473 #ifdef CONFIG_SMP
474 	__apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
475 #endif
476 }
477 
478 
479 /*
480  * The local apic timer can be used for any function which is CPU local.
481  */
482 static struct clock_event_device lapic_clockevent = {
483 	.name				= "lapic",
484 	.features			= CLOCK_EVT_FEAT_PERIODIC |
485 					  CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
486 					  | CLOCK_EVT_FEAT_DUMMY,
487 	.shift				= 32,
488 	.set_state_shutdown		= lapic_timer_shutdown,
489 	.set_state_periodic		= lapic_timer_set_periodic,
490 	.set_state_oneshot		= lapic_timer_set_oneshot,
491 	.set_state_oneshot_stopped	= lapic_timer_shutdown,
492 	.set_next_event			= lapic_next_event,
493 	.broadcast			= lapic_timer_broadcast,
494 	.rating				= 100,
495 	.irq				= -1,
496 };
497 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
498 
499 static const struct x86_cpu_id deadline_match[] __initconst = {
500 	X86_MATCH_VFM_STEPPINGS(INTEL_HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
501 	X86_MATCH_VFM_STEPPINGS(INTEL_HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
502 
503 	X86_MATCH_VFM(INTEL_BROADWELL_X,	0x0b000020),
504 
505 	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
506 	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
507 	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
508 	X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
509 
510 	X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
511 	X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
512 	X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
513 
514 	X86_MATCH_VFM(INTEL_HASWELL,		0x22),
515 	X86_MATCH_VFM(INTEL_HASWELL_L,		0x20),
516 	X86_MATCH_VFM(INTEL_HASWELL_G,		0x17),
517 
518 	X86_MATCH_VFM(INTEL_BROADWELL,		0x25),
519 	X86_MATCH_VFM(INTEL_BROADWELL_G,	0x17),
520 
521 	X86_MATCH_VFM(INTEL_SKYLAKE_L,		0xb2),
522 	X86_MATCH_VFM(INTEL_SKYLAKE,		0xb2),
523 
524 	X86_MATCH_VFM(INTEL_KABYLAKE_L,		0x52),
525 	X86_MATCH_VFM(INTEL_KABYLAKE,		0x52),
526 
527 	{},
528 };
529 
apic_validate_deadline_timer(void)530 static __init bool apic_validate_deadline_timer(void)
531 {
532 	const struct x86_cpu_id *m;
533 	u32 rev;
534 
535 	if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
536 		return false;
537 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
538 		return true;
539 
540 	m = x86_match_cpu(deadline_match);
541 	if (!m)
542 		return true;
543 
544 	rev = (u32)m->driver_data;
545 
546 	if (boot_cpu_data.microcode >= rev)
547 		return true;
548 
549 	setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
550 	pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
551 	       "please update microcode to version: 0x%x (or later)\n", rev);
552 	return false;
553 }
554 
555 /*
556  * Setup the local APIC timer for this CPU. Copy the initialized values
557  * of the boot CPU and register the clock event in the framework.
558  */
setup_APIC_timer(void)559 static void setup_APIC_timer(void)
560 {
561 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
562 
563 	if (this_cpu_has(X86_FEATURE_ARAT)) {
564 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
565 		/* Make LAPIC timer preferable over percpu HPET */
566 		lapic_clockevent.rating = 150;
567 	}
568 
569 	memcpy(levt, &lapic_clockevent, sizeof(*levt));
570 	levt->cpumask = cpumask_of(smp_processor_id());
571 
572 	if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
573 		levt->name = "lapic-deadline";
574 		levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
575 				    CLOCK_EVT_FEAT_DUMMY);
576 		levt->set_next_event = lapic_next_deadline;
577 		clockevents_config_and_register(levt,
578 						tsc_khz * (1000 / TSC_DIVISOR),
579 						0xF, ~0UL);
580 	} else
581 		clockevents_register_device(levt);
582 }
583 
584 /*
585  * Install the updated TSC frequency from recalibration at the TSC
586  * deadline clockevent devices.
587  */
__lapic_update_tsc_freq(void * info)588 static void __lapic_update_tsc_freq(void *info)
589 {
590 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
591 
592 	if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
593 		return;
594 
595 	clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
596 }
597 
lapic_update_tsc_freq(void)598 void lapic_update_tsc_freq(void)
599 {
600 	/*
601 	 * The clockevent device's ->mult and ->shift can both be
602 	 * changed. In order to avoid races, schedule the frequency
603 	 * update code on each CPU.
604 	 */
605 	on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
606 }
607 
608 /*
609  * In this functions we calibrate APIC bus clocks to the external timer.
610  *
611  * We want to do the calibration only once since we want to have local timer
612  * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
613  * frequency.
614  *
615  * This was previously done by reading the PIT/HPET and waiting for a wrap
616  * around to find out, that a tick has elapsed. I have a box, where the PIT
617  * readout is broken, so it never gets out of the wait loop again. This was
618  * also reported by others.
619  *
620  * Monitoring the jiffies value is inaccurate and the clockevents
621  * infrastructure allows us to do a simple substitution of the interrupt
622  * handler.
623  *
624  * The calibration routine also uses the pm_timer when possible, as the PIT
625  * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
626  * back to normal later in the boot process).
627  */
628 
629 #define LAPIC_CAL_LOOPS		(HZ/10)
630 
631 static __initdata int lapic_cal_loops = -1;
632 static __initdata long lapic_cal_t1, lapic_cal_t2;
633 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
634 static __initdata u32 lapic_cal_pm1, lapic_cal_pm2;
635 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
636 
637 /*
638  * Temporary interrupt handler and polled calibration function.
639  */
lapic_cal_handler(struct clock_event_device * dev)640 static void __init lapic_cal_handler(struct clock_event_device *dev)
641 {
642 	unsigned long long tsc = 0;
643 	long tapic = apic_read(APIC_TMCCT);
644 	u32 pm = acpi_pm_read_early();
645 
646 	if (boot_cpu_has(X86_FEATURE_TSC))
647 		tsc = rdtsc();
648 
649 	switch (lapic_cal_loops++) {
650 	case 0:
651 		lapic_cal_t1 = tapic;
652 		lapic_cal_tsc1 = tsc;
653 		lapic_cal_pm1 = pm;
654 		lapic_cal_j1 = jiffies;
655 		break;
656 
657 	case LAPIC_CAL_LOOPS:
658 		lapic_cal_t2 = tapic;
659 		lapic_cal_tsc2 = tsc;
660 		if (pm < lapic_cal_pm1)
661 			pm += ACPI_PM_OVRRUN;
662 		lapic_cal_pm2 = pm;
663 		lapic_cal_j2 = jiffies;
664 		break;
665 	}
666 }
667 
668 static int __init
calibrate_by_pmtimer(u32 deltapm,long * delta,long * deltatsc)669 calibrate_by_pmtimer(u32 deltapm, long *delta, long *deltatsc)
670 {
671 	const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
672 	const long pm_thresh = pm_100ms / 100;
673 	unsigned long mult;
674 	u64 res;
675 
676 #ifndef CONFIG_X86_PM_TIMER
677 	return -1;
678 #endif
679 
680 	apic_pr_verbose("... PM-Timer delta = %u\n", deltapm);
681 
682 	/* Check, if the PM timer is available */
683 	if (!deltapm)
684 		return -1;
685 
686 	mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
687 
688 	if (deltapm > (pm_100ms - pm_thresh) &&
689 	    deltapm < (pm_100ms + pm_thresh)) {
690 		apic_pr_verbose("... PM-Timer result ok\n");
691 		return 0;
692 	}
693 
694 	res = (((u64)deltapm) *  mult) >> 22;
695 	do_div(res, 1000000);
696 	pr_warn("APIC calibration not consistent with PM-Timer: %ldms instead of 100ms\n",
697 		(long)res);
698 
699 	/* Correct the lapic counter value */
700 	res = (((u64)(*delta)) * pm_100ms);
701 	do_div(res, deltapm);
702 	pr_info("APIC delta adjusted to PM-Timer: "
703 		"%lu (%ld)\n", (unsigned long)res, *delta);
704 	*delta = (long)res;
705 
706 	/* Correct the tsc counter value */
707 	if (boot_cpu_has(X86_FEATURE_TSC)) {
708 		res = (((u64)(*deltatsc)) * pm_100ms);
709 		do_div(res, deltapm);
710 		apic_pr_verbose("TSC delta adjusted to PM-Timer: %lu (%ld)\n",
711 				(unsigned long)res, *deltatsc);
712 		*deltatsc = (long)res;
713 	}
714 
715 	return 0;
716 }
717 
lapic_init_clockevent(void)718 static int __init lapic_init_clockevent(void)
719 {
720 	if (!lapic_timer_period)
721 		return -1;
722 
723 	/* Calculate the scaled math multiplication factor */
724 	lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
725 					TICK_NSEC, lapic_clockevent.shift);
726 	lapic_clockevent.max_delta_ns =
727 		clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
728 	lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
729 	lapic_clockevent.min_delta_ns =
730 		clockevent_delta2ns(0xF, &lapic_clockevent);
731 	lapic_clockevent.min_delta_ticks = 0xF;
732 
733 	return 0;
734 }
735 
apic_needs_pit(void)736 bool __init apic_needs_pit(void)
737 {
738 	/*
739 	 * If the frequencies are not known, PIT is required for both TSC
740 	 * and apic timer calibration.
