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/msr.h> |
63 | #include <asm/tsc.h> |
64 | #include <asm/hypervisor.h> |
65 | #include <asm/cpu_device_id.h> |
66 | #include <asm/intel-family.h> |
67 | #include <asm/irq_regs.h> |
68 | #include <asm/cpu.h> |
69 | |
70 | #include "local.h" |
71 | |
72 | /* Processor that is doing the boot up */ |
73 | u32 boot_cpu_physical_apicid __ro_after_init = BAD_APICID; |
74 | EXPORT_SYMBOL_GPL(boot_cpu_physical_apicid); |
75 | |
76 | u8 boot_cpu_apic_version __ro_after_init; |
77 | |
78 | /* |
79 | * This variable controls which CPUs receive external NMIs. By default, |
80 | * external NMIs are delivered only to the BSP. |
81 | */ |
82 | static int apic_extnmi __ro_after_init = APIC_EXTNMI_BSP; |
83 | |
84 | /* |
85 | * Hypervisor supports 15 bits of APIC ID in MSI Extended Destination ID |
86 | */ |
87 | static bool virt_ext_dest_id __ro_after_init; |
88 | |
89 | /* For parallel bootup. */ |
90 | unsigned long apic_mmio_base __ro_after_init; |
91 | |
92 | static inline bool apic_accessible(void) |
93 | { |
94 | return x2apic_mode || apic_mmio_base; |
95 | } |
96 | |
97 | #ifdef CONFIG_X86_32 |
98 | /* Local APIC was disabled by the BIOS and enabled by the kernel */ |
99 | static int enabled_via_apicbase __ro_after_init; |
100 | |
101 | /* |
102 | * Handle interrupt mode configuration register (IMCR). |
103 | * This register controls whether the interrupt signals |
104 | * that reach the BSP come from the master PIC or from the |
105 | * local APIC. Before entering Symmetric I/O Mode, either |
106 | * the BIOS or the operating system must switch out of |
107 | * PIC Mode by changing the IMCR. |
108 | */ |
109 | static inline void imcr_pic_to_apic(void) |
110 | { |
111 | /* NMI and 8259 INTR go through APIC */ |
112 | pc_conf_set(PC_CONF_MPS_IMCR, 0x01); |
113 | } |
114 | |
115 | static inline void imcr_apic_to_pic(void) |
116 | { |
117 | /* NMI and 8259 INTR go directly to BSP */ |
118 | pc_conf_set(PC_CONF_MPS_IMCR, 0x00); |
119 | } |
120 | #endif |
121 | |
122 | /* |
123 | * Knob to control our willingness to enable the local APIC. |
124 | * |
125 | * +1=force-enable |
126 | */ |
127 | static int force_enable_local_apic __initdata; |
128 | |
129 | /* |
130 | * APIC command line parameters |
131 | */ |
132 | static int __init parse_lapic(char *arg) |
133 | { |
134 | if (IS_ENABLED(CONFIG_X86_32) && !arg) |
135 | force_enable_local_apic = 1; |
136 | else if (arg && !strncmp(arg, "notscdeadline", 13)) |
137 | setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER); |
138 | return 0; |
139 | } |
140 | early_param("lapic", parse_lapic); |
141 | |
142 | #ifdef CONFIG_X86_64 |
143 | static int apic_calibrate_pmtmr __initdata; |
144 | static __init int setup_apicpmtimer(char *s) |
145 | { |
146 | apic_calibrate_pmtmr = 1; |
147 | notsc_setup(NULL); |
148 | return 1; |
149 | } |
150 | __setup("apicpmtimer", setup_apicpmtimer); |
151 | #endif |
152 | |
153 | static unsigned long mp_lapic_addr __ro_after_init; |
154 | bool apic_is_disabled __ro_after_init; |
155 | /* Disable local APIC timer from the kernel commandline or via dmi quirk */ |
156 | static int disable_apic_timer __initdata; |
157 | /* Local APIC timer works in C2 */ |
158 | int local_apic_timer_c2_ok __ro_after_init; |
159 | EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok); |
160 | |
161 | /* |
162 | * Debug level, exported for io_apic.c |
163 | */ |
164 | int apic_verbosity __ro_after_init; |
165 | |
166 | int pic_mode __ro_after_init; |
167 | |
168 | /* Have we found an MP table */ |
169 | int smp_found_config __ro_after_init; |
170 | |
171 | static struct resource lapic_resource = { |
172 | .name = "Local APIC", |
173 | .flags = IORESOURCE_MEM | IORESOURCE_BUSY, |
174 | }; |
175 | |
176 | unsigned int lapic_timer_period = 0; |
177 | |
178 | static void apic_pm_activate(void); |
179 | |
180 | /* |
181 | * Get the LAPIC version |
182 | */ |
183 | static inline int lapic_get_version(void) |
184 | { |
185 | return GET_APIC_VERSION(apic_read(APIC_LVR)); |
186 | } |
187 | |
188 | /* |
189 | * Check, if the APIC is integrated or a separate chip |
190 | */ |
191 | static inline int lapic_is_integrated(void) |
192 | { |
193 | return APIC_INTEGRATED(lapic_get_version()); |
194 | } |
195 | |
196 | /* |
197 | * Check, whether this is a modern or a first generation APIC |
198 | */ |
199 | static int modern_apic(void) |
200 | { |
201 | /* AMD systems use old APIC versions, so check the CPU */ |
202 | if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD && |
203 | boot_cpu_data.x86 >= 0xf) |
204 | return 1; |
205 | |
206 | /* Hygon systems use modern APIC */ |
207 | if (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) |
208 | return 1; |
209 | |
210 | return lapic_get_version() >= 0x14; |
211 | } |
212 | |
213 | /* |
214 | * right after this call apic become NOOP driven |
215 | * so apic->write/read doesn't do anything |
216 | */ |
217 | static void __init apic_disable(void) |
218 | { |
219 | apic_install_driver(driver: &apic_noop); |
220 | } |
221 | |
222 | void native_apic_icr_write(u32 low, u32 id) |
223 | { |
224 | unsigned long flags; |
225 | |
226 | local_irq_save(flags); |
227 | apic_write(APIC_ICR2, SET_XAPIC_DEST_FIELD(id)); |
228 | apic_write(APIC_ICR, val: low); |
229 | local_irq_restore(flags); |
230 | } |
231 | |
232 | u64 native_apic_icr_read(void) |
233 | { |
234 | u32 icr1, icr2; |
235 | |
236 | icr2 = apic_read(APIC_ICR2); |
237 | icr1 = apic_read(APIC_ICR); |
238 | |
239 | return icr1 | ((u64)icr2 << 32); |
240 | } |
241 | |
242 | /** |
243 | * lapic_get_maxlvt - get the maximum number of local vector table entries |
244 | */ |
245 | int lapic_get_maxlvt(void) |
246 | { |
247 | /* |
248 | * - we always have APIC integrated on 64bit mode |
249 | * - 82489DXs do not report # of LVT entries |
250 | */ |
251 | return lapic_is_integrated() ? GET_APIC_MAXLVT(apic_read(APIC_LVR)) : 2; |
252 | } |
253 | |
254 | /* |
255 | * Local APIC timer |
256 | */ |
257 | |
258 | /* Clock divisor */ |
259 | #define APIC_DIVISOR 16 |
260 | #define TSC_DIVISOR 8 |
261 | |
262 | /* i82489DX specific */ |
263 | #define I82489DX_BASE_DIVIDER (((0x2) << 18)) |
264 | |
265 | /* |
266 | * This function sets up the local APIC timer, with a timeout of |
267 | * 'clocks' APIC bus clock. During calibration we actually call |
268 | * this function twice on the boot CPU, once with a bogus timeout |
269 | * value, second time for real. The other (noncalibrating) CPUs |
270 | * call this function only once, with the real, calibrated value. |
271 | * |
272 | * We do reads before writes even if unnecessary, to get around the |
273 | * P5 APIC double write bug. |
274 | */ |
275 | static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen) |
276 | { |
277 | unsigned int lvtt_value, tmp_value; |
278 | |
279 | lvtt_value = LOCAL_TIMER_VECTOR; |
280 | if (!oneshot) |
281 | lvtt_value |= APIC_LVT_TIMER_PERIODIC; |
282 | else if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) |
283 | lvtt_value |= APIC_LVT_TIMER_TSCDEADLINE; |
284 | |
285 | /* |
286 | * The i82489DX APIC uses bit 18 and 19 for the base divider. This |
287 | * overlaps with bit 18 on integrated APICs, but is not documented |
288 | * in the SDM. No problem though. i82489DX equipped systems do not |
