Loading...
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 */
72u32 boot_cpu_physical_apicid __ro_after_init = BAD_APICID;
73EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
74
75u8 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 */
81static 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 */
86static bool virt_ext_dest_id __ro_after_init;
87
88/* For parallel bootup. */
89unsigned long apic_mmio_base __ro_after_init;
90
91static 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 */
98static 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 */
108static 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
114static 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 */
126static int force_enable_local_apic __initdata;
127
128/*
129 * APIC command line parameters
130 */
131static 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}
139early_param("lapic", parse_lapic);
140
141#ifdef CONFIG_X86_64
142static int apic_calibrate_pmtmr __initdata;
143static __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
152static unsigned long mp_lapic_addr __ro_after_init;
153bool apic_is_disabled __ro_after_init;
154/* Disable local APIC timer from the kernel commandline or via dmi quirk */
155static int disable_apic_timer __initdata;
156/* Local APIC timer works in C2 */
157int local_apic_timer_c2_ok __ro_after_init;
158EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
159
160/*
161 * Debug level, exported for io_apic.c
162 */
163int apic_verbosity __ro_after_init;
164
165int pic_mode __ro_after_init;
166
167/* Have we found an MP table */
168int smp_found_config __ro_after_init;
169
170static struct resource lapic_resource = {
171 .name = "Local APIC",
172 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
173};
174
175unsigned int lapic_timer_period = 0;
176
177static void apic_pm_activate(void);
178
179/*
180 * Get the LAPIC version
181 */
182static 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 */
190static 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 */
198static 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 */
216static void __init apic_disable(void)
217{
218 apic_install_driver(&apic_noop);
219}
220
221void 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
231u64 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 */
244int 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 */
274static 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
340static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
341
342static 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
349static 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
378int 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}
407EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
408
409/*
410 * Program the next event, relative to now
411 */
412static 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
419static 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
432static 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
444 /*
445 * Setting APIC_LVT_MASKED (above) should be enough to tell
446 * the hardware that this timer will never fire. But AMD
447 * erratum 411 and some Intel CPU behavior circa 2024 say
448 * otherwise. Time for belt and suspenders programming: mask
449 * the timer _and_ zero the counter registers:
450 */
451 if (v & APIC_LVT_TIMER_TSCDEADLINE)
452 wrmsrl(MSR_IA32_TSC_DEADLINE, 0);
453 else
454 apic_write(APIC_TMICT, 0);
455
456 return 0;
457}
458
459static inline int
460lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot)
461{
462 /* Lapic used as dummy for broadcast ? */
463 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
464 return 0;
465
466 __setup_APIC_LVTT(lapic_timer_period, oneshot, 1);
467 return 0;
468}
469
470static int lapic_timer_set_periodic(struct clock_event_device *evt)
471{
472 return lapic_timer_set_periodic_oneshot(evt, false);
473}
474
475static int lapic_timer_set_oneshot(struct clock_event_device *evt)
476{
477 return lapic_timer_set_periodic_oneshot(evt, true);
478}
479
480/*
481 * Local APIC timer broadcast function
482 */
483static void lapic_timer_broadcast(const struct cpumask *mask)
484{
485#ifdef CONFIG_SMP
486 __apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
487#endif
488}
489
490
491/*
492 * The local apic timer can be used for any function which is CPU local.
493 */
494static struct clock_event_device lapic_clockevent = {
495 .name = "lapic",
496 .features = CLOCK_EVT_FEAT_PERIODIC |
497 CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP
498 | CLOCK_EVT_FEAT_DUMMY,
499 .shift = 32,
500 .set_state_shutdown = lapic_timer_shutdown,
501 .set_state_periodic = lapic_timer_set_periodic,
502 .set_state_oneshot = lapic_timer_set_oneshot,
503 .set_state_oneshot_stopped = lapic_timer_shutdown,
504 .set_next_event = lapic_next_event,
505 .broadcast = lapic_timer_broadcast,
506 .rating = 100,
507 .irq = -1,
508};
509static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
510
511static const struct x86_cpu_id deadline_match[] __initconst = {
512 X86_MATCH_VFM_STEPPINGS(INTEL_HASWELL_X, X86_STEPPINGS(0x2, 0x2), 0x3a), /* EP */
513 X86_MATCH_VFM_STEPPINGS(INTEL_HASWELL_X, X86_STEPPINGS(0x4, 0x4), 0x0f), /* EX */
514
515 X86_MATCH_VFM(INTEL_BROADWELL_X, 0x0b000020),
516
517 X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x2, 0x2), 0x00000011),
518 X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x3, 0x3), 0x0700000e),
519 X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x4, 0x4), 0x0f00000c),
520 X86_MATCH_VFM_STEPPINGS(INTEL_BROADWELL_D, X86_STEPPINGS(0x5, 0x5), 0x0e000003),
521
522 X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x3, 0x3), 0x01000136),
523 X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x4, 0x4), 0x02000014),
524 X86_MATCH_VFM_STEPPINGS(INTEL_SKYLAKE_X, X86_STEPPINGS(0x5, 0xf), 0),
525
526 X86_MATCH_VFM(INTEL_HASWELL, 0x22),
527 X86_MATCH_VFM(INTEL_HASWELL_L, 0x20),
528 X86_MATCH_VFM(INTEL_HASWELL_G, 0x17),
529
530 X86_MATCH_VFM(INTEL_BROADWELL, 0x25),
531 X86_MATCH_VFM(INTEL_BROADWELL_G, 0x17),
532
533 X86_MATCH_VFM(INTEL_SKYLAKE_L, 0xb2),
534 X86_MATCH_VFM(INTEL_SKYLAKE, 0xb2),
535
536 X86_MATCH_VFM(INTEL_KABYLAKE_L, 0x52),
537 X86_MATCH_VFM(INTEL_KABYLAKE, 0x52),
538
539 {},
540};
541
542static __init bool apic_validate_deadline_timer(void)
543{
544 const struct x86_cpu_id *m;
545 u32 rev;
546
547 if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
548 return false;
549 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
550 return true;
551
552 m = x86_match_cpu(deadline_match);
553 if (!m)
554 return true;
555
556 rev = (u32)m->driver_data;
557
558 if (boot_cpu_data.microcode >= rev)
559 return true;
560
561 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
562 pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; "
563 "please update microcode to version: 0x%x (or later)\n", rev);
564 return false;
565}
566
567/*
568 * Setup the local APIC timer for this CPU. Copy the initialized values
569 * of the boot CPU and register the clock event in the framework.
570 */
571static void setup_APIC_timer(void)
572{
573 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
574
575 if (this_cpu_has(X86_FEATURE_ARAT)) {
576 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
577 /* Make LAPIC timer preferable over percpu HPET */
578 lapic_clockevent.rating = 150;
579 }
580
581 memcpy(levt, &lapic_clockevent, sizeof(*levt));
582 levt->cpumask = cpumask_of(smp_processor_id());
583
584 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
585 levt->name = "lapic-deadline";
586 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
587 CLOCK_EVT_FEAT_DUMMY);
588 levt->set_next_event = lapic_next_deadline;
589 clockevents_config_and_register(levt,
590 tsc_khz * (1000 / TSC_DIVISOR),
591 0xF, ~0UL);
592 } else
593 clockevents_register_device(levt);
594}
595
596/*
597 * Install the updated TSC frequency from recalibration at the TSC
598 * deadline clockevent devices.
599 */
600static void __lapic_update_tsc_freq(void *info)
601{
602 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
603
604 if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
605 return;
606
607 clockevents_update_freq(levt, tsc_khz * (1000 / TSC_DIVISOR));
608}
609
610void lapic_update_tsc_freq(void)
611{
612 /*
613 * The clockevent device's ->mult and ->shift can both be
614 * changed. In order to avoid races, schedule the frequency
615 * update code on each CPU.
616 */
617 on_each_cpu(__lapic_update_tsc_freq, NULL, 0);
618}
619
620/*
621 * In this functions we calibrate APIC bus clocks to the external timer.
622 *
623 * We want to do the calibration only once since we want to have local timer
624 * irqs synchronous. CPUs connected by the same APIC bus have the very same bus
625 * frequency.
626 *
627 * This was previously done by reading the PIT/HPET and waiting for a wrap
628 * around to find out, that a tick has elapsed. I have a box, where the PIT
629 * readout is broken, so it never gets out of the wait loop again. This was
630 * also reported by others.
631 *
632 * Monitoring the jiffies value is inaccurate and the clockevents
633 * infrastructure allows us to do a simple substitution of the interrupt
634 * handler.
635 *
636 * The calibration routine also uses the pm_timer when possible, as the PIT
637 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
638 * back to normal later in the boot process).
639 */
640
641#define LAPIC_CAL_LOOPS (HZ/10)
642
643static __initdata int lapic_cal_loops = -1;
644static __initdata long lapic_cal_t1, lapic_cal_t2;
645static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
646static __initdata u32 lapic_cal_pm1, lapic_cal_pm2;
647static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
648
649/*
650 * Temporary interrupt handler and polled calibration function.
651 */
652static void __init lapic_cal_handler(struct clock_event_device *dev)
653{
654 unsigned long long tsc = 0;
655 long tapic = apic_read(APIC_TMCCT);
656 u32 pm = acpi_pm_read_early();
657
658 if (boot_cpu_has(X86_FEATURE_TSC))
659 tsc = rdtsc();
660
661 switch (lapic_cal_loops++) {
662 case 0:
663 lapic_cal_t1 = tapic;
664 lapic_cal_tsc1 = tsc;
665 lapic_cal_pm1 = pm;
666 lapic_cal_j1 = jiffies;
667 break;
668
669 case LAPIC_CAL_LOOPS:
670 lapic_cal_t2 = tapic;
671 lapic_cal_tsc2 = tsc;
672 if (pm < lapic_cal_pm1)
673 pm += ACPI_PM_OVRRUN;
674 lapic_cal_pm2 = pm;
675 lapic_cal_j2 = jiffies;
676 break;
677 }
678}
679
680static int __init
681calibrate_by_pmtimer(u32 deltapm, long *delta, long *deltatsc)
682{
683 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
684 const long pm_thresh = pm_100ms / 100;
685 unsigned long mult;
686 u64 res;
687
688#ifndef CONFIG_X86_PM_TIMER
689 return -1;
690#endif
691
692 apic_pr_verbose("... PM-Timer delta = %u\n", deltapm);
693
694 /* Check, if the PM timer is available */
695 if (!deltapm)
696 return -1;
697
698 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
699
700 if (deltapm > (pm_100ms - pm_thresh) &&
701 deltapm < (pm_100ms + pm_thresh)) {
702 apic_pr_verbose("... PM-Timer result ok\n");
703 return 0;
704 }
705
706 res = (((u64)deltapm) * mult) >> 22;
707 do_div(res, 1000000);
708 pr_warn("APIC calibration not consistent with PM-Timer: %ldms instead of 100ms\n",
709 (long)res);
710
711 /* Correct the lapic counter value */
712 res = (((u64)(*delta)) * pm_100ms);
713 do_div(res, deltapm);
714 pr_info("APIC delta adjusted to PM-Timer: "
715 "%lu (%ld)\n", (unsigned long)res, *delta);
716 *delta = (long)res;
717
718 /* Correct the tsc counter value */
719 if (boot_cpu_has(X86_FEATURE_TSC)) {
720 res = (((u64)(*deltatsc)) * pm_100ms);
721 do_div(res, deltapm);
722 apic_pr_verbose("TSC delta adjusted to PM-Timer: %lu (%ld)\n",
723 (unsigned long)res, *deltatsc);
724 *deltatsc = (long)res;
725 }
726
727 return 0;
728}
729
730static int __init lapic_init_clockevent(void)
731{
732 if (!lapic_timer_period)
733 return -1;
734
735 /* Calculate the scaled math multiplication factor */
736 lapic_clockevent.mult = div_sc(lapic_timer_period/APIC_DIVISOR,
737 TICK_NSEC, lapic_clockevent.shift);
738 lapic_clockevent.max_delta_ns =
739 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
740 lapic_clockevent.max_delta_ticks = 0x7FFFFFFF;
741 lapic_clockevent.min_delta_ns =
742 clockevent_delta2ns(0xF, &lapic_clockevent);
743 lapic_clockevent.min_delta_ticks = 0xF;
744
745 return 0;
746}
747
748bool __init apic_needs_pit(void)
749{
750 /*
751 * If the frequencies are not known, PIT is required for both TSC
752 * and apic timer calibration.
753 */
754 if (!tsc_khz || !cpu_khz)
755 return true;
756
757 /* Is there an APIC at all or is it disabled? */
758 if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled)
759 return true;
760
761 /*
762 * If interrupt delivery mode is legacy PIC or virtual wire without
763 * configuration, the local APIC timer won't be set up. Make sure
764 * that the PIT is initialized.
765 */
766 if (apic_intr_mode == APIC_PIC ||
767 apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG)
768 return true;
769
770 /* Virt guests may lack ARAT, but still have DEADLINE */
771 if (!boot_cpu_has(X86_FEATURE_ARAT))
772 return true;
773
774 /* Deadline timer is based on TSC so no further PIT action required */
775 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
776 return false;
777
778 /* APIC timer disabled? */
779 if (disable_apic_timer)
780 return true;
781 /*
782 * The APIC timer frequency is known already, no PIT calibration
783 * required. If unknown, let the PIT be initialized.
784 */
785 return lapic_timer_period == 0;
786}
787
788static int __init calibrate_APIC_clock(void)
789{
790 struct clock_event_device *levt = this_cpu_ptr(&lapic_events);
791 u64 tsc_perj = 0, tsc_start = 0;
792 unsigned long jif_start;
793 unsigned long deltaj;
794 long delta, deltatsc;
795 int pm_referenced = 0;
796
797 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
798 return 0;
799
800 /*
801 * Check if lapic timer has already been calibrated by platform
802 * specific routine, such as tsc calibration code. If so just fill
803 * in the clockevent structure and return.
804 */
805 if (!lapic_init_clockevent()) {
806 apic_pr_verbose("lapic timer already calibrated %d\n", lapic_timer_period);
807 /*
808 * Direct calibration methods must have an always running
809 * local APIC timer, no need for broadcast timer.
810 */
811 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
812 return 0;
813 }
814
815 apic_pr_verbose("Using local APIC timer interrupts. Calibrating APIC timer ...\n");
816
817 /*
818 * There are platforms w/o global clockevent devices. Instead of
819 * making the calibration conditional on that, use a polling based
820 * approach everywhere.
821 */
822 local_irq_disable();
823
824 /*
825 * Setup the APIC counter to maximum. There is no way the lapic
826 * can underflow in the 100ms detection time frame
827 */
828 __setup_APIC_LVTT(0xffffffff, 0, 0);
829
830 /*
831 * Methods to terminate the calibration loop:
832 * 1) Global clockevent if available (jiffies)
833 * 2) TSC if available and frequency is known
834 */
835 jif_start = READ_ONCE(jiffies);
836
837 if (tsc_khz) {
838 tsc_start = rdtsc();
839 tsc_perj = div_u64((u64)tsc_khz * 1000, HZ);
840 }
841
842 /*
843 * Enable interrupts so the tick can fire, if a global
844 * clockevent device is available
845 */
846 local_irq_enable();
847
848 while (lapic_cal_loops <= LAPIC_CAL_LOOPS) {
849 /* Wait for a tick to elapse */
850 while (1) {
851 if (tsc_khz) {
852 u64 tsc_now = rdtsc();
853 if ((tsc_now - tsc_start) >= tsc_perj) {
854 tsc_start += tsc_perj;
855 break;
856 }
857 } else {
858 unsigned long jif_now = READ_ONCE(jiffies);
859
860 if (time_after(jif_now, jif_start)) {
861 jif_start = jif_now;
862 break;
863 }
864 }
865 cpu_relax();
866 }
867
868 /* Invoke the calibration routine */
869 local_irq_disable();
870 lapic_cal_handler(NULL);
871 local_irq_enable();
872 }
873
874 local_irq_disable();
875
876 /* Build delta t1-t2 as apic timer counts down */
877 delta = lapic_cal_t1 - lapic_cal_t2;
878 apic_pr_verbose("... lapic delta = %ld\n", delta);
879
880 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
881
882 /* we trust the PM based calibration if possible */
883 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
884 &delta, &deltatsc);
885
886 lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
887 lapic_init_clockevent();
888
889 apic_pr_verbose("..... delta %ld\n", delta);
890 apic_pr_verbose("..... mult: %u\n", lapic_clockevent.mult);
891 apic_pr_verbose("..... calibration result: %u\n", lapic_timer_period);
892
893 if (boot_cpu_has(X86_FEATURE_TSC)) {
894 apic_pr_verbose("..... CPU clock speed is %ld.%04ld MHz.\n",
895 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
896 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
897 }
898
899 apic_pr_verbose("..... host bus clock speed is %u.%04u MHz.\n",
900 lapic_timer_period / (1000000 / HZ),
901 lapic_timer_period % (1000000 / HZ));
902
903 /*
904 * Do a sanity check on the APIC calibration result
905 */
906 if (lapic_timer_period < (1000000 / HZ)) {
907 local_irq_enable();
908 pr_warn("APIC frequency too slow, disabling apic timer\n");
909 return -1;
910 }
911
912 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
913
914 /*
915 * PM timer calibration failed or not turned on so lets try APIC
916 * timer based calibration, if a global clockevent device is
917 * available.
918 */
919 if (!pm_referenced && global_clock_event) {
920 apic_pr_verbose("... verify APIC timer\n");
921
922 /*
923 * Setup the apic timer manually
924 */
925 levt->event_handler = lapic_cal_handler;
926 lapic_timer_set_periodic(levt);
927 lapic_cal_loops = -1;
928
929 /* Let the interrupts run */
930 local_irq_enable();
931
932 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
933 cpu_relax();
934
935 /* Stop the lapic timer */
936 local_irq_disable();
937 lapic_timer_shutdown(levt);
938
939 /* Jiffies delta */
940 deltaj = lapic_cal_j2 - lapic_cal_j1;
941 apic_pr_verbose("... jiffies delta = %lu\n", deltaj);
942
943 /* Check, if the jiffies result is consistent */
944 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
945 apic_pr_verbose("... jiffies result ok\n");
946 else
947 levt->features |= CLOCK_EVT_FEAT_DUMMY;
948 }
949 local_irq_enable();
950
951 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
952 pr_warn("APIC timer disabled due to verification failure\n");
953 return -1;
954 }
955
956 return 0;
957}
958
959/*
960 * Setup the boot APIC
961 *
962 * Calibrate and verify the result.
963 */
964void __init setup_boot_APIC_clock(void)
965{
966 /*
967 * The local apic timer can be disabled via the kernel
968 * commandline or from the CPU detection code. Register the lapic
969 * timer as a dummy clock event source on SMP systems, so the
970 * broadcast mechanism is used. On UP systems simply ignore it.