741 	 */
742 	if (!tsc_khz || !cpu_khz)
743 		return true;
744 
745 	/* Is there an APIC at all or is it disabled? */
746 	if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled)
747 		return true;
748 
749 	/*
750 	 * If interrupt delivery mode is legacy PIC or virtual wire without
751 	 * configuration, the local APIC timer won't be set up. Make sure
752 	 * that the PIT is initialized.
753 	 */
754 	if (apic_intr_mode == APIC_PIC ||
755 	    apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
756 		return true;
757 
758 	/* Virt guests may lack ARAT, but still have DEADLINE */
759 	if (!boot_cpu_has(X86_FEATURE_ARAT))
760 		return true;
761 
762 	/* Deadline timer is based on TSC so no further PIT action required */
763 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
764 		return false;
765 
766 	/* APIC timer disabled? */
767 	if (disable_apic_timer)
768 		return true;
769 	/*
770 	 * The APIC timer frequency is known already, no PIT calibration
771 	 * required. If unknown, let the PIT be initialized.
772 	 */
773 	return lapic_timer_period == 0;
774 }
775 
calibrate_APIC_clock(void)776 static int __init calibrate_APIC_clock(void)
777 {
778 	struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
779 	u64 tsc_perj = 0, tsc_start = 0;
780 	unsigned long jif_start;
781 	unsigned long deltaj;
782 	long delta, deltatsc;
783 	int pm_referenced = 0;
784 
785 	if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
786 		return 0;
787 
788 	/*
789 	 * Check if lapic timer has already been calibrated by platform
790 	 * specific routine, such as tsc calibration code. If so just fill
791 	 * in the clockevent structure and return.
792 	 */
793 	if (!lapic_init_clockevent()) {
794 		apic_pr_verbose("lapic timer already calibrated %d\n", lapic_timer_period);
795 		/*
796 		 * Direct calibration methods must have an always running
797 		 * local APIC timer, no need for broadcast timer.
798 		 */
799 		lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
800 		return 0;
801 	}
802 
803 	apic_pr_verbose("Using local APIC timer interrupts. Calibrating APIC timer ...\n");
804 
805 	/*
806 	 * There are platforms w/o global clockevent devices. Instead of
807 	 * making the calibration conditional on that, use a polling based
808 	 * approach everywhere.
809 	 */
810 	local_irq_disable();
811 
812 	/*
813 	 * Setup the APIC counter to maximum. There is no way the lapic
814 	 * can underflow in the 100ms detection time frame
815 	 */
816 	__setup_APIC_LVTT(0xffffffff, 0, 0);
817 
818 	/*
819 	 * Methods to terminate the calibration loop:
820 	 *  1) Global clockevent if available (jiffies)
821 	 *  2) TSC if available and frequency is known
822 	 */
823 	jif_start = READ_ONCE(jiffies);
824 
825 	if (tsc_khz) {
826 		tsc_start = rdtsc();
827 		tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
828 	}
829 
830 	/*
831 	 * Enable interrupts so the tick can fire, if a global
832 	 * clockevent device is available
833 	 */
834 	local_irq_enable();
835 
836 	while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
837 		/* Wait for a tick to elapse */
838 		while (1) {
839 			if (tsc_khz) {
840 				u64 tsc_now = rdtsc();
841 				if ((tsc_now - tsc_start) >= tsc_perj) {
842 					tsc_start += tsc_perj;
843 					break;
844 				}
845 			} else {
846 				unsigned long jif_now = READ_ONCE(jiffies);
847 
848 				if (time_after(jif_now, jif_start)) {
849 					jif_start = jif_now;
850 					break;
851 				}
852 			}
853 			cpu_relax();
854 		}
855 
856 		/* Invoke the calibration routine */
857 		local_irq_disable();
858 		lapic_cal_handler(NULL);
859 		local_irq_enable();
860 	}
861 
862 	local_irq_disable();
863 
864 	/* Build delta t1-t2 as apic timer counts down */
865 	delta = lapic_cal_t1 - lapic_cal_t2;
866 	apic_pr_verbose("... lapic delta = %ld\n", delta);
867 
868 	deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
869 
870 	/* we trust the PM based calibration if possible */
871 	pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
872 					&delta, &deltatsc);
873 
874 	lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
875 	lapic_init_clockevent();
876 
877 	apic_pr_verbose("..... delta %ld\n", delta);
878 	apic_pr_verbose("..... mult: %u\n", lapic_clockevent.mult);
879 	apic_pr_verbose("..... calibration result: %u\n", lapic_timer_period);
880 
881 	if (boot_cpu_has(X86_FEATURE_TSC)) {
882 		apic_pr_verbose("..... CPU clock speed is %ld.%04ld MHz.\n",
883 				(deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
884 				(deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
885 	}
886 
887 	apic_pr_verbose("..... host bus clock speed is %u.%04u MHz.\n",
888 			lapic_timer_period / (1000000 / HZ),
889 			lapic_timer_period % (1000000 / HZ));
890 
891 	/*
892 	 * Do a sanity check on the APIC calibration result
893 	 */
894 	if (lapic_timer_period < (1000000 / HZ)) {
895 		local_irq_enable();
896 		pr_warn("APIC frequency too slow, disabling apic timer\n");
897 		return -1;
898 	}
899 
900 	levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
901 
902 	/*
903 	 * PM timer calibration failed or not turned on so lets try APIC
904 	 * timer based calibration, if a global clockevent device is
905 	 * available.
906 	 */
907 	if (!pm_referenced && global_clock_event) {
908 		apic_pr_verbose("... verify APIC timer\n");
909 
910 		/*
911 		 * Setup the apic timer manually
912 		 */
913 		levt->event_handler = lapic_cal_handler;
914 		lapic_timer_set_periodic(levt);
915 		lapic_cal_loops = -1;
916 
917 		/* Let the interrupts run */
918 		local_irq_enable();
919 
920 		while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
921 			cpu_relax();
922 
923 		/* Stop the lapic timer */
924 		local_irq_disable();
925 		lapic_timer_shutdown(levt);
926 
927 		/* Jiffies delta */
928 		deltaj = lapic_cal_j2 - lapic_cal_j1;
929 		apic_pr_verbose("... jiffies delta = %lu\n", deltaj);
930 
931 		/* Check, if the jiffies result is consistent */
932 		if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
933 			apic_pr_verbose("... jiffies result ok\n");
934 		else
935 			levt->features |= CLOCK_EVT_FEAT_DUMMY;
936 	}
937 	local_irq_enable();
938 
939 	if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
940 		pr_warn("APIC timer disabled due to verification failure\n");
941 		return -1;
942 	}
943 
944 	return 0;
945 }
946 
947 /*
948  * Setup the boot APIC
949  *
950  * Calibrate and verify the result.
951  */
setup_boot_APIC_clock(void)952 void __init setup_boot_APIC_clock(void)
953 {
954 	/*
955 	 * The local apic timer can be disabled via the kernel
956 	 * commandline or from the CPU detection code. Register the lapic
957 	 * timer as a dummy clock event source on SMP systems, so the
958 	 * broadcast mechanism is used. On UP systems simply ignore it.
959 	 */
960 	if (disable_apic_timer) {
961 		pr_info("Disabling APIC timer\n");
962 		/* No broadcast on UP ! */
963 		if (num_possible_cpus() > 1) {
964 			lapic_clockevent.mult = 1;
965 			setup_APIC_timer();
966 		}
967 		return;
968 	}
969 
970 	if (calibrate_APIC_clock()) {
971 		/* No broadcast on UP ! */
972 		if (num_possible_cpus() > 1)
973 			setup_APIC_timer();
974 		return;
975 	}
976 
977 	/*
978 	 * If nmi_watchdog is set to IO_APIC, we need the
979 	 * PIT/HPET going.  Otherwise register lapic as a dummy
980 	 * device.
981 	 */
982 	lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
983 
984 	/* Setup the lapic or request the broadcast */
985 	setup_APIC_timer();
986 	amd_e400_c1e_apic_setup();
987 }
988 
setup_secondary_APIC_clock(void)989 void setup_secondary_APIC_clock(void)
990 {
991 	setup_APIC_timer();
992 	amd_e400_c1e_apic_setup();
993 }
994 
995 /*
996  * The guts of the apic timer interrupt
997  */
local_apic_timer_interrupt(void)998 static void local_apic_timer_interrupt(void)
999 {
1000 	struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1001 
1002 	/*
1003 	 * Normally we should not be here till LAPIC has been initialized but
1004 	 * in some cases like kdump, its possible that there is a pending LAPIC
1005 	 * timer interrupt from previous kernel's context and is delivered in
1006 	 * new kernel the moment interrupts are enabled.
1007 	 *
1008 	 * Interrupts are enabled early and LAPIC is setup much later, hence
1009 	 * its possible that when we get here evt->event_handler is NULL.
1010 	 * Check for event_handler being NULL and discard the interrupt as
1011 	 * spurious.
1012 	 */
1013 	if (!evt->event_handler) {
1014 		pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1015 			smp_processor_id());
1016 		/* Switch it off */
1017 		lapic_timer_shutdown(evt);
1018 		return;
1019 	}
1020 
1021 	/*
1022 	 * the NMI deadlock-detector uses this.