289 | * have TSC deadline timer. |
290 | */ |
291 | if (!lapic_is_integrated()) |
292 | lvtt_value |= I82489DX_BASE_DIVIDER; |
293 | |
294 | if (!irqen) |
295 | lvtt_value |= APIC_LVT_MASKED; |
296 | |
297 | apic_write(APIC_LVTT, val: lvtt_value); |
298 | |
299 | if (lvtt_value & APIC_LVT_TIMER_TSCDEADLINE) { |
300 | /* |
301 | * See Intel SDM: TSC-Deadline Mode chapter. In xAPIC mode, |
302 | * writing to the APIC LVTT and TSC_DEADLINE MSR isn't serialized. |
303 | * According to Intel, MFENCE can do the serialization here. |
304 | */ |
305 | asm volatile("mfence": : : "memory"); |
306 | return; |
307 | } |
308 | |
309 | /* |
310 | * Divide PICLK by 16 |
311 | */ |
312 | tmp_value = apic_read(APIC_TDCR); |
313 | apic_write(APIC_TDCR, |
314 | val: (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) | |
315 | APIC_TDR_DIV_16); |
316 | |
317 | if (!oneshot) |
318 | apic_write(APIC_TMICT, val: clocks / APIC_DIVISOR); |
319 | } |
320 | |
321 | /* |
322 | * Setup extended LVT, AMD specific |
323 | * |
324 | * Software should use the LVT offsets the BIOS provides. The offsets |
325 | * are determined by the subsystems using it like those for MCE |
326 | * threshold or IBS. On K8 only offset 0 (APIC500) and MCE interrupts |
327 | * are supported. Beginning with family 10h at least 4 offsets are |
328 | * available. |
329 | * |
330 | * Since the offsets must be consistent for all cores, we keep track |
331 | * of the LVT offsets in software and reserve the offset for the same |
332 | * vector also to be used on other cores. An offset is freed by |
333 | * setting the entry to APIC_EILVT_MASKED. |
334 | * |
335 | * If the BIOS is right, there should be no conflicts. Otherwise a |
336 | * "[Firmware Bug]: ..." error message is generated. However, if |
337 | * software does not properly determines the offsets, it is not |
338 | * necessarily a BIOS bug. |
339 | */ |
340 | |
341 | static atomic_t eilvt_offsets[APIC_EILVT_NR_MAX]; |
342 | |
343 | static inline int eilvt_entry_is_changeable(unsigned int old, unsigned int new) |
344 | { |
345 | return (old & APIC_EILVT_MASKED) |
346 | || (new == APIC_EILVT_MASKED) |
347 | || ((new & ~APIC_EILVT_MASKED) == old); |
348 | } |
349 | |
350 | static unsigned int reserve_eilvt_offset(int offset, unsigned int new) |
351 | { |
352 | unsigned int rsvd, vector; |
353 | |
354 | if (offset >= APIC_EILVT_NR_MAX) |
355 | return ~0; |
356 | |
357 | rsvd = atomic_read(v: &eilvt_offsets[offset]); |
358 | do { |
359 | vector = rsvd & ~APIC_EILVT_MASKED; /* 0: unassigned */ |
360 | if (vector && !eilvt_entry_is_changeable(old: vector, new)) |
361 | /* may not change if vectors are different */ |
362 | return rsvd; |
363 | } while (!atomic_try_cmpxchg(v: &eilvt_offsets[offset], old: &rsvd, new)); |
364 | |
365 | rsvd = new & ~APIC_EILVT_MASKED; |
366 | if (rsvd && rsvd != vector) |
367 | pr_info("LVT offset %d assigned for vector 0x%02x\n", |
368 | offset, rsvd); |
369 | |
370 | return new; |
371 | } |
372 | |
373 | /* |
374 | * If mask=1, the LVT entry does not generate interrupts while mask=0 |
375 | * enables the vector. See also the BKDGs. Must be called with |
376 | * preemption disabled. |
377 | */ |
378 | |
379 | int setup_APIC_eilvt(u8 offset, u8 vector, u8 msg_type, u8 mask) |
380 | { |
381 | unsigned long reg = APIC_EILVTn(offset); |
382 | unsigned int new, old, reserved; |
383 | |
384 | new = (mask << 16) | (msg_type << 8) | vector; |
385 | old = apic_read(reg); |
386 | reserved = reserve_eilvt_offset(offset, new); |
387 | |
388 | if (reserved != new) { |
389 | pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for " |
390 | "vector 0x%x, but the register is already in use for " |
391 | "vector 0x%x on another cpu\n", |
392 | smp_processor_id(), reg, offset, new, reserved); |
393 | return -EINVAL; |
394 | } |
395 | |
396 | if (!eilvt_entry_is_changeable(old, new)) { |
397 | pr_err(FW_BUG "cpu %d, try to use APIC%lX (LVT offset %d) for " |
398 | "vector 0x%x, but the register is already in use for " |
399 | "vector 0x%x on this cpu\n", |
400 | smp_processor_id(), reg, offset, new, old); |
401 | return -EBUSY; |
402 | } |
403 | |
404 | apic_write(reg, val: new); |
405 | |
406 | return 0; |
407 | } |
408 | EXPORT_SYMBOL_GPL(setup_APIC_eilvt); |
409 | |
410 | /* |
411 | * Program the next event, relative to now |
412 | */ |
413 | static int lapic_next_event(unsigned long delta, |
414 | struct clock_event_device *evt) |
415 | { |
416 | apic_write(APIC_TMICT, val: delta); |
417 | return 0; |
418 | } |
419 | |
420 | static int lapic_next_deadline(unsigned long delta, |
421 | struct clock_event_device *evt) |
422 | { |
423 | u64 tsc; |
424 | |
425 | /* This MSR is special and need a special fence: */ |
426 | weak_wrmsr_fence(); |
427 | |
428 | tsc = rdtsc(); |
429 | wrmsrq(MSR_IA32_TSC_DEADLINE, val: tsc + (((u64) delta) * TSC_DIVISOR)); |
430 | return 0; |
431 | } |
432 | |
433 | static int lapic_timer_shutdown(struct clock_event_device *evt) |
434 | { |
435 | unsigned int v; |
436 | |
437 | /* Lapic used as dummy for broadcast ? */ |
438 | if (evt->features & CLOCK_EVT_FEAT_DUMMY) |
439 | return 0; |
440 | |
441 | v = apic_read(APIC_LVTT); |
442 | v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); |
443 | apic_write(APIC_LVTT, val: v); |
444 | |
445 | /* |
446 | * Setting APIC_LVT_MASKED (above) should be enough to tell |
447 | * the hardware that this timer will never fire. But AMD |
448 | * erratum 411 and some Intel CPU behavior circa 2024 say |
449 | * otherwise. Time for belt and suspenders programming: mask |
450 | * the timer _and_ zero the counter registers: |
451 | */ |
452 | if (v & APIC_LVT_TIMER_TSCDEADLINE) |
453 | wrmsrq(MSR_IA32_TSC_DEADLINE, val: 0); |
454 | else |
455 | apic_write(APIC_TMICT, val: 0); |
456 | |
457 | return 0; |
458 | } |
459 | |
460 | static inline int |
461 | lapic_timer_set_periodic_oneshot(struct clock_event_device *evt, bool oneshot) |
462 | { |
463 | /* Lapic used as dummy for broadcast ? */ |
464 | if (evt->features & CLOCK_EVT_FEAT_DUMMY) |
465 | return 0; |
466 | |
467 | __setup_APIC_LVTT(clocks: lapic_timer_period, oneshot, irqen: 1); |
468 | return 0; |
469 | } |
470 | |
471 | static int lapic_timer_set_periodic(struct clock_event_device *evt) |
472 | { |
473 | return lapic_timer_set_periodic_oneshot(evt, oneshot: false); |
474 | } |
475 | |
476 | static int lapic_timer_set_oneshot(struct clock_event_device *evt) |
477 | { |
478 | return lapic_timer_set_periodic_oneshot(evt, oneshot: true); |
479 | } |
480 | |
481 | /* |
482 | * Local APIC timer broadcast function |
483 | */ |
484 | static void lapic_timer_broadcast(const struct cpumask *mask) |
485 | { |
486 | #ifdef CONFIG_SMP |
487 | __apic_send_IPI_mask(mask, LOCAL_TIMER_VECTOR); |
488 | #endif |
489 | } |
490 | |
491 | |
492 | /* |
493 | * The local apic timer can be used for any function which is CPU local. |
494 | */ |
495 | static struct clock_event_device lapic_clockevent = { |
496 | .name = "lapic", |
497 | .features = CLOCK_EVT_FEAT_PERIODIC | |
498 | CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP |
499 | | CLOCK_EVT_FEAT_DUMMY, |
500 | .shift = 32, |
501 | .set_state_shutdown = lapic_timer_shutdown, |
502 | .set_state_periodic = lapic_timer_set_periodic, |