971 */
972 if (disable_apic_timer) {
973 pr_info("Disabling APIC timer\n");
974 /* No broadcast on UP ! */
975 if (num_possible_cpus() > 1) {
976 lapic_clockevent.mult = 1;
977 setup_APIC_timer();
978 }
979 return;
980 }
981
982 if (calibrate_APIC_clock()) {
983 /* No broadcast on UP ! */
984 if (num_possible_cpus() > 1)
985 setup_APIC_timer();
986 return;
987 }
988
989 /*
990 * If nmi_watchdog is set to IO_APIC, we need the
991 * PIT/HPET going. Otherwise register lapic as a dummy
992 * device.
993 */
994 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
995
996 /* Setup the lapic or request the broadcast */
997 setup_APIC_timer();
998 amd_e400_c1e_apic_setup();
999}
1000
1001void setup_secondary_APIC_clock(void)
1002{
1003 setup_APIC_timer();
1004 amd_e400_c1e_apic_setup();
1005}
1006
1007/*
1008 * The guts of the apic timer interrupt
1009 */
1010static void local_apic_timer_interrupt(void)
1011{
1012 struct clock_event_device *evt = this_cpu_ptr(&lapic_events);
1013
1014 /*
1015 * Normally we should not be here till LAPIC has been initialized but
1016 * in some cases like kdump, its possible that there is a pending LAPIC
1017 * timer interrupt from previous kernel's context and is delivered in
1018 * new kernel the moment interrupts are enabled.
1019 *
1020 * Interrupts are enabled early and LAPIC is setup much later, hence
1021 * its possible that when we get here evt->event_handler is NULL.
1022 * Check for event_handler being NULL and discard the interrupt as
1023 * spurious.
1024 */
1025 if (!evt->event_handler) {
1026 pr_warn("Spurious LAPIC timer interrupt on cpu %d\n",
1027 smp_processor_id());
1028 /* Switch it off */
1029 lapic_timer_shutdown(evt);
1030 return;
1031 }
1032
1033 /*
1034 * the NMI deadlock-detector uses this.
1035 */
1036 inc_irq_stat(apic_timer_irqs);
1037
1038 evt->event_handler(evt);
1039}
1040
1041/*
1042 * Local APIC timer interrupt. This is the most natural way for doing
1043 * local interrupts, but local timer interrupts can be emulated by
1044 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1045 *
1046 * [ if a single-CPU system runs an SMP kernel then we call the local
1047 * interrupt as well. Thus we cannot inline the local irq ... ]
1048 */
1049DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt)
1050{
1051 struct pt_regs *old_regs = set_irq_regs(regs);
1052
1053 apic_eoi();
1054 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
1055 local_apic_timer_interrupt();
1056 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
1057
1058 set_irq_regs(old_regs);
1059}
1060
1061/*
1062 * Local APIC start and shutdown
1063 */
1064
1065/**
1066 * clear_local_APIC - shutdown the local APIC
1067 *
1068 * This is called, when a CPU is disabled and before rebooting, so the state of
1069 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
1070 * leftovers during boot.
1071 */
1072void clear_local_APIC(void)
1073{
1074 int maxlvt;
1075 u32 v;
1076
1077 if (!apic_accessible())
1078 return;
1079
1080 maxlvt = lapic_get_maxlvt();
1081 /*
1082 * Masking an LVT entry can trigger a local APIC error
1083 * if the vector is zero. Mask LVTERR first to prevent this.
1084 */
1085 if (maxlvt >= 3) {
1086 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1087 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1088 }
1089 /*
1090 * Careful: we have to set masks only first to deassert
1091 * any level-triggered sources.
1092 */
1093 v = apic_read(APIC_LVTT);
1094 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1095 v = apic_read(APIC_LVT0);
1096 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1097 v = apic_read(APIC_LVT1);
1098 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1099 if (maxlvt >= 4) {
1100 v = apic_read(APIC_LVTPC);
1101 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1102 }
1103
1104 /* lets not touch this if we didn't frob it */
1105#ifdef CONFIG_X86_THERMAL_VECTOR
1106 if (maxlvt >= 5) {
1107 v = apic_read(APIC_LVTTHMR);
1108 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1109 }
1110#endif
1111#ifdef CONFIG_X86_MCE_INTEL
1112 if (maxlvt >= 6) {
1113 v = apic_read(APIC_LVTCMCI);
1114 if (!(v & APIC_LVT_MASKED))
1115 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1116 }
1117#endif
1118
1119 /*
1120 * Clean APIC state for other OSs:
1121 */
1122 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1123 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1124 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1125 if (maxlvt >= 3)
1126 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1127 if (maxlvt >= 4)
1128 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1129
1130 /* Integrated APIC (!82489DX) ? */
1131 if (lapic_is_integrated()) {
1132 if (maxlvt > 3)
1133 /* Clear ESR due to Pentium errata 3AP and 11AP */
1134 apic_write(APIC_ESR, 0);
1135 apic_read(APIC_ESR);
1136 }
1137}
1138
1139/**
1140 * apic_soft_disable - Clears and software disables the local APIC on hotplug
1141 *
1142 * Contrary to disable_local_APIC() this does not touch the enable bit in
1143 * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC
1144 * bus would require a hardware reset as the APIC would lose track of bus
1145 * arbitration. On systems with FSB delivery APICBASE could be disabled,
1146 * but it has to be guaranteed that no interrupt is sent to the APIC while
1147 * in that state and it's not clear from the SDM whether it still responds
1148 * to INIT/SIPI messages. Stay on the safe side and use software disable.
1149 */
1150void apic_soft_disable(void)
1151{
1152 u32 value;
1153
1154 clear_local_APIC();
1155
1156 /* Soft disable APIC (implies clearing of registers for 82489DX!). */
1157 value = apic_read(APIC_SPIV);
1158 value &= ~APIC_SPIV_APIC_ENABLED;
1159 apic_write(APIC_SPIV, value);
1160}
1161
1162/**
1163 * disable_local_APIC - clear and disable the local APIC
1164 */
1165void disable_local_APIC(void)
1166{
1167 if (!apic_accessible())
1168 return;
1169
1170 apic_soft_disable();
1171
1172#ifdef CONFIG_X86_32
1173 /*
1174 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1175 * restore the disabled state.
1176 */
1177 if (enabled_via_apicbase) {
1178 unsigned int l, h;
1179
1180 rdmsr(MSR_IA32_APICBASE, l, h);
1181 l &= ~MSR_IA32_APICBASE_ENABLE;
1182 wrmsr(MSR_IA32_APICBASE, l, h);
1183 }
1184#endif
1185}
1186
1187/*
1188 * If Linux enabled the LAPIC against the BIOS default disable it down before
1189 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1190 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1191 * for the case where Linux didn't enable the LAPIC.
1192 */
1193void lapic_shutdown(void)
1194{
1195 unsigned long flags;
1196
1197 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1198 return;
1199
1200 local_irq_save(flags);
1201
1202#ifdef CONFIG_X86_32
1203 if (!enabled_via_apicbase)
1204 clear_local_APIC();
1205 else
1206#endif
1207 disable_local_APIC();
1208
1209
1210 local_irq_restore(flags);
1211}
1212
1213/**
1214 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1215 */
1216void __init sync_Arb_IDs(void)
1217{
1218 /*
1219 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1220 * needed on AMD.
1221 */
1222 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1223 return;
1224
1225 /*
1226 * Wait for idle.
1227 */
1228 apic_wait_icr_idle();
1229
1230 apic_pr_debug("Synchronizing Arb IDs.\n");
1231 apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG | APIC_DM_INIT);
1232}
1233
1234enum apic_intr_mode_id apic_intr_mode __ro_after_init;
1235
1236static int __init __apic_intr_mode_select(void)
1237{
1238 /* Check kernel option */
1239 if (apic_is_disabled) {
1240 pr_info("APIC disabled via kernel command line\n");
1241 return APIC_PIC;
1242 }
1243
1244 /* Check BIOS */
1245#ifdef CONFIG_X86_64
1246 /* On 64-bit, the APIC must be integrated, Check local APIC only */
1247 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1248 apic_is_disabled = true;
1249 pr_info("APIC disabled by BIOS\n");
1250 return APIC_PIC;
1251 }
1252#else
1253 /* On 32-bit, the APIC may be integrated APIC or 82489DX */
1254
1255 /* Neither 82489DX nor integrated APIC ? */
1256 if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) {
1257 apic_is_disabled = true;
1258 return APIC_PIC;
1259 }
1260
1261 /* If the BIOS pretends there is an integrated APIC ? */
1262 if (!boot_cpu_has(X86_FEATURE_APIC) &&
1263 APIC_INTEGRATED(boot_cpu_apic_version)) {
1264 apic_is_disabled = true;
1265 pr_err(FW_BUG "Local APIC not detected, force emulation\n");
1266 return APIC_PIC;
1267 }
1268#endif
1269
1270 /* Check MP table or ACPI MADT configuration */
1271 if (!smp_found_config) {
1272 disable_ioapic_support();
1273 if (!acpi_lapic) {
1274 pr_info("APIC: ACPI MADT or MP tables are not detected\n");
1275 return APIC_VIRTUAL_WIRE_NO_CONFIG;
1276 }
1277 return APIC_VIRTUAL_WIRE;
1278 }
1279
1280#ifdef CONFIG_SMP
1281 /* If SMP should be disabled, then really disable it! */
1282 if (!setup_max_cpus) {
1283 pr_info("APIC: SMP mode deactivated\n");
1284 return APIC_SYMMETRIC_IO_NO_ROUTING;
1285 }
1286#endif
1287
1288 return APIC_SYMMETRIC_IO;
1289}
1290
1291/* Select the interrupt delivery mode for the BSP */
1292void __init apic_intr_mode_select(void)
1293{
1294 apic_intr_mode = __apic_intr_mode_select();
1295}
1296
1297/*
1298 * An initial setup of the virtual wire mode.
1299 */
1300void __init init_bsp_APIC(void)
1301{
1302 unsigned int value;
1303
1304 /*
1305 * Don't do the setup now if we have a SMP BIOS as the
1306 * through-I/O-APIC virtual wire mode might be active.
1307 */
1308 if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC))
1309 return;
1310
1311 /*
1312 * Do not trust the local APIC being empty at bootup.
1313 */
1314 clear_local_APIC();
1315
1316 /*
1317 * Enable APIC.
1318 */
1319 value = apic_read(APIC_SPIV);
1320 value &= ~APIC_VECTOR_MASK;
1321 value |= APIC_SPIV_APIC_ENABLED;
1322
1323#ifdef CONFIG_X86_32
1324 /* This bit is reserved on P4/Xeon and should be cleared */
1325 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1326 (boot_cpu_data.x86 == 15))
1327 value &= ~APIC_SPIV_FOCUS_DISABLED;
1328 else
1329#endif
1330 value |= APIC_SPIV_FOCUS_DISABLED;
1331 value |= SPURIOUS_APIC_VECTOR;
1332 apic_write(APIC_SPIV, value);
1333
1334 /*
1335 * Set up the virtual wire mode.
1336 */
1337 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1338 value = APIC_DM_NMI;
1339 if (!lapic_is_integrated()) /* 82489DX */
1340 value |= APIC_LVT_LEVEL_TRIGGER;
1341 if (apic_extnmi == APIC_EXTNMI_NONE)
1342 value |= APIC_LVT_MASKED;
1343 apic_write(APIC_LVT1, value);
1344}
1345
1346static void __init apic_bsp_setup(bool upmode);
1347
1348/* Init the interrupt delivery mode for the BSP */
1349void __init apic_intr_mode_init(void)
1350{
1351 bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT);
1352
1353 switch (apic_intr_mode) {
1354 case APIC_PIC:
1355 pr_info("APIC: Keep in PIC mode(8259)\n");
1356 return;
1357 case APIC_VIRTUAL_WIRE:
1358 pr_info("APIC: Switch to virtual wire mode setup\n");
1359 break;
1360 case APIC_VIRTUAL_WIRE_NO_CONFIG:
1361 pr_info("APIC: Switch to virtual wire mode setup with no configuration\n");
1362 upmode = true;
1363 break;
1364 case APIC_SYMMETRIC_IO:
1365 pr_info("APIC: Switch to symmetric I/O mode setup\n");
1366 break;
1367 case APIC_SYMMETRIC_IO_NO_ROUTING:
1368 pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n");
1369 break;
1370 }
1371
1372 x86_64_probe_apic();
1373
1374 x86_32_install_bigsmp();
1375
1376 if (x86_platform.apic_post_init)
1377 x86_platform.apic_post_init();
1378
1379 apic_bsp_setup(upmode);
1380}
1381
1382static void lapic_setup_esr(void)
1383{
1384 unsigned int oldvalue, value, maxlvt;
1385
1386 if (!lapic_is_integrated()) {
1387 pr_info("No ESR for 82489DX.\n");
1388 return;
1389 }
1390
1391 if (apic->disable_esr) {
1392 /*
1393 * Something untraceable is creating bad interrupts on
1394 * secondary quads ... for the moment, just leave the
1395 * ESR disabled - we can't do anything useful with the
1396 * errors anyway - mbligh
1397 */
1398 pr_info("Leaving ESR disabled.\n");
1399 return;
1400 }
1401
1402 maxlvt = lapic_get_maxlvt();
1403 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1404 apic_write(APIC_ESR, 0);
1405 oldvalue = apic_read(APIC_ESR);
1406
1407 /* enables sending errors */
1408 value = ERROR_APIC_VECTOR;
1409 apic_write(APIC_LVTERR, value);
1410
1411 /*
1412 * spec says clear errors after enabling vector.
1413 */
1414 if (maxlvt > 3)
1415 apic_write(APIC_ESR, 0);
1416 value = apic_read(APIC_ESR);
1417 if (value != oldvalue) {
1418 apic_pr_verbose("ESR value before enabling vector: 0x%08x after: 0x%08x\n",
1419 oldvalue, value);
1420 }
1421}
1422
1423#define APIC_IR_REGS APIC_ISR_NR
1424#define APIC_IR_BITS (APIC_IR_REGS * 32)
1425#define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG)
1426
1427union apic_ir {
1428 unsigned long map[APIC_IR_MAPSIZE];
1429 u32 regs[APIC_IR_REGS];
1430};
1431
1432static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr)
1433{
1434 int i, bit;
1435
1436 /* Read the IRRs */
1437 for (i = 0; i < APIC_IR_REGS; i++)
1438 irr->regs[i] = apic_read(APIC_IRR + i * 0x10);
1439
1440 /* Read the ISRs */
1441 for (i = 0; i < APIC_IR_REGS; i++)
1442 isr->regs[i] = apic_read(APIC_ISR + i * 0x10);
1443
1444 /*
1445 * If the ISR map is not empty. ACK the APIC and run another round
1446 * to verify whether a pending IRR has been unblocked and turned
1447 * into a ISR.
1448 */
1449 if (!bitmap_empty(isr->map, APIC_IR_BITS)) {
1450 /*
1451 * There can be multiple ISR bits set when a high priority
1452 * interrupt preempted a lower priority one. Issue an ACK
1453 * per set bit.
1454 */
1455 for_each_set_bit(bit, isr->map, APIC_IR_BITS)
1456 apic_eoi();
1457 return true;
1458 }
1459
1460 return !bitmap_empty(irr->map, APIC_IR_BITS);
1461}
1462
1463/*
1464 * After a crash, we no longer service the interrupts and a pending
1465 * interrupt from previous kernel might still have ISR bit set.
1466 *
1467 * Most probably by now the CPU has serviced that pending interrupt and it
1468 * might not have done the apic_eoi() because it thought, interrupt
1469 * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear
1470 * the ISR bit and cpu thinks it has already serviced the interrupt. Hence
1471 * a vector might get locked. It was noticed for timer irq (vector
1472 * 0x31). Issue an extra EOI to clear ISR.
1473 *
1474 * If there are pending IRR bits they turn into ISR bits after a higher
1475 * priority ISR bit has been acked.
1476 */
1477static void apic_pending_intr_clear(void)
1478{
1479 union apic_ir irr, isr;
1480 unsigned int i;
1481
1482 /* 512 loops are way oversized and give the APIC a chance to obey. */
1483 for (i = 0; i < 512; i++) {
1484 if (!apic_check_and_ack(&irr, &isr))
1485 return;
1486 }
1487 /* Dump the IRR/ISR content if that failed */
1488 pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map);
1489}
1490
1491/**
1492 * setup_local_APIC - setup the local APIC
1493 *
1494 * Used to setup local APIC while initializing BSP or bringing up APs.
1495 * Always called with preemption disabled.
1496 */
1497static void setup_local_APIC(void)
1498{
1499 int cpu = smp_processor_id();
1500 unsigned int value;
1501
1502 if (apic_is_disabled) {
1503 disable_ioapic_support();
1504 return;
1505 }
1506
1507 /*
1508 * If this comes from kexec/kcrash the APIC might be enabled in
1509 * SPIV. Soft disable it before doing further initialization.
1510 */
1511 value = apic_read(APIC_SPIV);
1512 value &= ~APIC_SPIV_APIC_ENABLED;
1513 apic_write(APIC_SPIV, value);
1514
1515#ifdef CONFIG_X86_32
1516 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1517 if (lapic_is_integrated() && apic->disable_esr) {
1518 apic_write(APIC_ESR, 0);
1519 apic_write(APIC_ESR, 0);
1520 apic_write(APIC_ESR, 0);
1521 apic_write(APIC_ESR, 0);
1522 }
1523#endif
1524 /*
1525 * Intel recommends to set DFR, LDR and TPR before enabling
1526 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1527 * document number 292116).
1528 *
1529 * Except for APICs which operate in physical destination mode.
1530 */
1531 if (apic->init_apic_ldr)
1532 apic->init_apic_ldr();
1533
1534 /*
1535 * Set Task Priority to 'accept all except vectors 0-31'. An APIC
1536 * vector in the 16-31 range could be delivered if TPR == 0, but we
1537 * would think it's an exception and terrible things will happen. We
1538 * never change this later on.
1539 */
1540 value = apic_read(APIC_TASKPRI);
1541 value &= ~APIC_TPRI_MASK;
1542 value |= 0x10;
1543 apic_write(APIC_TASKPRI, value);
1544
1545 /* Clear eventually stale ISR/IRR bits */
1546 apic_pending_intr_clear();
1547
1548 /*
1549 * Now that we are all set up, enable the APIC
1550 */
1551 value = apic_read(APIC_SPIV);
1552 value &= ~APIC_VECTOR_MASK;
1553 /*
1554 * Enable APIC
1555 */
1556 value |= APIC_SPIV_APIC_ENABLED;
1557
1558#ifdef CONFIG_X86_32
1559 /*
1560 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1561 * certain networking cards. If high frequency interrupts are
1562 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1563 * entry is masked/unmasked at a high rate as well then sooner or
1564 * later IOAPIC line gets 'stuck', no more interrupts are received
1565 * from the device. If focus CPU is disabled then the hang goes
1566 * away, oh well :-(
1567 *
1568 * [ This bug can be reproduced easily with a level-triggered
1569 * PCI Ne2000 networking cards and PII/PIII processors, dual
1570 * BX chipset. ]
1571 */
1572 /*
1573 * Actually disabling the focus CPU check just makes the hang less
1574 * frequent as it makes the interrupt distribution model be more
1575 * like LRU than MRU (the short-term load is more even across CPUs).
