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 <xen/xen.h>
40
41 #include <asm/trace/irq_vectors.h>
42 #include <asm/irq_remapping.h>
43 #include <asm/pc-conf-reg.h>
44 #include <asm/perf_event.h>
45 #include <asm/x86_init.h>
46 #include <linux/atomic.h>
47 #include <asm/barrier.h>
48 #include <asm/mpspec.h>
49 #include <asm/i8259.h>
50 #include <asm/proto.h>
51 #include <asm/traps.h>
52 #include <asm/apic.h>
53 #include <asm/acpi.h>
54 #include <asm/io_apic.h>
55 #include <asm/desc.h>
56 #include <asm/hpet.h>
57 #include <asm/mtrr.h>
58 #include <asm/time.h>
59 #include <asm/smp.h>
60 #include <asm/mce.h>
61 #include <asm/tsc.h>
62 #include <asm/hypervisor.h>
63 #include <asm/cpu_device_id.h>
64 #include <asm/intel-family.h>
65 #include <asm/irq_regs.h>
66 #include <asm/cpu.h>
67
68 #include "local.h"
69
70 unsigned int num_processors;
71
72 unsigned disabled_cpus;
73
74 /* Processor that is doing the boot up */
75 unsigned int boot_cpu_physical_apicid __ro_after_init = -1U;
76 EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
77
78 u8 boot_cpu_apic_version __ro_after_init;
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 /* For parallel bootup. */
104 unsigned long apic_mmio_base __ro_after_init;
105
apic_accessible(void)106 static inline bool apic_accessible(void)
107 {
108 return x2apic_mode || apic_mmio_base;
109 }
110
111 /*
112 * Map cpu index to physical APIC ID
113 */
114 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
115 DEFINE_EARLY_PER_CPU_READ_MOSTLY(u32, x86_cpu_to_acpiid, U32_MAX);
116 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
117 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_acpiid);
118
119 #ifdef CONFIG_X86_32
120 /* Local APIC was disabled by the BIOS and enabled by the kernel */
121 static int enabled_via_apicbase __ro_after_init;
122
123 /*
124 * Handle interrupt mode configuration register (IMCR).
125 * This register controls whether the interrupt signals
126 * that reach the BSP come from the master PIC or from the
127 * local APIC. Before entering Symmetric I/O Mode, either
128 * the BIOS or the operating system must switch out of
129 * PIC Mode by changing the IMCR.
130 */
imcr_pic_to_apic(void)131 static inline void imcr_pic_to_apic(void)
132 {
133 /* NMI and 8259 INTR go through APIC */
134 pc_conf_set(PC_CONF_MPS_IMCR, 0x01);
135 }
136
imcr_apic_to_pic(void)137 static inline void imcr_apic_to_pic(void)
138 {
139 /* NMI and 8259 INTR go directly to BSP */
140 pc_conf_set(PC_CONF_MPS_IMCR, 0x00);
141 }
142 #endif
143
144 /*
145 * Knob to control our willingness to enable the local APIC.
146 *
147 * +1=force-enable
148 */
149 static int force_enable_local_apic __initdata;
150
151 /*
152 * APIC command line parameters
153 */
parse_lapic(char * arg)154 static int __init parse_lapic(char *arg)
155 {
156 if (IS_ENABLED(CONFIG_X86_32) && !arg)
157 force_enable_local_apic = 1;
158 else if (arg && !strncmp(arg, "notscdeadline", 13))
159 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
160 return 0;
161 }
162 early_param("lapic", parse_lapic);
163
164 #ifdef CONFIG_X86_64
165 static int apic_calibrate_pmtmr __initdata;
setup_apicpmtimer(char * s)166 static __init int setup_apicpmtimer(char *s)
167 {
168 apic_calibrate_pmtmr = 1;
169 notsc_setup(NULL);
170 return 1;
171 }
172 __setup("apicpmtimer", setup_apicpmtimer);
173 #endif
174
175 static unsigned long mp_lapic_addr __ro_after_init;
176 bool apic_is_disabled __ro_after_init;
177 /* Disable local APIC timer from the kernel commandline or via dmi quirk */
178 static int disable_apic_timer __initdata;
179 /* Local APIC timer works in C2 */
180 int local_apic_timer_c2_ok __ro_after_init;
181 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
182
183 /*
184 * Debug level, exported for io_apic.c
185 */
186 int apic_verbosity __ro_after_init;
187
188 int pic_mode __ro_after_init;
189
190 /* Have we found an MP table */
191 int smp_found_config __ro_after_init;
192
193 static struct resource lapic_resource = {
194 .name = "Local APIC",
195 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
196 };
197
198 unsigned int lapic_timer_period = 0;
199
200 static void apic_pm_activate(void);
201
202 /*
203 * Get the LAPIC version
204 */
lapic_get_version(void)205 static inline int lapic_get_version(void)
206 {
207 return GET_APIC_VERSION(apic_read(APIC_LVR));
208 }
209
210 /*
211 * Check, if the APIC is integrated or a separate chip
212 */
lapic_is_integrated(void)213 static inline int lapic_is_integrated(void)
214 {
215 return APIC_INTEGRATED(lapic_get_version());
216 }
217
218 /*
219 * Check, whether this is a modern or a first generation APIC
220 */
modern_apic(void)221 static int modern_apic(void)
222 {
223 /* AMD systems use old APIC versions, so check the CPU */
224 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
225 boot_cpu_data.x86 >= 0xf)
226 return 1;
227
228 /* Hygon systems use modern APIC */
229 if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
230 return 1;
231
232 return lapic_get_version() >= 0x14;
233 }
234
235 /*
236 * right after this call apic become NOOP driven
237 * so apic->write/read doesn't do anything
238 */
apic_disable(void)239 static void __init apic_disable(void)
240 {
241 apic_install_driver(&apic_noop);
242 }
243
native_apic_icr_write(u32 low,u32 id)244 void native_apic_icr_write(u32 low, u32 id)
245 {
246 unsigned long flags;
247
248 local_irq_save(flags);
249 apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id));
250 apic_write(APIC_ICR, low);
251 local_irq_restore(flags);
252 }
253
native_apic_icr_read(void)254 u64 native_apic_icr_read(void)
255 {
256 u32 icr1, icr2;
257
258 icr2 = apic_read(APIC_ICR2);
259 icr1 = apic_read(APIC_ICR);
260
261 return icr1 | ((u64)icr2 << 32);
262 }
263
264 #ifdef CONFIG_X86_32
265 /**
266 * get_physical_broadcast - Get number of physical broadcast IDs
267 */
get_physical_broadcast(void)268 int get_physical_broadcast(void)
269 {
270 return modern_apic() ? 0xff : 0xf;
271 }
272 #endif
273
274 /**
275 * lapic_get_maxlvt - get the maximum number of local vector table entries
276 */
lapic_get_maxlvt(void)277 int lapic_get_maxlvt(void)
278 {
279 /*
280 * - we always have APIC integrated on 64bit mode
281 * - 82489DXs do not report # of LVT entries
282 */
283 return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2;
284 }
285
286 /*
287 * Local APIC timer
288 */
289
290 /* Clock divisor */
291 #define APIC_DIVISOR 16
292 #define TSC_DIVISOR 8
293
294 /* i82489DX specific */
295 #define I82489DX_BASE_DIVIDER (((0x2) << 18))
296
297 /*
298 * This function sets up the local APIC timer, with a timeout of
299 * 'clocks' APIC bus clock. During calibration we actually call
300 * this function twice on the boot CPU, once with a bogus timeout
301 * value, second time for real. The other (noncalibrating) CPUs
302 * call this function only once, with the real, calibrated value.
303 *
304 * We do reads before writes even if unnecessary, to get around the
305 * P5 APIC double write bug.
306 */
__setup_APIC_LVTT(unsigned int clocks,int oneshot,int irqen)307 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
308 {
309 unsigned int lvtt_value, tmp_value;
310
311 lvtt_value = LOCAL_TIMER_VECTOR;
312 if (!oneshot)
313 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
314 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
315 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
316
317 /*
318 * The i82489DX APIC uses bit 18 and 19 for the base divider. This
319 * overlaps with bit 18 on integrated APICs, but is not documented
320 * in the SDM. No problem though. i82489DX equipped systems do not
321 * have TSC deadline timer.
322 */
323 if (!lapic_is_integrated())
324 lvtt_value |= I82489DX_BASE_DIVIDER;
325
326 if (!irqen)
327 lvtt_value |= APIC_LVT_MASKED;
328
329 apic_write(APIC_LVTT, lvtt_value);
330
331 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
332 /*
333 * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode,
334 * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized.
335 * According to Intel, MFENCE can do the serialization here.
336 */
337 asm volatile("mfence" : : : "memory");
338 return;
339 }
340
341 /*
342 * Divide PICLK by 16
343 */
344 tmp_value = apic_read(APIC_TDCR);
345 apic_write(APIC_TDCR,
346 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
347 APIC_TDR_DIV_16);
348
349 if (!oneshot)
350 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
351 }
352
353 /*
354 * Setup extended LVT, AMD specific
355 *
356 * Software should use the LVT offsets the BIOS provides. The offsets
357 * are determined by the subsystems using it like those for MCE
358 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
359 * are supported. Beginning with family 10h at least 4 offsets are
360 * available.
361 *
362 * Since the offsets must be consistent for all cores, we keep track
363 * of the LVT offsets in software and reserve the offset for the same
364 * vector also to be used on other cores. An offset is freed by
365 * setting the entry to APIC_EILVT_MASKED.
366 *
367 * If the BIOS is right, there should be no conflicts. Otherwise a
368 * "[Firmware Bug]: ..." error message is generated. However, if
369 * software does not properly determines the offsets, it is not
370 * necessarily a BIOS bug.
371 */
372
373 static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
374
eilvt_entry_is_changeable(unsigned int old,unsigned int new)375 static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
376 {
377 return (old & APIC_EILVT_MASKED)
378 || (new == APIC_EILVT_MASKED)
379 || ((new & ~APIC_EILVT_MASKED) == old);
380 }
381
reserve_eilvt_offset(int offset,unsigned int new)382 static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
383 {
384 unsigned int rsvd, vector;
385
386 if (offset >= APIC_EILVT_NR_MAX)
387 return ~0;
388
389 rsvd = atomic_read(&eilvt_offsets[offset]);
390 do {
391 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
392 if (vector && !eilvt_entry_is_changeable(vector, new))
393 /* may not change if vectors are different */
394 return rsvd;
395 } while (!atomic_try_cmpxchg(&eilvt_offsets[offset], &rsvd, new));
396
397 rsvd = new & ~APIC_EILVT_MASKED;
398 if (rsvd && rsvd != vector)
399 pr_info("LVT offset %d assigned for vector 0x%02x\n",
400 offset, rsvd);
401
402 return new;
403 }
404
405 /*
406 * If mask=1, the LVT entry does not generate interrupts while mask=0
407 * enables the vector. See also the BKDGs. Must be called with
408 * preemption disabled.