503 | .set_state_oneshot = lapic_timer_set_oneshot, |
504 | .set_state_oneshot_stopped = lapic_timer_shutdown, |
505 | .set_next_event = lapic_next_event, |
506 | .broadcast = lapic_timer_broadcast, |
507 | .rating = 100, |
508 | .irq = -1, |
509 | }; |
510 | static DEFINE_PER_CPU(struct clock_event_device, lapic_events); |
511 | |
512 | static const struct x86_cpu_id deadline_match[] __initconst = { |
513 | X86_MATCH_VFM_STEPS(INTEL_HASWELL_X, 0x2, 0x2, 0x3a), /* EP */ |
514 | X86_MATCH_VFM_STEPS(INTEL_HASWELL_X, 0x4, 0x4, 0x0f), /* EX */ |
515 | |
516 | X86_MATCH_VFM(INTEL_BROADWELL_X, 0x0b000020), |
517 | |
518 | X86_MATCH_VFM_STEPS(INTEL_BROADWELL_D, 0x2, 0x2, 0x00000011), |
519 | X86_MATCH_VFM_STEPS(INTEL_BROADWELL_D, 0x3, 0x3, 0x0700000e), |
520 | X86_MATCH_VFM_STEPS(INTEL_BROADWELL_D, 0x4, 0x4, 0x0f00000c), |
521 | X86_MATCH_VFM_STEPS(INTEL_BROADWELL_D, 0x5, 0x5, 0x0e000003), |
522 | |
523 | X86_MATCH_VFM_STEPS(INTEL_SKYLAKE_X, 0x3, 0x3, 0x01000136), |
524 | X86_MATCH_VFM_STEPS(INTEL_SKYLAKE_X, 0x4, 0x4, 0x02000014), |
525 | X86_MATCH_VFM_STEPS(INTEL_SKYLAKE_X, 0x5, 0xf, 0), |
526 | |
527 | X86_MATCH_VFM(INTEL_HASWELL, 0x22), |
528 | X86_MATCH_VFM(INTEL_HASWELL_L, 0x20), |
529 | X86_MATCH_VFM(INTEL_HASWELL_G, 0x17), |
530 | |
531 | X86_MATCH_VFM(INTEL_BROADWELL, 0x25), |
532 | X86_MATCH_VFM(INTEL_BROADWELL_G, 0x17), |
533 | |
534 | X86_MATCH_VFM(INTEL_SKYLAKE_L, 0xb2), |
535 | X86_MATCH_VFM(INTEL_SKYLAKE, 0xb2), |
536 | |
537 | X86_MATCH_VFM(INTEL_KABYLAKE_L, 0x52), |
538 | X86_MATCH_VFM(INTEL_KABYLAKE, 0x52), |
539 | |
540 | {}, |
541 | }; |
542 | |
543 | static __init bool apic_validate_deadline_timer(void) |
544 | { |
545 | const struct x86_cpu_id *m; |
546 | u32 rev; |
547 | |
548 | if (!boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) |
549 | return false; |
550 | if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) |
551 | return true; |
552 | |
553 | m = x86_match_cpu(match: deadline_match); |
554 | if (!m) |
555 | return true; |
556 | |
557 | rev = (u32)m->driver_data; |
558 | |
559 | if (boot_cpu_data.microcode >= rev) |
560 | return true; |
561 | |
562 | setup_clear_cpu_cap(X86_FEATURE_TSC_DEADLINE_TIMER); |
563 | pr_err(FW_BUG "TSC_DEADLINE disabled due to Errata; " |
564 | "please update microcode to version: 0x%x (or later)\n", rev); |
565 | return false; |
566 | } |
567 | |
568 | /* |
569 | * Setup the local APIC timer for this CPU. Copy the initialized values |
570 | * of the boot CPU and register the clock event in the framework. |
571 | */ |
572 | static void setup_APIC_timer(void) |
573 | { |
574 | struct clock_event_device *levt = this_cpu_ptr(&lapic_events); |
575 | |
576 | if (this_cpu_has(X86_FEATURE_ARAT)) { |
577 | lapic_clockevent.features &= ~CLOCK_EVT_FEAT_C3STOP; |
578 | /* Make LAPIC timer preferable over percpu HPET */ |
579 | lapic_clockevent.rating = 150; |
580 | } |
581 | |
582 | memcpy(levt, &lapic_clockevent, sizeof(*levt)); |
583 | levt->cpumask = cpumask_of(smp_processor_id()); |
584 | |
585 | if (this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) { |
586 | levt->name = "lapic-deadline"; |
587 | levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC | |
588 | CLOCK_EVT_FEAT_DUMMY); |
589 | levt->set_next_event = lapic_next_deadline; |
590 | clockevents_config_and_register(dev: levt, |
591 | freq: tsc_khz * (1000 / TSC_DIVISOR), |
592 | min_delta: 0xF, max_delta: ~0UL); |
593 | } else |
594 | clockevents_register_device(dev: levt); |
595 | } |
596 | |
597 | /* |
598 | * Install the updated TSC frequency from recalibration at the TSC |
599 | * deadline clockevent devices. |
600 | */ |
601 | static void __lapic_update_tsc_freq(void *info) |
602 | { |
603 | struct clock_event_device *levt = this_cpu_ptr(&lapic_events); |
604 | |
605 | if (!this_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) |
606 | return; |
607 | |
608 | clockevents_update_freq(ce: levt, freq: tsc_khz * (1000 / TSC_DIVISOR)); |
609 | } |
610 | |
611 | void lapic_update_tsc_freq(void) |
612 | { |
613 | /* |
614 | * The clockevent device's ->mult and ->shift can both be |
615 | * changed. In order to avoid races, schedule the frequency |
616 | * update code on each CPU. |
617 | */ |
618 | on_each_cpu(func: __lapic_update_tsc_freq, NULL, wait: 0); |
619 | } |
620 | |
621 | /* |
622 | * In this functions we calibrate APIC bus clocks to the external timer. |
623 | * |
624 | * We want to do the calibration only once since we want to have local timer |
625 | * irqs synchronous. CPUs connected by the same APIC bus have the very same bus |
626 | * frequency. |
627 | * |
628 | * This was previously done by reading the PIT/HPET and waiting for a wrap |
629 | * around to find out, that a tick has elapsed. I have a box, where the PIT |
630 | * readout is broken, so it never gets out of the wait loop again. This was |
631 | * also reported by others. |
632 | * |
633 | * Monitoring the jiffies value is inaccurate and the clockevents |
634 | * infrastructure allows us to do a simple substitution of the interrupt |
635 | * handler. |
636 | * |
637 | * The calibration routine also uses the pm_timer when possible, as the PIT |
638 | * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes |
639 | * back to normal later in the boot process). |
640 | */ |
641 | |
642 | #define LAPIC_CAL_LOOPS (HZ/10) |
643 | |
644 | static __initdata int lapic_cal_loops = -1; |
645 | static __initdata long lapic_cal_t1, lapic_cal_t2; |
646 | static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2; |
647 | static __initdata u32 lapic_cal_pm1, lapic_cal_pm2; |
648 | static __initdata unsigned long lapic_cal_j1, lapic_cal_j2; |
649 | |
650 | /* |
651 | * Temporary interrupt handler and polled calibration function. |
652 | */ |
653 | static void __init lapic_cal_handler(struct clock_event_device *dev) |
654 | { |
655 | unsigned long long tsc = 0; |
656 | long tapic = apic_read(APIC_TMCCT); |
657 | u32 pm = acpi_pm_read_early(); |
658 | |
659 | if (boot_cpu_has(X86_FEATURE_TSC)) |
660 | tsc = rdtsc(); |
661 | |
662 | switch (lapic_cal_loops++) { |
663 | case 0: |
664 | lapic_cal_t1 = tapic; |
665 | lapic_cal_tsc1 = tsc; |
666 | lapic_cal_pm1 = pm; |
667 | lapic_cal_j1 = jiffies; |
668 | break; |
669 | |
670 | case LAPIC_CAL_LOOPS: |
671 | lapic_cal_t2 = tapic; |
672 | lapic_cal_tsc2 = tsc; |
673 | if (pm < lapic_cal_pm1) |
674 | pm += ACPI_PM_OVRRUN; |
675 | lapic_cal_pm2 = pm; |
676 | lapic_cal_j2 = jiffies; |
677 | break; |
678 | } |
679 | } |
680 | |
681 | static int __init |
682 | calibrate_by_pmtimer(u32 deltapm, long *delta, long *deltatsc) |
683 | { |
684 | const long pm_100ms = PMTMR_TICKS_PER_SEC / 10; |
685 | const long pm_thresh = pm_100ms / 100; |
686 | unsigned long mult; |
687 | u64 res; |
688 | |
689 | #ifndef CONFIG_X86_PM_TIMER |
690 | return -1; |
691 | #endif |
692 | |
693 | apic_pr_verbose("... PM-Timer delta = %u\n", deltapm); |
694 | |
695 | /* Check, if the PM timer is available */ |
696 | if (!deltapm) |
697 | return -1; |
698 | |
699 | mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, shift_constant: 22); |
700 | |
701 | if (deltapm > (pm_100ms - pm_thresh) && |
702 | deltapm < (pm_100ms + pm_thresh)) { |
703 | apic_pr_verbose("... PM-Timer result ok\n"); |
704 | return 0; |