1023 	 */
1024 	inc_irq_stat(apic_timer_irqs);
1025 
1026 	evt->event_handler(evt);
1027 }
1028 
1029 /*
1030  * Local APIC timer interrupt. This is the most natural way for doing
1031  * local interrupts, but local timer interrupts can be emulated by
1032  * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1033  *
1034  * [ if a single-CPU system runs an SMP kernel then we call the local
1035  *   interrupt as well. Thus we cannot inline the local irq ... ]
1036  */
DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)1037 DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1038 {
1039 	struct pt_regs *old_regs = set_irq_regs(regs);
1040 
1041 	apic_eoi();
1042 	trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1043 	local_apic_timer_interrupt();
1044 	trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1045 
1046 	set_irq_regs(old_regs);
1047 }
1048 
1049 /*
1050  * Local APIC start and shutdown
1051  */
1052 
1053 /**
1054  * clear_local_APIC - shutdown the local APIC
1055  *
1056  * This is called, when a CPU is disabled and before rebooting, so the state of
1057  * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1058  * leftovers during boot.
1059  */
clear_local_APIC(void)1060 void clear_local_APIC(void)
1061 {
1062 	int maxlvt;
1063 	u32 v;
1064 
1065 	if (!apic_accessible())
1066 		return;
1067 
1068 	maxlvt = lapic_get_maxlvt();
1069 	/*
1070 	 * Masking an LVT entry can trigger a local APIC error
1071 	 * if the vector is zero. Mask LVTERR first to prevent this.
1072 	 */
1073 	if (maxlvt >= 3) {
1074 		v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1075 		apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1076 	}
1077 	/*
1078 	 * Careful: we have to set masks only first to deassert
1079 	 * any level-triggered sources.
1080 	 */
1081 	v = apic_read(APIC_LVTT);
1082 	apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1083 	v = apic_read(APIC_LVT0);
1084 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1085 	v = apic_read(APIC_LVT1);
1086 	apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1087 	if (maxlvt >= 4) {
1088 		v = apic_read(APIC_LVTPC);
1089 		apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1090 	}
1091 
1092 	/* lets not touch this if we didn't frob it */
1093 #ifdef CONFIG_X86_THERMAL_VECTOR
1094 	if (maxlvt >= 5) {
1095 		v = apic_read(APIC_LVTTHMR);
1096 		apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1097 	}
1098 #endif
1099 #ifdef CONFIG_X86_MCE_INTEL
1100 	if (maxlvt >= 6) {
1101 		v = apic_read(APIC_LVTCMCI);
1102 		if (!(v & APIC_LVT_MASKED))
1103 			apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1104 	}
1105 #endif
1106 
1107 	/*
1108 	 * Clean APIC state for other OSs:
1109 	 */
1110 	apic_write(APIC_LVTT, APIC_LVT_MASKED);
1111 	apic_write(APIC_LVT0, APIC_LVT_MASKED);
1112 	apic_write(APIC_LVT1, APIC_LVT_MASKED);
1113 	if (maxlvt >= 3)
1114 		apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1115 	if (maxlvt >= 4)
1116 		apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1117 
1118 	/* Integrated APIC (!82489DX) ? */
1119 	if (lapic_is_integrated()) {
1120 		if (maxlvt > 3)
1121 			/* Clear ESR due to Pentium errata 3AP and 11AP */
1122 			apic_write(APIC_ESR, 0);
1123 		apic_read(APIC_ESR);
1124 	}
1125 }
1126 
1127 /**
1128  * apic_soft_disable - Clears and software disables the local APIC on hotplug
1129  *
1130  * Contrary to disable_local_APIC() this does not touch the enable bit in
1131  * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1132  * bus would require a hardware reset as the APIC would lose track of bus
1133  * arbitration. On systems with FSB delivery APICBASE could be disabled,
1134  * but it has to be guaranteed that no interrupt is sent to the APIC while
1135  * in that state and it's not clear from the SDM whether it still responds
1136  * to INIT/SIPI messages. Stay on the safe side and use software disable.
1137  */
apic_soft_disable(void)1138 void apic_soft_disable(void)
1139 {
1140 	u32 value;
1141 
1142 	clear_local_APIC();
1143 
1144 	/* Soft disable APIC (implies clearing of registers for 82489DX!). */
1145 	value = apic_read(APIC_SPIV);
1146 	value &= ~APIC_SPIV_APIC_ENABLED;
1147 	apic_write(APIC_SPIV, value);
1148 }
1149 
1150 /**
1151  * disable_local_APIC - clear and disable the local APIC
1152  */
disable_local_APIC(void)1153 void disable_local_APIC(void)
1154 {
1155 	if (!apic_accessible())
1156 		return;
1157 
1158 	apic_soft_disable();
1159 
1160 #ifdef CONFIG_X86_32
1161 	/*
1162 	 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1163 	 * restore the disabled state.
1164 	 */
1165 	if (enabled_via_apicbase) {
1166 		unsigned int l, h;
1167 
1168 		rdmsr(MSR_IA32_APICBASE, l, h);
1169 		l &= ~MSR_IA32_APICBASE_ENABLE;
1170 		wrmsr(MSR_IA32_APICBASE, l, h);
1171 	}
1172 #endif
1173 }
1174 
1175 /*
1176  * If Linux enabled the LAPIC against the BIOS default disable it down before
1177  * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
1178  * not power-off.  Additionally clear all LVT entries before disable_local_APIC
1179  * for the case where Linux didn't enable the LAPIC.
1180  */
lapic_shutdown(void)1181 void lapic_shutdown(void)
1182 {
1183 	unsigned long flags;
1184 
1185 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1186 		return;
1187 
1188 	local_irq_save(flags);
1189 
1190 #ifdef CONFIG_X86_32
1191 	if (!enabled_via_apicbase)
1192 		clear_local_APIC();
1193 	else
1194 #endif
1195 		disable_local_APIC();
1196 
1197 
1198 	local_irq_restore(flags);
1199 }
1200 
1201 /**
1202  * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1203  */
sync_Arb_IDs(void)1204 void __init sync_Arb_IDs(void)
1205 {
1206 	/*
1207 	 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1208 	 * needed on AMD.
1209 	 */
1210 	if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1211 		return;
1212 
1213 	/*
1214 	 * Wait for idle.
1215 	 */
1216 	apic_wait_icr_idle();
1217 
1218 	apic_pr_debug("Synchronizing Arb IDs.\n");
1219 	apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG | APIC_DM_INIT);
1220 }
1221 
1222 enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1223 
__apic_intr_mode_select(void)1224 static int __init __apic_intr_mode_select(void)
1225 {
1226 	/* Check kernel option */
1227 	if (apic_is_disabled) {
1228 		pr_info("APIC disabled via kernel command line\n");
1229 		return APIC_PIC;
1230 	}
1231 
1232 	/* Check BIOS */
1233 #ifdef CONFIG_X86_64
1234 	/* On 64-bit, the APIC must be integrated, Check local APIC only */
1235 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1236 		apic_is_disabled = true;
1237 		pr_info("APIC disabled by BIOS\n");
1238 		return APIC_PIC;
1239 	}
1240 #else
1241 	/* On 32-bit, the APIC may be integrated APIC or 82489DX */
1242 
1243 	/* Neither 82489DX nor integrated APIC ? */
1244 	if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1245 		apic_is_disabled = true;
1246 		return APIC_PIC;
1247 	}
1248 
1249 	/* If the BIOS pretends there is an integrated APIC ? */
1250 	if (!boot_cpu_has(X86_FEATURE_APIC) &&
1251 		APIC_INTEGRATED(boot_cpu_apic_version)) {
1252 		apic_is_disabled = true;
1253 		pr_err(FW_BUG "Local APIC not detected, force emulation\n");
1254 		return APIC_PIC;
1255 	}
1256 #endif
1257 
1258 	/* Check MP table or ACPI MADT configuration */
1259 	if (!smp_found_config) {
1260 		disable_ioapic_support();
1261 		if (!acpi_lapic) {
1262 			pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1263 			return APIC_VIRTUAL_WIRE_NO_CONFIG;
1264 		}
1265 		return APIC_VIRTUAL_WIRE;
1266 	}
1267 
1268 #ifdef CONFIG_SMP
1269 	/* If SMP should be disabled, then really disable it! */
1270 	if (!setup_max_cpus) {
1271 		pr_info("APIC: SMP mode deactivated\n");
1272 		return APIC_SYMMETRIC_IO_NO_ROUTING;
1273 	}
1274 #endif
1275 
1276 	return APIC_SYMMETRIC_IO;
1277 }
1278 
1279 /* Select the interrupt delivery mode for the BSP */
apic_intr_mode_select(void)1280 void __init apic_intr_mode_select(void)
1281 {
1282 	apic_intr_mode = __apic_intr_mode_select();
1283 }
1284 
1285 /*
1286  * An initial setup of the virtual wire mode.
1287  */
init_bsp_APIC(void)1288 void __init init_bsp_APIC(void)
1289 {
1290 	unsigned int value;
1291 
1292 	/*
1293 	 * Don't do the setup now if we have a SMP BIOS as the
1294 	 * through-I/O-APIC virtual wire mode might be active.
1295 	 */
1296 	if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1297 		return;
1298 
1299 	/*
1300 	 * Do not trust the local APIC being empty at bootup.
1301 	 */
1302 	clear_local_APIC();
1303 
1304 	/*
1305 	 * Enable APIC.