409 */
410
setup_APIC_eilvt(u8 offset,u8 vector,u8 msg_type,u8 mask)411 int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
412 {
413 unsigned long reg = APIC_EILVTn(offset);
414 unsigned int new, old, reserved;
415
416 new = (mask << 16) | (msg_type << 8) | vector;
417 old = apic_read(reg);
418 reserved = reserve_eilvt_offset(offset, new);
419
420 if (reserved != new) {
421 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
422 "vector 0x%x, but the register is already in use for "
423 "vector 0x%x on another cpu\n",
424 smp_processor_id(), reg, offset, new, reserved);
425 return -EINVAL;
426 }
427
428 if (!eilvt_entry_is_changeable(old, new)) {
429 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
430 "vector 0x%x, but the register is already in use for "
431 "vector 0x%x on this cpu\n",
432 smp_processor_id(), reg, offset, new, old);
433 return -EBUSY;
434 }
435
436 apic_write(reg, new);
437
438 return 0;
439 }
440 EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
441
442 /*
443 * Program the next event, relative to now
444 */
lapic_next_event(unsigned long delta,struct clock_event_device * evt)445 static int lapic_next_event(unsigned long delta,
446 struct clock_event_device *evt)
447 {
448 apic_write(APIC_TMICT, delta);
449 return 0;
450 }
451
lapic_next_deadline(unsigned long delta,struct clock_event_device * evt)452 static int lapic_next_deadline(unsigned long delta,
453 struct clock_event_device *evt)
454 {
455 u64 tsc;
456
457 /* This MSR is special and need a special fence: */
458 weak_wrmsr_fence();
459
460 tsc = rdtsc();
461 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
462 return 0;
463 }
464
lapic_timer_shutdown(struct clock_event_device * evt)465 static int lapic_timer_shutdown(struct clock_event_device *evt)
466 {
467 unsigned int v;
468
469 /* Lapic used as dummy for broadcast ? */
470 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
471 return 0;
472
473 v = apic_read(APIC_LVTT);
474 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
475 apic_write(APIC_LVTT, v);
476 apic_write(APIC_TMICT, 0);
477 return 0;
478 }
479
480 static inline int
lapic_timer_set_periodic_oneshot(struct clock_event_device * evt,bool oneshot)481 lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
482 {
483 /* Lapic used as dummy for broadcast ? */
484 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
485 return 0;
486
487 __setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
488 return 0;
489 }
490
lapic_timer_set_periodic(struct clock_event_device * evt)491 static int lapic_timer_set_periodic(struct clock_event_device *evt)
492 {
493 return lapic_timer_set_periodic_oneshot(evt, false);
494 }
495
lapic_timer_set_oneshot(struct clock_event_device * evt)496 static int lapic_timer_set_oneshot(struct clock_event_device *evt)
497 {
498 return lapic_timer_set_periodic_oneshot(evt, true);
499 }
500
501 /*
502 * Local APIC timer broadcast function
503 */
lapic_timer_broadcast(const struct cpumask * mask)504 static void lapic_timer_broadcast(const struct cpumask *mask)
505 {
506 #ifdef CONFIG_SMP
507 __apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
508 #endif
509 }
510
511
512 /*
513 * The local apic timer can be used for any function which is CPU local.
514 */
515 static struct clock_event_device lapic_clockevent = {
516 .name = "lapic",
517 .features = CLOCK_EVT_FEAT_PERIODIC |
518 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
519 | CLOCK_EVT_FEAT_DUMMY,
520 .shift = 32,
521 .set_state_shutdown = lapic_timer_shutdown,
522 .set_state_periodic = lapic_timer_set_periodic,
523 .set_state_oneshot = lapic_timer_set_oneshot,
524 .set_state_oneshot_stopped = lapic_timer_shutdown,
525 .set_next_event = lapic_next_event,
526 .broadcast = lapic_timer_broadcast,
527 .rating = 100,
528 .irq = -1,
529 };
530 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
531
532 static const struct x86_cpu_id deadline_match[] __initconst = {
533 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
534 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
535
536 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_X, 0x0b000020),
537
538 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
539 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
540 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
541 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
542
543 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
544 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
545 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
546
547 X86_MATCH_INTEL_FAM6_MODEL( HASWELL, 0x22),
548 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_L, 0x20),
549 X86_MATCH_INTEL_FAM6_MODEL( HASWELL_G, 0x17),
550
551 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL, 0x25),
552 X86_MATCH_INTEL_FAM6_MODEL( BROADWELL_G, 0x17),
553
554 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE_L, 0xb2),
555 X86_MATCH_INTEL_FAM6_MODEL( SKYLAKE, 0xb2),
556
557 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE_L, 0x52),
558 X86_MATCH_INTEL_FAM6_MODEL( KABYLAKE, 0x52),
559
560 {},
561 };
562
apic_validate_deadline_timer(void)563 static __init bool apic_validate_deadline_timer(void)
564 {
565 const struct x86_cpu_id *m;
566 u32 rev;
567
568 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
569 return false;
570 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
571 return true;
572
573 m = x86_match_cpu(deadline_match);
574 if (!m)
575 return true;
576
577 rev = (u32)m->driver_data;
578
579 if (boot_cpu_data.microcode >= rev)
580 return true;
581
582 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
583 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
584 "please update microcode to version: 0x%x (or later)\n", rev);
585 return false;
586 }
587
588 /*
589 * Setup the local APIC timer for this CPU. Copy the initialized values
590 * of the boot CPU and register the clock event in the framework.
591 */
setup_APIC_timer(void)592 static void setup_APIC_timer(void)
593 {
594 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
595
596 if (this_cpu_has(X86_FEATURE_ARAT)) {
597 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
598 /* Make LAPIC timer preferable over percpu HPET */
599 lapic_clockevent.rating = 150;
600 }
601
602 memcpy(levt, &lapic_clockevent, sizeof(*levt));
603 levt->cpumask = cpumask_of(smp_processor_id());
604
605 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
606 levt->name = "lapic-deadline";
607 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
608 CLOCK_EVT_FEAT_DUMMY);
609 levt->set_next_event = lapic_next_deadline;
610 clockevents_config_and_register(levt,
611 tsc_khz * (1000 / TSC_DIVISOR),
612 0xF, ~0UL);
613 } else
614 clockevents_register_device(levt);
615 }
616
617 /*
618 * Install the updated TSC frequency from recalibration at the TSC
619 * deadline clockevent devices.
620 */
__lapic_update_tsc_freq(void * info)621 static void __lapic_update_tsc_freq(void *info)
622 {
623 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
624
625 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
626 return;
627
628 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
629 }
630
lapic_update_tsc_freq(void)631 void lapic_update_tsc_freq(void)
632 {
633 /*
634 * The clockevent device's ->mult and ->shift can both be
635 * changed. In order to avoid races, schedule the frequency
636 * update code on each CPU.
637 */
638 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
639 }
640
641 /*
642 * In this functions we calibrate APIC bus clocks to the external timer.
643 *
644 * We want to do the calibration only once since we want to have local timer
645 * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
646 * frequency.
647 *
648 * This was previously done by reading the PIT/HPET and waiting for a wrap
649 * around to find out, that a tick has elapsed. I have a box, where the PIT
650 * readout is broken, so it never gets out of the wait loop again. This was
651 * also reported by others.
652 *
653 * Monitoring the jiffies value is inaccurate and the clockevents
654 * infrastructure allows us to do a simple substitution of the interrupt
655 * handler.
656 *
657 * The calibration routine also uses the pm_timer when possible, as the PIT
658 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
659 * back to normal later in the boot process).
660 */
661
662 #define LAPIC_CAL_LOOPS (HZ/10)
663
664 static __initdata int lapic_cal_loops = -1;
665 static __initdata long lapic_cal_t1, lapic_cal_t2;
666 static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
667 static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
668 static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
669
670 /*
671 * Temporary interrupt handler and polled calibration function.
672 */
lapic_cal_handler(struct clock_event_device * dev)673 static void __init lapic_cal_handler(struct clock_event_device *dev)
674 {
675 unsigned long long tsc = 0;
676 long tapic = apic_read(APIC_TMCCT);
677 unsigned long pm = acpi_pm_read_early();
678
679 if (boot_cpu_has(X86_FEATURE_TSC))
680 tsc = rdtsc();
681
682 switch (lapic_cal_loops++) {
683 case 0:
684 lapic_cal_t1 = tapic;
685 lapic_cal_tsc1 = tsc;
686 lapic_cal_pm1 = pm;
687 lapic_cal_j1 = jiffies;
688 break;
689
690 case LAPIC_CAL_LOOPS:
691 lapic_cal_t2 = tapic;
692 lapic_cal_tsc2 = tsc;
693 if (pm < lapic_cal_pm1)
694 pm += ACPI_PM_OVRRUN;
695 lapic_cal_pm2 = pm;
696 lapic_cal_j2 = jiffies;
697 break;
698 }
699 }
700
701 static int __init
calibrate_by_pmtimer(long deltapm,long * delta,long * deltatsc)702 calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
703 {
704 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
705 const long pm_thresh = pm_100ms / 100;
706 unsigned long mult;
707 u64 res;
708
709 #ifndef CONFIG_X86_PM_TIMER
710 return -1;
711 #endif
712
713 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
714
715 /* Check, if the PM timer is available */
716 if (!deltapm)
717 return -1;
718
719 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
720
721 if (deltapm > (pm_100ms - pm_thresh) &&
722 deltapm < (pm_100ms + pm_thresh)) {
723 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
724 return 0;
725 }
726
727 res = (((u64)deltapm) * mult) >> 22;
728 do_div(res, 1000000);
729 pr_warn("APIC calibration not consistent "
730 "with PM-Timer: %ldms instead of 100ms\n", (long)res);
731
732 /* Correct the lapic counter value */
733 res = (((u64)(*delta)) * pm_100ms);
734 do_div(res, deltapm);
735 pr_info("APIC delta adjusted to PM-Timer: "
736 "%lu (%ld)\n", (unsigned long)res, *delta);
737 *delta = (long)res;
738
739 /* Correct the tsc counter value */
740 if (boot_cpu_has(X86_FEATURE_TSC)) {
741 res = (((u64)(*deltatsc)) * pm_100ms);
742 do_div(res, deltapm);
743 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
744 "PM-Timer: %lu (%ld)\n",
745 (unsigned long)res, *deltatsc);
746 *deltatsc = (long)res;
747 }
748
749 return 0;
750 }
751
lapic_init_clockevent(void)752 static int __init lapic_init_clockevent(void)
753 {
754 if (!lapic_timer_period)
755 return -1;
756
757 /* Calculate the scaled math multiplication factor */
758 lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
759 TICK_NSEC, lapic_clockevent.shift);
760 lapic_clockevent.max_delta_ns =
761 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
762 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
763 lapic_clockevent.min_delta_ns =
764 clockevent_delta2ns(0xF, &lapic_clockevent);
765 lapic_clockevent.min_delta_ticks = 0xF;
766
767 return 0;
768 }
769
apic_needs_pit(void)770 bool __init apic_needs_pit(void)
771 {
772 /*
773 * If the frequencies are not known, PIT is required for both TSC
774 * and apic timer calibration.