705 | } |
706 | |
707 | res = (((u64)deltapm) * mult) >> 22; |
708 | do_div(res, 1000000); |
709 | pr_warn("APIC calibration not consistent with PM-Timer: %ldms instead of 100ms\n", |
710 | (long)res); |
711 | |
712 | /* Correct the lapic counter value */ |
713 | res = (((u64)(*delta)) * pm_100ms); |
714 | do_div(res, deltapm); |
715 | pr_info("APIC delta adjusted to PM-Timer: " |
716 | "%lu (%ld)\n", (unsigned long)res, *delta); |
717 | *delta = (long)res; |
718 | |
719 | /* Correct the tsc counter value */ |
720 | if (boot_cpu_has(X86_FEATURE_TSC)) { |
721 | res = (((u64)(*deltatsc)) * pm_100ms); |
722 | do_div(res, deltapm); |
723 | apic_pr_verbose("TSC delta adjusted to PM-Timer: %lu (%ld)\n", |
724 | (unsigned long)res, *deltatsc); |
725 | *deltatsc = (long)res; |
726 | } |
727 | |
728 | return 0; |
729 | } |
730 | |
731 | static int __init lapic_init_clockevent(void) |
732 | { |
733 | if (!lapic_timer_period) |
734 | return -1; |
735 | |
736 | /* Calculate the scaled math multiplication factor */ |
737 | lapic_clockevent.mult = div_sc(ticks: lapic_timer_period/APIC_DIVISOR, |
738 | TICK_NSEC, shift: lapic_clockevent.shift); |
739 | lapic_clockevent.max_delta_ns = |
740 | clockevent_delta2ns(latch: 0x7FFFFFFF, evt: &lapic_clockevent); |
741 | lapic_clockevent.max_delta_ticks = 0x7FFFFFFF; |
742 | lapic_clockevent.min_delta_ns = |
743 | clockevent_delta2ns(latch: 0xF, evt: &lapic_clockevent); |
744 | lapic_clockevent.min_delta_ticks = 0xF; |
745 | |
746 | return 0; |
747 | } |
748 | |
749 | bool __init apic_needs_pit(void) |
750 | { |
751 | /* |
752 | * If the frequencies are not known, PIT is required for both TSC |
753 | * and apic timer calibration. |
754 | */ |
755 | if (!tsc_khz || !cpu_khz) |
756 | return true; |
757 | |
758 | /* Is there an APIC at all or is it disabled? */ |
759 | if (!boot_cpu_has(X86_FEATURE_APIC) || apic_is_disabled) |
760 | return true; |
761 | |
762 | /* |
763 | * If interrupt delivery mode is legacy PIC or virtual wire without |
764 | * configuration, the local APIC timer won't be set up. Make sure |
765 | * that the PIT is initialized. |
766 | */ |
767 | if (apic_intr_mode == APIC_PIC || |
768 | apic_intr_mode == APIC_VIRTUAL_WIRE_NO_CONFIG) |
769 | return true; |
770 | |
771 | /* Virt guests may lack ARAT, but still have DEADLINE */ |
772 | if (!boot_cpu_has(X86_FEATURE_ARAT)) |
773 | return true; |
774 | |
775 | /* Deadline timer is based on TSC so no further PIT action required */ |
776 | if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) |
777 | return false; |
778 | |
779 | /* APIC timer disabled? */ |
780 | if (disable_apic_timer) |
781 | return true; |
782 | /* |
783 | * The APIC timer frequency is known already, no PIT calibration |
784 | * required. If unknown, let the PIT be initialized. |
785 | */ |
786 | return lapic_timer_period == 0; |
787 | } |
788 | |
789 | static int __init calibrate_APIC_clock(void) |
790 | { |
791 | struct clock_event_device *levt = this_cpu_ptr(&lapic_events); |
792 | u64 tsc_perj = 0, tsc_start = 0; |
793 | unsigned long jif_start; |
794 | unsigned long deltaj; |
795 | long delta, deltatsc; |
796 | int pm_referenced = 0; |
797 | |
798 | if (boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER)) |
799 | return 0; |
800 | |
801 | /* |
802 | * Check if lapic timer has already been calibrated by platform |
803 | * specific routine, such as tsc calibration code. If so just fill |
804 | * in the clockevent structure and return. |
805 | */ |
806 | if (!lapic_init_clockevent()) { |
807 | apic_pr_verbose("lapic timer already calibrated %d\n", lapic_timer_period); |
808 | /* |
809 | * Direct calibration methods must have an always running |
810 | * local APIC timer, no need for broadcast timer. |
811 | */ |
812 | lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; |
813 | return 0; |
814 | } |
815 | |
816 | apic_pr_verbose("Using local APIC timer interrupts. Calibrating APIC timer ...\n"); |
817 | |
818 | /* |
819 | * There are platforms w/o global clockevent devices. Instead of |
820 | * making the calibration conditional on that, use a polling based |
821 | * approach everywhere. |
822 | */ |
823 | local_irq_disable(); |
824 | |
825 | /* |
826 | * Setup the APIC counter to maximum. There is no way the lapic |
827 | * can underflow in the 100ms detection time frame |
828 | */ |
829 | __setup_APIC_LVTT(clocks: 0xffffffff, oneshot: 0, irqen: 0); |
830 | |
831 | /* |
832 | * Methods to terminate the calibration loop: |
833 | * 1) Global clockevent if available (jiffies) |
834 | * 2) TSC if available and frequency is known |
835 | */ |
836 | jif_start = READ_ONCE(jiffies); |
837 | |
838 | if (tsc_khz) { |
839 | tsc_start = rdtsc(); |
840 | tsc_perj = div_u64(dividend: (u64)tsc_khz * 1000, HZ); |
841 | } |
842 | |
843 | /* |
844 | * Enable interrupts so the tick can fire, if a global |
845 | * clockevent device is available |
846 | */ |
847 | local_irq_enable(); |
848 | |
849 | while (lapic_cal_loops <= LAPIC_CAL_LOOPS) { |
850 | /* Wait for a tick to elapse */ |
851 | while (1) { |
852 | if (tsc_khz) { |
853 | u64 tsc_now = rdtsc(); |
854 | if ((tsc_now - tsc_start) >= tsc_perj) { |
855 | tsc_start += tsc_perj; |
856 | break; |
857 | } |
858 | } else { |
859 | unsigned long jif_now = READ_ONCE(jiffies); |
860 | |
861 | if (time_after(jif_now, jif_start)) { |
862 | jif_start = jif_now; |
863 | break; |
864 | } |
865 | } |
866 | cpu_relax(); |
867 | } |
868 | |
869 | /* Invoke the calibration routine */ |
870 | local_irq_disable(); |
871 | lapic_cal_handler(NULL); |
872 | local_irq_enable(); |
873 | } |
874 | |
875 | local_irq_disable(); |
876 | |
877 | /* Build delta t1-t2 as apic timer counts down */ |
878 | delta = lapic_cal_t1 - lapic_cal_t2; |
879 | apic_pr_verbose("... lapic delta = %ld\n", delta); |
880 | |
881 | deltatsc = (long)(lapic_cal_tsc2 - lapic_cal_tsc1); |
882 | |
883 | /* we trust the PM based calibration if possible */ |
884 | pm_referenced = !calibrate_by_pmtimer(deltapm: lapic_cal_pm2 - lapic_cal_pm1, |
885 | delta: &delta, deltatsc: &deltatsc); |
886 | |
887 | lapic_timer_period = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS; |
888 | lapic_init_clockevent(); |
889 | |
890 | apic_pr_verbose("..... delta %ld\n", delta); |
891 | apic_pr_verbose("..... mult: %u\n", lapic_clockevent.mult); |
892 | apic_pr_verbose("..... calibration result: %u\n", lapic_timer_period); |
893 | |
894 | if (boot_cpu_has(X86_FEATURE_TSC)) { |
895 | apic_pr_verbose("..... CPU clock speed is %ld.%04ld MHz.\n", |
896 | (deltatsc / LAPIC_CAL_LOOPS) / (1000000 / HZ), |
897 | (deltatsc / LAPIC_CAL_LOOPS) % (1000000 / HZ)); |
898 | } |
899 | |
900 | apic_pr_verbose("..... host bus clock speed is %u.%04u MHz.\n", |
901 | lapic_timer_period / (1000000 / HZ), |
902 | lapic_timer_period % (1000000 / HZ)); |
903 | |
904 | /* |
905 | * Do a sanity check on the APIC calibration result |
906 | */ |
907 | if (lapic_timer_period < (1000000 / HZ)) { |
908 | local_irq_enable(); |
909 | pr_warn("APIC frequency too slow, disabling apic timer\n"); |
910 | return -1; |
911 | } |
912 | |
913 | levt->features &= ~CLOCK_EVT_FEAT_DUMMY; |
914 | |
915 | /* |
916 | * PM timer calibration failed or not turned on so lets try APIC |
917 | * timer based calibration, if a global clockevent device is |