1306 	 */
1307 	value = apic_read(APIC_SPIV);
1308 	value &= ~APIC_VECTOR_MASK;
1309 	value |= APIC_SPIV_APIC_ENABLED;
1310 
1311 #ifdef CONFIG_X86_32
1312 	/* This bit is reserved on P4/Xeon and should be cleared */
1313 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1314 	    (boot_cpu_data.x86 == 15))
1315 		value &= ~APIC_SPIV_FOCUS_DISABLED;
1316 	else
1317 #endif
1318 		value |= APIC_SPIV_FOCUS_DISABLED;
1319 	value |= SPURIOUS_APIC_VECTOR;
1320 	apic_write(APIC_SPIV, value);
1321 
1322 	/*
1323 	 * Set up the virtual wire mode.
1324 	 */
1325 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
1326 	value = APIC_DM_NMI;
1327 	if (!lapic_is_integrated())		/* 82489DX */
1328 		value |= APIC_LVT_LEVEL_TRIGGER;
1329 	if (apic_extnmi == APIC_EXTNMI_NONE)
1330 		value |= APIC_LVT_MASKED;
1331 	apic_write(APIC_LVT1, value);
1332 }
1333 
1334 static void __init apic_bsp_setup(bool upmode);
1335 
1336 /* Init the interrupt delivery mode for the BSP */
apic_intr_mode_init(void)1337 void __init apic_intr_mode_init(void)
1338 {
1339 	bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1340 
1341 	switch (apic_intr_mode) {
1342 	case APIC_PIC:
1343 		pr_info("APIC: Keep in PIC mode(8259)\n");
1344 		return;
1345 	case APIC_VIRTUAL_WIRE:
1346 		pr_info("APIC: Switch to virtual wire mode setup\n");
1347 		break;
1348 	case APIC_VIRTUAL_WIRE_NO_CONFIG:
1349 		pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1350 		upmode = true;
1351 		break;
1352 	case APIC_SYMMETRIC_IO:
1353 		pr_info("APIC: Switch to symmetric I/O mode setup\n");
1354 		break;
1355 	case APIC_SYMMETRIC_IO_NO_ROUTING:
1356 		pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1357 		break;
1358 	}
1359 
1360 	x86_64_probe_apic();
1361 
1362 	x86_32_install_bigsmp();
1363 
1364 	if (x86_platform.apic_post_init)
1365 		x86_platform.apic_post_init();
1366 
1367 	apic_bsp_setup(upmode);
1368 }
1369 
lapic_setup_esr(void)1370 static void lapic_setup_esr(void)
1371 {
1372 	unsigned int oldvalue, value, maxlvt;
1373 
1374 	if (!lapic_is_integrated()) {
1375 		pr_info("No ESR for 82489DX.\n");
1376 		return;
1377 	}
1378 
1379 	if (apic->disable_esr) {
1380 		/*
1381 		 * Something untraceable is creating bad interrupts on
1382 		 * secondary quads ... for the moment, just leave the
1383 		 * ESR disabled - we can't do anything useful with the
1384 		 * errors anyway - mbligh
1385 		 */
1386 		pr_info("Leaving ESR disabled.\n");
1387 		return;
1388 	}
1389 
1390 	maxlvt = lapic_get_maxlvt();
1391 	if (maxlvt > 3)		/* Due to the Pentium erratum 3AP. */
1392 		apic_write(APIC_ESR, 0);
1393 	oldvalue = apic_read(APIC_ESR);
1394 
1395 	/* enables sending errors */
1396 	value = ERROR_APIC_VECTOR;
1397 	apic_write(APIC_LVTERR, value);
1398 
1399 	/*
1400 	 * spec says clear errors after enabling vector.
1401 	 */
1402 	if (maxlvt > 3)
1403 		apic_write(APIC_ESR, 0);
1404 	value = apic_read(APIC_ESR);
1405 	if (value != oldvalue) {
1406 		apic_pr_verbose("ESR value before enabling vector: 0x%08x  after: 0x%08x\n",
1407 				oldvalue, value);
1408 	}
1409 }
1410 
1411 #define APIC_IR_REGS		APIC_ISR_NR
1412 #define APIC_IR_BITS		(APIC_IR_REGS * 32)
1413 #define APIC_IR_MAPSIZE		(APIC_IR_BITS / BITS_PER_LONG)
1414 
1415 union apic_ir {
1416 	unsigned long	map[APIC_IR_MAPSIZE];
1417 	u32		regs[APIC_IR_REGS];
1418 };
1419 
apic_check_and_ack(union apic_ir * irr,union apic_ir * isr)1420 static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1421 {
1422 	int i, bit;
1423 
1424 	/* Read the IRRs */
1425 	for (i = 0; i < APIC_IR_REGS; i++)
1426 		irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1427 
1428 	/* Read the ISRs */
1429 	for (i = 0; i < APIC_IR_REGS; i++)
1430 		isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1431 
1432 	/*
1433 	 * If the ISR map is not empty. ACK the APIC and run another round
1434 	 * to verify whether a pending IRR has been unblocked and turned
1435 	 * into a ISR.
1436 	 */
1437 	if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1438 		/*
1439 		 * There can be multiple ISR bits set when a high priority
1440 		 * interrupt preempted a lower priority one. Issue an ACK
1441 		 * per set bit.
1442 		 */
1443 		for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1444 			apic_eoi();
1445 		return true;
1446 	}
1447 
1448 	return !bitmap_empty(irr->map, APIC_IR_BITS);
1449 }
1450 
1451 /*
1452  * After a crash, we no longer service the interrupts and a pending
1453  * interrupt from previous kernel might still have ISR bit set.
1454  *
1455  * Most probably by now the CPU has serviced that pending interrupt and it
1456  * might not have done the apic_eoi() because it thought, interrupt
1457  * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1458  * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1459  * a vector might get locked. It was noticed for timer irq (vector
1460  * 0x31). Issue an extra EOI to clear ISR.
1461  *
1462  * If there are pending IRR bits they turn into ISR bits after a higher
1463  * priority ISR bit has been acked.
1464  */
apic_pending_intr_clear(void)1465 static void apic_pending_intr_clear(void)
1466 {
1467 	union apic_ir irr, isr;
1468 	unsigned int i;
1469 
1470 	/* 512 loops are way oversized and give the APIC a chance to obey. */
1471 	for (i = 0; i < 512; i++) {
1472 		if (!apic_check_and_ack(&irr, &isr))
1473 			return;
1474 	}
1475 	/* Dump the IRR/ISR content if that failed */
1476 	pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1477 }
1478 
1479 /**
1480  * setup_local_APIC - setup the local APIC
1481  *
1482  * Used to setup local APIC while initializing BSP or bringing up APs.
1483  * Always called with preemption disabled.
1484  */
setup_local_APIC(void)1485 static void setup_local_APIC(void)
1486 {
1487 	int cpu = smp_processor_id();
1488 	unsigned int value;
1489 
1490 	if (apic_is_disabled) {
1491 		disable_ioapic_support();
1492 		return;
1493 	}
1494 
1495 	/*
1496 	 * If this comes from kexec/kcrash the APIC might be enabled in
1497 	 * SPIV. Soft disable it before doing further initialization.
1498 	 */
1499 	value = apic_read(APIC_SPIV);
1500 	value &= ~APIC_SPIV_APIC_ENABLED;
1501 	apic_write(APIC_SPIV, value);
1502 
1503 #ifdef CONFIG_X86_32
1504 	/* Pound the ESR really hard over the head with a big hammer - mbligh */
1505 	if (lapic_is_integrated() && apic->disable_esr) {
1506 		apic_write(APIC_ESR, 0);
1507 		apic_write(APIC_ESR, 0);
1508 		apic_write(APIC_ESR, 0);
1509 		apic_write(APIC_ESR, 0);
1510 	}
1511 #endif
1512 	/*
1513 	 * Intel recommends to set DFR, LDR and TPR before enabling
1514 	 * an APIC.  See e.g. "AP-388 82489DX User's Manual" (Intel
1515 	 * document number 292116).
1516 	 *
1517 	 * Except for APICs which operate in physical destination mode.
1518 	 */
1519 	if (apic->init_apic_ldr)
1520 		apic->init_apic_ldr();
1521 
1522 	/*
1523 	 * Set Task Priority to 'accept all except vectors 0-31'.  An APIC
1524 	 * vector in the 16-31 range could be delivered if TPR == 0, but we
1525 	 * would think it's an exception and terrible things will happen.  We
1526 	 * never change this later on.
1527 	 */
1528 	value = apic_read(APIC_TASKPRI);
1529 	value &= ~APIC_TPRI_MASK;
1530 	value |= 0x10;
1531 	apic_write(APIC_TASKPRI, value);
1532 
1533 	/* Clear eventually stale ISR/IRR bits */
1534 	apic_pending_intr_clear();
1535 
1536 	/*
1537 	 * Now that we are all set up, enable the APIC
1538 	 */
1539 	value = apic_read(APIC_SPIV);
1540 	value &= ~APIC_VECTOR_MASK;
1541 	/*
1542 	 * Enable APIC
1543 	 */
1544 	value |= APIC_SPIV_APIC_ENABLED;
1545 
1546 #ifdef CONFIG_X86_32
1547 	/*
1548 	 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1549 	 * certain networking cards. If high frequency interrupts are
1550 	 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1551 	 * entry is masked/unmasked at a high rate as well then sooner or
1552 	 * later IOAPIC line gets 'stuck', no more interrupts are received
1553 	 * from the device. If focus CPU is disabled then the hang goes
1554 	 * away, oh well :-(
1555 	 *
1556 	 * [ This bug can be reproduced easily with a level-triggered
1557 	 *   PCI Ne2000 networking cards and PII/PIII processors, dual
1558 	 *   BX chipset. ]
1559 	 */
1560 	/*
1561 	 * Actually disabling the focus CPU check just makes the hang less
1562 	 * frequent as it makes the interrupt distribution model be more
1563 	 * like LRU than MRU (the short-term load is more even across CPUs).