775 */
776 if (!tsc_khz || !cpu_khz)
777 return true;
778
779 /* Is there an APIC at all or is it disabled? */
780 if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled)
781 return true;
782
783 /*
784 * If interrupt delivery mode is legacy PIC or virtual wire without
785 * configuration, the local APIC timer wont be set up. Make sure
786 * that the PIT is initialized.
787 */
788 if (apic_intr_mode == APIC_PIC ||
789 apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
790 return true;
791
792 /* Virt guests may lack ARAT, but still have DEADLINE */
793 if (!boot_cpu_has(X86_FEATURE_ARAT))
794 return true;
795
796 /* Deadline timer is based on TSC so no further PIT action required */
797 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
798 return false;
799
800 /* APIC timer disabled? */
801 if (disable_apic_timer)
802 return true;
803 /*
804 * The APIC timer frequency is known already, no PIT calibration
805 * required. If unknown, let the PIT be initialized.
806 */
807 return lapic_timer_period == 0;
808 }
809
calibrate_APIC_clock(void)810 static int __init calibrate_APIC_clock(void)
811 {
812 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
813 u64 tsc_perj = 0, tsc_start = 0;
814 unsigned long jif_start;
815 unsigned long deltaj;
816 long delta, deltatsc;
817 int pm_referenced = 0;
818
819 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
820 return 0;
821
822 /*
823 * Check if lapic timer has already been calibrated by platform
824 * specific routine, such as tsc calibration code. If so just fill
825 * in the clockevent structure and return.
826 */
827 if (!lapic_init_clockevent()) {
828 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
829 lapic_timer_period);
830 /*
831 * Direct calibration methods must have an always running
832 * local APIC timer, no need for broadcast timer.
833 */
834 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
835 return 0;
836 }
837
838 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
839 "calibrating APIC timer ...\n");
840
841 /*
842 * There are platforms w/o global clockevent devices. Instead of
843 * making the calibration conditional on that, use a polling based
844 * approach everywhere.
845 */
846 local_irq_disable();
847
848 /*
849 * Setup the APIC counter to maximum. There is no way the lapic
850 * can underflow in the 100ms detection time frame
851 */
852 __setup_APIC_LVTT(0xffffffff, 0, 0);
853
854 /*
855 * Methods to terminate the calibration loop:
856 * 1) Global clockevent if available (jiffies)
857 * 2) TSC if available and frequency is known
858 */
859 jif_start = READ_ONCE(jiffies);
860
861 if (tsc_khz) {
862 tsc_start = rdtsc();
863 tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
864 }
865
866 /*
867 * Enable interrupts so the tick can fire, if a global
868 * clockevent device is available
869 */
870 local_irq_enable();
871
872 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
873 /* Wait for a tick to elapse */
874 while (1) {
875 if (tsc_khz) {
876 u64 tsc_now = rdtsc();
877 if ((tsc_now - tsc_start) >= tsc_perj) {
878 tsc_start += tsc_perj;
879 break;
880 }
881 } else {
882 unsigned long jif_now = READ_ONCE(jiffies);
883
884 if (time_after(jif_now, jif_start)) {
885 jif_start = jif_now;
886 break;
887 }
888 }
889 cpu_relax();
890 }
891
892 /* Invoke the calibration routine */
893 local_irq_disable();
894 lapic_cal_handler(NULL);
895 local_irq_enable();
896 }
897
898 local_irq_disable();
899
900 /* Build delta t1-t2 as apic timer counts down */
901 delta = lapic_cal_t1 - lapic_cal_t2;
902 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
903
904 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
905
906 /* we trust the PM based calibration if possible */
907 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
908 &delta, &deltatsc);
909
910 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
911 lapic_init_clockevent();
912
913 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
914 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
915 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
916 lapic_timer_period);
917
918 if (boot_cpu_has(X86_FEATURE_TSC)) {
919 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
920 "%ld.%04ld MHz.\n",
921 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
922 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
923 }
924
925 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
926 "%u.%04u MHz.\n",
927 lapic_timer_period / (1000000 / HZ),
928 lapic_timer_period % (1000000 / HZ));
929
930 /*
931 * Do a sanity check on the APIC calibration result
932 */
933 if (lapic_timer_period < (1000000 / HZ)) {
934 local_irq_enable();
935 pr_warn("APIC frequency too slow, disabling apic timer\n");
936 return -1;
937 }
938
939 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
940
941 /*
942 * PM timer calibration failed or not turned on so lets try APIC
943 * timer based calibration, if a global clockevent device is
944 * available.
945 */
946 if (!pm_referenced && global_clock_event) {
947 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
948
949 /*
950 * Setup the apic timer manually
951 */
952 levt->event_handler = lapic_cal_handler;
953 lapic_timer_set_periodic(levt);
954 lapic_cal_loops = -1;
955
956 /* Let the interrupts run */
957 local_irq_enable();
958
959 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
960 cpu_relax();
961
962 /* Stop the lapic timer */
963 local_irq_disable();
964 lapic_timer_shutdown(levt);
965
966 /* Jiffies delta */
967 deltaj = lapic_cal_j2 - lapic_cal_j1;
968 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
969
970 /* Check, if the jiffies result is consistent */
971 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
972 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
973 else
974 levt->features |= CLOCK_EVT_FEAT_DUMMY;
975 }
976 local_irq_enable();
977
978 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
979 pr_warn("APIC timer disabled due to verification failure\n");
980 return -1;
981 }
982
983 return 0;
984 }
985
986 /*
987 * Setup the boot APIC
988 *
989 * Calibrate and verify the result.
990 */
setup_boot_APIC_clock(void)991 void __init setup_boot_APIC_clock(void)
992 {
993 /*
994 * The local apic timer can be disabled via the kernel
995 * commandline or from the CPU detection code. Register the lapic
996 * timer as a dummy clock event source on SMP systems, so the
997 * broadcast mechanism is used. On UP systems simply ignore it.
998 */
999 if (disable_apic_timer) {
1000 pr_info("Disabling APIC timer\n");
1001 /* No broadcast on UP ! */
1002 if (num_possible_cpus() > 1) {
1003 lapic_clockevent.mult = 1;
1004 setup_APIC_timer();
1005 }
1006 return;
1007 }
1008
1009 if (calibrate_APIC_clock()) {
1010 /* No broadcast on UP ! */
1011 if (num_possible_cpus() > 1)
1012 setup_APIC_timer();
1013 return;
1014 }
1015
1016 /*
1017 * If nmi_watchdog is set to IO_APIC, we need the
1018 * PIT/HPET going. Otherwise register lapic as a dummy
1019 * device.
1020 */
1021 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
1022
1023 /* Setup the lapic or request the broadcast */
1024 setup_APIC_timer();
1025 amd_e400_c1e_apic_setup();
1026 }
1027
setup_secondary_APIC_clock(void)1028 void setup_secondary_APIC_clock(void)
1029 {
1030 setup_APIC_timer();
1031 amd_e400_c1e_apic_setup();
1032 }
1033
1034 /*
1035 * The guts of the apic timer interrupt
1036 */
local_apic_timer_interrupt(void)1037 static void local_apic_timer_interrupt(void)
1038 {
1039 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1040
1041 /*
1042 * Normally we should not be here till LAPIC has been initialized but
1043 * in some cases like kdump, its possible that there is a pending LAPIC
1044 * timer interrupt from previous kernel's context and is delivered in
1045 * new kernel the moment interrupts are enabled.
1046 *
1047 * Interrupts are enabled early and LAPIC is setup much later, hence
1048 * its possible that when we get here evt->event_handler is NULL.
1049 * Check for event_handler being NULL and discard the interrupt as
1050 * spurious.
1051 */
1052 if (!evt->event_handler) {
1053 pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1054 smp_processor_id());
1055 /* Switch it off */
1056 lapic_timer_shutdown(evt);
1057 return;
1058 }
1059
1060 /*
1061 * the NMI deadlock-detector uses this.
1062 */
1063 inc_irq_stat(apic_timer_irqs);
1064
1065 evt->event_handler(evt);
1066 }
1067
1068 /*
1069 * Local APIC timer interrupt. This is the most natural way for doing
1070 * local interrupts, but local timer interrupts can be emulated by
1071 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1072 *
1073 * [ if a single-CPU system runs an SMP kernel then we call the local
1074 * interrupt as well. Thus we cannot inline the local irq ... ]
1075 */
DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)1076 DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1077 {
1078 struct pt_regs *old_regs = set_irq_regs(regs);
1079
1080 apic_eoi();
1081 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1082 local_apic_timer_interrupt();
1083 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1084
1085 set_irq_regs(old_regs);
1086 }
1087
1088 /*
1089 * Local APIC start and shutdown
1090 */
1091
1092 /**
1093 * clear_local_APIC - shutdown the local APIC
1094 *
1095 * This is called, when a CPU is disabled and before rebooting, so the state of
1096 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1097 * leftovers during boot.
1098 */
clear_local_APIC(void)1099 void clear_local_APIC(void)
1100 {
1101 int maxlvt;
1102 u32 v;
1103
1104 if (!apic_accessible())
1105 return;
1106
1107 maxlvt = lapic_get_maxlvt();
1108 /*
1109 * Masking an LVT entry can trigger a local APIC error
1110 * if the vector is zero. Mask LVTERR first to prevent this.
1111 */
1112 if (maxlvt >= 3) {
1113 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1114 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1115 }
1116 /*
1117 * Careful: we have to set masks only first to deassert
1118 * any level-triggered sources.