918 | * available. |
919 | */ |
920 | if (!pm_referenced && global_clock_event) { |
921 | apic_pr_verbose("... verify APIC timer\n"); |
922 | |
923 | /* |
924 | * Setup the apic timer manually |
925 | */ |
926 | levt->event_handler = lapic_cal_handler; |
927 | lapic_timer_set_periodic(evt: levt); |
928 | lapic_cal_loops = -1; |
929 | |
930 | /* Let the interrupts run */ |
931 | local_irq_enable(); |
932 | |
933 | while (lapic_cal_loops <= LAPIC_CAL_LOOPS) |
934 | cpu_relax(); |
935 | |
936 | /* Stop the lapic timer */ |
937 | local_irq_disable(); |
938 | lapic_timer_shutdown(evt: levt); |
939 | |
940 | /* Jiffies delta */ |
941 | deltaj = lapic_cal_j2 - lapic_cal_j1; |
942 | apic_pr_verbose("... jiffies delta = %lu\n", deltaj); |
943 | |
944 | /* Check, if the jiffies result is consistent */ |
945 | if (deltaj >= LAPIC_CAL_LOOPS-2 && deltaj <= LAPIC_CAL_LOOPS+2) |
946 | apic_pr_verbose("... jiffies result ok\n"); |
947 | else |
948 | levt->features |= CLOCK_EVT_FEAT_DUMMY; |
949 | } |
950 | local_irq_enable(); |
951 | |
952 | if (levt->features & CLOCK_EVT_FEAT_DUMMY) { |
953 | pr_warn("APIC timer disabled due to verification failure\n"); |
954 | return -1; |
955 | } |
956 | |
957 | return 0; |
958 | } |
959 | |
960 | /* |
961 | * Setup the boot APIC |
962 | * |
963 | * Calibrate and verify the result. |
964 | */ |
965 | void __init setup_boot_APIC_clock(void) |
966 | { |
967 | /* |
968 | * The local apic timer can be disabled via the kernel |
969 | * commandline or from the CPU detection code. Register the lapic |
970 | * timer as a dummy clock event source on SMP systems, so the |
971 | * broadcast mechanism is used. On UP systems simply ignore it. |
972 | */ |
973 | if (disable_apic_timer) { |
974 | pr_info("Disabling APIC timer\n"); |
975 | /* No broadcast on UP ! */ |
976 | if (num_possible_cpus() > 1) { |
977 | lapic_clockevent.mult = 1; |
978 | setup_APIC_timer(); |
979 | } |
980 | return; |
981 | } |
982 | |
983 | if (calibrate_APIC_clock()) { |
984 | /* No broadcast on UP ! */ |
985 | if (num_possible_cpus() > 1) |
986 | setup_APIC_timer(); |
987 | return; |
988 | } |
989 | |
990 | /* |
991 | * If nmi_watchdog is set to IO_APIC, we need the |
992 | * PIT/HPET going. Otherwise register lapic as a dummy |
993 | * device. |
994 | */ |
995 | lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY; |
996 | |
997 | /* Setup the lapic or request the broadcast */ |
998 | setup_APIC_timer(); |
999 | amd_e400_c1e_apic_setup(); |
1000 | } |
1001 | |
1002 | void setup_secondary_APIC_clock(void) |
1003 | { |
1004 | setup_APIC_timer(); |
1005 | amd_e400_c1e_apic_setup(); |
1006 | } |
1007 | |
1008 | /* |
1009 | * The guts of the apic timer interrupt |
1010 | */ |
1011 | static void local_apic_timer_interrupt(void) |
1012 | { |
1013 | struct clock_event_device *evt = this_cpu_ptr(&lapic_events); |
1014 | |
1015 | /* |
1016 | * Normally we should not be here till LAPIC has been initialized but |
1017 | * in some cases like kdump, its possible that there is a pending LAPIC |
1018 | * timer interrupt from previous kernel's context and is delivered in |
1019 | * new kernel the moment interrupts are enabled. |
1020 | * |
1021 | * Interrupts are enabled early and LAPIC is setup much later, hence |
1022 | * its possible that when we get here evt->event_handler is NULL. |
1023 | * Check for event_handler being NULL and discard the interrupt as |
1024 | * spurious. |
1025 | */ |
1026 | if (!evt->event_handler) { |
1027 | pr_warn("Spurious LAPIC timer interrupt on cpu %d\n", |
1028 | smp_processor_id()); |
1029 | /* Switch it off */ |
1030 | lapic_timer_shutdown(evt); |
1031 | return; |
1032 | } |
1033 | |
1034 | /* |
1035 | * the NMI deadlock-detector uses this. |
1036 | */ |
1037 | inc_irq_stat(apic_timer_irqs); |
1038 | |
1039 | evt->event_handler(evt); |
1040 | } |
1041 | |
1042 | /* |
1043 | * Local APIC timer interrupt. This is the most natural way for doing |
1044 | * local interrupts, but local timer interrupts can be emulated by |
1045 | * broadcast interrupts too. [in case the hw doesn't support APIC timers] |
1046 | * |
1047 | * [ if a single-CPU system runs an SMP kernel then we call the local |
1048 | * interrupt as well. Thus we cannot inline the local irq ... ] |
1049 | */ |
1050 | DEFINE_IDTENTRY_SYSVEC(sysvec_apic_timer_interrupt) |
1051 | { |
1052 | struct pt_regs *old_regs = set_irq_regs(regs); |
1053 | |
1054 | apic_eoi(); |
1055 | trace_local_timer_entry(LOCAL_TIMER_VECTOR); |
1056 | local_apic_timer_interrupt(); |
1057 | trace_local_timer_exit(LOCAL_TIMER_VECTOR); |
1058 | |
1059 | set_irq_regs(old_regs); |
1060 | } |
1061 | |
1062 | /* |
1063 | * Local APIC start and shutdown |
1064 | */ |
1065 | |
1066 | /** |
1067 | * clear_local_APIC - shutdown the local APIC |
1068 | * |
1069 | * This is called, when a CPU is disabled and before rebooting, so the state of |
1070 | * the local APIC has no dangling leftovers. Also used to cleanout any BIOS |
1071 | * leftovers during boot. |
1072 | */ |
1073 | void clear_local_APIC(void) |
1074 | { |
1075 | int maxlvt; |
1076 | u32 v; |
1077 | |
1078 | if (!apic_accessible()) |
1079 | return; |
1080 | |
1081 | maxlvt = lapic_get_maxlvt(); |
1082 | /* |
1083 | * Masking an LVT entry can trigger a local APIC error |
1084 | * if the vector is zero. Mask LVTERR first to prevent this. |
1085 | */ |
1086 | if (maxlvt >= 3) { |
1087 | v = ERROR_APIC_VECTOR; /* any non-zero vector will do */ |
1088 | apic_write(APIC_LVTERR, val: v | APIC_LVT_MASKED); |
1089 | } |
1090 | /* |
1091 | * Careful: we have to set masks only first to deassert |
1092 | * any level-triggered sources. |
1093 | */ |
1094 | v = apic_read(APIC_LVTT); |
1095 | apic_write(APIC_LVTT, val: v | APIC_LVT_MASKED); |
1096 | v = apic_read(APIC_LVT0); |
1097 | apic_write(APIC_LVT0, val: v | APIC_LVT_MASKED); |
1098 | v = apic_read(APIC_LVT1); |
1099 | apic_write(APIC_LVT1, val: v | APIC_LVT_MASKED); |
1100 | if (maxlvt >= 4) { |
1101 | v = apic_read(APIC_LVTPC); |
1102 | apic_write(APIC_LVTPC, val: v | APIC_LVT_MASKED); |
1103 | } |
1104 | |
1105 | /* lets not touch this if we didn't frob it */ |
1106 | #ifdef CONFIG_X86_THERMAL_VECTOR |
1107 | if (maxlvt >= 5) { |
1108 | v = apic_read(APIC_LVTTHMR); |
1109 | apic_write(APIC_LVTTHMR, val: v | APIC_LVT_MASKED); |
1110 | } |
1111 | #endif |
1112 | #ifdef CONFIG_X86_MCE_INTEL |
1113 | if (maxlvt >= 6) { |
1114 | v = apic_read(APIC_LVTCMCI); |
1115 | if (!(v & APIC_LVT_MASKED)) |
1116 | apic_write(APIC_LVTCMCI, val: v | APIC_LVT_MASKED); |
1117 | } |
1118 | #endif |
1119 | |
1120 | /* |
1121 | * Clean APIC state for other OSs: |
1122 | */ |
1123 | apic_write(APIC_LVTT, APIC_LVT_MASKED); |
1124 | apic_write(APIC_LVT0, APIC_LVT_MASKED); |
1125 | apic_write(APIC_LVT1, APIC_LVT_MASKED); |
1126 | if (maxlvt >= 3) |
1127 | apic_write(APIC_LVTERR, APIC_LVT_MASKED); |
1128 | if (maxlvt >= 4) |
1129 | apic_write(APIC_LVTPC, APIC_LVT_MASKED); |
1130 | |
1131 | /* Integrated APIC (!82489DX) ? */ |
1132 | if (lapic_is_integrated()) { |
1133 | if (maxlvt > 3) |
1134 | /* Clear ESR due to Pentium errata 3AP and 11AP */ |
1135 | apic_write(APIC_ESR, val: 0); |
1136 | apic_read(APIC_ESR); |
1137 | } |
1138 | } |
1139 | |
1140 | /** |