1576 */
1577
1578 /*
1579 * - enable focus processor (bit==0)
1580 * - 64bit mode always use processor focus
1581 * so no need to set it
1582 */
1583 value &= ~APIC_SPIV_FOCUS_DISABLED;
1584#endif
1585
1586 /*
1587 * Set spurious IRQ vector
1588 */
1589 value |= SPURIOUS_APIC_VECTOR;
1590 apic_write(APIC_SPIV, value);
1591
1592 perf_events_lapic_init();
1593
1594 /*
1595 * Set up LVT0, LVT1:
1596 *
1597 * set up through-local-APIC on the boot CPU's LINT0. This is not
1598 * strictly necessary in pure symmetric-IO mode, but sometimes
1599 * we delegate interrupts to the 8259A.
1600 */
1601 /*
1602 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1603 */
1604 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1605 if (!cpu && (pic_mode || !value || ioapic_is_disabled)) {
1606 value = APIC_DM_EXTINT;
1607 apic_pr_verbose("Enabled ExtINT on CPU#%d\n", cpu);
1608 } else {
1609 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1610 apic_pr_verbose("Masked ExtINT on CPU#%d\n", cpu);
1611 }
1612 apic_write(APIC_LVT0, value);
1613
1614 /*
1615 * Only the BSP sees the LINT1 NMI signal by default. This can be
1616 * modified by apic_extnmi= boot option.
1617 */
1618 if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) ||
1619 apic_extnmi == APIC_EXTNMI_ALL)
1620 value = APIC_DM_NMI;
1621 else
1622 value = APIC_DM_NMI | APIC_LVT_MASKED;
1623
1624 /* Is 82489DX ? */
1625 if (!lapic_is_integrated())
1626 value |= APIC_LVT_LEVEL_TRIGGER;
1627 apic_write(APIC_LVT1, value);
1628
1629#ifdef CONFIG_X86_MCE_INTEL
1630 /* Recheck CMCI information after local APIC is up on CPU #0 */
1631 if (!cpu)
1632 cmci_recheck();
1633#endif
1634}
1635
1636static void end_local_APIC_setup(void)
1637{
1638 lapic_setup_esr();
1639
1640#ifdef CONFIG_X86_32
1641 {
1642 unsigned int value;
1643 /* Disable the local apic timer */
1644 value = apic_read(APIC_LVTT);
1645 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1646 apic_write(APIC_LVTT, value);
1647 }
1648#endif
1649
1650 apic_pm_activate();
1651}
1652
1653/*
1654 * APIC setup function for application processors. Called from smpboot.c
1655 */
1656void apic_ap_setup(void)
1657{
1658 setup_local_APIC();
1659 end_local_APIC_setup();
1660}
1661
1662static __init void apic_read_boot_cpu_id(bool x2apic)
1663{
1664 /*
1665 * This can be invoked from check_x2apic() before the APIC has been
1666 * selected. But that code knows for sure that the BIOS enabled
1667 * X2APIC.
1668 */
1669 if (x2apic) {
1670 boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID);
1671 boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR));
1672 } else {
1673 boot_cpu_physical_apicid = read_apic_id();
1674 boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR));
1675 }
1676 topology_register_boot_apic(boot_cpu_physical_apicid);
1677 x86_32_probe_bigsmp_early();
1678}
1679
1680#ifdef CONFIG_X86_X2APIC
1681int x2apic_mode;
1682EXPORT_SYMBOL_GPL(x2apic_mode);
1683
1684enum {
1685 X2APIC_OFF,
1686 X2APIC_DISABLED,
1687 /* All states below here have X2APIC enabled */
1688 X2APIC_ON,
1689 X2APIC_ON_LOCKED
1690};
1691static int x2apic_state;
1692
1693static bool x2apic_hw_locked(void)
1694{
1695 u64 x86_arch_cap_msr;
1696 u64 msr;
1697
1698 x86_arch_cap_msr = x86_read_arch_cap_msr();
1699 if (x86_arch_cap_msr & ARCH_CAP_XAPIC_DISABLE) {
1700 rdmsrl(MSR_IA32_XAPIC_DISABLE_STATUS, msr);
1701 return (msr & LEGACY_XAPIC_DISABLED);
1702 }
1703 return false;
1704}
1705
1706static void __x2apic_disable(void)
1707{
1708 u64 msr;
1709
1710 if (!boot_cpu_has(X86_FEATURE_APIC))
1711 return;
1712
1713 rdmsrl(MSR_IA32_APICBASE, msr);
1714 if (!(msr & X2APIC_ENABLE))
1715 return;
1716 /* Disable xapic and x2apic first and then reenable xapic mode */
1717 wrmsrl(MSR_IA32_APICBASE, msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1718 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1719 printk_once(KERN_INFO "x2apic disabled\n");
1720}
1721
1722static void __x2apic_enable(void)
1723{
1724 u64 msr;
1725
1726 rdmsrl(MSR_IA32_APICBASE, msr);
1727 if (msr & X2APIC_ENABLE)
1728 return;
1729 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1730 printk_once(KERN_INFO "x2apic enabled\n");
1731}
1732
1733static int __init setup_nox2apic(char *str)
1734{
1735 if (x2apic_enabled()) {
1736 u32 apicid = native_apic_msr_read(APIC_ID);
1737
1738 if (apicid >= 255) {
1739 pr_warn("Apicid: %08x, cannot enforce nox2apic\n",
1740 apicid);
1741 return 0;
1742 }
1743 if (x2apic_hw_locked()) {
1744 pr_warn("APIC locked in x2apic mode, can't disable\n");
1745 return 0;
1746 }
1747 pr_warn("x2apic already enabled.\n");
1748 __x2apic_disable();
1749 }
1750 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1751 x2apic_state = X2APIC_DISABLED;
1752 x2apic_mode = 0;
1753 return 0;
1754}
1755early_param("nox2apic", setup_nox2apic);
1756
1757/* Called from cpu_init() to enable x2apic on (secondary) cpus */
1758void x2apic_setup(void)
1759{
1760 /*
1761 * Try to make the AP's APIC state match that of the BSP, but if the
1762 * BSP is unlocked and the AP is locked then there is a state mismatch.
1763 * Warn about the mismatch in case a GP fault occurs due to a locked AP
1764 * trying to be turned off.
1765 */
1766 if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked())
1767 pr_warn("x2apic lock mismatch between BSP and AP.\n");
1768 /*
1769 * If x2apic is not in ON or LOCKED state, disable it if already enabled
1770 * from BIOS.
1771 */
1772 if (x2apic_state < X2APIC_ON) {
1773 __x2apic_disable();
1774 return;
1775 }
1776 __x2apic_enable();
1777}
1778
1779static __init void apic_set_fixmap(bool read_apic);
1780
1781static __init void x2apic_disable(void)
1782{
1783 u32 x2apic_id;
1784
1785 if (x2apic_state < X2APIC_ON)
1786 return;
1787
1788 x2apic_id = read_apic_id();
1789 if (x2apic_id >= 255)
1790 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1791
1792 if (x2apic_hw_locked()) {
1793 pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id);
1794 return;
1795 }
1796
1797 __x2apic_disable();
1798
1799 x2apic_mode = 0;
1800 x2apic_state = X2APIC_DISABLED;
1801
1802 /*
1803 * Don't reread the APIC ID as it was already done from
1804 * check_x2apic() and the APIC driver still is a x2APIC variant,
1805 * which fails to do the read after x2APIC was disabled.
1806 */
1807 apic_set_fixmap(false);
1808}
1809
1810static __init void x2apic_enable(void)
1811{
1812 if (x2apic_state != X2APIC_OFF)
1813 return;
1814
1815 x2apic_mode = 1;
1816 x2apic_state = X2APIC_ON;
1817 __x2apic_enable();
1818}
1819
1820static __init void try_to_enable_x2apic(int remap_mode)
1821{
1822 if (x2apic_state == X2APIC_DISABLED)
1823 return;
1824
1825 if (remap_mode != IRQ_REMAP_X2APIC_MODE) {
1826 u32 apic_limit = 255;
1827
1828 /*
1829 * Using X2APIC without IR is not architecturally supported
1830 * on bare metal but may be supported in guests.
1831 */
1832 if (!x86_init.hyper.x2apic_available()) {
1833 pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n");
1834 x2apic_disable();
1835 return;
1836 }
1837
1838 /*
1839 * If the hypervisor supports extended destination ID in
1840 * MSI, that increases the maximum APIC ID that can be
1841 * used for non-remapped IRQ domains.
1842 */
1843 if (x86_init.hyper.msi_ext_dest_id()) {
1844 virt_ext_dest_id = 1;
1845 apic_limit = 32767;
1846 }
1847
1848 /*
1849 * Without IR, all CPUs can be addressed by IOAPIC/MSI only
1850 * in physical mode, and CPUs with an APIC ID that cannot
1851 * be addressed must not be brought online.
1852 */
1853 x2apic_set_max_apicid(apic_limit);
1854 x2apic_phys = 1;
1855 }
1856 x2apic_enable();
1857}
1858
1859void __init check_x2apic(void)
1860{
1861 if (x2apic_enabled()) {
1862 pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n");
1863 x2apic_mode = 1;
1864 if (x2apic_hw_locked())
1865 x2apic_state = X2APIC_ON_LOCKED;
1866 else
1867 x2apic_state = X2APIC_ON;
1868 apic_read_boot_cpu_id(true);
1869 } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) {
1870 x2apic_state = X2APIC_DISABLED;
1871 }
1872}
1873#else /* CONFIG_X86_X2APIC */
1874void __init check_x2apic(void)
1875{
1876 if (!apic_is_x2apic_enabled())
1877 return;
1878 /*
1879 * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC?
1880 */
1881 pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n");
1882 pr_err("Disabling APIC, expect reduced performance and functionality.\n");
1883
1884 apic_is_disabled = true;
1885 setup_clear_cpu_cap(X86_FEATURE_APIC);
1886}
1887
1888static inline void try_to_enable_x2apic(int remap_mode) { }
1889static inline void __x2apic_enable(void) { }
1890#endif /* !CONFIG_X86_X2APIC */
1891
1892void __init enable_IR_x2apic(void)
1893{
1894 unsigned long flags;
1895 int ret, ir_stat;
1896
1897 if (ioapic_is_disabled) {
1898 pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n");
1899 return;
1900 }
1901
1902 ir_stat = irq_remapping_prepare();
1903 if (ir_stat < 0 && !x2apic_supported())
1904 return;
1905
1906 ret = save_ioapic_entries();
1907 if (ret) {
1908 pr_info("Saving IO-APIC state failed: %d\n", ret);
1909 return;
1910 }
1911
1912 local_irq_save(flags);
1913 legacy_pic->mask_all();
1914 mask_ioapic_entries();
1915
1916 /* If irq_remapping_prepare() succeeded, try to enable it */
1917 if (ir_stat >= 0)
1918 ir_stat = irq_remapping_enable();
1919 /* ir_stat contains the remap mode or an error code */
1920 try_to_enable_x2apic(ir_stat);
1921
1922 if (ir_stat < 0)
1923 restore_ioapic_entries();
1924 legacy_pic->restore_mask();
1925 local_irq_restore(flags);
1926}
1927
1928#ifdef CONFIG_X86_64
1929/*
1930 * Detect and enable local APICs on non-SMP boards.
1931 * Original code written by Keir Fraser.
1932 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1933 * not correctly set up (usually the APIC timer won't work etc.)
1934 */
1935static bool __init detect_init_APIC(void)
1936{
1937 if (!boot_cpu_has(X86_FEATURE_APIC)) {
1938 pr_info("No local APIC present\n");
1939 return false;
1940 }
1941
1942 register_lapic_address(APIC_DEFAULT_PHYS_BASE);
1943 return true;
1944}
1945#else
1946
1947static bool __init apic_verify(unsigned long addr)
1948{
1949 u32 features, h, l;
1950
1951 /*
1952 * The APIC feature bit should now be enabled
1953 * in `cpuid'
1954 */
1955 features = cpuid_edx(1);
1956 if (!(features & (1 << X86_FEATURE_APIC))) {
1957 pr_warn("Could not enable APIC!\n");
1958 return false;
1959 }
1960 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1961
1962 /* The BIOS may have set up the APIC at some other address */
1963 if (boot_cpu_data.x86 >= 6) {
1964 rdmsr(MSR_IA32_APICBASE, l, h);
1965 if (l & MSR_IA32_APICBASE_ENABLE)
1966 addr = l & MSR_IA32_APICBASE_BASE;
1967 }
1968
1969 register_lapic_address(addr);
1970 pr_info("Found and enabled local APIC!\n");
1971 return true;
1972}
1973
1974bool __init apic_force_enable(unsigned long addr)
1975{
1976 u32 h, l;
1977
1978 if (apic_is_disabled)
1979 return false;
1980
1981 /*
1982 * Some BIOSes disable the local APIC in the APIC_BASE
1983 * MSR. This can only be done in software for Intel P6 or later
1984 * and AMD K7 (Model > 1) or later.
1985 */
1986 if (boot_cpu_data.x86 >= 6) {
1987 rdmsr(MSR_IA32_APICBASE, l, h);
1988 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1989 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1990 l &= ~MSR_IA32_APICBASE_BASE;
1991 l |= MSR_IA32_APICBASE_ENABLE | addr;
1992 wrmsr(MSR_IA32_APICBASE, l, h);
1993 enabled_via_apicbase = 1;
1994 }
1995 }
1996 return apic_verify(addr);
1997}
1998
1999/*
2000 * Detect and initialize APIC
2001 */
2002static bool __init detect_init_APIC(void)
2003{
2004 /* Disabled by kernel option? */
2005 if (apic_is_disabled)
2006 return false;
2007
2008 switch (boot_cpu_data.x86_vendor) {
2009 case X86_VENDOR_AMD:
2010 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
2011 (boot_cpu_data.x86 >= 15))
2012 break;
2013 goto no_apic;
2014 case X86_VENDOR_HYGON:
2015 break;
2016 case X86_VENDOR_INTEL:
2017 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
2018 (boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)))
2019 break;
2020 goto no_apic;
2021 default:
2022 goto no_apic;
2023 }
2024
2025 if (!boot_cpu_has(X86_FEATURE_APIC)) {
2026 /*
2027 * Over-ride BIOS and try to enable the local APIC only if
2028 * "lapic" specified.
2029 */
2030 if (!force_enable_local_apic) {
2031 pr_info("Local APIC disabled by BIOS -- "
2032 "you can enable it with \"lapic\"\n");
2033 return false;
2034 }
2035 if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE))
2036 return false;
2037 } else {
2038 if (!apic_verify(APIC_DEFAULT_PHYS_BASE))
2039 return false;
2040 }
2041
2042 apic_pm_activate();
2043
2044 return true;
2045
2046no_apic:
2047 pr_info("No local APIC present or hardware disabled\n");
2048 return false;
2049}
2050#endif
2051
2052/**
2053 * init_apic_mappings - initialize APIC mappings
2054 */
2055void __init init_apic_mappings(void)
2056{
2057 if (apic_validate_deadline_timer())
2058 pr_info("TSC deadline timer available\n");
2059
2060 if (x2apic_mode)
2061 return;
2062
2063 if (!smp_found_config) {
2064 if (!detect_init_APIC()) {
2065 pr_info("APIC: disable apic facility\n");
2066 apic_disable();
2067 }
2068 }
2069}
2070
2071static __init void apic_set_fixmap(bool read_apic)
2072{
2073 set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr);
2074 apic_mmio_base = APIC_BASE;
2075 apic_pr_verbose("Mapped APIC to %16lx (%16lx)\n", apic_mmio_base, mp_lapic_addr);
2076 if (read_apic)
2077 apic_read_boot_cpu_id(false);
2078}
2079
2080void __init register_lapic_address(unsigned long address)
2081{
2082 /* This should only happen once */
2083 WARN_ON_ONCE(mp_lapic_addr);
2084 mp_lapic_addr = address;
2085
2086 if (!x2apic_mode)
2087 apic_set_fixmap(true);
2088}
2089
2090/*
2091 * Local APIC interrupts
2092 */
2093
2094/*
2095 * Common handling code for spurious_interrupt and spurious_vector entry
2096 * points below. No point in allowing the compiler to inline it twice.
2097 */
2098static noinline void handle_spurious_interrupt(u8 vector)
2099{
2100 u32 v;
2101
2102 trace_spurious_apic_entry(vector);
2103
2104 inc_irq_stat(irq_spurious_count);
2105
2106 /*
2107 * If this is a spurious interrupt then do not acknowledge
2108 */
2109 if (vector == SPURIOUS_APIC_VECTOR) {
2110 /* See SDM vol 3 */
2111 pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n",
2112 smp_processor_id());
2113 goto out;
2114 }
2115
2116 /*
2117 * If it is a vectored one, verify it's set in the ISR. If set,
2118 * acknowledge it.
2119 */
2120 v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1));
2121 if (v & (1 << (vector & 0x1f))) {
2122 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n",
2123 vector, smp_processor_id());
2124 apic_eoi();
2125 } else {
2126 pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n",
2127 vector, smp_processor_id());
2128 }
2129out:
2130 trace_spurious_apic_exit(vector);
2131}
2132
2133/**
2134 * spurious_interrupt - Catch all for interrupts raised on unused vectors
2135 * @regs: Pointer to pt_regs on stack
2136 * @vector: The vector number
2137 *
2138 * This is invoked from ASM entry code to catch all interrupts which
2139 * trigger on an entry which is routed to the common_spurious idtentry
2140 * point.
2141 */
2142DEFINE_IDTENTRY_IRQ(spurious_interrupt)
2143{
2144 handle_spurious_interrupt(vector);
2145}
2146
2147DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt)
2148{
2149 handle_spurious_interrupt(SPURIOUS_APIC_VECTOR);
2150}
2151
2152/*
2153 * This interrupt should never happen with our APIC/SMP architecture
2154 */
2155DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt)
2156{
2157 static const char * const error_interrupt_reason[] = {
2158 "Send CS error", /* APIC Error Bit 0 */
2159 "Receive CS error", /* APIC Error Bit 1 */
2160 "Send accept error", /* APIC Error Bit 2 */
2161 "Receive accept error", /* APIC Error Bit 3 */
2162 "Redirectable IPI", /* APIC Error Bit 4 */
2163 "Send illegal vector", /* APIC Error Bit 5 */
2164 "Received illegal vector", /* APIC Error Bit 6 */
2165 "Illegal register address", /* APIC Error Bit 7 */
2166 };
2167 u32 v, i = 0;
2168
2169 trace_error_apic_entry(ERROR_APIC_VECTOR);
2170
2171 /* First tickle the hardware, only then report what went on. -- REW */
2172 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2173 apic_write(APIC_ESR, 0);
2174 v = apic_read(APIC_ESR);
2175 apic_eoi();
2176 atomic_inc(&irq_err_count);
2177
2178 apic_pr_debug("APIC error on CPU%d: %02x", smp_processor_id(), v);
2179
2180 v &= 0xff;
2181 while (v) {
2182 if (v & 0x1)
2183 apic_pr_debug_cont(" : %s", error_interrupt_reason[i]);
2184 i++;
2185 v >>= 1;
2186 }
2187
2188 apic_pr_debug_cont("\n");
2189
2190 trace_error_apic_exit(ERROR_APIC_VECTOR);
2191}
2192
2193/**
2194 * connect_bsp_APIC - attach the APIC to the interrupt system
2195 */
2196static void __init connect_bsp_APIC(void)
2197{
2198#ifdef CONFIG_X86_32
2199 if (pic_mode) {
2200 /*
2201 * Do not trust the local APIC being empty at bootup.