1119 */
1120 v = apic_read(APIC_LVTT);
1121 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1122 v = apic_read(APIC_LVT0);
1123 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1124 v = apic_read(APIC_LVT1);
1125 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1126 if (maxlvt >= 4) {
1127 v = apic_read(APIC_LVTPC);
1128 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1129 }
1130
1131 /* lets not touch this if we didn't frob it */
1132 #ifdef CONFIG_X86_THERMAL_VECTOR
1133 if (maxlvt >= 5) {
1134 v = apic_read(APIC_LVTTHMR);
1135 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1136 }
1137 #endif
1138 #ifdef CONFIG_X86_MCE_INTEL
1139 if (maxlvt >= 6) {
1140 v = apic_read(APIC_LVTCMCI);
1141 if (!(v & APIC_LVT_MASKED))
1142 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1143 }
1144 #endif
1145
1146 /*
1147 * Clean APIC state for other OSs:
1148 */
1149 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1150 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1151 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1152 if (maxlvt >= 3)
1153 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1154 if (maxlvt >= 4)
1155 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1156
1157 /* Integrated APIC (!82489DX) ? */
1158 if (lapic_is_integrated()) {
1159 if (maxlvt > 3)
1160 /* Clear ESR due to Pentium errata 3AP and 11AP */
1161 apic_write(APIC_ESR, 0);
1162 apic_read(APIC_ESR);
1163 }
1164 }
1165
1166 /**
1167 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1168 *
1169 * Contrary to disable_local_APIC() this does not touch the enable bit in
1170 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1171 * bus would require a hardware reset as the APIC would lose track of bus
1172 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1173 * but it has to be guaranteed that no interrupt is sent to the APIC while
1174 * in that state and it's not clear from the SDM whether it still responds
1175 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1176 */
apic_soft_disable(void)1177 void apic_soft_disable(void)
1178 {
1179 u32 value;
1180
1181 clear_local_APIC();
1182
1183 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1184 value = apic_read(APIC_SPIV);
1185 value &= ~APIC_SPIV_APIC_ENABLED;
1186 apic_write(APIC_SPIV, value);
1187 }
1188
1189 /**
1190 * disable_local_APIC - clear and disable the local APIC
1191 */
disable_local_APIC(void)1192 void disable_local_APIC(void)
1193 {
1194 if (!apic_accessible())
1195 return;
1196
1197 apic_soft_disable();
1198
1199 #ifdef CONFIG_X86_32
1200 /*
1201 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1202 * restore the disabled state.
1203 */
1204 if (enabled_via_apicbase) {
1205 unsigned int l, h;
1206
1207 rdmsr(MSR_IA32_APICBASE, l, h);
1208 l &= ~MSR_IA32_APICBASE_ENABLE;
1209 wrmsr(MSR_IA32_APICBASE, l, h);
1210 }
1211 #endif
1212 }
1213
1214 /*
1215 * If Linux enabled the LAPIC against the BIOS default disable it down before
1216 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1217 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1218 * for the case where Linux didn't enable the LAPIC.
1219 */
lapic_shutdown(void)1220 void lapic_shutdown(void)
1221 {
1222 unsigned long flags;
1223
1224 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1225 return;
1226
1227 local_irq_save(flags);
1228
1229 #ifdef CONFIG_X86_32
1230 if (!enabled_via_apicbase)
1231 clear_local_APIC();
1232 else
1233 #endif
1234 disable_local_APIC();
1235
1236
1237 local_irq_restore(flags);
1238 }
1239
1240 /**
1241 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1242 */
sync_Arb_IDs(void)1243 void __init sync_Arb_IDs(void)
1244 {
1245 /*
1246 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1247 * needed on AMD.
1248 */
1249 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1250 return;
1251
1252 /*
1253 * Wait for idle.
1254 */
1255 apic_wait_icr_idle();
1256
1257 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1258 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1259 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1260 }
1261
1262 enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1263
__apic_intr_mode_select(void)1264 static int __init __apic_intr_mode_select(void)
1265 {
1266 /* Check kernel option */
1267 if (apic_is_disabled) {
1268 pr_info("APIC disabled via kernel command line\n");
1269 return APIC_PIC;
1270 }
1271
1272 /* Check BIOS */
1273 #ifdef CONFIG_X86_64
1274 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1275 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1276 apic_is_disabled = true;
1277 pr_info("APIC disabled by BIOS\n");
1278 return APIC_PIC;
1279 }
1280 #else
1281 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1282
1283 /* Neither 82489DX nor integrated APIC ? */
1284 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1285 apic_is_disabled = true;
1286 return APIC_PIC;
1287 }
1288
1289 /* If the BIOS pretends there is an integrated APIC ? */
1290 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1291 APIC_INTEGRATED(boot_cpu_apic_version)) {
1292 apic_is_disabled = true;
1293 pr_err(FW_BUG "Local APIC not detected, force emulation\n");
1294 return APIC_PIC;
1295 }
1296 #endif
1297
1298 /* Check MP table or ACPI MADT configuration */
1299 if (!smp_found_config) {
1300 disable_ioapic_support();
1301 if (!acpi_lapic) {
1302 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1303 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1304 }
1305 return APIC_VIRTUAL_WIRE;
1306 }
1307
1308 #ifdef CONFIG_SMP
1309 /* If SMP should be disabled, then really disable it! */
1310 if (!setup_max_cpus) {
1311 pr_info("APIC: SMP mode deactivated\n");
1312 return APIC_SYMMETRIC_IO_NO_ROUTING;
1313 }
1314 #endif
1315
1316 return APIC_SYMMETRIC_IO;
1317 }
1318
1319 /* Select the interrupt delivery mode for the BSP */
apic_intr_mode_select(void)1320 void __init apic_intr_mode_select(void)
1321 {
1322 apic_intr_mode = __apic_intr_mode_select();
1323 }
1324
1325 /*
1326 * An initial setup of the virtual wire mode.
1327 */
init_bsp_APIC(void)1328 void __init init_bsp_APIC(void)
1329 {
1330 unsigned int value;
1331
1332 /*
1333 * Don't do the setup now if we have a SMP BIOS as the
1334 * through-I/O-APIC virtual wire mode might be active.
1335 */
1336 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1337 return;
1338
1339 /*
1340 * Do not trust the local APIC being empty at bootup.
1341 */
1342 clear_local_APIC();
1343
1344 /*
1345 * Enable APIC.
1346 */
1347 value = apic_read(APIC_SPIV);
1348 value &= ~APIC_VECTOR_MASK;
1349 value |= APIC_SPIV_APIC_ENABLED;
1350
1351 #ifdef CONFIG_X86_32
1352 /* This bit is reserved on P4/Xeon and should be cleared */
1353 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1354 (boot_cpu_data.x86 == 15))
1355 value &= ~APIC_SPIV_FOCUS_DISABLED;
1356 else
1357 #endif
1358 value |= APIC_SPIV_FOCUS_DISABLED;
1359 value |= SPURIOUS_APIC_VECTOR;
1360 apic_write(APIC_SPIV, value);
1361
1362 /*
1363 * Set up the virtual wire mode.
1364 */
1365 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1366 value = APIC_DM_NMI;
1367 if (!lapic_is_integrated()) /* 82489DX */
1368 value |= APIC_LVT_LEVEL_TRIGGER;
1369 if (apic_extnmi == APIC_EXTNMI_NONE)
1370 value |= APIC_LVT_MASKED;
1371 apic_write(APIC_LVT1, value);
1372 }
1373
1374 static void __init apic_bsp_setup(bool upmode);
1375
1376 /* Init the interrupt delivery mode for the BSP */
apic_intr_mode_init(void)1377 void __init apic_intr_mode_init(void)
1378 {
1379 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1380
1381 switch (apic_intr_mode) {
1382 case APIC_PIC:
1383 pr_info("APIC: Keep in PIC mode(8259)\n");
1384 return;
1385 case APIC_VIRTUAL_WIRE:
1386 pr_info("APIC: Switch to virtual wire mode setup\n");
1387 break;
1388 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1389 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1390 upmode = true;
1391 break;
1392 case APIC_SYMMETRIC_IO:
1393 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1394 break;
1395 case APIC_SYMMETRIC_IO_NO_ROUTING:
1396 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1397 break;
1398 }
1399
1400 x86_64_probe_apic();
1401
1402 x86_32_install_bigsmp();
1403
1404 if (x86_platform.apic_post_init)
1405 x86_platform.apic_post_init();
1406
1407 apic_bsp_setup(upmode);
1408 }
1409
lapic_setup_esr(void)1410 static void lapic_setup_esr(void)
1411 {
1412 unsigned int oldvalue, value, maxlvt;
1413
1414 if (!lapic_is_integrated()) {
1415 pr_info("No ESR for 82489DX.\n");
1416 return;
1417 }
1418
1419 if (apic->disable_esr) {
1420 /*
1421 * Something untraceable is creating bad interrupts on
1422 * secondary quads ... for the moment, just leave the
1423 * ESR disabled - we can't do anything useful with the
1424 * errors anyway - mbligh
1425 */
1426 pr_info("Leaving ESR disabled.\n");
1427 return;
1428 }
1429
1430 maxlvt = lapic_get_maxlvt();
1431 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1432 apic_write(APIC_ESR, 0);
1433 oldvalue = apic_read(APIC_ESR);
1434
1435 /* enables sending errors */
1436 value = ERROR_APIC_VECTOR;
1437 apic_write(APIC_LVTERR, value);
1438
1439 /*
1440 * spec says clear errors after enabling vector.
1441 */
1442 if (maxlvt > 3)
1443 apic_write(APIC_ESR, 0);
1444 value = apic_read(APIC_ESR);
1445 if (value != oldvalue)
1446 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1447 "vector: 0x%08x after: 0x%08x\n",
1448 oldvalue, value);
1449 }
1450
1451 #define APIC_IR_REGS APIC_ISR_NR
1452 #define APIC_IR_BITS (APIC_IR_REGS * 32)
1453 #define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1454
1455 union apic_ir {
1456 unsigned long map[APIC_IR_MAPSIZE];
1457 u32 regs[APIC_IR_REGS];
1458 };
1459
apic_check_and_ack(union apic_ir * irr,union apic_ir * isr)1460 static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1461 {
1462 int i, bit;
1463
1464 /* Read the IRRs */
1465 for (i = 0; i < APIC_IR_REGS; i++)
1466 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1467
1468 /* Read the ISRs */
1469 for (i = 0; i < APIC_IR_REGS; i++)
1470 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1471
1472 /*
1473 * If the ISR map is not empty. ACK the APIC and run another round
1474 * to verify whether a pending IRR has been unblocked and turned
1475 * into a ISR.
1476 */
1477 if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1478 /*
1479 * There can be multiple ISR bits set when a high priority
1480 * interrupt preempted a lower priority one. Issue an ACK
1481 * per set bit.
1482 */
1483 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1484 apic_eoi();
1485 return true;
1486 }
1487
1488 return !bitmap_empty(irr->map, APIC_IR_BITS);
1489 }
1490
1491 /*
1492 * After a crash, we no longer service the interrupts and a pending
1493 * interrupt from previous kernel might still have ISR bit set.
1494 *
1495 * Most probably by now the CPU has serviced that pending interrupt and it
1496 * might not have done the apic_eoi() because it thought, interrupt
1497 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1498 * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1499 * a vector might get locked. It was noticed for timer irq (vector
1500 * 0x31). Issue an extra EOI to clear ISR.
1501 *
1502 * If there are pending IRR bits they turn into ISR bits after a higher
1503 * priority ISR bit has been acked.