1564 	 */
1565 
1566 	/*
1567 	 * - enable focus processor (bit==0)
1568 	 * - 64bit mode always use processor focus
1569 	 *   so no need to set it
1570 	 */
1571 	value &= ~APIC_SPIV_FOCUS_DISABLED;
1572 #endif
1573 
1574 	/*
1575 	 * Set spurious IRQ vector
1576 	 */
1577 	value |= SPURIOUS_APIC_VECTOR;
1578 	apic_write(APIC_SPIV, value);
1579 
1580 	perf_events_lapic_init();
1581 
1582 	/*
1583 	 * Set up LVT0, LVT1:
1584 	 *
1585 	 * set up through-local-APIC on the boot CPU's LINT0. This is not
1586 	 * strictly necessary in pure symmetric-IO mode, but sometimes
1587 	 * we delegate interrupts to the 8259A.
1588 	 */
1589 	/*
1590 	 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1591 	 */
1592 	value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1593 	if (!cpu && (pic_mode || !value || ioapic_is_disabled)) {
1594 		value = APIC_DM_EXTINT;
1595 		apic_pr_verbose("Enabled ExtINT on CPU#%d\n", cpu);
1596 	} else {
1597 		value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1598 		apic_pr_verbose("Masked ExtINT on CPU#%d\n", cpu);
1599 	}
1600 	apic_write(APIC_LVT0, value);
1601 
1602 	/*
1603 	 * Only the BSP sees the LINT1 NMI signal by default. This can be
1604 	 * modified by apic_extnmi= boot option.
1605 	 */
1606 	if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1607 	    apic_extnmi == APIC_EXTNMI_ALL)
1608 		value = APIC_DM_NMI;
1609 	else
1610 		value = APIC_DM_NMI | APIC_LVT_MASKED;
1611 
1612 	/* Is 82489DX ? */
1613 	if (!lapic_is_integrated())
1614 		value |= APIC_LVT_LEVEL_TRIGGER;
1615 	apic_write(APIC_LVT1, value);
1616 
1617 #ifdef CONFIG_X86_MCE_INTEL
1618 	/* Recheck CMCI information after local APIC is up on CPU #0 */
1619 	if (!cpu)
1620 		cmci_recheck();
1621 #endif
1622 }
1623 
end_local_APIC_setup(void)1624 static void end_local_APIC_setup(void)
1625 {
1626 	lapic_setup_esr();
1627 
1628 #ifdef CONFIG_X86_32
1629 	{
1630 		unsigned int value;
1631 		/* Disable the local apic timer */
1632 		value = apic_read(APIC_LVTT);
1633 		value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1634 		apic_write(APIC_LVTT, value);
1635 	}
1636 #endif
1637 
1638 	apic_pm_activate();
1639 }
1640 
1641 /*
1642  * APIC setup function for application processors. Called from smpboot.c
1643  */
apic_ap_setup(void)1644 void apic_ap_setup(void)
1645 {
1646 	setup_local_APIC();
1647 	end_local_APIC_setup();
1648 }
1649 
apic_read_boot_cpu_id(bool x2apic)1650 static __init void apic_read_boot_cpu_id(bool x2apic)
1651 {
1652 	/*
1653 	 * This can be invoked from check_x2apic() before the APIC has been
1654 	 * selected. But that code knows for sure that the BIOS enabled
1655 	 * X2APIC.
1656 	 */
1657 	if (x2apic) {
1658 		boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID);
1659 		boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR));
1660 	} else {
1661 		boot_cpu_physical_apicid = read_apic_id();
1662 		boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1663 	}
1664 	topology_register_boot_apic(boot_cpu_physical_apicid);
1665 	x86_32_probe_bigsmp_early();
1666 }
1667 
1668 #ifdef CONFIG_X86_X2APIC
1669 int x2apic_mode;
1670 EXPORT_SYMBOL_GPL(x2apic_mode);
1671 
1672 enum {
1673 	X2APIC_OFF,
1674 	X2APIC_DISABLED,
1675 	/* All states below here have X2APIC enabled */
1676 	X2APIC_ON,
1677 	X2APIC_ON_LOCKED
1678 };
1679 static int x2apic_state;
1680 
x2apic_hw_locked(void)1681 static bool x2apic_hw_locked(void)
1682 {
1683 	u64 x86_arch_cap_msr;
1684 	u64 msr;
1685 
1686 	x86_arch_cap_msr = x86_read_arch_cap_msr();
1687 	if (x86_arch_cap_msr & ARCH_CAP_XAPIC_DISABLE) {
1688 		rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1689 		return (msr & LEGACY_XAPIC_DISABLED);
1690 	}
1691 	return false;
1692 }
1693 
__x2apic_disable(void)1694 static void __x2apic_disable(void)
1695 {
1696 	u64 msr;
1697 
1698 	if (!boot_cpu_has(X86_FEATURE_APIC))
1699 		return;
1700 
1701 	rdmsrl(MSR_IA32_APICBASE, msr);
1702 	if (!(msr & X2APIC_ENABLE))
1703 		return;
1704 	/* Disable xapic and x2apic first and then reenable xapic mode */
1705 	wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1706 	wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1707 	printk_once(KERN_INFO "x2apic disabled\n");
1708 }
1709 
__x2apic_enable(void)1710 static void __x2apic_enable(void)
1711 {
1712 	u64 msr;
1713 
1714 	rdmsrl(MSR_IA32_APICBASE, msr);
1715 	if (msr & X2APIC_ENABLE)
1716 		return;
1717 	wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1718 	printk_once(KERN_INFO "x2apic enabled\n");
1719 }
1720 
setup_nox2apic(char * str)1721 static int __init setup_nox2apic(char *str)
1722 {
1723 	if (x2apic_enabled()) {
1724 		u32 apicid = native_apic_msr_read(APIC_ID);
1725 
1726 		if (apicid >= 255) {
1727 			pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1728 				apicid);
1729 			return 0;
1730 		}
1731 		if (x2apic_hw_locked()) {
1732 			pr_warn("APIC locked in x2apic mode, can't disable\n");
1733 			return 0;
1734 		}
1735 		pr_warn("x2apic already enabled.\n");
1736 		__x2apic_disable();
1737 	}
1738 	setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1739 	x2apic_state = X2APIC_DISABLED;
1740 	x2apic_mode = 0;
1741 	return 0;
1742 }
1743 early_param("nox2apic", setup_nox2apic);
1744 
1745 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
x2apic_setup(void)1746 void x2apic_setup(void)
1747 {
1748 	/*
1749 	 * Try to make the AP's APIC state match that of the BSP,  but if the
1750 	 * BSP is unlocked and the AP is locked then there is a state mismatch.
1751 	 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1752 	 * trying to be turned off.
1753 	 */
1754 	if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1755 		pr_warn("x2apic lock mismatch between BSP and AP.\n");
1756 	/*
1757 	 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1758 	 * from BIOS.
1759 	 */
1760 	if (x2apic_state < X2APIC_ON) {
1761 		__x2apic_disable();
1762 		return;
1763 	}
1764 	__x2apic_enable();
1765 }
1766 
1767 static __init void apic_set_fixmap(bool read_apic);
1768 
x2apic_disable(void)1769 static __init void x2apic_disable(void)
1770 {
1771 	u32 x2apic_id;
1772 
1773 	if (x2apic_state < X2APIC_ON)
1774 		return;
1775 
1776 	x2apic_id = read_apic_id();
1777 	if (x2apic_id >= 255)
1778 		panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1779 
1780 	if (x2apic_hw_locked()) {
1781 		pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1782 		return;
1783 	}
1784 
1785 	__x2apic_disable();
1786 
1787 	x2apic_mode = 0;
1788 	x2apic_state = X2APIC_DISABLED;
1789 
1790 	/*
1791 	 * Don't reread the APIC ID as it was already done from
1792 	 * check_x2apic() and the APIC driver still is a x2APIC variant,
1793 	 * which fails to do the read after x2APIC was disabled.
1794 	 */
1795 	apic_set_fixmap(false);
1796 }
1797 
x2apic_enable(void)1798 static __init void x2apic_enable(void)
1799 {
1800 	if (x2apic_state != X2APIC_OFF)
1801 		return;
1802 
1803 	x2apic_mode = 1;
1804 	x2apic_state = X2APIC_ON;
1805 	__x2apic_enable();
1806 }
1807 
try_to_enable_x2apic(int remap_mode)1808 static __init void try_to_enable_x2apic(int remap_mode)
1809 {
1810 	if (x2apic_state == X2APIC_DISABLED)
1811 		return;
1812 
1813 	if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1814 		u32 apic_limit = 255;
1815 
1816 		/*
1817 		 * Using X2APIC without IR is not architecturally supported
1818 		 * on bare metal but may be supported in guests.
1819 		 */
1820 		if (!x86_init.hyper.x2apic_available()) {
1821 			pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1822 			x2apic_disable();
1823 			return;
1824 		}
1825 
1826 		/*
1827 		 * If the hypervisor supports extended destination ID in
1828 		 * MSI, that increases the maximum APIC ID that can be
1829 		 * used for non-remapped IRQ domains.