2202 */
2203 clear_local_APIC();
2204 /*
2205 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2206 * local APIC to INT and NMI lines.
2207 */
2208 apic_pr_verbose("Leaving PIC mode, enabling APIC mode.\n");
2209 imcr_pic_to_apic();
2210 }
2211#endif
2212}
2213
2214/**
2215 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2216 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2217 *
2218 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2219 * APIC is disabled.
2220 */
2221void disconnect_bsp_APIC(int virt_wire_setup)
2222{
2223 unsigned int value;
2224
2225#ifdef CONFIG_X86_32
2226 if (pic_mode) {
2227 /*
2228 * Put the board back into PIC mode (has an effect only on
2229 * certain older boards). Note that APIC interrupts, including
2230 * IPIs, won't work beyond this point! The only exception are
2231 * INIT IPIs.
2232 */
2233 apic_pr_verbose("Disabling APIC mode, entering PIC mode.\n");
2234 imcr_apic_to_pic();
2235 return;
2236 }
2237#endif
2238
2239 /* Go back to Virtual Wire compatibility mode */
2240
2241 /* For the spurious interrupt use vector F, and enable it */
2242 value = apic_read(APIC_SPIV);
2243 value &= ~APIC_VECTOR_MASK;
2244 value |= APIC_SPIV_APIC_ENABLED;
2245 value |= 0xf;
2246 apic_write(APIC_SPIV, value);
2247
2248 if (!virt_wire_setup) {
2249 /*
2250 * For LVT0 make it edge triggered, active high,
2251 * external and enabled
2252 */
2253 value = apic_read(APIC_LVT0);
2254 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2255 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2256 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2257 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2258 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2259 apic_write(APIC_LVT0, value);
2260 } else {
2261 /* Disable LVT0 */
2262 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2263 }
2264
2265 /*
2266 * For LVT1 make it edge triggered, active high,
2267 * nmi and enabled
2268 */
2269 value = apic_read(APIC_LVT1);
2270 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2271 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2272 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2273 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2274 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2275 apic_write(APIC_LVT1, value);
2276}
2277
2278void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg,
2279 bool dmar)
2280{
2281 memset(msg, 0, sizeof(*msg));
2282
2283 msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
2284 msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical;
2285 msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF;
2286
2287 msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED;
2288 msg->arch_data.vector = cfg->vector;
2289
2290 msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
2291 /*
2292 * Only the IOMMU itself can use the trick of putting destination
2293 * APIC ID into the high bits of the address. Anything else would
2294 * just be writing to memory if it tried that, and needs IR to
2295 * address APICs which can't be addressed in the normal 32-bit
2296 * address range at 0xFFExxxxx. That is typically just 8 bits, but
2297 * some hypervisors allow the extended destination ID field in bits
2298 * 5-11 to be used, giving support for 15 bits of APIC IDs in total.
2299 */
2300 if (dmar)
2301 msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8;
2302 else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000)
2303 msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8;
2304 else
2305 WARN_ON_ONCE(cfg->dest_apicid > 0xFF);
2306}
2307
2308u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid)
2309{
2310 u32 dest = msg->arch_addr_lo.destid_0_7;
2311
2312 if (extid)
2313 dest |= msg->arch_addr_hi.destid_8_31 << 8;
2314 return dest;
2315}
2316EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid);
2317
2318static void __init apic_bsp_up_setup(void)
2319{
2320 reset_phys_cpu_present_map(boot_cpu_physical_apicid);
2321}
2322
2323/**
2324 * apic_bsp_setup - Setup function for local apic and io-apic
2325 * @upmode: Force UP mode (for APIC_init_uniprocessor)
2326 */
2327static void __init apic_bsp_setup(bool upmode)
2328{
2329 connect_bsp_APIC();
2330 if (upmode)
2331 apic_bsp_up_setup();
2332 setup_local_APIC();
2333
2334 enable_IO_APIC();
2335 end_local_APIC_setup();
2336 irq_remap_enable_fault_handling();
2337 setup_IO_APIC();
2338 lapic_update_legacy_vectors();
2339}
2340
2341#ifdef CONFIG_UP_LATE_INIT
2342void __init up_late_init(void)
2343{
2344 if (apic_intr_mode == APIC_PIC)
2345 return;
2346
2347 /* Setup local timer */
2348 x86_init.timers.setup_percpu_clockev();
2349}
2350#endif
2351
2352/*
2353 * Power management
2354 */
2355#ifdef CONFIG_PM
2356
2357static struct {
2358 /*
2359 * 'active' is true if the local APIC was enabled by us and
2360 * not the BIOS; this signifies that we are also responsible
2361 * for disabling it before entering apm/acpi suspend
2362 */
2363 int active;
2364 /* r/w apic fields */
2365 u32 apic_id;
2366 unsigned int apic_taskpri;
2367 unsigned int apic_ldr;
2368 unsigned int apic_dfr;
2369 unsigned int apic_spiv;
2370 unsigned int apic_lvtt;
2371 unsigned int apic_lvtpc;
2372 unsigned int apic_lvt0;
2373 unsigned int apic_lvt1;
2374 unsigned int apic_lvterr;
2375 unsigned int apic_tmict;
2376 unsigned int apic_tdcr;
2377 unsigned int apic_thmr;
2378 unsigned int apic_cmci;
2379} apic_pm_state;
2380
2381static int lapic_suspend(void)
2382{
2383 unsigned long flags;
2384 int maxlvt;
2385
2386 if (!apic_pm_state.active)
2387 return 0;
2388
2389 maxlvt = lapic_get_maxlvt();
2390
2391 apic_pm_state.apic_id = apic_read(APIC_ID);
2392 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2393 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2394 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2395 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2396 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2397 if (maxlvt >= 4)
2398 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2399 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2400 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2401 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2402 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2403 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2404#ifdef CONFIG_X86_THERMAL_VECTOR
2405 if (maxlvt >= 5)
2406 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2407#endif
2408#ifdef CONFIG_X86_MCE_INTEL
2409 if (maxlvt >= 6)
2410 apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
2411#endif
2412
2413 local_irq_save(flags);
2414
2415 /*
2416 * Mask IOAPIC before disabling the local APIC to prevent stale IRR
2417 * entries on some implementations.
2418 */
2419 mask_ioapic_entries();
2420
2421 disable_local_APIC();
2422
2423 irq_remapping_disable();
2424
2425 local_irq_restore(flags);
2426 return 0;
2427}
2428
2429static void lapic_resume(void)
2430{
2431 unsigned int l, h;
2432 unsigned long flags;
2433 int maxlvt;
2434
2435 if (!apic_pm_state.active)
2436 return;
2437
2438 local_irq_save(flags);
2439
2440 /*
2441 * IO-APIC and PIC have their own resume routines.
2442 * We just mask them here to make sure the interrupt
2443 * subsystem is completely quiet while we enable x2apic
2444 * and interrupt-remapping.
2445 */
2446 mask_ioapic_entries();
2447 legacy_pic->mask_all();
2448
2449 if (x2apic_mode) {
2450 __x2apic_enable();
2451 } else {
2452 /*
2453 * Make sure the APICBASE points to the right address
2454 *
2455 * FIXME! This will be wrong if we ever support suspend on
2456 * SMP! We'll need to do this as part of the CPU restore!
2457 */
2458 if (boot_cpu_data.x86 >= 6) {
2459 rdmsr(MSR_IA32_APICBASE, l, h);
2460 l &= ~MSR_IA32_APICBASE_BASE;
2461 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2462 wrmsr(MSR_IA32_APICBASE, l, h);
2463 }
2464 }
2465
2466 maxlvt = lapic_get_maxlvt();
2467 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2468 apic_write(APIC_ID, apic_pm_state.apic_id);
2469 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2470 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2471 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2472 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2473 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2474 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2475#ifdef CONFIG_X86_THERMAL_VECTOR
2476 if (maxlvt >= 5)
2477 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2478#endif
2479#ifdef CONFIG_X86_MCE_INTEL
2480 if (maxlvt >= 6)
2481 apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
2482#endif
2483 if (maxlvt >= 4)
2484 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2485 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2486 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2487 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2488 apic_write(APIC_ESR, 0);
2489 apic_read(APIC_ESR);
2490 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2491 apic_write(APIC_ESR, 0);
2492 apic_read(APIC_ESR);
2493
2494 irq_remapping_reenable(x2apic_mode);
2495
2496 local_irq_restore(flags);
2497}
2498
2499/*
2500 * This device has no shutdown method - fully functioning local APICs
2501 * are needed on every CPU up until machine_halt/restart/poweroff.
2502 */
2503
2504static struct syscore_ops lapic_syscore_ops = {
2505 .resume = lapic_resume,
2506 .suspend = lapic_suspend,
2507};
2508
2509static void apic_pm_activate(void)
2510{
2511 apic_pm_state.active = 1;
2512}
2513
2514static int __init init_lapic_sysfs(void)
2515{
2516 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2517 if (boot_cpu_has(X86_FEATURE_APIC))
2518 register_syscore_ops(&lapic_syscore_ops);
2519
2520 return 0;
2521}
2522
2523/* local apic needs to resume before other devices access its registers. */
2524core_initcall(init_lapic_sysfs);
2525
2526#else /* CONFIG_PM */
2527
2528static void apic_pm_activate(void) { }
2529
2530#endif /* CONFIG_PM */
2531
2532#ifdef CONFIG_X86_64
2533
2534static int multi_checked;
2535static int multi;
2536
2537static int set_multi(const struct dmi_system_id *d)
2538{
2539 if (multi)
2540 return 0;
2541 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2542 multi = 1;
2543 return 0;
2544}
2545
2546static const struct dmi_system_id multi_dmi_table[] = {
2547 {
2548 .callback = set_multi,
2549 .ident = "IBM System Summit2",
2550 .matches = {
2551 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2552 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2553 },
2554 },
2555 {}
2556};
2557
2558static void dmi_check_multi(void)
2559{
2560 if (multi_checked)
2561 return;
2562
2563 dmi_check_system(multi_dmi_table);
2564 multi_checked = 1;
2565}
2566
2567/*
2568 * apic_is_clustered_box() -- Check if we can expect good TSC
2569 *
2570 * Thus far, the major user of this is IBM's Summit2 series:
2571 * Clustered boxes may have unsynced TSC problems if they are
2572 * multi-chassis.
2573 * Use DMI to check them
2574 */
2575int apic_is_clustered_box(void)
2576{
2577 dmi_check_multi();
2578 return multi;
2579}
2580#endif
2581
2582/*
2583 * APIC command line parameters
2584 */
2585static int __init setup_disableapic(char *arg)
2586{
2587 apic_is_disabled = true;
2588 setup_clear_cpu_cap(X86_FEATURE_APIC);
2589 return 0;
2590}
2591early_param("disableapic", setup_disableapic);
2592
2593/* same as disableapic, for compatibility */
2594static int __init setup_nolapic(char *arg)
2595{
2596 return setup_disableapic(arg);
2597}
2598early_param("nolapic", setup_nolapic);
2599
2600static int __init parse_lapic_timer_c2_ok(char *arg)
2601{
2602 local_apic_timer_c2_ok = 1;
2603 return 0;
2604}
2605early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2606
2607static int __init parse_disable_apic_timer(char *arg)
2608{
2609 disable_apic_timer = 1;
2610 return 0;
2611}
2612early_param("noapictimer", parse_disable_apic_timer);
2613
2614static int __init parse_nolapic_timer(char *arg)
2615{
2616 disable_apic_timer = 1;
2617 return 0;
2618}
2619early_param("nolapic_timer", parse_nolapic_timer);
2620
2621static int __init apic_set_verbosity(char *arg)
2622{
2623 if (!arg) {
2624 if (IS_ENABLED(CONFIG_X86_32))
2625 return -EINVAL;
2626
2627 ioapic_is_disabled = false;
2628 return 0;
2629 }
2630
2631 if (strcmp("debug", arg) == 0)
2632 apic_verbosity = APIC_DEBUG;
2633 else if (strcmp("verbose", arg) == 0)
2634 apic_verbosity = APIC_VERBOSE;
2635#ifdef CONFIG_X86_64
2636 else {
2637 pr_warn("APIC Verbosity level %s not recognised"
2638 " use apic=verbose or apic=debug\n", arg);
2639 return -EINVAL;
2640 }
2641#endif
2642
2643 return 0;
2644}
2645early_param("apic", apic_set_verbosity);
2646
2647static int __init lapic_insert_resource(void)
2648{
2649 if (!apic_mmio_base)
2650 return -1;
2651
2652 /* Put local APIC into the resource map. */
2653 lapic_resource.start = apic_mmio_base;
2654 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2655 insert_resource(&iomem_resource, &lapic_resource);
2656
2657 return 0;
2658}
2659
2660/*
2661 * need call insert after e820__reserve_resources()
2662 * that is using request_resource
2663 */
2664late_initcall(lapic_insert_resource);
2665
2666static int __init apic_set_extnmi(char *arg)
2667{
2668 if (!arg)
2669 return -EINVAL;
2670
2671 if (!strncmp("all", arg, 3))
2672 apic_extnmi = APIC_EXTNMI_ALL;
2673 else if (!strncmp("none", arg, 4))
2674 apic_extnmi = APIC_EXTNMI_NONE;
2675 else if (!strncmp("bsp", arg, 3))
2676 apic_extnmi = APIC_EXTNMI_BSP;
2677 else {
2678 pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg);
2679 return -EINVAL;
2680 }
2681
2682 return 0;
2683}
2684early_param("apic_extnmi", apic_set_extnmi);
1/*
2 * Local APIC handling, local APIC timers
3 *
4 * (c) 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
5 *
6 * Fixes
7 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
8 * thanks to Eric Gilmore
9 * and Rolf G. Tews
10 * for testing these extensively.
11 * Maciej W. Rozycki : Various updates and fixes.
12 * Mikael Pettersson : Power Management for UP-APIC.
13 * Pavel Machek and
14 * Mikael Pettersson : PM converted to driver model.
15 */
16
17#include <linux/perf_event.h>
18#include <linux/kernel_stat.h>
19#include <linux/mc146818rtc.h>
20#include <linux/acpi_pmtmr.h>
21#include <linux/clockchips.h>
22#include <linux/interrupt.h>
23#include <linux/bootmem.h>
24#include <linux/ftrace.h>
25#include <linux/ioport.h>
26#include <linux/module.h>
27#include <linux/syscore_ops.h>
28#include <linux/delay.h>
29#include <linux/timex.h>
30#include <linux/i8253.h>
31#include <linux/dmar.h>
32#include <linux/init.h>
33#include <linux/cpu.h>
34#include <linux/dmi.h>
35#include <linux/smp.h>
36#include <linux/mm.h>
37
38#include <asm/trace/irq_vectors.h>
39#include <asm/irq_remapping.h>
40#include <asm/perf_event.h>
41#include <asm/x86_init.h>
42#include <asm/pgalloc.h>
43#include <linux/atomic.h>
44#include <asm/mpspec.h>
45#include <asm/i8259.h>
46#include <asm/proto.h>
47#include <asm/apic.h>
48#include <asm/io_apic.h>
49#include <asm/desc.h>
50#include <asm/hpet.h>
51#include <asm/idle.h>
52#include <asm/mtrr.h>
53#include <asm/time.h>
54#include <asm/smp.h>
55#include <asm/mce.h>
56#include <asm/tsc.h>
57#include <asm/hypervisor.h>
58
59unsigned int num_processors;
60
61unsigned disabled_cpus;
62
63/* Processor that is doing the boot up */
64unsigned int boot_cpu_physical_apicid = -1U;
65EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid);
66
67/*
68 * The highest APIC ID seen during enumeration.
69 */
70unsigned int max_physical_apicid;
71
72/*
73 * Bitmask of physically existing CPUs:
74 */
75physid_mask_t phys_cpu_present_map;
76
77/*
78 * Processor to be disabled specified by kernel parameter
79 * disable_cpu_apicid=<int>, mostly used for the kdump 2nd kernel to
80 * avoid undefined behaviour caused by sending INIT from AP to BSP.
81 */
82static unsigned int disabled_cpu_apicid __read_mostly = BAD_APICID;
83
84/*
85 * Map cpu index to physical APIC ID
86 */
87DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_cpu_to_apicid, BAD_APICID);
88DEFINE_EARLY_PER_CPU_READ_MOSTLY(u16, x86_bios_cpu_apicid, BAD_APICID);
89EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
90EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
91
92#ifdef CONFIG_X86_32
93
94/*
95 * On x86_32, the mapping between cpu and logical apicid may vary
96 * depending on apic in use. The following early percpu variable is
97 * used for the mapping. This is where the behaviors of x86_64 and 32
98 * actually diverge. Let's keep it ugly for now.
99 */
100DEFINE_EARLY_PER_CPU_READ_MOSTLY(int, x86_cpu_to_logical_apicid, BAD_APICID);
101
102/* Local APIC was disabled by the BIOS and enabled by the kernel */
103static int enabled_via_apicbase;
104
105/*
106 * Handle interrupt mode configuration register (IMCR).
107 * This register controls whether the interrupt signals
108 * that reach the BSP come from the master PIC or from the
109 * local APIC. Before entering Symmetric I/O Mode, either
110 * the BIOS or the operating system must switch out of
111 * PIC Mode by changing the IMCR.
112 */
113static inline void imcr_pic_to_apic(void)
114{
115 /* select IMCR register */
116 outb(0x70, 0x22);
117 /* NMI and 8259 INTR go through APIC */
118 outb(0x01, 0x23);
119}
120
121static inline void imcr_apic_to_pic(void)
122{
123 /* select IMCR register */
124 outb(0x70, 0x22);
125 /* NMI and 8259 INTR go directly to BSP */
126 outb(0x00, 0x23);
127}
128#endif
129
130/*
131 * Knob to control our willingness to enable the local APIC.