1141 | * apic_soft_disable - Clears and software disables the local APIC on hotplug |
1142 | * |
1143 | * Contrary to disable_local_APIC() this does not touch the enable bit in |
1144 | * MSR_IA32_APICBASE. Clearing that bit on systems based on the 3 wire APIC |
1145 | * bus would require a hardware reset as the APIC would lose track of bus |
1146 | * arbitration. On systems with FSB delivery APICBASE could be disabled, |
1147 | * but it has to be guaranteed that no interrupt is sent to the APIC while |
1148 | * in that state and it's not clear from the SDM whether it still responds |
1149 | * to INIT/SIPI messages. Stay on the safe side and use software disable. |
1150 | */ |
1151 | void apic_soft_disable(void) |
1152 | { |
1153 | u32 value; |
1154 | |
1155 | clear_local_APIC(); |
1156 | |
1157 | /* Soft disable APIC (implies clearing of registers for 82489DX!). */ |
1158 | value = apic_read(APIC_SPIV); |
1159 | value &= ~APIC_SPIV_APIC_ENABLED; |
1160 | apic_write(APIC_SPIV, val: value); |
1161 | } |
1162 | |
1163 | /** |
1164 | * disable_local_APIC - clear and disable the local APIC |
1165 | */ |
1166 | void disable_local_APIC(void) |
1167 | { |
1168 | if (!apic_accessible()) |
1169 | return; |
1170 | |
1171 | apic_soft_disable(); |
1172 | |
1173 | #ifdef CONFIG_X86_32 |
1174 | /* |
1175 | * When LAPIC was disabled by the BIOS and enabled by the kernel, |
1176 | * restore the disabled state. |
1177 | */ |
1178 | if (enabled_via_apicbase) { |
1179 | unsigned int l, h; |
1180 | |
1181 | rdmsr(MSR_IA32_APICBASE, l, h); |
1182 | l &= ~MSR_IA32_APICBASE_ENABLE; |
1183 | wrmsr(MSR_IA32_APICBASE, l, h); |
1184 | } |
1185 | #endif |
1186 | } |
1187 | |
1188 | /* |
1189 | * If Linux enabled the LAPIC against the BIOS default disable it down before |
1190 | * re-entering the BIOS on shutdown. Otherwise the BIOS may get confused and |
1191 | * not power-off. Additionally clear all LVT entries before disable_local_APIC |
1192 | * for the case where Linux didn't enable the LAPIC. |
1193 | */ |
1194 | void lapic_shutdown(void) |
1195 | { |
1196 | unsigned long flags; |
1197 | |
1198 | if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config()) |
1199 | return; |
1200 | |
1201 | local_irq_save(flags); |
1202 | |
1203 | #ifdef CONFIG_X86_32 |
1204 | if (!enabled_via_apicbase) |
1205 | clear_local_APIC(); |
1206 | else |
1207 | #endif |
1208 | disable_local_APIC(); |
1209 | |
1210 | |
1211 | local_irq_restore(flags); |
1212 | } |
1213 | |
1214 | /** |
1215 | * sync_Arb_IDs - synchronize APIC bus arbitration IDs |
1216 | */ |
1217 | void __init sync_Arb_IDs(void) |
1218 | { |
1219 | /* |
1220 | * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not |
1221 | * needed on AMD. |
1222 | */ |
1223 | if (modern_apic() || boot_cpu_data.x86_vendor == X86_VENDOR_AMD) |
1224 | return; |
1225 | |
1226 | /* |
1227 | * Wait for idle. |
1228 | */ |
1229 | apic_wait_icr_idle(); |
1230 | |
1231 | apic_pr_debug("Synchronizing Arb IDs.\n"); |
1232 | apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG | APIC_DM_INIT); |
1233 | } |
1234 | |
1235 | enum apic_intr_mode_id apic_intr_mode __ro_after_init; |
1236 | |
1237 | static int __init __apic_intr_mode_select(void) |
1238 | { |
1239 | /* Check kernel option */ |
1240 | if (apic_is_disabled) { |
1241 | pr_info("APIC disabled via kernel command line\n"); |
1242 | return APIC_PIC; |
1243 | } |
1244 | |
1245 | /* Check BIOS */ |
1246 | #ifdef CONFIG_X86_64 |
1247 | /* On 64-bit, the APIC must be integrated, Check local APIC only */ |
1248 | if (!boot_cpu_has(X86_FEATURE_APIC)) { |
1249 | apic_is_disabled = true; |
1250 | pr_info("APIC disabled by BIOS\n"); |
1251 | return APIC_PIC; |
1252 | } |
1253 | #else |
1254 | /* On 32-bit, the APIC may be integrated APIC or 82489DX */ |
1255 | |
1256 | /* Neither 82489DX nor integrated APIC ? */ |
1257 | if (!boot_cpu_has(X86_FEATURE_APIC) && !smp_found_config) { |
1258 | apic_is_disabled = true; |
1259 | return APIC_PIC; |
1260 | } |
1261 | |
1262 | /* If the BIOS pretends there is an integrated APIC ? */ |
1263 | if (!boot_cpu_has(X86_FEATURE_APIC) && |
1264 | APIC_INTEGRATED(boot_cpu_apic_version)) { |
1265 | apic_is_disabled = true; |
1266 | pr_err(FW_BUG "Local APIC not detected, force emulation\n"); |
1267 | return APIC_PIC; |
1268 | } |
1269 | #endif |
1270 | |
1271 | /* Check MP table or ACPI MADT configuration */ |
1272 | if (!smp_found_config) { |
1273 | disable_ioapic_support(); |
1274 | if (!acpi_lapic) { |
1275 | pr_info("APIC: ACPI MADT or MP tables are not detected\n"); |
1276 | return APIC_VIRTUAL_WIRE_NO_CONFIG; |
1277 | } |
1278 | return APIC_VIRTUAL_WIRE; |
1279 | } |
1280 | |
1281 | #ifdef CONFIG_SMP |
1282 | /* If SMP should be disabled, then really disable it! */ |
1283 | if (!setup_max_cpus) { |
1284 | pr_info("APIC: SMP mode deactivated\n"); |
1285 | return APIC_SYMMETRIC_IO_NO_ROUTING; |
1286 | } |
1287 | #endif |
1288 | |
1289 | return APIC_SYMMETRIC_IO; |
1290 | } |
1291 | |
1292 | /* Select the interrupt delivery mode for the BSP */ |
1293 | void __init apic_intr_mode_select(void) |
1294 | { |
1295 | apic_intr_mode = __apic_intr_mode_select(); |
1296 | } |
1297 | |
1298 | /* |
1299 | * An initial setup of the virtual wire mode. |
1300 | */ |
1301 | void __init init_bsp_APIC(void) |
1302 | { |
1303 | unsigned int value; |
1304 | |
1305 | /* |
1306 | * Don't do the setup now if we have a SMP BIOS as the |
1307 | * through-I/O-APIC virtual wire mode might be active. |
1308 | */ |
1309 | if (smp_found_config || !boot_cpu_has(X86_FEATURE_APIC)) |
1310 | return; |
1311 | |
1312 | /* |
1313 | * Do not trust the local APIC being empty at bootup. |
1314 | */ |
1315 | clear_local_APIC(); |
1316 | |
1317 | /* |
1318 | * Enable APIC. |
1319 | */ |
1320 | value = apic_read(APIC_SPIV); |
1321 | value &= ~APIC_VECTOR_MASK; |
1322 | value |= APIC_SPIV_APIC_ENABLED; |
1323 | |
1324 | #ifdef CONFIG_X86_32 |
1325 | /* This bit is reserved on P4/Xeon and should be cleared */ |
1326 | if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && |
1327 | (boot_cpu_data.x86 == 15)) |
1328 | value &= ~APIC_SPIV_FOCUS_DISABLED; |
1329 | else |
1330 | #endif |
1331 | value |= APIC_SPIV_FOCUS_DISABLED; |
1332 | value |= SPURIOUS_APIC_VECTOR; |
1333 | apic_write(APIC_SPIV, val: value); |
1334 | |
1335 | /* |
1336 | * Set up the virtual wire mode. |
1337 | */ |
1338 | apic_write(APIC_LVT0, APIC_DM_EXTINT); |
1339 | value = APIC_DM_NMI; |
1340 | if (!lapic_is_integrated()) /* 82489DX */ |
1341 | value |= APIC_LVT_LEVEL_TRIGGER; |
1342 | if (apic_extnmi == APIC_EXTNMI_NONE) |
1343 | value |= APIC_LVT_MASKED; |
1344 | apic_write(APIC_LVT1, val: value); |
1345 | } |
1346 | |
1347 | static void __init apic_bsp_setup(bool upmode); |
1348 | |
1349 | /* Init the interrupt delivery mode for the BSP */ |
1350 | void __init apic_intr_mode_init(void) |
1351 | { |
1352 | bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT); |
1353 | |
1354 | switch (apic_intr_mode) { |
1355 | case APIC_PIC: |
1356 | pr_info("APIC: Keep in PIC mode(8259)\n"); |
1357 | return; |
1358 | case APIC_VIRTUAL_WIRE: |
1359 | pr_info("APIC: Switch to virtual wire mode setup\n"); |
1360 | break; |
1361 | case APIC_VIRTUAL_WIRE_NO_CONFIG: |
1362 | pr_info("APIC: Switch to virtual wire mode setup with no configuration\n"); |