1504 */
apic_pending_intr_clear(void)1505 static void apic_pending_intr_clear(void)
1506 {
1507 union apic_ir irr, isr;
1508 unsigned int i;
1509
1510 /* 512 loops are way oversized and give the APIC a chance to obey. */
1511 for (i = 0; i < 512; i++) {
1512 if (!apic_check_and_ack(&irr, &isr))
1513 return;
1514 }
1515 /* Dump the IRR/ISR content if that failed */
1516 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1517 }
1518
1519 /**
1520 * setup_local_APIC - setup the local APIC
1521 *
1522 * Used to setup local APIC while initializing BSP or bringing up APs.
1523 * Always called with preemption disabled.
1524 */
setup_local_APIC(void)1525 static void setup_local_APIC(void)
1526 {
1527 int cpu = smp_processor_id();
1528 unsigned int value;
1529
1530 if (apic_is_disabled) {
1531 disable_ioapic_support();
1532 return;
1533 }
1534
1535 /*
1536 * If this comes from kexec/kcrash the APIC might be enabled in
1537 * SPIV. Soft disable it before doing further initialization.
1538 */
1539 value = apic_read(APIC_SPIV);
1540 value &= ~APIC_SPIV_APIC_ENABLED;
1541 apic_write(APIC_SPIV, value);
1542
1543 #ifdef CONFIG_X86_32
1544 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1545 if (lapic_is_integrated() && apic->disable_esr) {
1546 apic_write(APIC_ESR, 0);
1547 apic_write(APIC_ESR, 0);
1548 apic_write(APIC_ESR, 0);
1549 apic_write(APIC_ESR, 0);
1550 }
1551 #endif
1552 /* Validate that the APIC is registered if required */
1553 BUG_ON(apic->apic_id_registered && !apic->apic_id_registered());
1554
1555 /*
1556 * Intel recommends to set DFR, LDR and TPR before enabling
1557 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1558 * document number 292116).
1559 *
1560 * Except for APICs which operate in physical destination mode.
1561 */
1562 if (apic->init_apic_ldr)
1563 apic->init_apic_ldr();
1564
1565 /*
1566 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1567 * vector in the 16-31 range could be delivered if TPR == 0, but we
1568 * would think it's an exception and terrible things will happen. We
1569 * never change this later on.
1570 */
1571 value = apic_read(APIC_TASKPRI);
1572 value &= ~APIC_TPRI_MASK;
1573 value |= 0x10;
1574 apic_write(APIC_TASKPRI, value);
1575
1576 /* Clear eventually stale ISR/IRR bits */
1577 apic_pending_intr_clear();
1578
1579 /*
1580 * Now that we are all set up, enable the APIC
1581 */
1582 value = apic_read(APIC_SPIV);
1583 value &= ~APIC_VECTOR_MASK;
1584 /*
1585 * Enable APIC
1586 */
1587 value |= APIC_SPIV_APIC_ENABLED;
1588
1589 #ifdef CONFIG_X86_32
1590 /*
1591 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1592 * certain networking cards. If high frequency interrupts are
1593 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1594 * entry is masked/unmasked at a high rate as well then sooner or
1595 * later IOAPIC line gets 'stuck', no more interrupts are received
1596 * from the device. If focus CPU is disabled then the hang goes
1597 * away, oh well :-(
1598 *
1599 * [ This bug can be reproduced easily with a level-triggered
1600 * PCI Ne2000 networking cards and PII/PIII processors, dual
1601 * BX chipset. ]
1602 */
1603 /*
1604 * Actually disabling the focus CPU check just makes the hang less
1605 * frequent as it makes the interrupt distribution model be more
1606 * like LRU than MRU (the short-term load is more even across CPUs).
1607 */
1608
1609 /*
1610 * - enable focus processor (bit==0)
1611 * - 64bit mode always use processor focus
1612 * so no need to set it
1613 */
1614 value &= ~APIC_SPIV_FOCUS_DISABLED;
1615 #endif
1616
1617 /*
1618 * Set spurious IRQ vector
1619 */
1620 value |= SPURIOUS_APIC_VECTOR;
1621 apic_write(APIC_SPIV, value);
1622
1623 perf_events_lapic_init();
1624
1625 /*
1626 * Set up LVT0, LVT1:
1627 *
1628 * set up through-local-APIC on the boot CPU's LINT0. This is not
1629 * strictly necessary in pure symmetric-IO mode, but sometimes
1630 * we delegate interrupts to the 8259A.
1631 */
1632 /*
1633 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1634 */
1635 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1636 if (!cpu && (pic_mode || !value || ioapic_is_disabled)) {
1637 value = APIC_DM_EXTINT;
1638 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1639 } else {
1640 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1641 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1642 }
1643 apic_write(APIC_LVT0, value);
1644
1645 /*
1646 * Only the BSP sees the LINT1 NMI signal by default. This can be
1647 * modified by apic_extnmi= boot option.
1648 */
1649 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1650 apic_extnmi == APIC_EXTNMI_ALL)
1651 value = APIC_DM_NMI;
1652 else
1653 value = APIC_DM_NMI | APIC_LVT_MASKED;
1654
1655 /* Is 82489DX ? */
1656 if (!lapic_is_integrated())
1657 value |= APIC_LVT_LEVEL_TRIGGER;
1658 apic_write(APIC_LVT1, value);
1659
1660 #ifdef CONFIG_X86_MCE_INTEL
1661 /* Recheck CMCI information after local APIC is up on CPU #0 */
1662 if (!cpu)
1663 cmci_recheck();
1664 #endif
1665 }
1666
end_local_APIC_setup(void)1667 static void end_local_APIC_setup(void)
1668 {
1669 lapic_setup_esr();
1670
1671 #ifdef CONFIG_X86_32
1672 {
1673 unsigned int value;
1674 /* Disable the local apic timer */
1675 value = apic_read(APIC_LVTT);
1676 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1677 apic_write(APIC_LVTT, value);
1678 }
1679 #endif
1680
1681 apic_pm_activate();
1682 }
1683
1684 /*
1685 * APIC setup function for application processors. Called from smpboot.c
1686 */
apic_ap_setup(void)1687 void apic_ap_setup(void)
1688 {
1689 setup_local_APIC();
1690 end_local_APIC_setup();
1691 }
1692
1693 static __init void cpu_set_boot_apic(void);
1694
apic_read_boot_cpu_id(bool x2apic)1695 static __init void apic_read_boot_cpu_id(bool x2apic)
1696 {
1697 /*
1698 * This can be invoked from check_x2apic() before the APIC has been
1699 * selected. But that code knows for sure that the BIOS enabled
1700 * X2APIC.
1701 */
1702 if (x2apic) {
1703 boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID);
1704 boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR));
1705 } else {
1706 boot_cpu_physical_apicid = read_apic_id();
1707 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1708 }
1709 cpu_set_boot_apic();
1710 }
1711
1712 #ifdef CONFIG_X86_X2APIC
1713 int x2apic_mode;
1714 EXPORT_SYMBOL_GPL(x2apic_mode);
1715
1716 enum {
1717 X2APIC_OFF,
1718 X2APIC_DISABLED,
1719 /* All states below here have X2APIC enabled */
1720 X2APIC_ON,
1721 X2APIC_ON_LOCKED
1722 };
1723 static int x2apic_state;
1724
x2apic_hw_locked(void)1725 static bool x2apic_hw_locked(void)
1726 {
1727 u64 x86_arch_cap_msr;
1728 u64 msr;
1729
1730 x86_arch_cap_msr = x86_read_arch_cap_msr();
1731 if (x86_arch_cap_msr & ARCH_CAP_XAPIC_DISABLE) {
1732 rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1733 return (msr & LEGACY_XAPIC_DISABLED);
1734 }
1735 return false;
1736 }
1737
__x2apic_disable(void)1738 static void __x2apic_disable(void)
1739 {
1740 u64 msr;
1741
1742 if (!boot_cpu_has(X86_FEATURE_APIC))
1743 return;
1744
1745 rdmsrl(MSR_IA32_APICBASE, msr);
1746 if (!(msr & X2APIC_ENABLE))
1747 return;
1748 /* Disable xapic and x2apic first and then reenable xapic mode */
1749 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1750 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1751 printk_once(KERN_INFO "x2apic disabled\n");
1752 }
1753
__x2apic_enable(void)1754 static void __x2apic_enable(void)
1755 {
1756 u64 msr;
1757
1758 rdmsrl(MSR_IA32_APICBASE, msr);
1759 if (msr & X2APIC_ENABLE)
1760 return;
1761 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1762 printk_once(KERN_INFO "x2apic enabled\n");
1763 }
1764
setup_nox2apic(char * str)1765 static int __init setup_nox2apic(char *str)
1766 {
1767 if (x2apic_enabled()) {
1768 int apicid = native_apic_msr_read(APIC_ID);
1769
1770 if (apicid >= 255) {
1771 pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1772 apicid);
1773 return 0;
1774 }
1775 if (x2apic_hw_locked()) {
1776 pr_warn("APIC locked in x2apic mode, can't disable\n");
1777 return 0;
1778 }
1779 pr_warn("x2apic already enabled.\n");
1780 __x2apic_disable();
1781 }
1782 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1783 x2apic_state = X2APIC_DISABLED;
1784 x2apic_mode = 0;
1785 return 0;
1786 }
1787 early_param("nox2apic", setup_nox2apic);
1788
1789 /* Called from cpu_init() to enable x2apic on (secondary) cpus */
x2apic_setup(void)1790 void x2apic_setup(void)
1791 {
1792 /*
1793 * Try to make the AP's APIC state match that of the BSP, but if the
1794 * BSP is unlocked and the AP is locked then there is a state mismatch.
1795 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1796 * trying to be turned off.
1797 */
1798 if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1799 pr_warn("x2apic lock mismatch between BSP and AP.\n");
1800 /*
1801 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1802 * from BIOS.
1803 */
1804 if (x2apic_state < X2APIC_ON) {
1805 __x2apic_disable();
1806 return;
1807 }
1808 __x2apic_enable();
1809 }
1810
1811 static __init void apic_set_fixmap(bool read_apic);
1812
x2apic_disable(void)1813 static __init void x2apic_disable(void)
1814 {
1815 u32 x2apic_id, state = x2apic_state;
1816
1817 x2apic_mode = 0;
1818 x2apic_state = X2APIC_DISABLED;
1819
1820 if (state != X2APIC_ON)
1821 return;
1822
1823 x2apic_id = read_apic_id();
1824 if (x2apic_id >= 255)
1825 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1826
1827 if (x2apic_hw_locked()) {
1828 pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1829 return;
1830 }
1831
1832 __x2apic_disable();
1833 /*
1834 * Don't reread the APIC ID as it was already done from
1835 * check_x2apic() and the APIC driver still is a x2APIC variant,
1836 * which fails to do the read after x2APIC was disabled.