132 *
133 * +1=force-enable
134 */
135static int force_enable_local_apic __initdata;
136
137/* Control whether x2APIC mode is enabled or not */
138static bool nox2apic __initdata;
139
140/*
141 * APIC command line parameters
142 */
143static int __init parse_lapic(char *arg)
144{
145 if (config_enabled(CONFIG_X86_32) && !arg)
146 force_enable_local_apic = 1;
147 else if (arg && !strncmp(arg, "notscdeadline", 13))
148 setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER);
149 return 0;
150}
151early_param("lapic", parse_lapic);
152
153#ifdef CONFIG_X86_64
154static int apic_calibrate_pmtmr __initdata;
155static __init int setup_apicpmtimer(char *s)
156{
157 apic_calibrate_pmtmr = 1;
158 notsc_setup(NULL);
159 return 0;
160}
161__setup("apicpmtimer", setup_apicpmtimer);
162#endif
163
164int x2apic_mode;
165#ifdef CONFIG_X86_X2APIC
166/* x2apic enabled before OS handover */
167int x2apic_preenabled;
168static int x2apic_disabled;
169static int __init setup_nox2apic(char *str)
170{
171 if (x2apic_enabled()) {
172 int apicid = native_apic_msr_read(APIC_ID);
173
174 if (apicid >= 255) {
175 pr_warning("Apicid: %08x, cannot enforce nox2apic\n",
176 apicid);
177 return 0;
178 }
179
180 pr_warning("x2apic already enabled. will disable it\n");
181 } else
182 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
183
184 nox2apic = true;
185
186 return 0;
187}
188early_param("nox2apic", setup_nox2apic);
189#endif
190
191unsigned long mp_lapic_addr;
192int disable_apic;
193/* Disable local APIC timer from the kernel commandline or via dmi quirk */
194static int disable_apic_timer __initdata;
195/* Local APIC timer works in C2 */
196int local_apic_timer_c2_ok;
197EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
198
199int first_system_vector = 0xfe;
200
201/*
202 * Debug level, exported for io_apic.c
203 */
204unsigned int apic_verbosity;
205
206int pic_mode;
207
208/* Have we found an MP table */
209int smp_found_config;
210
211static struct resource lapic_resource = {
212 .name = "Local APIC",
213 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
214};
215
216unsigned int lapic_timer_frequency = 0;
217
218static void apic_pm_activate(void);
219
220static unsigned long apic_phys;
221
222/*
223 * Get the LAPIC version
224 */
225static inline int lapic_get_version(void)
226{
227 return GET_APIC_VERSION(apic_read(APIC_LVR));
228}
229
230/*
231 * Check, if the APIC is integrated or a separate chip
232 */
233static inline int lapic_is_integrated(void)
234{
235#ifdef CONFIG_X86_64
236 return 1;
237#else
238 return APIC_INTEGRATED(lapic_get_version());
239#endif
240}
241
242/*
243 * Check, whether this is a modern or a first generation APIC
244 */
245static int modern_apic(void)
246{
247 /* AMD systems use old APIC versions, so check the CPU */
248 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
249 boot_cpu_data.x86 >= 0xf)
250 return 1;
251 return lapic_get_version() >= 0x14;
252}
253
254/*
255 * right after this call apic become NOOP driven
256 * so apic->write/read doesn't do anything
257 */
258static void __init apic_disable(void)
259{
260 pr_info("APIC: switched to apic NOOP\n");
261 apic = &apic_noop;
262}
263
264void native_apic_wait_icr_idle(void)
265{
266 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
267 cpu_relax();
268}
269
270u32 native_safe_apic_wait_icr_idle(void)
271{
272 u32 send_status;
273 int timeout;
274
275 timeout = 0;
276 do {
277 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
278 if (!send_status)
279 break;
280 inc_irq_stat(icr_read_retry_count);
281 udelay(100);
282 } while (timeout++ < 1000);
283
284 return send_status;
285}
286
287void native_apic_icr_write(u32 low, u32 id)
288{
289 unsigned long flags;
290
291 local_irq_save(flags);
292 apic_write(APIC_ICR2, SET_APIC_DEST_FIELD(id));
293 apic_write(APIC_ICR, low);
294 local_irq_restore(flags);
295}
296
297u64 native_apic_icr_read(void)
298{
299 u32 icr1, icr2;
300
301 icr2 = apic_read(APIC_ICR2);
302 icr1 = apic_read(APIC_ICR);
303
304 return icr1 | ((u64)icr2 << 32);
305}
306
307#ifdef CONFIG_X86_32
308/**
309 * get_physical_broadcast - Get number of physical broadcast IDs
310 */
311int get_physical_broadcast(void)
312{
313 return modern_apic() ? 0xff : 0xf;
314}
315#endif
316
317/**
318 * lapic_get_maxlvt - get the maximum number of local vector table entries
319 */
320int lapic_get_maxlvt(void)
321{
322 unsigned int v;
323
324 v = apic_read(APIC_LVR);
325 /*
326 * - we always have APIC integrated on 64bit mode
327 * - 82489DXs do not report # of LVT entries
328 */
329 return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
330}
331
332/*
333 * Local APIC timer
334 */
335
336/* Clock divisor */
337#define APIC_DIVISOR 16
338#define TSC_DIVISOR 32
339
340/*
341 * This function sets up the local APIC timer, with a timeout of
342 * 'clocks' APIC bus clock. During calibration we actually call
343 * this function twice on the boot CPU, once with a bogus timeout
344 * value, second time for real. The other (noncalibrating) CPUs
345 * call this function only once, with the real, calibrated value.
346 *
347 * We do reads before writes even if unnecessary, to get around the
348 * P5 APIC double write bug.
349 */
350static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
351{
352 unsigned int lvtt_value, tmp_value;
353
354 lvtt_value = LOCAL_TIMER_VECTOR;
355 if (!oneshot)
356 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
357 else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER))
358 lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE;
359
360 if (!lapic_is_integrated())
361 lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
362
363 if (!irqen)
364 lvtt_value |= APIC_LVT_MASKED;
365
366 apic_write(APIC_LVTT, lvtt_value);
367
368 if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) {
369 printk_once(KERN_DEBUG "TSC deadline timer enabled\n");
370 return;
371 }
372
373 /*
374 * Divide PICLK by 16
375 */
376 tmp_value = apic_read(APIC_TDCR);
377 apic_write(APIC_TDCR,
378 (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
379 APIC_TDR_DIV_16);
380
381 if (!oneshot)
382 apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
383}
384
385/*
386 * Setup extended LVT, AMD specific
387 *
388 * Software should use the LVT offsets the BIOS provides. The offsets
389 * are determined by the subsystems using it like those for MCE
390 * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts
391 * are supported. Beginning with family 10h at least 4 offsets are
392 * available.
393 *
394 * Since the offsets must be consistent for all cores, we keep track
395 * of the LVT offsets in software and reserve the offset for the same
396 * vector also to be used on other cores. An offset is freed by
397 * setting the entry to APIC_EILVT_MASKED.
398 *
399 * If the BIOS is right, there should be no conflicts. Otherwise a
400 * "[Firmware Bug]: ..." error message is generated. However, if
401 * software does not properly determines the offsets, it is not
402 * necessarily a BIOS bug.
403 */
404
405static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX];
406
407static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new)
408{
409 return (old & APIC_EILVT_MASKED)
410 || (new == APIC_EILVT_MASKED)
411 || ((new & ~APIC_EILVT_MASKED) == old);
412}
413
414static unsigned int reserve_eilvt_offset(int offset, unsigned int new)
415{
416 unsigned int rsvd, vector;
417
418 if (offset >= APIC_EILVT_NR_MAX)
419 return ~0;
420
421 rsvd = atomic_read(&eilvt_offsets[offset]);
422 do {
423 vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */
424 if (vector && !eilvt_entry_is_changeable(vector, new))
425 /* may not change if vectors are different */
426 return rsvd;
427 rsvd = atomic_cmpxchg(&eilvt_offsets[offset], rsvd, new);
428 } while (rsvd != new);
429
430 rsvd &= ~APIC_EILVT_MASKED;
431 if (rsvd && rsvd != vector)
432 pr_info("LVT offset %d assigned for vector 0x%02x\n",
433 offset, rsvd);
434
435 return new;
436}
437
438/*
439 * If mask=1, the LVT entry does not generate interrupts while mask=0
440 * enables the vector. See also the BKDGs. Must be called with
441 * preemption disabled.
442 */
443
444int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask)
445{
446 unsigned long reg = APIC_EILVTn(offset);
447 unsigned int new, old, reserved;
448
449 new = (mask << 16) | (msg_type << 8) | vector;
450 old = apic_read(reg);
451 reserved = reserve_eilvt_offset(offset, new);
452
453 if (reserved != new) {
454 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
455 "vector 0x%x, but the register is already in use for "
456 "vector 0x%x on another cpu\n",
457 smp_processor_id(), reg, offset, new, reserved);
458 return -EINVAL;
459 }
460
461 if (!eilvt_entry_is_changeable(old, new)) {
462 pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for "
463 "vector 0x%x, but the register is already in use for "
464 "vector 0x%x on this cpu\n",
465 smp_processor_id(), reg, offset, new, old);
466 return -EBUSY;
467 }
468
469 apic_write(reg, new);
470
471 return 0;
472}
473EXPORT_SYMBOL_GPL(setup_APIC_eilvt);
474
475/*
476 * Program the next event, relative to now
477 */
478static int lapic_next_event(unsigned long delta,
479 struct clock_event_device *evt)
480{
481 apic_write(APIC_TMICT, delta);
482 return 0;
483}
484
485static int lapic_next_deadline(unsigned long delta,
486 struct clock_event_device *evt)
487{
488 u64 tsc;
489
490 rdtscll(tsc);
491 wrmsrl(MSR_IA32_TSC_DEADLINE, tsc + (((u64) delta) * TSC_DIVISOR));
492 return 0;
493}
494
495/*
496 * Setup the lapic timer in periodic or oneshot mode
497 */
498static void lapic_timer_setup(enum clock_event_mode mode,
499 struct clock_event_device *evt)
500{
501 unsigned long flags;
502 unsigned int v;
503
504 /* Lapic used as dummy for broadcast ? */
505 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
506 return;
507
508 local_irq_save(flags);
509
510 switch (mode) {
511 case CLOCK_EVT_MODE_PERIODIC:
512 case CLOCK_EVT_MODE_ONESHOT:
513 __setup_APIC_LVTT(lapic_timer_frequency,
514 mode != CLOCK_EVT_MODE_PERIODIC, 1);
515 break;
516 case CLOCK_EVT_MODE_UNUSED:
517 case CLOCK_EVT_MODE_SHUTDOWN:
518 v = apic_read(APIC_LVTT);
519 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
520 apic_write(APIC_LVTT, v);
521 apic_write(APIC_TMICT, 0);
522 break;
523 case CLOCK_EVT_MODE_RESUME:
524 /* Nothing to do here */
525 break;
526 }
527
528 local_irq_restore(flags);
529}
530
531/*
532 * Local APIC timer broadcast function
533 */
534static void lapic_timer_broadcast(const struct cpumask *mask)
535{
536#ifdef CONFIG_SMP
537 apic->send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
538#endif
539}
540
541
542/*
543 * The local apic timer can be used for any function which is CPU local.
544 */
545static struct clock_event_device lapic_clockevent = {
546 .name = "lapic",
547 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
548 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
549 .shift = 32,
550 .set_mode = lapic_timer_setup,
551 .set_next_event = lapic_next_event,
552 .broadcast = lapic_timer_broadcast,
553 .rating = 100,
554 .irq = -1,
555};
556static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
557
558/*
559 * Setup the local APIC timer for this CPU. Copy the initialized values
560 * of the boot CPU and register the clock event in the framework.
561 */
562static void setup_APIC_timer(void)
563{
564 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
565
566 if (this_cpu_has(X86_FEATURE_ARAT)) {
567 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP;
568 /* Make LAPIC timer preferrable over percpu HPET */
569 lapic_clockevent.rating = 150;
570 }
571
572 memcpy(levt, &lapic_clockevent, sizeof(*levt));
573 levt->cpumask = cpumask_of(smp_processor_id());
574
575 if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
576 levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
577 CLOCK_EVT_FEAT_DUMMY);
578 levt->set_next_event = lapic_next_deadline;
579 clockevents_config_and_register(levt,
580 (tsc_khz / TSC_DIVISOR) * 1000,
581 0xF, ~0UL);
582 } else
583 clockevents_register_device(levt);
584}
585
586/*
587 * In this functions we calibrate APIC bus clocks to the external timer.
588 *
589 * We want to do the calibration only once since we want to have local timer
590 * irqs syncron. CPUs connected by the same APIC bus have the very same bus
591 * frequency.
592 *
593 * This was previously done by reading the PIT/HPET and waiting for a wrap
594 * around to find out, that a tick has elapsed. I have a box, where the PIT
595 * readout is broken, so it never gets out of the wait loop again. This was
596 * also reported by others.
597 *
598 * Monitoring the jiffies value is inaccurate and the clockevents
599 * infrastructure allows us to do a simple substitution of the interrupt
600 * handler.
601 *
602 * The calibration routine also uses the pm_timer when possible, as the PIT
603 * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
604 * back to normal later in the boot process).
605 */
606
607#define LAPIC_CAL_LOOPS (HZ/10)
608
609static __initdata int lapic_cal_loops = -1;
610static __initdata long lapic_cal_t1, lapic_cal_t2;
611static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
612static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
613static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
614
615/*
616 * Temporary interrupt handler.
617 */
618static void __init lapic_cal_handler(struct clock_event_device *dev)
619{
620 unsigned long long tsc = 0;
621 long tapic = apic_read(APIC_TMCCT);
622 unsigned long pm = acpi_pm_read_early();
623
624 if (cpu_has_tsc)
625 rdtscll(tsc);
626
627 switch (lapic_cal_loops++) {
628 case 0:
629 lapic_cal_t1 = tapic;
630 lapic_cal_tsc1 = tsc;
631 lapic_cal_pm1 = pm;
632 lapic_cal_j1 = jiffies;
633 break;
634
635 case LAPIC_CAL_LOOPS:
636 lapic_cal_t2 = tapic;
637 lapic_cal_tsc2 = tsc;
638 if (pm < lapic_cal_pm1)
639 pm += ACPI_PM_OVRRUN;
640 lapic_cal_pm2 = pm;
641 lapic_cal_j2 = jiffies;
642 break;
643 }
644}
645
646static int __init
647calibrate_by_pmtimer(long deltapm, long *delta, long *deltatsc)
648{
649 const long pm_100ms = PMTMR_TICKS_PER_SEC / 10;
650 const long pm_thresh = pm_100ms / 100;
651 unsigned long mult;
652 u64 res;
653
654#ifndef CONFIG_X86_PM_TIMER
655 return -1;
656#endif
657
658 apic_printk(APIC_VERBOSE, "... PM-Timer delta = %ld\n", deltapm);
659
660 /* Check, if the PM timer is available */
661 if (!deltapm)
662 return -1;
663
664 mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
665
666 if (deltapm > (pm_100ms - pm_thresh) &&
667 deltapm < (pm_100ms + pm_thresh)) {
668 apic_printk(APIC_VERBOSE, "... PM-Timer result ok\n");
669 return 0;
670 }
671
672 res = (((u64)deltapm) * mult) >> 22;
673 do_div(res, 1000000);
674 pr_warning("APIC calibration not consistent "
675 "with PM-Timer: %ldms instead of 100ms\n",(long)res);
676
677 /* Correct the lapic counter value */
678 res = (((u64)(*delta)) * pm_100ms);
679 do_div(res, deltapm);
680 pr_info("APIC delta adjusted to PM-Timer: "
681 "%lu (%ld)\n", (unsigned long)res, *delta);
682 *delta = (long)res;
683
684 /* Correct the tsc counter value */
685 if (cpu_has_tsc) {
686 res = (((u64)(*deltatsc)) * pm_100ms);
687 do_div(res, deltapm);
688 apic_printk(APIC_VERBOSE, "TSC delta adjusted to "
689 "PM-Timer: %lu (%ld)\n",
690 (unsigned long)res, *deltatsc);
691 *deltatsc = (long)res;
692 }
693
694 return 0;
695}
696
697static int __init calibrate_APIC_clock(void)
698{
699 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
700 void (*real_handler)(struct clock_event_device *dev);
701 unsigned long deltaj;
702 long delta, deltatsc;
703 int pm_referenced = 0;
704
705 /**
706 * check if lapic timer has already been calibrated by platform
707 * specific routine, such as tsc calibration code. if so, we just fill
708 * in the clockevent structure and return.
709 */
710
711 if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) {
712 return 0;
713 } else if (lapic_timer_frequency) {
714 apic_printk(APIC_VERBOSE, "lapic timer already calibrated %d\n",
715 lapic_timer_frequency);
716 lapic_clockevent.mult = div_sc(lapic_timer_frequency/APIC_DIVISOR,
717 TICK_NSEC, lapic_clockevent.shift);
718 lapic_clockevent.max_delta_ns =
719 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
720 lapic_clockevent.min_delta_ns =
721 clockevent_delta2ns(0xF, &lapic_clockevent);
722 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
723 return 0;
724 }
725
726 apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
727 "calibrating APIC timer ...\n");
728
729 local_irq_disable();
730
731 /* Replace the global interrupt handler */
732 real_handler = global_clock_event->event_handler;
733 global_clock_event->event_handler = lapic_cal_handler;
734
735 /*
736 * Setup the APIC counter to maximum. There is no way the lapic
737 * can underflow in the 100ms detection time frame
738 */
739 __setup_APIC_LVTT(0xffffffff, 0, 0);
740
741 /* Let the interrupts run */
742 local_irq_enable();
743
744 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
745 cpu_relax();
746
747 local_irq_disable();
748
749 /* Restore the real event handler */
750 global_clock_event->event_handler = real_handler;
751
752 /* Build delta t1-t2 as apic timer counts down */
753 delta = lapic_cal_t1 - lapic_cal_t2;
754 apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
755
756 deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
757
758 /* we trust the PM based calibration if possible */
759 pm_referenced = !calibrate_by_pmtimer(lapic_cal_pm2 - lapic_cal_pm1,
760 &delta, &deltatsc);
761
762 /* Calculate the scaled math multiplication factor */
763 lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS,
764 lapic_clockevent.shift);
765 lapic_clockevent.max_delta_ns =
766 clockevent_delta2ns(0x7FFFFFFF, &lapic_clockevent);
767 lapic_clockevent.min_delta_ns =
768 clockevent_delta2ns(0xF, &lapic_clockevent);
769
770 lapic_timer_frequency = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
771
772 apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
773 apic_printk(APIC_VERBOSE, "..... mult: %u\n", lapic_clockevent.mult);
774 apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
775 lapic_timer_frequency);
776
777 if (cpu_has_tsc) {
778 apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
779 "%ld.%04ld MHz.\n",
780 (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ),
781 (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ));
782 }
783
784 apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
785 "%u.%04u MHz.\n",
786 lapic_timer_frequency / (1000000 / HZ),
787 lapic_timer_frequency % (1000000 / HZ));
788
789 /*
790 * Do a sanity check on the APIC calibration result
791 */
792 if (lapic_timer_frequency < (1000000 / HZ)) {
793 local_irq_enable();
794 pr_warning("APIC frequency too slow, disabling apic timer\n");
795 return -1;
796 }
797
798 levt->features &= ~CLOCK_EVT_FEAT_DUMMY;
799
800 /*
801 * PM timer calibration failed or not turned on
802 * so lets try APIC timer based calibration
803 */
804 if (!pm_referenced) {
805 apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
806
807 /*
808 * Setup the apic timer manually
809 */
810 levt->event_handler = lapic_cal_handler;
811 lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
812 lapic_cal_loops = -1;
813
814 /* Let the interrupts run */
815 local_irq_enable();
816
817 while (lapic_cal_loops <= LAPIC_CAL_LOOPS)
818 cpu_relax();
819
820 /* Stop the lapic timer */
821 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
822
823 /* Jiffies delta */
824 deltaj = lapic_cal_j2 - lapic_cal_j1;
825 apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
826
827 /* Check, if the jiffies result is consistent */
828 if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2)
829 apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
830 else
831 levt->features |= CLOCK_EVT_FEAT_DUMMY;
832 } else
833 local_irq_enable();
834
835 if (levt->features & CLOCK_EVT_FEAT_DUMMY) {
836 pr_warning("APIC timer disabled due to verification failure\n");
837 return -1;
838 }
839
840 return 0;
841}
842
843/*
844 * Setup the boot APIC
845 *
846 * Calibrate and verify the result.