1363 | upmode = true; |
1364 | break; |
1365 | case APIC_SYMMETRIC_IO: |
1366 | pr_info("APIC: Switch to symmetric I/O mode setup\n"); |
1367 | break; |
1368 | case APIC_SYMMETRIC_IO_NO_ROUTING: |
1369 | pr_info("APIC: Switch to symmetric I/O mode setup in no SMP routine\n"); |
1370 | break; |
1371 | } |
1372 | |
1373 | x86_64_probe_apic(); |
1374 | |
1375 | if (x86_platform.apic_post_init) |
1376 | x86_platform.apic_post_init(); |
1377 | |
1378 | apic_bsp_setup(upmode); |
1379 | } |
1380 | |
1381 | static void lapic_setup_esr(void) |
1382 | { |
1383 | unsigned int oldvalue, value, maxlvt; |
1384 | |
1385 | if (!lapic_is_integrated()) { |
1386 | pr_info("No ESR for 82489DX.\n"); |
1387 | return; |
1388 | } |
1389 | |
1390 | if (apic->disable_esr) { |
1391 | /* |
1392 | * Something untraceable is creating bad interrupts on |
1393 | * secondary quads ... for the moment, just leave the |
1394 | * ESR disabled - we can't do anything useful with the |
1395 | * errors anyway - mbligh |
1396 | */ |
1397 | pr_info("Leaving ESR disabled.\n"); |
1398 | return; |
1399 | } |
1400 | |
1401 | maxlvt = lapic_get_maxlvt(); |
1402 | if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ |
1403 | apic_write(APIC_ESR, val: 0); |
1404 | oldvalue = apic_read(APIC_ESR); |
1405 | |
1406 | /* enables sending errors */ |
1407 | value = ERROR_APIC_VECTOR; |
1408 | apic_write(APIC_LVTERR, val: value); |
1409 | |
1410 | /* |
1411 | * spec says clear errors after enabling vector. |
1412 | */ |
1413 | if (maxlvt > 3) |
1414 | apic_write(APIC_ESR, val: 0); |
1415 | value = apic_read(APIC_ESR); |
1416 | if (value != oldvalue) { |
1417 | apic_pr_verbose("ESR value before enabling vector: 0x%08x after: 0x%08x\n", |
1418 | oldvalue, value); |
1419 | } |
1420 | } |
1421 | |
1422 | #define APIC_IR_REGS APIC_ISR_NR |
1423 | #define APIC_IR_BITS (APIC_IR_REGS * 32) |
1424 | #define APIC_IR_MAPSIZE (APIC_IR_BITS / BITS_PER_LONG) |
1425 | |
1426 | union apic_ir { |
1427 | unsigned long map[APIC_IR_MAPSIZE]; |
1428 | u32 regs[APIC_IR_REGS]; |
1429 | }; |
1430 | |
1431 | static bool apic_check_and_ack(union apic_ir *irr, union apic_ir *isr) |
1432 | { |
1433 | int i, bit; |
1434 | |
1435 | /* Read the IRRs */ |
1436 | for (i = 0; i < APIC_IR_REGS; i++) |
1437 | irr->regs[i] = apic_read(APIC_IRR + i * 0x10); |
1438 | |
1439 | /* Read the ISRs */ |
1440 | for (i = 0; i < APIC_IR_REGS; i++) |
1441 | isr->regs[i] = apic_read(APIC_ISR + i * 0x10); |
1442 | |
1443 | /* |
1444 | * If the ISR map is not empty. ACK the APIC and run another round |
1445 | * to verify whether a pending IRR has been unblocked and turned |
1446 | * into a ISR. |
1447 | */ |
1448 | if (!bitmap_empty(src: isr->map, APIC_IR_BITS)) { |
1449 | /* |
1450 | * There can be multiple ISR bits set when a high priority |
1451 | * interrupt preempted a lower priority one. Issue an ACK |
1452 | * per set bit. |
1453 | */ |
1454 | for_each_set_bit(bit, isr->map, APIC_IR_BITS) |
1455 | apic_eoi(); |
1456 | return true; |
1457 | } |
1458 | |
1459 | return !bitmap_empty(src: irr->map, APIC_IR_BITS); |
1460 | } |
1461 | |
1462 | /* |
1463 | * After a crash, we no longer service the interrupts and a pending |
1464 | * interrupt from previous kernel might still have ISR bit set. |
1465 | * |
1466 | * Most probably by now the CPU has serviced that pending interrupt and it |
1467 | * might not have done the apic_eoi() because it thought, interrupt |
1468 | * came from i8259 as ExtInt. LAPIC did not get EOI so it does not clear |
1469 | * the ISR bit and cpu thinks it has already serviced the interrupt. Hence |
1470 | * a vector might get locked. It was noticed for timer irq (vector |
1471 | * 0x31). Issue an extra EOI to clear ISR. |
1472 | * |
1473 | * If there are pending IRR bits they turn into ISR bits after a higher |
1474 | * priority ISR bit has been acked. |
1475 | */ |
1476 | static void apic_pending_intr_clear(void) |
1477 | { |
1478 | union apic_ir irr, isr; |
1479 | unsigned int i; |
1480 | |
1481 | /* 512 loops are way oversized and give the APIC a chance to obey. */ |
1482 | for (i = 0; i < 512; i++) { |
1483 | if (!apic_check_and_ack(irr: &irr, isr: &isr)) |
1484 | return; |
1485 | } |
1486 | /* Dump the IRR/ISR content if that failed */ |
1487 | pr_warn("APIC: Stale IRR: %256pb ISR: %256pb\n", irr.map, isr.map); |
1488 | } |
1489 | |
1490 | /** |
1491 | * setup_local_APIC - setup the local APIC |
1492 | * |
1493 | * Used to setup local APIC while initializing BSP or bringing up APs. |
1494 | * Always called with preemption disabled. |
1495 | */ |
1496 | static void setup_local_APIC(void) |
1497 | { |
1498 | int cpu = smp_processor_id(); |
1499 | unsigned int value; |
1500 | |
1501 | if (apic_is_disabled) { |
1502 | disable_ioapic_support(); |
1503 | return; |
1504 | } |
1505 | |
1506 | /* |
1507 | * If this comes from kexec/kcrash the APIC might be enabled in |
1508 | * SPIV. Soft disable it before doing further initialization. |
1509 | */ |
1510 | value = apic_read(APIC_SPIV); |
1511 | value &= ~APIC_SPIV_APIC_ENABLED; |
1512 | apic_write(APIC_SPIV, val: value); |
1513 | |
1514 | #ifdef CONFIG_X86_32 |
1515 | /* Pound the ESR really hard over the head with a big hammer - mbligh */ |
1516 | if (lapic_is_integrated() && apic->disable_esr) { |
1517 | apic_write(APIC_ESR, 0); |
1518 | apic_write(APIC_ESR, 0); |
1519 | apic_write(APIC_ESR, 0); |
1520 | apic_write(APIC_ESR, 0); |
1521 | } |
1522 | #endif |
1523 | /* |
1524 | * Intel recommends to set DFR, LDR and TPR before enabling |
1525 | * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel |
1526 | * document number 292116). |
1527 | * |
1528 | * Except for APICs which operate in physical destination mode. |
1529 | */ |
1530 | if (apic->init_apic_ldr) |
1531 | apic->init_apic_ldr(); |
1532 | |
1533 | /* |
1534 | * Set Task Priority to 'accept all except vectors 0-31'. An APIC |
1535 | * vector in the 16-31 range could be delivered if TPR == 0, but we |
1536 | * would think it's an exception and terrible things will happen. We |
1537 | * never change this later on. |
1538 | */ |
1539 | value = apic_read(APIC_TASKPRI); |
1540 | value &= ~APIC_TPRI_MASK; |
1541 | value |= 0x10; |
1542 | apic_write(APIC_TASKPRI, val: value); |
1543 | |
1544 | /* Clear eventually stale ISR/IRR bits */ |
1545 | apic_pending_intr_clear(); |
1546 | |
1547 | /* |
1548 | * Now that we are all set up, enable the APIC |
1549 | */ |
1550 | value = apic_read(APIC_SPIV); |
1551 | value &= ~APIC_VECTOR_MASK; |
1552 | /* |
1553 | * Enable APIC |
1554 | */ |
1555 | value |= APIC_SPIV_APIC_ENABLED; |
1556 | |
1557 | #ifdef CONFIG_X86_32 |
1558 | /* |
1559 | * Some unknown Intel IO/APIC (or APIC) errata is biting us with |
1560 | * certain networking cards. If high frequency interrupts are |
1561 | * happening on a particular IOAPIC pin, plus the IOAPIC routing |
1562 | * entry is masked/unmasked at a high rate as well then sooner or |
1563 | * later IOAPIC line gets 'stuck', no more interrupts are received |
1564 | * from the device. If focus CPU is disabled then the hang goes |
1565 | * away, oh well :-( |
1566 | * |
1567 | * [ This bug can be reproduced easily with a level-triggered |
1568 | * PCI Ne2000 networking cards and PII/PIII processors, dual |
1569 | * BX chipset. ] |