1830 		 */
1831 		if (x86_init.hyper.msi_ext_dest_id()) {
1832 			virt_ext_dest_id = 1;
1833 			apic_limit = 32767;
1834 		}
1835 
1836 		/*
1837 		 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1838 		 * in physical mode, and CPUs with an APIC ID that cannot
1839 		 * be addressed must not be brought online.
1840 		 */
1841 		x2apic_set_max_apicid(apic_limit);
1842 		x2apic_phys = 1;
1843 	}
1844 	x2apic_enable();
1845 }
1846 
check_x2apic(void)1847 void __init check_x2apic(void)
1848 {
1849 	if (x2apic_enabled()) {
1850 		pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1851 		x2apic_mode = 1;
1852 		if (x2apic_hw_locked())
1853 			x2apic_state = X2APIC_ON_LOCKED;
1854 		else
1855 			x2apic_state = X2APIC_ON;
1856 		apic_read_boot_cpu_id(true);
1857 	} else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1858 		x2apic_state = X2APIC_DISABLED;
1859 	}
1860 }
1861 #else /* CONFIG_X86_X2APIC */
check_x2apic(void)1862 void __init check_x2apic(void)
1863 {
1864 	if (!apic_is_x2apic_enabled())
1865 		return;
1866 	/*
1867 	 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1868 	 */
1869 	pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1870 	pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1871 
1872 	apic_is_disabled = true;
1873 	setup_clear_cpu_cap(X86_FEATURE_APIC);
1874 }
1875 
try_to_enable_x2apic(int remap_mode)1876 static inline void try_to_enable_x2apic(int remap_mode) { }
__x2apic_enable(void)1877 static inline void __x2apic_enable(void) { }
1878 #endif /* !CONFIG_X86_X2APIC */
1879 
enable_IR_x2apic(void)1880 void __init enable_IR_x2apic(void)
1881 {
1882 	unsigned long flags;
1883 	int ret, ir_stat;
1884 
1885 	if (ioapic_is_disabled) {
1886 		pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1887 		return;
1888 	}
1889 
1890 	ir_stat = irq_remapping_prepare();
1891 	if (ir_stat < 0 && !x2apic_supported())
1892 		return;
1893 
1894 	ret = save_ioapic_entries();
1895 	if (ret) {
1896 		pr_info("Saving IO-APIC state failed: %d\n", ret);
1897 		return;
1898 	}
1899 
1900 	local_irq_save(flags);
1901 	legacy_pic->mask_all();
1902 	mask_ioapic_entries();
1903 
1904 	/* If irq_remapping_prepare() succeeded, try to enable it */
1905 	if (ir_stat >= 0)
1906 		ir_stat = irq_remapping_enable();
1907 	/* ir_stat contains the remap mode or an error code */
1908 	try_to_enable_x2apic(ir_stat);
1909 
1910 	if (ir_stat < 0)
1911 		restore_ioapic_entries();
1912 	legacy_pic->restore_mask();
1913 	local_irq_restore(flags);
1914 }
1915 
1916 #ifdef CONFIG_X86_64
1917 /*
1918  * Detect and enable local APICs on non-SMP boards.
1919  * Original code written by Keir Fraser.
1920  * On AMD64 we trust the BIOS - if it says no APIC it is likely
1921  * not correctly set up (usually the APIC timer won't work etc.)
1922  */
detect_init_APIC(void)1923 static bool __init detect_init_APIC(void)
1924 {
1925 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
1926 		pr_info("No local APIC present\n");
1927 		return false;
1928 	}
1929 
1930 	register_lapic_address(APIC_DEFAULT_PHYS_BASE);
1931 	return true;
1932 }
1933 #else
1934 
apic_verify(unsigned long addr)1935 static bool __init apic_verify(unsigned long addr)
1936 {
1937 	u32 features, h, l;
1938 
1939 	/*
1940 	 * The APIC feature bit should now be enabled
1941 	 * in `cpuid'
1942 	 */
1943 	features = cpuid_edx(1);
1944 	if (!(features & (1 << X86_FEATURE_APIC))) {
1945 		pr_warn("Could not enable APIC!\n");
1946 		return false;
1947 	}
1948 	set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1949 
1950 	/* The BIOS may have set up the APIC at some other address */
1951 	if (boot_cpu_data.x86 >= 6) {
1952 		rdmsr(MSR_IA32_APICBASE, l, h);
1953 		if (l & MSR_IA32_APICBASE_ENABLE)
1954 			addr = l & MSR_IA32_APICBASE_BASE;
1955 	}
1956 
1957 	register_lapic_address(addr);
1958 	pr_info("Found and enabled local APIC!\n");
1959 	return true;
1960 }
1961 
apic_force_enable(unsigned long addr)1962 bool __init apic_force_enable(unsigned long addr)
1963 {
1964 	u32 h, l;
1965 
1966 	if (apic_is_disabled)
1967 		return false;
1968 
1969 	/*
1970 	 * Some BIOSes disable the local APIC in the APIC_BASE
1971 	 * MSR. This can only be done in software for Intel P6 or later
1972 	 * and AMD K7 (Model > 1) or later.
1973 	 */
1974 	if (boot_cpu_data.x86 >= 6) {
1975 		rdmsr(MSR_IA32_APICBASE, l, h);
1976 		if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1977 			pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1978 			l &= ~MSR_IA32_APICBASE_BASE;
1979 			l |= MSR_IA32_APICBASE_ENABLE | addr;
1980 			wrmsr(MSR_IA32_APICBASE, l, h);
1981 			enabled_via_apicbase = 1;
1982 		}
1983 	}
1984 	return apic_verify(addr);
1985 }
1986 
1987 /*
1988  * Detect and initialize APIC
1989  */
detect_init_APIC(void)1990 static bool __init detect_init_APIC(void)
1991 {
1992 	/* Disabled by kernel option? */
1993 	if (apic_is_disabled)
1994 		return false;
1995 
1996 	switch (boot_cpu_data.x86_vendor) {
1997 	case X86_VENDOR_AMD:
1998 		if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1999 		    (boot_cpu_data.x86 >= 15))
2000 			break;
2001 		goto no_apic;
2002 	case X86_VENDOR_HYGON:
2003 		break;
2004 	case X86_VENDOR_INTEL:
2005 		if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2006 		    (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2007 			break;
2008 		goto no_apic;
2009 	default:
2010 		goto no_apic;
2011 	}
2012 
2013 	if (!boot_cpu_has(X86_FEATURE_APIC)) {
2014 		/*
2015 		 * Over-ride BIOS and try to enable the local APIC only if
2016 		 * "lapic" specified.
2017 		 */
2018 		if (!force_enable_local_apic) {
2019 			pr_info("Local APIC disabled by BIOS -- "
2020 				"you can enable it with \"lapic\"\n");
2021 			return false;
2022 		}
2023 		if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2024 			return false;
2025 	} else {
2026 		if (!apic_verify(APIC_DEFAULT_PHYS_BASE))
2027 			return false;
2028 	}
2029 
2030 	apic_pm_activate();
2031 
2032 	return true;
2033 
2034 no_apic:
2035 	pr_info("No local APIC present or hardware disabled\n");
2036 	return false;
2037 }
2038 #endif
2039 
2040 /**
2041  * init_apic_mappings - initialize APIC mappings
2042  */
init_apic_mappings(void)2043 void __init init_apic_mappings(void)
2044 {
2045 	if (apic_validate_deadline_timer())
2046 		pr_info("TSC deadline timer available\n");
2047 
2048 	if (x2apic_mode)
2049 		return;
2050 
2051 	if (!smp_found_config) {
2052 		if (!detect_init_APIC()) {
2053 			pr_info("APIC: disable apic facility\n");
2054 			apic_disable();
2055 		}
2056 	}
2057 }
2058 
apic_set_fixmap(bool read_apic)2059 static __init void apic_set_fixmap(bool read_apic)
2060 {
2061 	set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr);
2062 	apic_mmio_base = APIC_BASE;
2063 	apic_pr_verbose("Mapped APIC to %16lx (%16lx)\n", apic_mmio_base, mp_lapic_addr);
2064 	if (read_apic)
2065 		apic_read_boot_cpu_id(false);
2066 }
2067 
register_lapic_address(unsigned long address)2068 void __init register_lapic_address(unsigned long address)
2069 {
2070 	/* This should only happen once */
2071 	WARN_ON_ONCE(mp_lapic_addr);
2072 	mp_lapic_addr = address;
2073 
2074 	if (!x2apic_mode)
2075 		apic_set_fixmap(true);
2076 }
2077 
2078 /*
2079  * Local APIC interrupts
2080  */
2081 
2082 /*
2083  * Common handling code for spurious_interrupt and spurious_vector entry
2084  * points below. No point in allowing the compiler to inline it twice.
2085  */
handle_spurious_interrupt(u8 vector)2086 static noinline void handle_spurious_interrupt(u8 vector)
2087 {
2088 	u32 v;
2089 
2090 	trace_spurious_apic_entry(vector);
2091 
2092 	inc_irq_stat(irq_spurious_count);
2093 
2094 	/*
2095 	 * If this is a spurious interrupt then do not acknowledge
2096 	 */
2097 	if (vector == SPURIOUS_APIC_VECTOR) {
2098 		/* See SDM vol 3 */
2099 		pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2100 			smp_processor_id());
2101 		goto out;
2102 	}
2103 
2104 	/*
2105 	 * If it is a vectored one, verify it's set in the ISR. If set,
2106 	 * acknowledge it.