847 */
848void __init setup_boot_APIC_clock(void)
849{
850 /*
851 * The local apic timer can be disabled via the kernel
852 * commandline or from the CPU detection code. Register the lapic
853 * timer as a dummy clock event source on SMP systems, so the
854 * broadcast mechanism is used. On UP systems simply ignore it.
855 */
856 if (disable_apic_timer) {
857 pr_info("Disabling APIC timer\n");
858 /* No broadcast on UP ! */
859 if (num_possible_cpus() > 1) {
860 lapic_clockevent.mult = 1;
861 setup_APIC_timer();
862 }
863 return;
864 }
865
866 if (calibrate_APIC_clock()) {
867 /* No broadcast on UP ! */
868 if (num_possible_cpus() > 1)
869 setup_APIC_timer();
870 return;
871 }
872
873 /*
874 * If nmi_watchdog is set to IO_APIC, we need the
875 * PIT/HPET going. Otherwise register lapic as a dummy
876 * device.
877 */
878 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
879
880 /* Setup the lapic or request the broadcast */
881 setup_APIC_timer();
882}
883
884void setup_secondary_APIC_clock(void)
885{
886 setup_APIC_timer();
887}
888
889/*
890 * The guts of the apic timer interrupt
891 */
892static void local_apic_timer_interrupt(void)
893{
894 int cpu = smp_processor_id();
895 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
896
897 /*
898 * Normally we should not be here till LAPIC has been initialized but
899 * in some cases like kdump, its possible that there is a pending LAPIC
900 * timer interrupt from previous kernel's context and is delivered in
901 * new kernel the moment interrupts are enabled.
902 *
903 * Interrupts are enabled early and LAPIC is setup much later, hence
904 * its possible that when we get here evt->event_handler is NULL.
905 * Check for event_handler being NULL and discard the interrupt as
906 * spurious.
907 */
908 if (!evt->event_handler) {
909 pr_warning("Spurious LAPIC timer interrupt on cpu %d\n", cpu);
910 /* Switch it off */
911 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
912 return;
913 }
914
915 /*
916 * the NMI deadlock-detector uses this.
917 */
918 inc_irq_stat(apic_timer_irqs);
919
920 evt->event_handler(evt);
921}
922
923/*
924 * Local APIC timer interrupt. This is the most natural way for doing
925 * local interrupts, but local timer interrupts can be emulated by
926 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
927 *
928 * [ if a single-CPU system runs an SMP kernel then we call the local
929 * interrupt as well. Thus we cannot inline the local irq ... ]
930 */
931__visible void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs)
932{
933 struct pt_regs *old_regs = set_irq_regs(regs);
934
935 /*
936 * NOTE! We'd better ACK the irq immediately,
937 * because timer handling can be slow.
938 *
939 * update_process_times() expects us to have done irq_enter().
940 * Besides, if we don't timer interrupts ignore the global
941 * interrupt lock, which is the WrongThing (tm) to do.
942 */
943 entering_ack_irq();
944 local_apic_timer_interrupt();
945 exiting_irq();
946
947 set_irq_regs(old_regs);
948}
949
950__visible void __irq_entry smp_trace_apic_timer_interrupt(struct pt_regs *regs)
951{
952 struct pt_regs *old_regs = set_irq_regs(regs);
953
954 /*
955 * NOTE! We'd better ACK the irq immediately,
956 * because timer handling can be slow.
957 *
958 * update_process_times() expects us to have done irq_enter().
959 * Besides, if we don't timer interrupts ignore the global
960 * interrupt lock, which is the WrongThing (tm) to do.
961 */
962 entering_ack_irq();
963 trace_local_timer_entry(LOCAL_TIMER_VECTOR);
964 local_apic_timer_interrupt();
965 trace_local_timer_exit(LOCAL_TIMER_VECTOR);
966 exiting_irq();
967
968 set_irq_regs(old_regs);
969}
970
971int setup_profiling_timer(unsigned int multiplier)
972{
973 return -EINVAL;
974}
975
976/*
977 * Local APIC start and shutdown
978 */
979
980/**
981 * clear_local_APIC - shutdown the local APIC
982 *
983 * This is called, when a CPU is disabled and before rebooting, so the state of
984 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
985 * leftovers during boot.
986 */
987void clear_local_APIC(void)
988{
989 int maxlvt;
990 u32 v;
991
992 /* APIC hasn't been mapped yet */
993 if (!x2apic_mode && !apic_phys)
994 return;
995
996 maxlvt = lapic_get_maxlvt();
997 /*
998 * Masking an LVT entry can trigger a local APIC error
999 * if the vector is zero. Mask LVTERR first to prevent this.
1000 */
1001 if (maxlvt >= 3) {
1002 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
1003 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
1004 }
1005 /*
1006 * Careful: we have to set masks only first to deassert
1007 * any level-triggered sources.
1008 */
1009 v = apic_read(APIC_LVTT);
1010 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
1011 v = apic_read(APIC_LVT0);
1012 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1013 v = apic_read(APIC_LVT1);
1014 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
1015 if (maxlvt >= 4) {
1016 v = apic_read(APIC_LVTPC);
1017 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
1018 }
1019
1020 /* lets not touch this if we didn't frob it */
1021#ifdef CONFIG_X86_THERMAL_VECTOR
1022 if (maxlvt >= 5) {
1023 v = apic_read(APIC_LVTTHMR);
1024 apic_write(APIC_LVTTHMR, v | APIC_LVT_MASKED);
1025 }
1026#endif
1027#ifdef CONFIG_X86_MCE_INTEL
1028 if (maxlvt >= 6) {
1029 v = apic_read(APIC_LVTCMCI);
1030 if (!(v & APIC_LVT_MASKED))
1031 apic_write(APIC_LVTCMCI, v | APIC_LVT_MASKED);
1032 }
1033#endif
1034
1035 /*
1036 * Clean APIC state for other OSs:
1037 */
1038 apic_write(APIC_LVTT, APIC_LVT_MASKED);
1039 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1040 apic_write(APIC_LVT1, APIC_LVT_MASKED);
1041 if (maxlvt >= 3)
1042 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
1043 if (maxlvt >= 4)
1044 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
1045
1046 /* Integrated APIC (!82489DX) ? */
1047 if (lapic_is_integrated()) {
1048 if (maxlvt > 3)
1049 /* Clear ESR due to Pentium errata 3AP and 11AP */
1050 apic_write(APIC_ESR, 0);
1051 apic_read(APIC_ESR);
1052 }
1053}
1054
1055/**
1056 * disable_local_APIC - clear and disable the local APIC
1057 */
1058void disable_local_APIC(void)
1059{
1060 unsigned int value;
1061
1062 /* APIC hasn't been mapped yet */
1063 if (!x2apic_mode && !apic_phys)
1064 return;
1065
1066 clear_local_APIC();
1067
1068 /*
1069 * Disable APIC (implies clearing of registers
1070 * for 82489DX!).
1071 */
1072 value = apic_read(APIC_SPIV);
1073 value &= ~APIC_SPIV_APIC_ENABLED;
1074 apic_write(APIC_SPIV, value);
1075
1076#ifdef CONFIG_X86_32
1077 /*
1078 * When LAPIC was disabled by the BIOS and enabled by the kernel,
1079 * restore the disabled state.
1080 */
1081 if (enabled_via_apicbase) {
1082 unsigned int l, h;
1083
1084 rdmsr(MSR_IA32_APICBASE, l, h);
1085 l &= ~MSR_IA32_APICBASE_ENABLE;
1086 wrmsr(MSR_IA32_APICBASE, l, h);
1087 }
1088#endif
1089}
1090
1091/*
1092 * If Linux enabled the LAPIC against the BIOS default disable it down before
1093 * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and
1094 * not power-off. Additionally clear all LVT entries before disable_local_APIC
1095 * for the case where Linux didn't enable the LAPIC.
1096 */
1097void lapic_shutdown(void)
1098{
1099 unsigned long flags;
1100
1101 if (!cpu_has_apic && !apic_from_smp_config())
1102 return;
1103
1104 local_irq_save(flags);
1105
1106#ifdef CONFIG_X86_32
1107 if (!enabled_via_apicbase)
1108 clear_local_APIC();
1109 else
1110#endif
1111 disable_local_APIC();
1112
1113
1114 local_irq_restore(flags);
1115}
1116
1117/*
1118 * This is to verify that we're looking at a real local APIC.
1119 * Check these against your board if the CPUs aren't getting
1120 * started for no apparent reason.
1121 */
1122int __init verify_local_APIC(void)
1123{
1124 unsigned int reg0, reg1;
1125
1126 /*
1127 * The version register is read-only in a real APIC.
1128 */
1129 reg0 = apic_read(APIC_LVR);
1130 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
1131 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
1132 reg1 = apic_read(APIC_LVR);
1133 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
1134
1135 /*
1136 * The two version reads above should print the same
1137 * numbers. If the second one is different, then we
1138 * poke at a non-APIC.
1139 */
1140 if (reg1 != reg0)
1141 return 0;
1142
1143 /*
1144 * Check if the version looks reasonably.
1145 */
1146 reg1 = GET_APIC_VERSION(reg0);
1147 if (reg1 == 0x00 || reg1 == 0xff)
1148 return 0;
1149 reg1 = lapic_get_maxlvt();
1150 if (reg1 < 0x02 || reg1 == 0xff)
1151 return 0;
1152
1153 /*
1154 * The ID register is read/write in a real APIC.
1155 */
1156 reg0 = apic_read(APIC_ID);
1157 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
1158 apic_write(APIC_ID, reg0 ^ apic->apic_id_mask);
1159 reg1 = apic_read(APIC_ID);
1160 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
1161 apic_write(APIC_ID, reg0);
1162 if (reg1 != (reg0 ^ apic->apic_id_mask))
1163 return 0;
1164
1165 /*
1166 * The next two are just to see if we have sane values.
1167 * They're only really relevant if we're in Virtual Wire
1168 * compatibility mode, but most boxes are anymore.
1169 */
1170 reg0 = apic_read(APIC_LVT0);
1171 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
1172 reg1 = apic_read(APIC_LVT1);
1173 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
1174
1175 return 1;
1176}
1177
1178/**
1179 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
1180 */
1181void __init sync_Arb_IDs(void)
1182{
1183 /*
1184 * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
1185 * needed on AMD.
1186 */
1187 if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
1188 return;
1189
1190 /*
1191 * Wait for idle.
1192 */
1193 apic_wait_icr_idle();
1194
1195 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
1196 apic_write(APIC_ICR, APIC_DEST_ALLINC |
1197 APIC_INT_LEVELTRIG | APIC_DM_INIT);
1198}
1199
1200/*
1201 * An initial setup of the virtual wire mode.
1202 */
1203void __init init_bsp_APIC(void)
1204{
1205 unsigned int value;
1206
1207 /*
1208 * Don't do the setup now if we have a SMP BIOS as the
1209 * through-I/O-APIC virtual wire mode might be active.
1210 */
1211 if (smp_found_config || !cpu_has_apic)
1212 return;
1213
1214 /*
1215 * Do not trust the local APIC being empty at bootup.
1216 */
1217 clear_local_APIC();
1218
1219 /*
1220 * Enable APIC.
1221 */
1222 value = apic_read(APIC_SPIV);
1223 value &= ~APIC_VECTOR_MASK;
1224 value |= APIC_SPIV_APIC_ENABLED;
1225
1226#ifdef CONFIG_X86_32
1227 /* This bit is reserved on P4/Xeon and should be cleared */
1228 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
1229 (boot_cpu_data.x86 == 15))
1230 value &= ~APIC_SPIV_FOCUS_DISABLED;
1231 else
1232#endif
1233 value |= APIC_SPIV_FOCUS_DISABLED;
1234 value |= SPURIOUS_APIC_VECTOR;
1235 apic_write(APIC_SPIV, value);
1236
1237 /*
1238 * Set up the virtual wire mode.
1239 */
1240 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1241 value = APIC_DM_NMI;
1242 if (!lapic_is_integrated()) /* 82489DX */
1243 value |= APIC_LVT_LEVEL_TRIGGER;
1244 apic_write(APIC_LVT1, value);
1245}
1246
1247static void lapic_setup_esr(void)
1248{
1249 unsigned int oldvalue, value, maxlvt;
1250
1251 if (!lapic_is_integrated()) {
1252 pr_info("No ESR for 82489DX.\n");
1253 return;
1254 }
1255
1256 if (apic->disable_esr) {
1257 /*
1258 * Something untraceable is creating bad interrupts on
1259 * secondary quads ... for the moment, just leave the
1260 * ESR disabled - we can't do anything useful with the
1261 * errors anyway - mbligh
1262 */
1263 pr_info("Leaving ESR disabled.\n");
1264 return;
1265 }
1266
1267 maxlvt = lapic_get_maxlvt();
1268 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1269 apic_write(APIC_ESR, 0);
1270 oldvalue = apic_read(APIC_ESR);
1271
1272 /* enables sending errors */
1273 value = ERROR_APIC_VECTOR;
1274 apic_write(APIC_LVTERR, value);
1275
1276 /*
1277 * spec says clear errors after enabling vector.
1278 */
1279 if (maxlvt > 3)
1280 apic_write(APIC_ESR, 0);
1281 value = apic_read(APIC_ESR);
1282 if (value != oldvalue)
1283 apic_printk(APIC_VERBOSE, "ESR value before enabling "
1284 "vector: 0x%08x after: 0x%08x\n",
1285 oldvalue, value);
1286}
1287
1288/**
1289 * setup_local_APIC - setup the local APIC
1290 *
1291 * Used to setup local APIC while initializing BSP or bringin up APs.
1292 * Always called with preemption disabled.
1293 */
1294void setup_local_APIC(void)
1295{
1296 int cpu = smp_processor_id();
1297 unsigned int value, queued;
1298 int i, j, acked = 0;
1299 unsigned long long tsc = 0, ntsc;
1300 long long max_loops = cpu_khz;
1301
1302 if (cpu_has_tsc)
1303 rdtscll(tsc);
1304
1305 if (disable_apic) {
1306 disable_ioapic_support();
1307 return;
1308 }
1309
1310#ifdef CONFIG_X86_32
1311 /* Pound the ESR really hard over the head with a big hammer - mbligh */
1312 if (lapic_is_integrated() && apic->disable_esr) {
1313 apic_write(APIC_ESR, 0);
1314 apic_write(APIC_ESR, 0);
1315 apic_write(APIC_ESR, 0);
1316 apic_write(APIC_ESR, 0);
1317 }
1318#endif
1319 perf_events_lapic_init();
1320
1321 /*
1322 * Double-check whether this APIC is really registered.
1323 * This is meaningless in clustered apic mode, so we skip it.
1324 */
1325 BUG_ON(!apic->apic_id_registered());
1326
1327 /*
1328 * Intel recommends to set DFR, LDR and TPR before enabling
1329 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
1330 * document number 292116). So here it goes...
1331 */
1332 apic->init_apic_ldr();
1333
1334#ifdef CONFIG_X86_32
1335 /*
1336 * APIC LDR is initialized. If logical_apicid mapping was
1337 * initialized during get_smp_config(), make sure it matches the
1338 * actual value.
1339 */
1340 i = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
1341 WARN_ON(i != BAD_APICID && i != logical_smp_processor_id());
1342 /* always use the value from LDR */
1343 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
1344 logical_smp_processor_id();
1345
1346 /*
1347 * Some NUMA implementations (NUMAQ) don't initialize apicid to
1348 * node mapping during NUMA init. Now that logical apicid is
1349 * guaranteed to be known, give it another chance. This is already
1350 * a bit too late - percpu allocation has already happened without
1351 * proper NUMA affinity.
1352 */
1353 if (apic->x86_32_numa_cpu_node)
1354 set_apicid_to_node(early_per_cpu(x86_cpu_to_apicid, cpu),
1355 apic->x86_32_numa_cpu_node(cpu));
1356#endif
1357
1358 /*
1359 * Set Task Priority to 'accept all'. We never change this
1360 * later on.
1361 */
1362 value = apic_read(APIC_TASKPRI);
1363 value &= ~APIC_TPRI_MASK;
1364 apic_write(APIC_TASKPRI, value);
1365
1366 /*
1367 * After a crash, we no longer service the interrupts and a pending
1368 * interrupt from previous kernel might still have ISR bit set.
1369 *
1370 * Most probably by now CPU has serviced that pending interrupt and
1371 * it might not have done the ack_APIC_irq() because it thought,
1372 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
1373 * does not clear the ISR bit and cpu thinks it has already serivced
1374 * the interrupt. Hence a vector might get locked. It was noticed
1375 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
1376 */
1377 do {
1378 queued = 0;
1379 for (i = APIC_ISR_NR - 1; i >= 0; i--)
1380 queued |= apic_read(APIC_IRR + i*0x10);
1381
1382 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
1383 value = apic_read(APIC_ISR + i*0x10);
1384 for (j = 31; j >= 0; j--) {
1385 if (value & (1<<j)) {
1386 ack_APIC_irq();
1387 acked++;
1388 }
1389 }
1390 }
1391 if (acked > 256) {
1392 printk(KERN_ERR "LAPIC pending interrupts after %d EOI\n",
1393 acked);
1394 break;
1395 }
1396 if (queued) {
1397 if (cpu_has_tsc) {
1398 rdtscll(ntsc);
1399 max_loops = (cpu_khz << 10) - (ntsc - tsc);
1400 } else
1401 max_loops--;
1402 }
1403 } while (queued && max_loops > 0);
1404 WARN_ON(max_loops <= 0);
1405
1406 /*
1407 * Now that we are all set up, enable the APIC
1408 */
1409 value = apic_read(APIC_SPIV);
1410 value &= ~APIC_VECTOR_MASK;
1411 /*
1412 * Enable APIC
1413 */
1414 value |= APIC_SPIV_APIC_ENABLED;
1415
1416#ifdef CONFIG_X86_32
1417 /*
1418 * Some unknown Intel IO/APIC (or APIC) errata is biting us with
1419 * certain networking cards. If high frequency interrupts are
1420 * happening on a particular IOAPIC pin, plus the IOAPIC routing
1421 * entry is masked/unmasked at a high rate as well then sooner or
1422 * later IOAPIC line gets 'stuck', no more interrupts are received
1423 * from the device. If focus CPU is disabled then the hang goes
1424 * away, oh well :-(
1425 *
1426 * [ This bug can be reproduced easily with a level-triggered
1427 * PCI Ne2000 networking cards and PII/PIII processors, dual
1428 * BX chipset. ]
1429 */
1430 /*
1431 * Actually disabling the focus CPU check just makes the hang less
1432 * frequent as it makes the interrupt distributon model be more
1433 * like LRU than MRU (the short-term load is more even across CPUs).