1837 */
1838 apic_set_fixmap(false);
1839 }
1840
x2apic_enable(void)1841 static __init void x2apic_enable(void)
1842 {
1843 if (x2apic_state != X2APIC_OFF)
1844 return;
1845
1846 x2apic_mode = 1;
1847 x2apic_state = X2APIC_ON;
1848 __x2apic_enable();
1849 }
1850
try_to_enable_x2apic(int remap_mode)1851 static __init void try_to_enable_x2apic(int remap_mode)
1852 {
1853 if (x2apic_state == X2APIC_DISABLED)
1854 return;
1855
1856 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1857 u32 apic_limit = 255;
1858
1859 /*
1860 * Using X2APIC without IR is not architecturally supported
1861 * on bare metal but may be supported in guests.
1862 */
1863 if (!x86_init.hyper.x2apic_available()) {
1864 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1865 x2apic_disable();
1866 return;
1867 }
1868
1869 /*
1870 * If the hypervisor supports extended destination ID in
1871 * MSI, that increases the maximum APIC ID that can be
1872 * used for non-remapped IRQ domains.
1873 */
1874 if (x86_init.hyper.msi_ext_dest_id()) {
1875 virt_ext_dest_id = 1;
1876 apic_limit = 32767;
1877 }
1878
1879 /*
1880 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1881 * in physical mode, and CPUs with an APIC ID that cannot
1882 * be addressed must not be brought online.
1883 */
1884 x2apic_set_max_apicid(apic_limit);
1885 x2apic_phys = 1;
1886 }
1887 x2apic_enable();
1888 }
1889
check_x2apic(void)1890 void __init check_x2apic(void)
1891 {
1892 if (x2apic_enabled()) {
1893 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1894 x2apic_mode = 1;
1895 if (x2apic_hw_locked())
1896 x2apic_state = X2APIC_ON_LOCKED;
1897 else
1898 x2apic_state = X2APIC_ON;
1899 apic_read_boot_cpu_id(true);
1900 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1901 x2apic_state = X2APIC_DISABLED;
1902 }
1903 }
1904 #else /* CONFIG_X86_X2APIC */
check_x2apic(void)1905 void __init check_x2apic(void)
1906 {
1907 if (!apic_is_x2apic_enabled())
1908 return;
1909 /*
1910 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1911 */
1912 pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1913 pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1914
1915 apic_is_disabled = true;
1916 setup_clear_cpu_cap(X86_FEATURE_APIC);
1917 }
1918
try_to_enable_x2apic(int remap_mode)1919 static inline void try_to_enable_x2apic(int remap_mode) { }
__x2apic_enable(void)1920 static inline void __x2apic_enable(void) { }
1921 #endif /* !CONFIG_X86_X2APIC */
1922
enable_IR_x2apic(void)1923 void __init enable_IR_x2apic(void)
1924 {
1925 unsigned long flags;
1926 int ret, ir_stat;
1927
1928 if (ioapic_is_disabled) {
1929 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1930 return;
1931 }
1932
1933 ir_stat = irq_remapping_prepare();
1934 if (ir_stat < 0 && !x2apic_supported())
1935 return;
1936
1937 ret = save_ioapic_entries();
1938 if (ret) {
1939 pr_info("Saving IO-APIC state failed: %d\n", ret);
1940 return;
1941 }
1942
1943 local_irq_save(flags);
1944 legacy_pic->mask_all();
1945 mask_ioapic_entries();
1946
1947 /* If irq_remapping_prepare() succeeded, try to enable it */
1948 if (ir_stat >= 0)
1949 ir_stat = irq_remapping_enable();
1950 /* ir_stat contains the remap mode or an error code */
1951 try_to_enable_x2apic(ir_stat);
1952
1953 if (ir_stat < 0)
1954 restore_ioapic_entries();
1955 legacy_pic->restore_mask();
1956 local_irq_restore(flags);
1957 }
1958
1959 #ifdef CONFIG_X86_64
1960 /*
1961 * Detect and enable local APICs on non-SMP boards.
1962 * Original code written by Keir Fraser.
1963 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1964 * not correctly set up (usually the APIC timer won't work etc.)
1965 */
detect_init_APIC(void)1966 static bool __init detect_init_APIC(void)
1967 {
1968 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1969 pr_info("No local APIC present\n");
1970 return false;
1971 }
1972
1973 register_lapic_address(APIC_DEFAULT_PHYS_BASE);
1974 return true;
1975 }
1976 #else
1977
apic_verify(unsigned long addr)1978 static bool __init apic_verify(unsigned long addr)
1979 {
1980 u32 features, h, l;
1981
1982 /*
1983 * The APIC feature bit should now be enabled
1984 * in `cpuid'
1985 */
1986 features = cpuid_edx(1);
1987 if (!(features & (1 << X86_FEATURE_APIC))) {
1988 pr_warn("Could not enable APIC!\n");
1989 return false;
1990 }
1991 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1992
1993 /* The BIOS may have set up the APIC at some other address */
1994 if (boot_cpu_data.x86 >= 6) {
1995 rdmsr(MSR_IA32_APICBASE, l, h);
1996 if (l & MSR_IA32_APICBASE_ENABLE)
1997 addr = l & MSR_IA32_APICBASE_BASE;
1998 }
1999
2000 register_lapic_address(addr);
2001 pr_info("Found and enabled local APIC!\n");
2002 return true;
2003 }
2004
apic_force_enable(unsigned long addr)2005 bool __init apic_force_enable(unsigned long addr)
2006 {
2007 u32 h, l;
2008
2009 if (apic_is_disabled)
2010 return false;
2011
2012 /*
2013 * Some BIOSes disable the local APIC in the APIC_BASE
2014 * MSR. This can only be done in software for Intel P6 or later
2015 * and AMD K7 (Model > 1) or later.
2016 */
2017 if (boot_cpu_data.x86 >= 6) {
2018 rdmsr(MSR_IA32_APICBASE, l, h);
2019 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
2020 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
2021 l &= ~MSR_IA32_APICBASE_BASE;
2022 l |= MSR_IA32_APICBASE_ENABLE | addr;
2023 wrmsr(MSR_IA32_APICBASE, l, h);
2024 enabled_via_apicbase = 1;
2025 }
2026 }
2027 return apic_verify(addr);
2028 }
2029
2030 /*
2031 * Detect and initialize APIC
2032 */
detect_init_APIC(void)2033 static bool __init detect_init_APIC(void)
2034 {
2035 /* Disabled by kernel option? */
2036 if (apic_is_disabled)
2037 return false;
2038
2039 switch (boot_cpu_data.x86_vendor) {
2040 case X86_VENDOR_AMD:
2041 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2042 (boot_cpu_data.x86 >= 15))
2043 break;
2044 goto no_apic;
2045 case X86_VENDOR_HYGON:
2046 break;
2047 case X86_VENDOR_INTEL:
2048 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2049 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2050 break;
2051 goto no_apic;
2052 default:
2053 goto no_apic;
2054 }
2055
2056 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2057 /*
2058 * Over-ride BIOS and try to enable the local APIC only if
2059 * "lapic" specified.
2060 */
2061 if (!force_enable_local_apic) {
2062 pr_info("Local APIC disabled by BIOS -- "
2063 "you can enable it with \"lapic\"\n");
2064 return false;
2065 }
2066 if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2067 return false;
2068 } else {
2069 if (!apic_verify(APIC_DEFAULT_PHYS_BASE))
2070 return false;
2071 }
2072
2073 apic_pm_activate();
2074
2075 return true;
2076
2077 no_apic:
2078 pr_info("No local APIC present or hardware disabled\n");
2079 return false;
2080 }
2081 #endif
2082
2083 /**
2084 * init_apic_mappings - initialize APIC mappings
2085 */
init_apic_mappings(void)2086 void __init init_apic_mappings(void)
2087 {
2088 if (apic_validate_deadline_timer())
2089 pr_info("TSC deadline timer available\n");
2090
2091 if (x2apic_mode)
2092 return;
2093
2094 if (!smp_found_config) {
2095 if (!detect_init_APIC()) {
2096 pr_info("APIC: disable apic facility\n");
2097 apic_disable();
2098 }
2099 num_processors = 1;
2100 }
2101 }
2102
apic_set_fixmap(bool read_apic)2103 static __init void apic_set_fixmap(bool read_apic)
2104 {
2105 set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr);
2106 apic_mmio_base = APIC_BASE;
2107 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
2108 apic_mmio_base, mp_lapic_addr);
2109 if (read_apic)
2110 apic_read_boot_cpu_id(false);
2111 }
2112
register_lapic_address(unsigned long address)2113 void __init register_lapic_address(unsigned long address)
2114 {
2115 /* This should only happen once */
2116 WARN_ON_ONCE(mp_lapic_addr);
2117 mp_lapic_addr = address;
2118
2119 if (!x2apic_mode)
2120 apic_set_fixmap(true);
2121 }
2122
2123 /*
2124 * Local APIC interrupts
2125 */
2126
2127 /*
2128 * Common handling code for spurious_interrupt and spurious_vector entry
2129 * points below. No point in allowing the compiler to inline it twice.
2130 */
handle_spurious_interrupt(u8 vector)2131 static noinline void handle_spurious_interrupt(u8 vector)
2132 {
2133 u32 v;
2134
2135 trace_spurious_apic_entry(vector);
2136
2137 inc_irq_stat(irq_spurious_count);
2138
2139 /*
2140 * If this is a spurious interrupt then do not acknowledge
2141 */
2142 if (vector == SPURIOUS_APIC_VECTOR) {
2143 /* See SDM vol 3 */
2144 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2145 smp_processor_id());
2146 goto out;
2147 }
2148
2149 /*
2150 * If it is a vectored one, verify it's set in the ISR. If set,
2151 * acknowledge it.
2152 */
2153 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2154 if (v & (1 << (vector & 0x1f))) {
2155 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2156 vector, smp_processor_id());
2157 apic_eoi();
2158 } else {
2159 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2160 vector, smp_processor_id());
2161 }
2162 out:
2163 trace_spurious_apic_exit(vector);
2164 }
2165
2166 /**
2167 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2168 * @regs: Pointer to pt_regs on stack
2169 * @vector: The vector number
2170 *
2171 * This is invoked from ASM entry code to catch all interrupts which
2172 * trigger on an entry which is routed to the common_spurious idtentry
2173 * point.