2107 	 */
2108 	v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2109 	if (v & (1 << (vector & 0x1f))) {
2110 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2111 			vector, smp_processor_id());
2112 		apic_eoi();
2113 	} else {
2114 		pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2115 			vector, smp_processor_id());
2116 	}
2117 out:
2118 	trace_spurious_apic_exit(vector);
2119 }
2120 
2121 /**
2122  * spurious_interrupt - Catch all for interrupts raised on unused vectors
2123  * @regs:	Pointer to pt_regs on stack
2124  * @vector:	The vector number
2125  *
2126  * This is invoked from ASM entry code to catch all interrupts which
2127  * trigger on an entry which is routed to the common_spurious idtentry
2128  * point.
2129  */
DEFINE_IDTENTRY_IRQ(spurious_interrupt)2130 DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2131 {
2132 	handle_spurious_interrupt(vector);
2133 }
2134 
DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)2135 DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2136 {
2137 	handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2138 }
2139 
2140 /*
2141  * This interrupt should never happen with our APIC/SMP architecture
2142  */
DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)2143 DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2144 {
2145 	static const char * const error_interrupt_reason[] = {
2146 		"Send CS error",		/* APIC Error Bit 0 */
2147 		"Receive CS error",		/* APIC Error Bit 1 */
2148 		"Send accept error",		/* APIC Error Bit 2 */
2149 		"Receive accept error",		/* APIC Error Bit 3 */
2150 		"Redirectable IPI",		/* APIC Error Bit 4 */
2151 		"Send illegal vector",		/* APIC Error Bit 5 */
2152 		"Received illegal vector",	/* APIC Error Bit 6 */
2153 		"Illegal register address",	/* APIC Error Bit 7 */
2154 	};
2155 	u32 v, i = 0;
2156 
2157 	trace_error_apic_entry(ERROR_APIC_VECTOR);
2158 
2159 	/* First tickle the hardware, only then report what went on. -- REW */
2160 	if (lapic_get_maxlvt() > 3)	/* Due to the Pentium erratum 3AP. */
2161 		apic_write(APIC_ESR, 0);
2162 	v = apic_read(APIC_ESR);
2163 	apic_eoi();
2164 	atomic_inc(&irq_err_count);
2165 
2166 	apic_pr_debug("APIC error on CPU%d: %02x", smp_processor_id(), v);
2167 
2168 	v &= 0xff;
2169 	while (v) {
2170 		if (v & 0x1)
2171 			apic_pr_debug_cont(" : %s", error_interrupt_reason[i]);
2172 		i++;
2173 		v >>= 1;
2174 	}
2175 
2176 	apic_pr_debug_cont("\n");
2177 
2178 	trace_error_apic_exit(ERROR_APIC_VECTOR);
2179 }
2180 
2181 /**
2182  * connect_bsp_APIC - attach the APIC to the interrupt system
2183  */
connect_bsp_APIC(void)2184 static void __init connect_bsp_APIC(void)
2185 {
2186 #ifdef CONFIG_X86_32
2187 	if (pic_mode) {
2188 		/*
2189 		 * Do not trust the local APIC being empty at bootup.
2190 		 */
2191 		clear_local_APIC();
2192 		/*
2193 		 * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
2194 		 * local APIC to INT and NMI lines.
2195 		 */
2196 		apic_pr_verbose("Leaving PIC mode, enabling APIC mode.\n");
2197 		imcr_pic_to_apic();
2198 	}
2199 #endif
2200 }
2201 
2202 /**
2203  * disconnect_bsp_APIC - detach the APIC from the interrupt system
2204  * @virt_wire_setup:	indicates, whether virtual wire mode is selected
2205  *
2206  * Virtual wire mode is necessary to deliver legacy interrupts even when the
2207  * APIC is disabled.
2208  */
disconnect_bsp_APIC(int virt_wire_setup)2209 void disconnect_bsp_APIC(int virt_wire_setup)
2210 {
2211 	unsigned int value;
2212 
2213 #ifdef CONFIG_X86_32
2214 	if (pic_mode) {
2215 		/*
2216 		 * Put the board back into PIC mode (has an effect only on
2217 		 * certain older boards).  Note that APIC interrupts, including
2218 		 * IPIs, won't work beyond this point!  The only exception are
2219 		 * INIT IPIs.
2220 		 */
2221 		apic_pr_verbose("Disabling APIC mode, entering PIC mode.\n");
2222 		imcr_apic_to_pic();
2223 		return;
2224 	}
2225 #endif
2226 
2227 	/* Go back to Virtual Wire compatibility mode */
2228 
2229 	/* For the spurious interrupt use vector F, and enable it */
2230 	value = apic_read(APIC_SPIV);
2231 	value &= ~APIC_VECTOR_MASK;
2232 	value |= APIC_SPIV_APIC_ENABLED;
2233 	value |= 0xf;
2234 	apic_write(APIC_SPIV, value);
2235 
2236 	if (!virt_wire_setup) {
2237 		/*
2238 		 * For LVT0 make it edge triggered, active high,
2239 		 * external and enabled
2240 		 */
2241 		value = apic_read(APIC_LVT0);
2242 		value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2243 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2244 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2245 		value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2246 		value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2247 		apic_write(APIC_LVT0, value);
2248 	} else {
2249 		/* Disable LVT0 */
2250 		apic_write(APIC_LVT0, APIC_LVT_MASKED);
2251 	}
2252 
2253 	/*
2254 	 * For LVT1 make it edge triggered, active high,
2255 	 * nmi and enabled
2256 	 */
2257 	value = apic_read(APIC_LVT1);
2258 	value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2259 			APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2260 			APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2261 	value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2262 	value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2263 	apic_write(APIC_LVT1, value);
2264 }
2265 
__irq_msi_compose_msg(struct irq_cfg * cfg,struct msi_msg * msg,bool dmar)2266 void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2267 			   bool dmar)
2268 {
2269 	memset(msg, 0, sizeof(*msg));
2270 
2271 	msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2272 	msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2273 	msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2274 
2275 	msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2276 	msg->arch_data.vector = cfg->vector;
2277 
2278 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2279 	/*
2280 	 * Only the IOMMU itself can use the trick of putting destination
2281 	 * APIC ID into the high bits of the address. Anything else would
2282 	 * just be writing to memory if it tried that, and needs IR to
2283 	 * address APICs which can't be addressed in the normal 32-bit
2284 	 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2285 	 * some hypervisors allow the extended destination ID field in bits
2286 	 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2287 	 */
2288 	if (dmar)
2289 		msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2290 	else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2291 		msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2292 	else
2293 		WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2294 }
2295 
x86_msi_msg_get_destid(struct msi_msg * msg,bool extid)2296 u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2297 {
2298 	u32 dest = msg->arch_addr_lo.destid_0_7;
2299 
2300 	if (extid)
2301 		dest |= msg->arch_addr_hi.destid_8_31 << 8;
2302 	return dest;
2303 }
2304 EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2305 
apic_bsp_up_setup(void)2306 static void __init apic_bsp_up_setup(void)
2307 {
2308 	reset_phys_cpu_present_map(boot_cpu_physical_apicid);
2309 }
2310 
2311 /**
2312  * apic_bsp_setup - Setup function for local apic and io-apic
2313  * @upmode:		Force UP mode (for APIC_init_uniprocessor)
2314  */
apic_bsp_setup(bool upmode)2315 static void __init apic_bsp_setup(bool upmode)
2316 {
2317 	connect_bsp_APIC();
2318 	if (upmode)
2319 		apic_bsp_up_setup();
2320 	setup_local_APIC();
2321 
2322 	enable_IO_APIC();
2323 	end_local_APIC_setup();
2324 	irq_remap_enable_fault_handling();
2325 	setup_IO_APIC();
2326 	lapic_update_legacy_vectors();
2327 }
2328 
2329 #ifdef CONFIG_UP_LATE_INIT
up_late_init(void)2330 void __init up_late_init(void)
2331 {
2332 	if (apic_intr_mode == APIC_PIC)
2333 		return;
2334 
2335 	/* Setup local timer */
2336 	x86_init.timers.setup_percpu_clockev();
2337 }
2338 #endif
2339 
2340 /*
2341  * Power management
2342  */
2343 #ifdef CONFIG_PM
2344 
2345 static struct {
2346 	/*
2347 	 * 'active' is true if the local APIC was enabled by us and
2348 	 * not the BIOS; this signifies that we are also responsible
2349 	 * for disabling it before entering apm/acpi suspend
2350 	 */
2351 	int active;
2352 	/* r/w apic fields */
2353 	u32 apic_id;
2354 	unsigned int apic_taskpri;
2355 	unsigned int apic_ldr;
2356 	unsigned int apic_dfr;
2357 	unsigned int apic_spiv;
2358 	unsigned int apic_lvtt;
2359 	unsigned int apic_lvtpc;
2360 	unsigned int apic_lvt0;
2361 	unsigned int apic_lvt1;
2362 	unsigned int apic_lvterr;
2363 	unsigned int apic_tmict;
2364 	unsigned int apic_tdcr;
2365 	unsigned int apic_thmr;
2366 	unsigned int apic_cmci;
2367 } apic_pm_state;
2368 
lapic_suspend(void)2369 static int lapic_suspend(void)
2370 {
2371 	unsigned long flags;
2372 	int maxlvt;
2373 
2374 	if (!apic_pm_state.active)
2375 		return 0;
2376 
2377 	maxlvt = lapic_get_maxlvt();
2378 
2379 	apic_pm_state.apic_id = apic_read(APIC_ID);
2380 	apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2381 	apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2382 	apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2383 	apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2384 	apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2385 	if (maxlvt >= 4)
2386 		apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2387 	apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2388 	apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2389 	apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2390 	apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2391 	apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2392 #ifdef CONFIG_X86_THERMAL_VECTOR
2393 	if (maxlvt >= 5)
2394 		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2395 #endif
2396 #ifdef CONFIG_X86_MCE_INTEL
2397 	if (maxlvt >= 6)
2398 		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2399 #endif
2400 
2401 	local_irq_save(flags);
2402 
2403 	/*
2404 	 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2405 	 * entries on some implementations.