1434 * See also the comment in end_level_ioapic_irq(). --macro
1435 */
1436
1437 /*
1438 * - enable focus processor (bit==0)
1439 * - 64bit mode always use processor focus
1440 * so no need to set it
1441 */
1442 value &= ~APIC_SPIV_FOCUS_DISABLED;
1443#endif
1444
1445 /*
1446 * Set spurious IRQ vector
1447 */
1448 value |= SPURIOUS_APIC_VECTOR;
1449 apic_write(APIC_SPIV, value);
1450
1451 /*
1452 * Set up LVT0, LVT1:
1453 *
1454 * set up through-local-APIC on the BP's LINT0. This is not
1455 * strictly necessary in pure symmetric-IO mode, but sometimes
1456 * we delegate interrupts to the 8259A.
1457 */
1458 /*
1459 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
1460 */
1461 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
1462 if (!cpu && (pic_mode || !value)) {
1463 value = APIC_DM_EXTINT;
1464 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", cpu);
1465 } else {
1466 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
1467 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", cpu);
1468 }
1469 apic_write(APIC_LVT0, value);
1470
1471 /*
1472 * only the BP should see the LINT1 NMI signal, obviously.
1473 */
1474 if (!cpu)
1475 value = APIC_DM_NMI;
1476 else
1477 value = APIC_DM_NMI | APIC_LVT_MASKED;
1478 if (!lapic_is_integrated()) /* 82489DX */
1479 value |= APIC_LVT_LEVEL_TRIGGER;
1480 apic_write(APIC_LVT1, value);
1481
1482#ifdef CONFIG_X86_MCE_INTEL
1483 /* Recheck CMCI information after local APIC is up on CPU #0 */
1484 if (!cpu)
1485 cmci_recheck();
1486#endif
1487}
1488
1489void end_local_APIC_setup(void)
1490{
1491 lapic_setup_esr();
1492
1493#ifdef CONFIG_X86_32
1494 {
1495 unsigned int value;
1496 /* Disable the local apic timer */
1497 value = apic_read(APIC_LVTT);
1498 value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
1499 apic_write(APIC_LVTT, value);
1500 }
1501#endif
1502
1503 apic_pm_activate();
1504}
1505
1506void __init bsp_end_local_APIC_setup(void)
1507{
1508 end_local_APIC_setup();
1509
1510 /*
1511 * Now that local APIC setup is completed for BP, configure the fault
1512 * handling for interrupt remapping.
1513 */
1514 irq_remap_enable_fault_handling();
1515
1516}
1517
1518#ifdef CONFIG_X86_X2APIC
1519/*
1520 * Need to disable xapic and x2apic at the same time and then enable xapic mode
1521 */
1522static inline void __disable_x2apic(u64 msr)
1523{
1524 wrmsrl(MSR_IA32_APICBASE,
1525 msr & ~(X2APIC_ENABLE | XAPIC_ENABLE));
1526 wrmsrl(MSR_IA32_APICBASE, msr & ~X2APIC_ENABLE);
1527}
1528
1529static __init void disable_x2apic(void)
1530{
1531 u64 msr;
1532
1533 if (!cpu_has_x2apic)
1534 return;
1535
1536 rdmsrl(MSR_IA32_APICBASE, msr);
1537 if (msr & X2APIC_ENABLE) {
1538 u32 x2apic_id = read_apic_id();
1539
1540 if (x2apic_id >= 255)
1541 panic("Cannot disable x2apic, id: %08x\n", x2apic_id);
1542
1543 pr_info("Disabling x2apic\n");
1544 __disable_x2apic(msr);
1545
1546 if (nox2apic) {
1547 clear_cpu_cap(&cpu_data(0), X86_FEATURE_X2APIC);
1548 setup_clear_cpu_cap(X86_FEATURE_X2APIC);
1549 }
1550
1551 x2apic_disabled = 1;
1552 x2apic_mode = 0;
1553
1554 register_lapic_address(mp_lapic_addr);
1555 }
1556}
1557
1558void check_x2apic(void)
1559{
1560 if (x2apic_enabled()) {
1561 pr_info("x2apic enabled by BIOS, switching to x2apic ops\n");
1562 x2apic_preenabled = x2apic_mode = 1;
1563 }
1564}
1565
1566void enable_x2apic(void)
1567{
1568 u64 msr;
1569
1570 rdmsrl(MSR_IA32_APICBASE, msr);
1571 if (x2apic_disabled) {
1572 __disable_x2apic(msr);
1573 return;
1574 }
1575
1576 if (!x2apic_mode)
1577 return;
1578
1579 if (!(msr & X2APIC_ENABLE)) {
1580 printk_once(KERN_INFO "Enabling x2apic\n");
1581 wrmsrl(MSR_IA32_APICBASE, msr | X2APIC_ENABLE);
1582 }
1583}
1584#endif /* CONFIG_X86_X2APIC */
1585
1586int __init enable_IR(void)
1587{
1588#ifdef CONFIG_IRQ_REMAP
1589 if (!irq_remapping_supported()) {
1590 pr_debug("intr-remapping not supported\n");
1591 return -1;
1592 }
1593
1594 if (!x2apic_preenabled && skip_ioapic_setup) {
1595 pr_info("Skipped enabling intr-remap because of skipping "
1596 "io-apic setup\n");
1597 return -1;
1598 }
1599
1600 return irq_remapping_enable();
1601#endif
1602 return -1;
1603}
1604
1605void __init enable_IR_x2apic(void)
1606{
1607 unsigned long flags;
1608 int ret, x2apic_enabled = 0;
1609 int hardware_init_ret;
1610
1611 /* Make sure irq_remap_ops are initialized */
1612 setup_irq_remapping_ops();
1613
1614 hardware_init_ret = irq_remapping_prepare();
1615 if (hardware_init_ret && !x2apic_supported())
1616 return;
1617
1618 ret = save_ioapic_entries();
1619 if (ret) {
1620 pr_info("Saving IO-APIC state failed: %d\n", ret);
1621 return;
1622 }
1623
1624 local_irq_save(flags);
1625 legacy_pic->mask_all();
1626 mask_ioapic_entries();
1627
1628 if (x2apic_preenabled && nox2apic)
1629 disable_x2apic();
1630
1631 if (hardware_init_ret)
1632 ret = -1;
1633 else
1634 ret = enable_IR();
1635
1636 if (!x2apic_supported())
1637 goto skip_x2apic;
1638
1639 if (ret < 0) {
1640 /* IR is required if there is APIC ID > 255 even when running
1641 * under KVM
1642 */
1643 if (max_physical_apicid > 255 ||
1644 !hypervisor_x2apic_available()) {
1645 if (x2apic_preenabled)
1646 disable_x2apic();
1647 goto skip_x2apic;
1648 }
1649 /*
1650 * without IR all CPUs can be addressed by IOAPIC/MSI
1651 * only in physical mode
1652 */
1653 x2apic_force_phys();
1654 }
1655
1656 if (ret == IRQ_REMAP_XAPIC_MODE) {
1657 pr_info("x2apic not enabled, IRQ remapping is in xapic mode\n");
1658 goto skip_x2apic;
1659 }
1660
1661 x2apic_enabled = 1;
1662
1663 if (x2apic_supported() && !x2apic_mode) {
1664 x2apic_mode = 1;
1665 enable_x2apic();
1666 pr_info("Enabled x2apic\n");
1667 }
1668
1669skip_x2apic:
1670 if (ret < 0) /* IR enabling failed */
1671 restore_ioapic_entries();
1672 legacy_pic->restore_mask();
1673 local_irq_restore(flags);
1674}
1675
1676#ifdef CONFIG_X86_64
1677/*
1678 * Detect and enable local APICs on non-SMP boards.
1679 * Original code written by Keir Fraser.
1680 * On AMD64 we trust the BIOS - if it says no APIC it is likely
1681 * not correctly set up (usually the APIC timer won't work etc.)
1682 */
1683static int __init detect_init_APIC(void)
1684{
1685 if (!cpu_has_apic) {
1686 pr_info("No local APIC present\n");
1687 return -1;
1688 }
1689
1690 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1691 return 0;
1692}
1693#else
1694
1695static int __init apic_verify(void)
1696{
1697 u32 features, h, l;
1698
1699 /*
1700 * The APIC feature bit should now be enabled
1701 * in `cpuid'
1702 */
1703 features = cpuid_edx(1);
1704 if (!(features & (1 << X86_FEATURE_APIC))) {
1705 pr_warning("Could not enable APIC!\n");
1706 return -1;
1707 }
1708 set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1709 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
1710
1711 /* The BIOS may have set up the APIC at some other address */
1712 if (boot_cpu_data.x86 >= 6) {
1713 rdmsr(MSR_IA32_APICBASE, l, h);
1714 if (l & MSR_IA32_APICBASE_ENABLE)
1715 mp_lapic_addr = l & MSR_IA32_APICBASE_BASE;
1716 }
1717
1718 pr_info("Found and enabled local APIC!\n");
1719 return 0;
1720}
1721
1722int __init apic_force_enable(unsigned long addr)
1723{
1724 u32 h, l;
1725
1726 if (disable_apic)
1727 return -1;
1728
1729 /*
1730 * Some BIOSes disable the local APIC in the APIC_BASE
1731 * MSR. This can only be done in software for Intel P6 or later
1732 * and AMD K7 (Model > 1) or later.
1733 */
1734 if (boot_cpu_data.x86 >= 6) {
1735 rdmsr(MSR_IA32_APICBASE, l, h);
1736 if (!(l & MSR_IA32_APICBASE_ENABLE)) {
1737 pr_info("Local APIC disabled by BIOS -- reenabling.\n");
1738 l &= ~MSR_IA32_APICBASE_BASE;
1739 l |= MSR_IA32_APICBASE_ENABLE | addr;
1740 wrmsr(MSR_IA32_APICBASE, l, h);
1741 enabled_via_apicbase = 1;
1742 }
1743 }
1744 return apic_verify();
1745}
1746
1747/*
1748 * Detect and initialize APIC
1749 */
1750static int __init detect_init_APIC(void)
1751{
1752 /* Disabled by kernel option? */
1753 if (disable_apic)
1754 return -1;
1755
1756 switch (boot_cpu_data.x86_vendor) {
1757 case X86_VENDOR_AMD:
1758 if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
1759 (boot_cpu_data.x86 >= 15))
1760 break;
1761 goto no_apic;
1762 case X86_VENDOR_INTEL:
1763 if (boot_cpu_data.x86 == 6 || boot_cpu_data.x86 == 15 ||
1764 (boot_cpu_data.x86 == 5 && cpu_has_apic))
1765 break;
1766 goto no_apic;
1767 default:
1768 goto no_apic;
1769 }
1770
1771 if (!cpu_has_apic) {
1772 /*
1773 * Over-ride BIOS and try to enable the local APIC only if
1774 * "lapic" specified.
1775 */
1776 if (!force_enable_local_apic) {
1777 pr_info("Local APIC disabled by BIOS -- "
1778 "you can enable it with \"lapic\"\n");
1779 return -1;
1780 }
1781 if (apic_force_enable(APIC_DEFAULT_PHYS_BASE))
1782 return -1;
1783 } else {
1784 if (apic_verify())
1785 return -1;
1786 }
1787
1788 apic_pm_activate();
1789
1790 return 0;
1791
1792no_apic:
1793 pr_info("No local APIC present or hardware disabled\n");
1794 return -1;
1795}
1796#endif
1797
1798/**
1799 * init_apic_mappings - initialize APIC mappings
1800 */
1801void __init init_apic_mappings(void)
1802{
1803 unsigned int new_apicid;
1804
1805 if (x2apic_mode) {
1806 boot_cpu_physical_apicid = read_apic_id();
1807 return;
1808 }
1809
1810 /* If no local APIC can be found return early */
1811 if (!smp_found_config && detect_init_APIC()) {
1812 /* lets NOP'ify apic operations */
1813 pr_info("APIC: disable apic facility\n");
1814 apic_disable();
1815 } else {
1816 apic_phys = mp_lapic_addr;
1817
1818 /*
1819 * acpi lapic path already maps that address in
1820 * acpi_register_lapic_address()
1821 */
1822 if (!acpi_lapic && !smp_found_config)
1823 register_lapic_address(apic_phys);
1824 }
1825
1826 /*
1827 * Fetch the APIC ID of the BSP in case we have a
1828 * default configuration (or the MP table is broken).
1829 */
1830 new_apicid = read_apic_id();
1831 if (boot_cpu_physical_apicid != new_apicid) {
1832 boot_cpu_physical_apicid = new_apicid;
1833 /*
1834 * yeah -- we lie about apic_version
1835 * in case if apic was disabled via boot option
1836 * but it's not a problem for SMP compiled kernel
1837 * since smp_sanity_check is prepared for such a case
1838 * and disable smp mode
1839 */
1840 apic_version[new_apicid] =
1841 GET_APIC_VERSION(apic_read(APIC_LVR));
1842 }
1843}
1844
1845void __init register_lapic_address(unsigned long address)
1846{
1847 mp_lapic_addr = address;
1848
1849 if (!x2apic_mode) {
1850 set_fixmap_nocache(FIX_APIC_BASE, address);
1851 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
1852 APIC_BASE, mp_lapic_addr);
1853 }
1854 if (boot_cpu_physical_apicid == -1U) {
1855 boot_cpu_physical_apicid = read_apic_id();
1856 apic_version[boot_cpu_physical_apicid] =
1857 GET_APIC_VERSION(apic_read(APIC_LVR));
1858 }
1859}
1860
1861/*
1862 * This initializes the IO-APIC and APIC hardware if this is
1863 * a UP kernel.
1864 */
1865int apic_version[MAX_LOCAL_APIC];
1866
1867int __init APIC_init_uniprocessor(void)
1868{
1869 if (disable_apic) {
1870 pr_info("Apic disabled\n");
1871 return -1;
1872 }
1873#ifdef CONFIG_X86_64
1874 if (!cpu_has_apic) {
1875 disable_apic = 1;
1876 pr_info("Apic disabled by BIOS\n");
1877 return -1;
1878 }
1879#else
1880 if (!smp_found_config && !cpu_has_apic)
1881 return -1;
1882
1883 /*
1884 * Complain if the BIOS pretends there is one.
1885 */
1886 if (!cpu_has_apic &&
1887 APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
1888 pr_err("BIOS bug, local APIC 0x%x not detected!...\n",
1889 boot_cpu_physical_apicid);
1890 return -1;
1891 }
1892#endif
1893
1894 default_setup_apic_routing();
1895
1896 verify_local_APIC();
1897 connect_bsp_APIC();
1898
1899#ifdef CONFIG_X86_64
1900 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
1901#else
1902 /*
1903 * Hack: In case of kdump, after a crash, kernel might be booting
1904 * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
1905 * might be zero if read from MP tables. Get it from LAPIC.
1906 */
1907# ifdef CONFIG_CRASH_DUMP
1908 boot_cpu_physical_apicid = read_apic_id();
1909# endif
1910#endif
1911 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
1912 setup_local_APIC();
1913
1914#ifdef CONFIG_X86_IO_APIC
1915 /*
1916 * Now enable IO-APICs, actually call clear_IO_APIC
1917 * We need clear_IO_APIC before enabling error vector
1918 */
1919 if (!skip_ioapic_setup && nr_ioapics)
1920 enable_IO_APIC();
1921#endif
1922
1923 bsp_end_local_APIC_setup();
1924
1925#ifdef CONFIG_X86_IO_APIC
1926 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1927 setup_IO_APIC();
1928 else {
1929 nr_ioapics = 0;
1930 }
1931#endif
1932
1933 x86_init.timers.setup_percpu_clockev();
1934 return 0;
1935}
1936
1937/*
1938 * Local APIC interrupts
1939 */
1940
1941/*
1942 * This interrupt should _never_ happen with our APIC/SMP architecture
1943 */
1944static inline void __smp_spurious_interrupt(void)
1945{
1946 u32 v;
1947
1948 /*
1949 * Check if this really is a spurious interrupt and ACK it
1950 * if it is a vectored one. Just in case...
1951 * Spurious interrupts should not be ACKed.
1952 */
1953 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1954 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1955 ack_APIC_irq();
1956
1957 inc_irq_stat(irq_spurious_count);
1958
1959 /* see sw-dev-man vol 3, chapter 7.4.13.5 */
1960 pr_info("spurious APIC interrupt on CPU#%d, "
1961 "should never happen.\n", smp_processor_id());
1962}
1963
1964__visible void smp_spurious_interrupt(struct pt_regs *regs)
1965{
1966 entering_irq();
1967 __smp_spurious_interrupt();
1968 exiting_irq();
1969}
1970
1971__visible void smp_trace_spurious_interrupt(struct pt_regs *regs)
1972{
1973 entering_irq();
1974 trace_spurious_apic_entry(SPURIOUS_APIC_VECTOR);
1975 __smp_spurious_interrupt();
1976 trace_spurious_apic_exit(SPURIOUS_APIC_VECTOR);
1977 exiting_irq();
1978}
1979
1980/*
1981 * This interrupt should never happen with our APIC/SMP architecture
1982 */
1983static inline void __smp_error_interrupt(struct pt_regs *regs)
1984{
1985 u32 v;
1986 u32 i = 0;
1987 static const char * const error_interrupt_reason[] = {
1988 "Send CS error", /* APIC Error Bit 0 */
1989 "Receive CS error", /* APIC Error Bit 1 */
1990 "Send accept error", /* APIC Error Bit 2 */
1991 "Receive accept error", /* APIC Error Bit 3 */
1992 "Redirectable IPI", /* APIC Error Bit 4 */
1993 "Send illegal vector", /* APIC Error Bit 5 */
1994 "Received illegal vector", /* APIC Error Bit 6 */
1995 "Illegal register address", /* APIC Error Bit 7 */
1996 };
1997
1998 /* First tickle the hardware, only then report what went on. -- REW */
1999 if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */
2000 apic_write(APIC_ESR, 0);
2001 v = apic_read(APIC_ESR);
2002 ack_APIC_irq();
2003 atomic_inc(&irq_err_count);
2004
2005 apic_printk(APIC_DEBUG, KERN_DEBUG "APIC error on CPU%d: %02x",
2006 smp_processor_id(), v);
2007
2008 v &= 0xff;
2009 while (v) {
2010 if (v & 0x1)
2011 apic_printk(APIC_DEBUG, KERN_CONT " : %s", error_interrupt_reason[i]);
2012 i++;
2013 v >>= 1;
2014 }
2015
2016 apic_printk(APIC_DEBUG, KERN_CONT "\n");
2017
2018}
2019
2020__visible void smp_error_interrupt(struct pt_regs *regs)
2021{
2022 entering_irq();
2023 __smp_error_interrupt(regs);
2024 exiting_irq();
2025}
2026
2027__visible void smp_trace_error_interrupt(struct pt_regs *regs)
2028{
2029 entering_irq();
2030 trace_error_apic_entry(ERROR_APIC_VECTOR);
2031 __smp_error_interrupt(regs);
2032 trace_error_apic_exit(ERROR_APIC_VECTOR);
2033 exiting_irq();
2034}
2035
2036/**
2037 * connect_bsp_APIC - attach the APIC to the interrupt system
2038 */
2039void __init connect_bsp_APIC(void)
2040{
2041#ifdef CONFIG_X86_32
2042 if (pic_mode) {
2043 /*
2044 * Do not trust the local APIC being empty at bootup.