2174 */
DEFINE_IDTENTRY_IRQ(spurious_interrupt)2175 DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2176 {
2177 handle_spurious_interrupt(vector);
2178 }
2179
DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)2180 DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2181 {
2182 handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2183 }
2184
2185 /*
2186 * This interrupt should never happen with our APIC/SMP architecture
2187 */
DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)2188 DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2189 {
2190 static const char * const error_interrupt_reason[] = {
2191 "Send CS error", /* APIC Error Bit 0 */
2192 "Receive CS error", /* APIC Error Bit 1 */
2193 "Send accept error", /* APIC Error Bit 2 */
2194 "Receive accept error", /* APIC Error Bit 3 */
2195 "Redirectable IPI", /* APIC Error Bit 4 */
2196 "Send illegal vector", /* APIC Error Bit 5 */
2197 "Received illegal vector", /* APIC Error Bit 6 */
2198 "Illegal register address", /* APIC Error Bit 7 */
2199 };
2200 u32 v, i = 0;
2201
2202 trace_error_apic_entry(ERROR_APIC_VECTOR);
2203
2204 /* First tickle the hardware, only then report what went on. -- REW */
2205 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2206 apic_write(APIC_ESR, 0);
2207 v = apic_read(APIC_ESR);
2208 apic_eoi();
2209 atomic_inc(&irq_err_count);
2210
2211 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2212 smp_processor_id(), v);
2213
2214 v &= 0xff;
2215 while (v) {
2216 if (v & 0x1)
2217 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2218 i++;
2219 v >>= 1;
2220 }
2221
2222 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2223
2224 trace_error_apic_exit(ERROR_APIC_VECTOR);
2225 }
2226
2227 /**
2228 * connect_bsp_APIC - attach the APIC to the interrupt system
2229 */
connect_bsp_APIC(void)2230 static void __init connect_bsp_APIC(void)
2231 {
2232 #ifdef CONFIG_X86_32
2233 if (pic_mode) {
2234 /*
2235 * Do not trust the local APIC being empty at bootup.
2236 */
2237 clear_local_APIC();
2238 /*
2239 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2240 * local APIC to INT and NMI lines.
2241 */
2242 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2243 "enabling APIC mode.\n");
2244 imcr_pic_to_apic();
2245 }
2246 #endif
2247 }
2248
2249 /**
2250 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2251 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2252 *
2253 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2254 * APIC is disabled.
2255 */
disconnect_bsp_APIC(int virt_wire_setup)2256 void disconnect_bsp_APIC(int virt_wire_setup)
2257 {
2258 unsigned int value;
2259
2260 #ifdef CONFIG_X86_32
2261 if (pic_mode) {
2262 /*
2263 * Put the board back into PIC mode (has an effect only on
2264 * certain older boards). Note that APIC interrupts, including
2265 * IPIs, won't work beyond this point! The only exception are
2266 * INIT IPIs.
2267 */
2268 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2269 "entering PIC mode.\n");
2270 imcr_apic_to_pic();
2271 return;
2272 }
2273 #endif
2274
2275 /* Go back to Virtual Wire compatibility mode */
2276
2277 /* For the spurious interrupt use vector F, and enable it */
2278 value = apic_read(APIC_SPIV);
2279 value &= ~APIC_VECTOR_MASK;
2280 value |= APIC_SPIV_APIC_ENABLED;
2281 value |= 0xf;
2282 apic_write(APIC_SPIV, value);
2283
2284 if (!virt_wire_setup) {
2285 /*
2286 * For LVT0 make it edge triggered, active high,
2287 * external and enabled
2288 */
2289 value = apic_read(APIC_LVT0);
2290 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2291 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2292 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2293 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2294 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2295 apic_write(APIC_LVT0, value);
2296 } else {
2297 /* Disable LVT0 */
2298 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2299 }
2300
2301 /*
2302 * For LVT1 make it edge triggered, active high,
2303 * nmi and enabled
2304 */
2305 value = apic_read(APIC_LVT1);
2306 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2307 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2308 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2309 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2310 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2311 apic_write(APIC_LVT1, value);
2312 }
2313
2314 /*
2315 * The number of allocated logical CPU IDs. Since logical CPU IDs are allocated
2316 * contiguously, it equals to current allocated max logical CPU ID plus 1.
2317 * All allocated CPU IDs should be in the [0, nr_logical_cpuids) range,
2318 * so the maximum of nr_logical_cpuids is nr_cpu_ids.
2319 *
2320 * NOTE: Reserve 0 for BSP.
2321 */
2322 static int nr_logical_cpuids = 1;
2323
2324 /*
2325 * Used to store mapping between logical CPU IDs and APIC IDs.
2326 */
2327 int cpuid_to_apicid[] = {
2328 [0 ... NR_CPUS - 1] = -1,
2329 };
2330
arch_match_cpu_phys_id(int cpu,u64 phys_id)2331 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
2332 {
2333 return phys_id == cpuid_to_apicid[cpu];
2334 }
2335
2336 #ifdef CONFIG_SMP
cpu_mark_primary_thread(unsigned int cpu,unsigned int apicid)2337 static void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid)
2338 {
2339 /* Isolate the SMT bit(s) in the APICID and check for 0 */
2340 u32 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
2341
2342 if (smp_num_siblings == 1 || !(apicid & mask))
2343 cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
2344 }
2345
2346 /*
2347 * Due to the utter mess of CPUID evaluation smp_num_siblings is not valid
2348 * during early boot. Initialize the primary thread mask before SMP
2349 * bringup.
2350 */
smp_init_primary_thread_mask(void)2351 static int __init smp_init_primary_thread_mask(void)
2352 {
2353 unsigned int cpu;
2354
2355 /*
2356 * XEN/PV provides either none or useless topology information.
2357 * Pretend that all vCPUs are primary threads.
2358 */
2359 if (xen_pv_domain()) {
2360 cpumask_copy(&__cpu_primary_thread_mask, cpu_possible_mask);
2361 return 0;
2362 }
2363
2364 for (cpu = 0; cpu < nr_logical_cpuids; cpu++)
2365 cpu_mark_primary_thread(cpu, cpuid_to_apicid[cpu]);
2366 return 0;
2367 }
2368 early_initcall(smp_init_primary_thread_mask);
2369 #else
cpu_mark_primary_thread(unsigned int cpu,unsigned int apicid)2370 static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
2371 #endif
2372
2373 /*
2374 * Should use this API to allocate logical CPU IDs to keep nr_logical_cpuids
2375 * and cpuid_to_apicid[] synchronized.
2376 */
allocate_logical_cpuid(int apicid)2377 static int allocate_logical_cpuid(int apicid)
2378 {
2379 int i;
2380
2381 /*
2382 * cpuid <-> apicid mapping is persistent, so when a cpu is up,
2383 * check if the kernel has allocated a cpuid for it.
2384 */
2385 for (i = 0; i < nr_logical_cpuids; i++) {
2386 if (cpuid_to_apicid[i] == apicid)
2387 return i;
2388 }
2389
2390 /* Allocate a new cpuid. */
2391 if (nr_logical_cpuids >= nr_cpu_ids) {
2392 WARN_ONCE(1, "APIC: NR_CPUS/possible_cpus limit of %u reached. "
2393 "Processor %d/0x%x and the rest are ignored.\n",
2394 nr_cpu_ids, nr_logical_cpuids, apicid);
2395 return -EINVAL;
2396 }
2397
2398 cpuid_to_apicid[nr_logical_cpuids] = apicid;
2399 return nr_logical_cpuids++;
2400 }
2401
cpu_update_apic(int cpu,int apicid)2402 static void cpu_update_apic(int cpu, int apicid)
2403 {
2404 #if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2405 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2406 #endif
2407 set_cpu_possible(cpu, true);
2408 physid_set(apicid, phys_cpu_present_map);
2409 set_cpu_present(cpu, true);
2410 num_processors++;
2411
2412 if (system_state != SYSTEM_BOOTING)
2413 cpu_mark_primary_thread(cpu, apicid);
2414 }
2415
cpu_set_boot_apic(void)2416 static __init void cpu_set_boot_apic(void)
2417 {
2418 cpuid_to_apicid[0] = boot_cpu_physical_apicid;
2419 cpu_update_apic(0, boot_cpu_physical_apicid);
2420 x86_32_probe_bigsmp_early();
2421 }
2422
generic_processor_info(int apicid)2423 int generic_processor_info(int apicid)
2424 {
2425 int cpu, max = nr_cpu_ids;
2426
2427 /* The boot CPU must be set before MADT/MPTABLE parsing happens */
2428 if (cpuid_to_apicid[0] == BAD_APICID)
2429 panic("Boot CPU APIC not registered yet\n");
2430
2431 if (apicid == boot_cpu_physical_apicid)
2432 return 0;
2433
2434 if (disabled_cpu_apicid == apicid) {
2435 int thiscpu = num_processors + disabled_cpus;
2436
2437 pr_warn("APIC: Disabling requested cpu. Processor %d/0x%x ignored.\n",
2438 thiscpu, apicid);
2439
2440 disabled_cpus++;
2441 return -ENODEV;
2442 }
2443
2444 if (num_processors >= nr_cpu_ids) {
2445 int thiscpu = max + disabled_cpus;
2446
2447 pr_warn("APIC: NR_CPUS/possible_cpus limit of %i reached. "
2448 "Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2449
2450 disabled_cpus++;
2451 return -EINVAL;
2452 }
2453
2454 cpu = allocate_logical_cpuid(apicid);
2455 if (cpu < 0) {
2456 disabled_cpus++;
2457 return -EINVAL;
2458 }
2459
2460 cpu_update_apic(cpu, apicid);
2461 return cpu;
2462 }
2463
2464
__irq_msi_compose_msg(struct irq_cfg * cfg,struct msi_msg * msg,bool dmar)2465 void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2466 bool dmar)
2467 {
2468 memset(msg, 0, sizeof(*msg));
2469
2470 msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2471 msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2472 msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2473
2474 msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2475 msg->arch_data.vector = cfg->vector;
2476
2477 msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2478 /*
2479 * Only the IOMMU itself can use the trick of putting destination
2480 * APIC ID into the high bits of the address. Anything else would
2481 * just be writing to memory if it tried that, and needs IR to
2482 * address APICs which can't be addressed in the normal 32-bit
2483 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2484 * some hypervisors allow the extended destination ID field in bits
2485 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2486 */
2487 if (dmar)
2488 msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2489 else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2490 msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2491 else
2492 WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2493 }
2494
x86_msi_msg_get_destid(struct msi_msg * msg,bool extid)2495 u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2496 {
2497 u32 dest = msg->arch_addr_lo.destid_0_7;
2498
2499 if (extid)
2500 dest |= msg->arch_addr_hi.destid_8_31 << 8;
2501 return dest;
2502 }
2503 EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2504
apic_bsp_up_setup(void)2505 static void __init apic_bsp_up_setup(void)
2506 {
2507 #ifdef CONFIG_X86_64
2508 apic_write(APIC_ID, apic->set_apic_id(boot_cpu_physical_apicid));
2509 #endif
2510 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
2511 }
2512
2513 /**
2514 * apic_bsp_setup - Setup function for local apic and io-apic
2515 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2516 */
apic_bsp_setup(bool upmode)2517 static void __init apic_bsp_setup(bool upmode)
2518 {
2519 connect_bsp_APIC();
2520 if (upmode)
2521 apic_bsp_up_setup();
2522 setup_local_APIC();
2523
2524 enable_IO_APIC();
2525 end_local_APIC_setup();
2526 irq_remap_enable_fault_handling();
2527 setup_IO_APIC();
2528 lapic_update_legacy_vectors();
2529 }
2530
2531 #ifdef CONFIG_UP_LATE_INIT
up_late_init(void)2532 void __init up_late_init(void)
2533 {
2534 if (apic_intr_mode == APIC_PIC)
2535 return;
2536
2537 /* Setup local timer */
2538 x86_init.timers.setup_percpu_clockev();
2539 }
2540 #endif
2541
2542 /*
2543 * Power management
2544 */
2545 #ifdef CONFIG_PM
2546
2547 static struct {
2548 /*
2549 * 'active' is true if the local APIC was enabled by us and
2550 * not the BIOS; this signifies that we are also responsible
2551 * for disabling it before entering apm/acpi suspend
2552 */
2553 int active;
2554 /* r/w apic fields */
2555 unsigned int apic_id;
2556 unsigned int apic_taskpri;
2557 unsigned int apic_ldr;
2558 unsigned int apic_dfr;
2559 unsigned int apic_spiv;
2560 unsigned int apic_lvtt;
2561 unsigned int apic_lvtpc;
2562 unsigned int apic_lvt0;
2563 unsigned int apic_lvt1;
2564 unsigned int apic_lvterr;
2565 unsigned int apic_tmict;
2566 unsigned int apic_tdcr;
2567 unsigned int apic_thmr;
2568 unsigned int apic_cmci;
2569 } apic_pm_state;
2570
lapic_suspend(void)2571 static int lapic_suspend(void)
2572 {
2573 unsigned long flags;
2574 int maxlvt;
2575
2576 if (!apic_pm_state.active)
2577 return 0;
2578
2579 maxlvt = lapic_get_maxlvt();
2580
2581 apic_pm_state.apic_id = apic_read(APIC_ID);
2582 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2583 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2584 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2585 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2586 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2587 if (maxlvt >= 4)
2588 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2589 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2590 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2591 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2592 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2593 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2594 #ifdef CONFIG_X86_THERMAL_VECTOR
2595 if (maxlvt >= 5)
2596 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2597 #endif
2598 #ifdef CONFIG_X86_MCE_INTEL
2599 if (maxlvt >= 6)
2600 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2601 #endif
2602
2603 local_irq_save(flags);
2604
2605 /*
2606 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2607 * entries on some implementations.