2406 	 */
2407 	mask_ioapic_entries();
2408 
2409 	disable_local_APIC();
2410 
2411 	irq_remapping_disable();
2412 
2413 	local_irq_restore(flags);
2414 	return 0;
2415 }
2416 
lapic_resume(void)2417 static void lapic_resume(void)
2418 {
2419 	unsigned int l, h;
2420 	unsigned long flags;
2421 	int maxlvt;
2422 
2423 	if (!apic_pm_state.active)
2424 		return;
2425 
2426 	local_irq_save(flags);
2427 
2428 	/*
2429 	 * IO-APIC and PIC have their own resume routines.
2430 	 * We just mask them here to make sure the interrupt
2431 	 * subsystem is completely quiet while we enable x2apic
2432 	 * and interrupt-remapping.
2433 	 */
2434 	mask_ioapic_entries();
2435 	legacy_pic->mask_all();
2436 
2437 	if (x2apic_mode) {
2438 		__x2apic_enable();
2439 	} else {
2440 		/*
2441 		 * Make sure the APICBASE points to the right address
2442 		 *
2443 		 * FIXME! This will be wrong if we ever support suspend on
2444 		 * SMP! We'll need to do this as part of the CPU restore!
2445 		 */
2446 		if (boot_cpu_data.x86 >= 6) {
2447 			rdmsr(MSR_IA32_APICBASE, l, h);
2448 			l &= ~MSR_IA32_APICBASE_BASE;
2449 			l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2450 			wrmsr(MSR_IA32_APICBASE, l, h);
2451 		}
2452 	}
2453 
2454 	maxlvt = lapic_get_maxlvt();
2455 	apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2456 	apic_write(APIC_ID, apic_pm_state.apic_id);
2457 	apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2458 	apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2459 	apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2460 	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2461 	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2462 	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2463 #ifdef CONFIG_X86_THERMAL_VECTOR
2464 	if (maxlvt >= 5)
2465 		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2466 #endif
2467 #ifdef CONFIG_X86_MCE_INTEL
2468 	if (maxlvt >= 6)
2469 		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2470 #endif
2471 	if (maxlvt >= 4)
2472 		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2473 	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2474 	apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2475 	apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2476 	apic_write(APIC_ESR, 0);
2477 	apic_read(APIC_ESR);
2478 	apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2479 	apic_write(APIC_ESR, 0);
2480 	apic_read(APIC_ESR);
2481 
2482 	irq_remapping_reenable(x2apic_mode);
2483 
2484 	local_irq_restore(flags);
2485 }
2486 
2487 /*
2488  * This device has no shutdown method - fully functioning local APICs
2489  * are needed on every CPU up until machine_halt/restart/poweroff.
2490  */
2491 
2492 static struct syscore_ops lapic_syscore_ops = {
2493 	.resume		= lapic_resume,
2494 	.suspend	= lapic_suspend,
2495 };
2496 
apic_pm_activate(void)2497 static void apic_pm_activate(void)
2498 {
2499 	apic_pm_state.active = 1;
2500 }
2501 
init_lapic_sysfs(void)2502 static int __init init_lapic_sysfs(void)
2503 {
2504 	/* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2505 	if (boot_cpu_has(X86_FEATURE_APIC))
2506 		register_syscore_ops(&lapic_syscore_ops);
2507 
2508 	return 0;
2509 }
2510 
2511 /* local apic needs to resume before other devices access its registers. */
2512 core_initcall(init_lapic_sysfs);
2513 
2514 #else	/* CONFIG_PM */
2515 
apic_pm_activate(void)2516 static void apic_pm_activate(void) { }
2517 
2518 #endif	/* CONFIG_PM */
2519 
2520 #ifdef CONFIG_X86_64
2521 
2522 static int multi_checked;
2523 static int multi;
2524 
set_multi(const struct dmi_system_id * d)2525 static int set_multi(const struct dmi_system_id *d)
2526 {
2527 	if (multi)
2528 		return 0;
2529 	pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2530 	multi = 1;
2531 	return 0;
2532 }
2533 
2534 static const struct dmi_system_id multi_dmi_table[] = {
2535 	{
2536 		.callback = set_multi,
2537 		.ident = "IBM System Summit2",
2538 		.matches = {
2539 			DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2540 			DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2541 		},
2542 	},
2543 	{}
2544 };
2545 
dmi_check_multi(void)2546 static void dmi_check_multi(void)
2547 {
2548 	if (multi_checked)
2549 		return;
2550 
2551 	dmi_check_system(multi_dmi_table);
2552 	multi_checked = 1;
2553 }
2554 
2555 /*
2556  * apic_is_clustered_box() -- Check if we can expect good TSC
2557  *
2558  * Thus far, the major user of this is IBM's Summit2 series:
2559  * Clustered boxes may have unsynced TSC problems if they are
2560  * multi-chassis.
2561  * Use DMI to check them
2562  */
apic_is_clustered_box(void)2563 int apic_is_clustered_box(void)
2564 {
2565 	dmi_check_multi();
2566 	return multi;
2567 }
2568 #endif
2569 
2570 /*
2571  * APIC command line parameters
2572  */
setup_disableapic(char * arg)2573 static int __init setup_disableapic(char *arg)
2574 {
2575 	apic_is_disabled = true;
2576 	setup_clear_cpu_cap(X86_FEATURE_APIC);
2577 	return 0;
2578 }
2579 early_param("disableapic", setup_disableapic);
2580 
2581 /* same as disableapic, for compatibility */
setup_nolapic(char * arg)2582 static int __init setup_nolapic(char *arg)
2583 {
2584 	return setup_disableapic(arg);
2585 }
2586 early_param("nolapic", setup_nolapic);
2587 
parse_lapic_timer_c2_ok(char * arg)2588 static int __init parse_lapic_timer_c2_ok(char *arg)
2589 {
2590 	local_apic_timer_c2_ok = 1;
2591 	return 0;
2592 }
2593 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2594 
parse_disable_apic_timer(char * arg)2595 static int __init parse_disable_apic_timer(char *arg)
2596 {
2597 	disable_apic_timer = 1;
2598 	return 0;
2599 }
2600 early_param("noapictimer", parse_disable_apic_timer);
2601 
parse_nolapic_timer(char * arg)2602 static int __init parse_nolapic_timer(char *arg)
2603 {
2604 	disable_apic_timer = 1;
2605 	return 0;
2606 }
2607 early_param("nolapic_timer", parse_nolapic_timer);
2608 
apic_set_verbosity(char * arg)2609 static int __init apic_set_verbosity(char *arg)
2610 {
2611 	if (!arg)  {
2612 		if (IS_ENABLED(CONFIG_X86_32))
2613 			return -EINVAL;
2614 
2615 		ioapic_is_disabled = false;
2616 		return 0;
2617 	}
2618 
2619 	if (strcmp("debug", arg) == 0)
2620 		apic_verbosity = APIC_DEBUG;
2621 	else if (strcmp("verbose", arg) == 0)
2622 		apic_verbosity = APIC_VERBOSE;
2623 #ifdef CONFIG_X86_64
2624 	else {
2625 		pr_warn("APIC Verbosity level %s not recognised"
2626 			" use apic=verbose or apic=debug\n", arg);
2627 		return -EINVAL;
2628 	}
2629 #endif
2630 
2631 	return 0;
2632 }
2633 early_param("apic", apic_set_verbosity);
2634 
lapic_insert_resource(void)2635 static int __init lapic_insert_resource(void)
2636 {
2637 	if (!apic_mmio_base)
2638 		return -1;
2639 
2640 	/* Put local APIC into the resource map. */
2641 	lapic_resource.start = apic_mmio_base;
2642 	lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2643 	insert_resource(&iomem_resource, &lapic_resource);
2644 
2645 	return 0;
2646 }
2647 
2648 /*
2649  * need call insert after e820__reserve_resources()
2650  * that is using request_resource
2651  */
2652 late_initcall(lapic_insert_resource);
2653 
apic_set_extnmi(char * arg)2654 static int __init apic_set_extnmi(char *arg)
2655 {
2656 	if (!arg)
2657 		return -EINVAL;
2658 
2659 	if (!strncmp("all", arg, 3))
2660 		apic_extnmi = APIC_EXTNMI_ALL;
2661 	else if (!strncmp("none", arg, 4))
2662 		apic_extnmi = APIC_EXTNMI_NONE;
2663 	else if (!strncmp("bsp", arg, 3))
2664 		apic_extnmi = APIC_EXTNMI_BSP;
2665 	else {
2666 		pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2667 		return -EINVAL;
2668 	}
2669 
2670 	return 0;
2671 }
2672 early_param("apic_extnmi", apic_set_extnmi);
2673