2045 */
2046 clear_local_APIC();
2047 /*
2048 * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's
2049 * local APIC to INT and NMI lines.
2050 */
2051 apic_printk(APIC_VERBOSE, "leaving PIC mode, "
2052 "enabling APIC mode.\n");
2053 imcr_pic_to_apic();
2054 }
2055#endif
2056 if (apic->enable_apic_mode)
2057 apic->enable_apic_mode();
2058}
2059
2060/**
2061 * disconnect_bsp_APIC - detach the APIC from the interrupt system
2062 * @virt_wire_setup: indicates, whether virtual wire mode is selected
2063 *
2064 * Virtual wire mode is necessary to deliver legacy interrupts even when the
2065 * APIC is disabled.
2066 */
2067void disconnect_bsp_APIC(int virt_wire_setup)
2068{
2069 unsigned int value;
2070
2071#ifdef CONFIG_X86_32
2072 if (pic_mode) {
2073 /*
2074 * Put the board back into PIC mode (has an effect only on
2075 * certain older boards). Note that APIC interrupts, including
2076 * IPIs, won't work beyond this point! The only exception are
2077 * INIT IPIs.
2078 */
2079 apic_printk(APIC_VERBOSE, "disabling APIC mode, "
2080 "entering PIC mode.\n");
2081 imcr_apic_to_pic();
2082 return;
2083 }
2084#endif
2085
2086 /* Go back to Virtual Wire compatibility mode */
2087
2088 /* For the spurious interrupt use vector F, and enable it */
2089 value = apic_read(APIC_SPIV);
2090 value &= ~APIC_VECTOR_MASK;
2091 value |= APIC_SPIV_APIC_ENABLED;
2092 value |= 0xf;
2093 apic_write(APIC_SPIV, value);
2094
2095 if (!virt_wire_setup) {
2096 /*
2097 * For LVT0 make it edge triggered, active high,
2098 * external and enabled
2099 */
2100 value = apic_read(APIC_LVT0);
2101 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2102 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2103 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2104 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2105 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
2106 apic_write(APIC_LVT0, value);
2107 } else {
2108 /* Disable LVT0 */
2109 apic_write(APIC_LVT0, APIC_LVT_MASKED);
2110 }
2111
2112 /*
2113 * For LVT1 make it edge triggered, active high,
2114 * nmi and enabled
2115 */
2116 value = apic_read(APIC_LVT1);
2117 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
2118 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
2119 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
2120 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
2121 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
2122 apic_write(APIC_LVT1, value);
2123}
2124
2125int generic_processor_info(int apicid, int version)
2126{
2127 int cpu, max = nr_cpu_ids;
2128 bool boot_cpu_detected = physid_isset(boot_cpu_physical_apicid,
2129 phys_cpu_present_map);
2130
2131 /*
2132 * boot_cpu_physical_apicid is designed to have the apicid
2133 * returned by read_apic_id(), i.e, the apicid of the
2134 * currently booting-up processor. However, on some platforms,
2135 * it is temporarily modified by the apicid reported as BSP
2136 * through MP table. Concretely:
2137 *
2138 * - arch/x86/kernel/mpparse.c: MP_processor_info()
2139 * - arch/x86/mm/amdtopology.c: amd_numa_init()
2140 *
2141 * This function is executed with the modified
2142 * boot_cpu_physical_apicid. So, disabled_cpu_apicid kernel
2143 * parameter doesn't work to disable APs on kdump 2nd kernel.
2144 *
2145 * Since fixing handling of boot_cpu_physical_apicid requires
2146 * another discussion and tests on each platform, we leave it
2147 * for now and here we use read_apic_id() directly in this
2148 * function, generic_processor_info().
2149 */
2150 if (disabled_cpu_apicid != BAD_APICID &&
2151 disabled_cpu_apicid != read_apic_id() &&
2152 disabled_cpu_apicid == apicid) {
2153 int thiscpu = num_processors + disabled_cpus;
2154
2155 pr_warning("APIC: Disabling requested cpu."
2156 " Processor %d/0x%x ignored.\n",
2157 thiscpu, apicid);
2158
2159 disabled_cpus++;
2160 return -ENODEV;
2161 }
2162
2163 /*
2164 * If boot cpu has not been detected yet, then only allow upto
2165 * nr_cpu_ids - 1 processors and keep one slot free for boot cpu
2166 */
2167 if (!boot_cpu_detected && num_processors >= nr_cpu_ids - 1 &&
2168 apicid != boot_cpu_physical_apicid) {
2169 int thiscpu = max + disabled_cpus - 1;
2170
2171 pr_warning(
2172 "ACPI: NR_CPUS/possible_cpus limit of %i almost"
2173 " reached. Keeping one slot for boot cpu."
2174 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2175
2176 disabled_cpus++;
2177 return -ENODEV;
2178 }
2179
2180 if (num_processors >= nr_cpu_ids) {
2181 int thiscpu = max + disabled_cpus;
2182
2183 pr_warning(
2184 "ACPI: NR_CPUS/possible_cpus limit of %i reached."
2185 " Processor %d/0x%x ignored.\n", max, thiscpu, apicid);
2186
2187 disabled_cpus++;
2188 return -EINVAL;
2189 }
2190
2191 num_processors++;
2192 if (apicid == boot_cpu_physical_apicid) {
2193 /*
2194 * x86_bios_cpu_apicid is required to have processors listed
2195 * in same order as logical cpu numbers. Hence the first
2196 * entry is BSP, and so on.
2197 * boot_cpu_init() already hold bit 0 in cpu_present_mask
2198 * for BSP.
2199 */
2200 cpu = 0;
2201 } else
2202 cpu = cpumask_next_zero(-1, cpu_present_mask);
2203
2204 /*
2205 * Validate version
2206 */
2207 if (version == 0x0) {
2208 pr_warning("BIOS bug: APIC version is 0 for CPU %d/0x%x, fixing up to 0x10\n",
2209 cpu, apicid);
2210 version = 0x10;
2211 }
2212 apic_version[apicid] = version;
2213
2214 if (version != apic_version[boot_cpu_physical_apicid]) {
2215 pr_warning("BIOS bug: APIC version mismatch, boot CPU: %x, CPU %d: version %x\n",
2216 apic_version[boot_cpu_physical_apicid], cpu, version);
2217 }
2218
2219 physid_set(apicid, phys_cpu_present_map);
2220 if (apicid > max_physical_apicid)
2221 max_physical_apicid = apicid;
2222
2223#if defined(CONFIG_SMP) || defined(CONFIG_X86_64)
2224 early_per_cpu(x86_cpu_to_apicid, cpu) = apicid;
2225 early_per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
2226#endif
2227#ifdef CONFIG_X86_32
2228 early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
2229 apic->x86_32_early_logical_apicid(cpu);
2230#endif
2231 set_cpu_possible(cpu, true);
2232 set_cpu_present(cpu, true);
2233
2234 return cpu;
2235}
2236
2237int hard_smp_processor_id(void)
2238{
2239 return read_apic_id();
2240}
2241
2242void default_init_apic_ldr(void)
2243{
2244 unsigned long val;
2245
2246 apic_write(APIC_DFR, APIC_DFR_VALUE);
2247 val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
2248 val |= SET_APIC_LOGICAL_ID(1UL << smp_processor_id());
2249 apic_write(APIC_LDR, val);
2250}
2251
2252int default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
2253 const struct cpumask *andmask,
2254 unsigned int *apicid)
2255{
2256 unsigned int cpu;
2257
2258 for_each_cpu_and(cpu, cpumask, andmask) {
2259 if (cpumask_test_cpu(cpu, cpu_online_mask))
2260 break;
2261 }
2262
2263 if (likely(cpu < nr_cpu_ids)) {
2264 *apicid = per_cpu(x86_cpu_to_apicid, cpu);
2265 return 0;
2266 }
2267
2268 return -EINVAL;
2269}
2270
2271/*
2272 * Override the generic EOI implementation with an optimized version.
2273 * Only called during early boot when only one CPU is active and with
2274 * interrupts disabled, so we know this does not race with actual APIC driver
2275 * use.
2276 */
2277void __init apic_set_eoi_write(void (*eoi_write)(u32 reg, u32 v))
2278{
2279 struct apic **drv;
2280
2281 for (drv = __apicdrivers; drv < __apicdrivers_end; drv++) {
2282 /* Should happen once for each apic */
2283 WARN_ON((*drv)->eoi_write == eoi_write);
2284 (*drv)->eoi_write = eoi_write;
2285 }
2286}
2287
2288/*
2289 * Power management
2290 */
2291#ifdef CONFIG_PM
2292
2293static struct {
2294 /*
2295 * 'active' is true if the local APIC was enabled by us and
2296 * not the BIOS; this signifies that we are also responsible
2297 * for disabling it before entering apm/acpi suspend
2298 */
2299 int active;
2300 /* r/w apic fields */
2301 unsigned int apic_id;
2302 unsigned int apic_taskpri;
2303 unsigned int apic_ldr;
2304 unsigned int apic_dfr;
2305 unsigned int apic_spiv;
2306 unsigned int apic_lvtt;
2307 unsigned int apic_lvtpc;
2308 unsigned int apic_lvt0;
2309 unsigned int apic_lvt1;
2310 unsigned int apic_lvterr;
2311 unsigned int apic_tmict;
2312 unsigned int apic_tdcr;
2313 unsigned int apic_thmr;
2314} apic_pm_state;
2315
2316static int lapic_suspend(void)
2317{
2318 unsigned long flags;
2319 int maxlvt;
2320
2321 if (!apic_pm_state.active)
2322 return 0;
2323
2324 maxlvt = lapic_get_maxlvt();
2325
2326 apic_pm_state.apic_id = apic_read(APIC_ID);
2327 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
2328 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
2329 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
2330 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
2331 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
2332 if (maxlvt >= 4)
2333 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
2334 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
2335 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
2336 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
2337 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
2338 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
2339#ifdef CONFIG_X86_THERMAL_VECTOR
2340 if (maxlvt >= 5)
2341 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
2342#endif
2343
2344 local_irq_save(flags);
2345 disable_local_APIC();
2346
2347 irq_remapping_disable();
2348
2349 local_irq_restore(flags);
2350 return 0;
2351}
2352
2353static void lapic_resume(void)
2354{
2355 unsigned int l, h;
2356 unsigned long flags;
2357 int maxlvt;
2358
2359 if (!apic_pm_state.active)
2360 return;
2361
2362 local_irq_save(flags);
2363
2364 /*
2365 * IO-APIC and PIC have their own resume routines.
2366 * We just mask them here to make sure the interrupt
2367 * subsystem is completely quiet while we enable x2apic
2368 * and interrupt-remapping.
2369 */
2370 mask_ioapic_entries();
2371 legacy_pic->mask_all();
2372
2373 if (x2apic_mode)
2374 enable_x2apic();
2375 else {
2376 /*
2377 * Make sure the APICBASE points to the right address
2378 *
2379 * FIXME! This will be wrong if we ever support suspend on
2380 * SMP! We'll need to do this as part of the CPU restore!
2381 */
2382 if (boot_cpu_data.x86 >= 6) {
2383 rdmsr(MSR_IA32_APICBASE, l, h);
2384 l &= ~MSR_IA32_APICBASE_BASE;
2385 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
2386 wrmsr(MSR_IA32_APICBASE, l, h);
2387 }
2388 }
2389
2390 maxlvt = lapic_get_maxlvt();
2391 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
2392 apic_write(APIC_ID, apic_pm_state.apic_id);
2393 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
2394 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
2395 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
2396 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
2397 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
2398 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
2399#if defined(CONFIG_X86_MCE_INTEL)
2400 if (maxlvt >= 5)
2401 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
2402#endif
2403 if (maxlvt >= 4)
2404 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
2405 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
2406 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
2407 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
2408 apic_write(APIC_ESR, 0);
2409 apic_read(APIC_ESR);
2410 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
2411 apic_write(APIC_ESR, 0);
2412 apic_read(APIC_ESR);
2413
2414 irq_remapping_reenable(x2apic_mode);
2415
2416 local_irq_restore(flags);
2417}
2418
2419/*
2420 * This device has no shutdown method - fully functioning local APICs
2421 * are needed on every CPU up until machine_halt/restart/poweroff.
2422 */
2423
2424static struct syscore_ops lapic_syscore_ops = {
2425 .resume = lapic_resume,
2426 .suspend = lapic_suspend,
2427};
2428
2429static void apic_pm_activate(void)
2430{
2431 apic_pm_state.active = 1;
2432}
2433
2434static int __init init_lapic_sysfs(void)
2435{
2436 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
2437 if (cpu_has_apic)
2438 register_syscore_ops(&lapic_syscore_ops);
2439
2440 return 0;
2441}
2442
2443/* local apic needs to resume before other devices access its registers. */
2444core_initcall(init_lapic_sysfs);
2445
2446#else /* CONFIG_PM */
2447
2448static void apic_pm_activate(void) { }
2449
2450#endif /* CONFIG_PM */
2451
2452#ifdef CONFIG_X86_64
2453
2454static int apic_cluster_num(void)
2455{
2456 int i, clusters, zeros;
2457 unsigned id;
2458 u16 *bios_cpu_apicid;
2459 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
2460
2461 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
2462 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
2463
2464 for (i = 0; i < nr_cpu_ids; i++) {
2465 /* are we being called early in kernel startup? */
2466 if (bios_cpu_apicid) {
2467 id = bios_cpu_apicid[i];
2468 } else if (i < nr_cpu_ids) {
2469 if (cpu_present(i))
2470 id = per_cpu(x86_bios_cpu_apicid, i);
2471 else
2472 continue;
2473 } else
2474 break;
2475
2476 if (id != BAD_APICID)
2477 __set_bit(APIC_CLUSTERID(id), clustermap);
2478 }
2479
2480 /* Problem: Partially populated chassis may not have CPUs in some of
2481 * the APIC clusters they have been allocated. Only present CPUs have
2482 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
2483 * Since clusters are allocated sequentially, count zeros only if
2484 * they are bounded by ones.
2485 */
2486 clusters = 0;
2487 zeros = 0;
2488 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
2489 if (test_bit(i, clustermap)) {
2490 clusters += 1 + zeros;
2491 zeros = 0;
2492 } else
2493 ++zeros;
2494 }
2495
2496 return clusters;
2497}
2498
2499static int multi_checked;
2500static int multi;
2501
2502static int set_multi(const struct dmi_system_id *d)
2503{
2504 if (multi)
2505 return 0;
2506 pr_info("APIC: %s detected, Multi Chassis\n", d->ident);
2507 multi = 1;
2508 return 0;
2509}
2510
2511static const struct dmi_system_id multi_dmi_table[] = {
2512 {
2513 .callback = set_multi,
2514 .ident = "IBM System Summit2",
2515 .matches = {
2516 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
2517 DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"),
2518 },
2519 },
2520 {}
2521};
2522
2523static void dmi_check_multi(void)
2524{
2525 if (multi_checked)
2526 return;
2527
2528 dmi_check_system(multi_dmi_table);
2529 multi_checked = 1;
2530}
2531
2532/*
2533 * apic_is_clustered_box() -- Check if we can expect good TSC
2534 *
2535 * Thus far, the major user of this is IBM's Summit2 series:
2536 * Clustered boxes may have unsynced TSC problems if they are
2537 * multi-chassis.
2538 * Use DMI to check them
2539 */
2540int apic_is_clustered_box(void)
2541{
2542 dmi_check_multi();
2543 if (multi)
2544 return 1;
2545
2546 if (!is_vsmp_box())
2547 return 0;
2548
2549 /*
2550 * ScaleMP vSMPowered boxes have one cluster per board and TSCs are
2551 * not guaranteed to be synced between boards
2552 */
2553 if (apic_cluster_num() > 1)
2554 return 1;
2555
2556 return 0;
2557}
2558#endif
2559
2560/*
2561 * APIC command line parameters
2562 */
2563static int __init setup_disableapic(char *arg)
2564{
2565 disable_apic = 1;
2566 setup_clear_cpu_cap(X86_FEATURE_APIC);
2567 return 0;
2568}
2569early_param("disableapic", setup_disableapic);
2570
2571/* same as disableapic, for compatibility */
2572static int __init setup_nolapic(char *arg)
2573{
2574 return setup_disableapic(arg);
2575}
2576early_param("nolapic", setup_nolapic);
2577
2578static int __init parse_lapic_timer_c2_ok(char *arg)
2579{
2580 local_apic_timer_c2_ok = 1;
2581 return 0;
2582}
2583early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
2584
2585static int __init parse_disable_apic_timer(char *arg)
2586{
2587 disable_apic_timer = 1;
2588 return 0;
2589}
2590early_param("noapictimer", parse_disable_apic_timer);
2591
2592static int __init parse_nolapic_timer(char *arg)
2593{
2594 disable_apic_timer = 1;
2595 return 0;
2596}
2597early_param("nolapic_timer", parse_nolapic_timer);
2598
2599static int __init apic_set_verbosity(char *arg)
2600{
2601 if (!arg) {
2602#ifdef CONFIG_X86_64
2603 skip_ioapic_setup = 0;
2604 return 0;
2605#endif
2606 return -EINVAL;
2607 }
2608
2609 if (strcmp("debug", arg) == 0)
2610 apic_verbosity = APIC_DEBUG;
2611 else if (strcmp("verbose", arg) == 0)
2612 apic_verbosity = APIC_VERBOSE;
2613 else {
2614 pr_warning("APIC Verbosity level %s not recognised"
2615 " use apic=verbose or apic=debug\n", arg);
2616 return -EINVAL;
2617 }
2618
2619 return 0;
2620}
2621early_param("apic", apic_set_verbosity);
2622
2623static int __init lapic_insert_resource(void)
2624{
2625 if (!apic_phys)
2626 return -1;
2627
2628 /* Put local APIC into the resource map. */
2629 lapic_resource.start = apic_phys;
2630 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
2631 insert_resource(&iomem_resource, &lapic_resource);
2632
2633 return 0;
2634}
2635
2636/*
2637 * need call insert after e820_reserve_resources()
2638 * that is using request_resource
2639 */
2640late_initcall(lapic_insert_resource);
2641
2642static int __init apic_set_disabled_cpu_apicid(char *arg)
2643{
2644 if (!arg || !get_option(&arg, &disabled_cpu_apicid))
2645 return -EINVAL;
2646
2647 return 0;
2648}
2649early_param("disable_cpu_apicid", apic_set_disabled_cpu_apicid);