2608 */
2609 mask_ioapic_entries();
2610
2611 disable_local_APIC();
2612
2613 irq_remapping_disable();
2614
2615 local_irq_restore(flags);
2616 return 0;
2617 }
2618
lapic_resume(void)2619 static void lapic_resume(void)
2620 {
2621 unsigned int l, h;
2622 unsigned long flags;
2623 int maxlvt;
2624
2625 if (!apic_pm_state.active)
2626 return;
2627
2628 local_irq_save(flags);
2629
2630 /*
2631 * IO-APIC and PIC have their own resume routines.
2632 * We just mask them here to make sure the interrupt
2633 * subsystem is completely quiet while we enable x2apic
2634 * and interrupt-remapping.
2635 */
2636 mask_ioapic_entries();
2637 legacy_pic->mask_all();
2638
2639 if (x2apic_mode) {
2640 __x2apic_enable();
2641 } else {
2642 /*
2643 * Make sure the APICBASE points to the right address
2644 *
2645 * FIXME! This will be wrong if we ever support suspend on
2646 * SMP! We'll need to do this as part of the CPU restore!
2647 */
2648 if (boot_cpu_data.x86 >= 6) {
2649 rdmsr(MSR_IA32_APICBASE, l, h);
2650 l &= ~MSR_IA32_APICBASE_BASE;
2651 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2652 wrmsr(MSR_IA32_APICBASE, l, h);
2653 }
2654 }
2655
2656 maxlvt = lapic_get_maxlvt();
2657 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2658 apic_write(APIC_ID, apic_pm_state.apic_id);
2659 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2660 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2661 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2662 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2663 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2664 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2665 #ifdef CONFIG_X86_THERMAL_VECTOR
2666 if (maxlvt >= 5)
2667 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2668 #endif
2669 #ifdef CONFIG_X86_MCE_INTEL
2670 if (maxlvt >= 6)
2671 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2672 #endif
2673 if (maxlvt >= 4)
2674 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2675 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2676 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2677 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2678 apic_write(APIC_ESR, 0);
2679 apic_read(APIC_ESR);
2680 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2681 apic_write(APIC_ESR, 0);
2682 apic_read(APIC_ESR);
2683
2684 irq_remapping_reenable(x2apic_mode);
2685
2686 local_irq_restore(flags);
2687 }
2688
2689 /*
2690 * This device has no shutdown method - fully functioning local APICs
2691 * are needed on every CPU up until machine_halt/restart/poweroff.
2692 */
2693
2694 static struct syscore_ops lapic_syscore_ops = {
2695 .resume = lapic_resume,
2696 .suspend = lapic_suspend,
2697 };
2698
apic_pm_activate(void)2699 static void apic_pm_activate(void)
2700 {
2701 apic_pm_state.active = 1;
2702 }
2703
init_lapic_sysfs(void)2704 static int __init init_lapic_sysfs(void)
2705 {
2706 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2707 if (boot_cpu_has(X86_FEATURE_APIC))
2708 register_syscore_ops(&lapic_syscore_ops);
2709
2710 return 0;
2711 }
2712
2713 /* local apic needs to resume before other devices access its registers. */
2714 core_initcall(init_lapic_sysfs);
2715
2716 #else /* CONFIG_PM */
2717
apic_pm_activate(void)2718 static void apic_pm_activate(void) { }
2719
2720 #endif /* CONFIG_PM */
2721
2722 #ifdef CONFIG_X86_64
2723
2724 static int multi_checked;
2725 static int multi;
2726
set_multi(const struct dmi_system_id * d)2727 static int set_multi(const struct dmi_system_id *d)
2728 {
2729 if (multi)
2730 return 0;
2731 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2732 multi = 1;
2733 return 0;
2734 }
2735
2736 static const struct dmi_system_id multi_dmi_table[] = {
2737 {
2738 .callback = set_multi,
2739 .ident = "IBM System Summit2",
2740 .matches = {
2741 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2742 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2743 },
2744 },
2745 {}
2746 };
2747
dmi_check_multi(void)2748 static void dmi_check_multi(void)
2749 {
2750 if (multi_checked)
2751 return;
2752
2753 dmi_check_system(multi_dmi_table);
2754 multi_checked = 1;
2755 }
2756
2757 /*
2758 * apic_is_clustered_box() -- Check if we can expect good TSC
2759 *
2760 * Thus far, the major user of this is IBM's Summit2 series:
2761 * Clustered boxes may have unsynced TSC problems if they are
2762 * multi-chassis.
2763 * Use DMI to check them
2764 */
apic_is_clustered_box(void)2765 int apic_is_clustered_box(void)
2766 {
2767 dmi_check_multi();
2768 return multi;
2769 }
2770 #endif
2771
2772 /*
2773 * APIC command line parameters
2774 */
setup_disableapic(char * arg)2775 static int __init setup_disableapic(char *arg)
2776 {
2777 apic_is_disabled = true;
2778 setup_clear_cpu_cap(X86_FEATURE_APIC);
2779 return 0;
2780 }
2781 early_param("disableapic", setup_disableapic);
2782
2783 /* same as disableapic, for compatibility */
setup_nolapic(char * arg)2784 static int __init setup_nolapic(char *arg)
2785 {
2786 return setup_disableapic(arg);
2787 }
2788 early_param("nolapic", setup_nolapic);
2789
parse_lapic_timer_c2_ok(char * arg)2790 static int __init parse_lapic_timer_c2_ok(char *arg)
2791 {
2792 local_apic_timer_c2_ok = 1;
2793 return 0;
2794 }
2795 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2796
parse_disable_apic_timer(char * arg)2797 static int __init parse_disable_apic_timer(char *arg)
2798 {
2799 disable_apic_timer = 1;
2800 return 0;
2801 }
2802 early_param("noapictimer", parse_disable_apic_timer);
2803
parse_nolapic_timer(char * arg)2804 static int __init parse_nolapic_timer(char *arg)
2805 {
2806 disable_apic_timer = 1;
2807 return 0;
2808 }
2809 early_param("nolapic_timer", parse_nolapic_timer);
2810
apic_set_verbosity(char * arg)2811 static int __init apic_set_verbosity(char *arg)
2812 {
2813 if (!arg) {
2814 if (IS_ENABLED(CONFIG_X86_32))
2815 return -EINVAL;
2816
2817 ioapic_is_disabled = false;
2818 return 0;
2819 }
2820
2821 if (strcmp("debug", arg) == 0)
2822 apic_verbosity = APIC_DEBUG;
2823 else if (strcmp("verbose", arg) == 0)
2824 apic_verbosity = APIC_VERBOSE;
2825 #ifdef CONFIG_X86_64
2826 else {
2827 pr_warn("APIC Verbosity level %s not recognised"
2828 " use apic=verbose or apic=debug\n", arg);
2829 return -EINVAL;
2830 }
2831 #endif
2832
2833 return 0;
2834 }
2835 early_param("apic", apic_set_verbosity);
2836
lapic_insert_resource(void)2837 static int __init lapic_insert_resource(void)
2838 {
2839 if (!apic_mmio_base)
2840 return -1;
2841
2842 /* Put local APIC into the resource map. */
2843 lapic_resource.start = apic_mmio_base;
2844 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2845 insert_resource(&iomem_resource, &lapic_resource);
2846
2847 return 0;
2848 }
2849
2850 /*
2851 * need call insert after e820__reserve_resources()
2852 * that is using request_resource
2853 */
2854 late_initcall(lapic_insert_resource);
2855
apic_set_disabled_cpu_apicid(char * arg)2856 static int __init apic_set_disabled_cpu_apicid(char *arg)
2857 {
2858 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2859 return -EINVAL;
2860
2861 return 0;
2862 }
2863 early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);
2864
apic_set_extnmi(char * arg)2865 static int __init apic_set_extnmi(char *arg)
2866 {
2867 if (!arg)
2868 return -EINVAL;
2869
2870 if (!strncmp("all", arg, 3))
2871 apic_extnmi = APIC_EXTNMI_ALL;
2872 else if (!strncmp("none", arg, 4))
2873 apic_extnmi = APIC_EXTNMI_NONE;
2874 else if (!strncmp("bsp", arg, 3))
2875 apic_extnmi = APIC_EXTNMI_BSP;
2876 else {
2877 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2878 return -EINVAL;
2879 }
2880
2881 return 0;
2882 }
2883 early_param("apic_extnmi", apic_set_extnmi);
2884