1570 | */ |
1571 | /* |
1572 | * Actually disabling the focus CPU check just makes the hang less |
1573 | * frequent as it makes the interrupt distribution model be more |
1574 | * like LRU than MRU (the short-term load is more even across CPUs). |
1575 | */ |
1576 | |
1577 | /* |
1578 | * - enable focus processor (bit==0) |
1579 | * - 64bit mode always use processor focus |
1580 | * so no need to set it |
1581 | */ |
1582 | value &= ~APIC_SPIV_FOCUS_DISABLED; |
1583 | #endif |
1584 | |
1585 | /* |
1586 | * Set spurious IRQ vector |
1587 | */ |
1588 | value |= SPURIOUS_APIC_VECTOR; |
1589 | apic_write(APIC_SPIV, val: value); |
1590 | |
1591 | perf_events_lapic_init(); |
1592 | |
1593 | /* |
1594 | * Set up LVT0, LVT1: |
1595 | * |
1596 | * set up through-local-APIC on the boot CPU's LINT0. This is not |
1597 | * strictly necessary in pure symmetric-IO mode, but sometimes |
1598 | * we delegate interrupts to the 8259A. |
1599 | */ |
1600 | /* |
1601 | * TODO: set up through-local-APIC from through-I/O-APIC? --macro |
1602 | */ |
1603 | value = apic_read(APIC_LVT0) & APIC_LVT_MASKED; |
1604 | if (!cpu && (pic_mode || !value || ioapic_is_disabled)) { |
1605 | value = APIC_DM_EXTINT; |
1606 | apic_pr_verbose("Enabled ExtINT on CPU#%d\n", cpu); |
1607 | } else { |
1608 | value = APIC_DM_EXTINT | APIC_LVT_MASKED; |
1609 | apic_pr_verbose("Masked ExtINT on CPU#%d\n", cpu); |
1610 | } |
1611 | apic_write(APIC_LVT0, val: value); |
1612 | |
1613 | /* |
1614 | * Only the BSP sees the LINT1 NMI signal by default. This can be |
1615 | * modified by apic_extnmi= boot option. |
1616 | */ |
1617 | if ((!cpu && apic_extnmi != APIC_EXTNMI_NONE) || |
1618 | apic_extnmi == APIC_EXTNMI_ALL) |
1619 | value = APIC_DM_NMI; |
1620 | else |
1621 | value = APIC_DM_NMI | APIC_LVT_MASKED; |
1622 | |
1623 | /* Is 82489DX ? */ |
1624 | if (!lapic_is_integrated()) |
1625 | value |= APIC_LVT_LEVEL_TRIGGER; |
1626 | apic_write(APIC_LVT1, val: value); |
1627 | |
1628 | #ifdef CONFIG_X86_MCE_INTEL |
1629 | /* Recheck CMCI information after local APIC is up on CPU #0 */ |
1630 | if (!cpu) |
1631 | cmci_recheck(); |
1632 | #endif |
1633 | } |
1634 | |
1635 | static void end_local_APIC_setup(void) |
1636 | { |
1637 | lapic_setup_esr(); |
1638 | |
1639 | #ifdef CONFIG_X86_32 |
1640 | { |
1641 | unsigned int value; |
1642 | /* Disable the local apic timer */ |
1643 | value = apic_read(APIC_LVTT); |
1644 | value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR); |
1645 | apic_write(APIC_LVTT, value); |
1646 | } |
1647 | #endif |
1648 | |
1649 | apic_pm_activate(); |
1650 | } |
1651 | |
1652 | /* |
1653 | * APIC setup function for application processors. Called from smpboot.c |
1654 | */ |
1655 | void apic_ap_setup(void) |
1656 | { |
1657 | setup_local_APIC(); |
1658 | end_local_APIC_setup(); |
1659 | } |
1660 | |
1661 | static __init void apic_read_boot_cpu_id(bool x2apic) |
1662 | { |
1663 | /* |
1664 | * This can be invoked from check_x2apic() before the APIC has been |
1665 | * selected. But that code knows for sure that the BIOS enabled |
1666 | * X2APIC. |
1667 | */ |
1668 | if (x2apic) { |
1669 | boot_cpu_physical_apicid = native_apic_msr_read(APIC_ID); |
1670 | boot_cpu_apic_version = GET_APIC_VERSION(native_apic_msr_read(APIC_LVR)); |
1671 | } else { |
1672 | boot_cpu_physical_apicid = read_apic_id(); |
1673 | boot_cpu_apic_version = GET_APIC_VERSION(apic_read(APIC_LVR)); |
1674 | } |
1675 | topology_register_boot_apic(apic_id: boot_cpu_physical_apicid); |
1676 | } |
1677 | |
1678 | #ifdef CONFIG_X86_X2APIC |
1679 | int x2apic_mode; |
1680 | EXPORT_SYMBOL_GPL(x2apic_mode); |
1681 | |
1682 | enum { |
1683 | X2APIC_OFF, |
1684 | X2APIC_DISABLED, |
1685 | /* All states below here have X2APIC enabled */ |
1686 | X2APIC_ON, |
1687 | X2APIC_ON_LOCKED |
1688 | }; |
1689 | static int x2apic_state; |
1690 | |
1691 | static bool x2apic_hw_locked(void) |
1692 | { |
1693 | u64 x86_arch_cap_msr; |
1694 | u64 msr; |
1695 | |
1696 | x86_arch_cap_msr = x86_read_arch_cap_msr(); |
1697 | if (x86_arch_cap_msr & ARCH_CAP_XAPIC_DISABLE) { |
1698 | rdmsrq(MSR_IA32_XAPIC_DISABLE_STATUS, msr); |
1699 | return (msr & LEGACY_XAPIC_DISABLED); |
1700 | } |
1701 | return false; |
1702 | } |
1703 | |
1704 | static void __x2apic_disable(void) |
1705 | { |
1706 | u64 msr; |
1707 | |
1708 | if (!boot_cpu_has(X86_FEATURE_APIC)) |
1709 | return; |
1710 | |
1711 | rdmsrq(MSR_IA32_APICBASE, msr); |
1712 | if (!(msr & X2APIC_ENABLE)) |
1713 | return; |
1714 | /* Disable xapic and x2apic first and then reenable xapic mode */ |
1715 | wrmsrq(MSR_IA32_APICBASE, val: msr & ~(X2APIC_ENABLE | XAPIC_ENABLE)); |
1716 | wrmsrq(MSR_IA32_APICBASE, val: msr & ~X2APIC_ENABLE); |
1717 | printk_once(KERN_INFO "x2apic disabled\n"); |
1718 | } |
1719 | |
1720 | static void __x2apic_enable(void) |
1721 | { |
1722 | u64 msr; |
1723 | |
1724 | rdmsrq(MSR_IA32_APICBASE, msr); |
1725 | if (msr & X2APIC_ENABLE) |
1726 | return; |
1727 | wrmsrq(MSR_IA32_APICBASE, val: msr | X2APIC_ENABLE); |
1728 | printk_once(KERN_INFO "x2apic enabled\n"); |
1729 | } |
1730 | |
1731 | static int __init setup_nox2apic(char *str) |
1732 | { |
1733 | if (x2apic_enabled()) { |
1734 | u32 apicid = native_apic_msr_read(APIC_ID); |
1735 | |
1736 | if (apicid >= 255) { |
1737 | pr_warn("Apicid: %08x, cannot enforce nox2apic\n", |
1738 | apicid); |
1739 | return 0; |
1740 | } |
1741 | if (x2apic_hw_locked()) { |
1742 | pr_warn("APIC locked in x2apic mode, can't disable\n"); |
1743 | return 0; |
1744 | } |
1745 | pr_warn("x2apic already enabled.\n"); |
1746 | __x2apic_disable(); |
1747 | } |
1748 | setup_clear_cpu_cap(X86_FEATURE_X2APIC); |
1749 | x2apic_state = X2APIC_DISABLED; |
1750 | x2apic_mode = 0; |
1751 | return 0; |
1752 | } |
1753 | early_param("nox2apic", setup_nox2apic); |
1754 | |
1755 | /* Called from cpu_init() to enable x2apic on (secondary) cpus */ |
1756 | void x2apic_setup(void) |
1757 | { |
1758 | /* |
1759 | * Try to make the AP's APIC state match that of the BSP, but if the |
1760 | * BSP is unlocked and the AP is locked then there is a state mismatch. |
1761 | * Warn about the mismatch in case a GP fault occurs due to a locked AP |
1762 | * trying to be turned off. |
1763 | */ |
1764 | if (x2apic_state != X2APIC_ON_LOCKED && x2apic_hw_locked()) |
1765 | pr_warn("x2apic lock mismatch between BSP and AP.\n"); |
1766 | /* |
1767 | * If x2apic is not in ON or LOCKED state, disable it if already enabled |
1768 | * from BIOS. |
1769 | */ |
1770 | if (x2apic_state < X2APIC_ON) { |
1771 | __x2apic_disable(); |
1772 | return; |
1773 | } |
1774 | __x2apic_enable(); |
1775 | } |
1776 | |
1777 | static __init void apic_set_fixmap(bool read_apic); |
1778 | |
1779 | static __init void x2apic_disable(void) |
1780 | { |
1781 | u32 x2apic_id; |
1782 | |
1783 | if (x2apic_state < X2APIC_ON) |
1784 | return; |
1785 | |
1786 | x2apic_id = read_apic_id(); |
1787 | if (x2apic_id >= 255) |
1788 | panic(fmt: "Cannot disable x2apic, id: %08x\n", x2apic_id); |
1789 | |
1790 | if (x2apic_hw_locked()) { |
1791 | pr_warn("Cannot disable locked x2apic, id: %08x\n", x2apic_id); |
1792 | return; |
1793 | } |
1794 | |
1795 | __x2apic_disable(); |
1796 | |
1797 | x2apic_mode = 0; |
1798 | x2apic_state = X2APIC_DISABLED; |
1799 | |
1800 | /* |
1801 | * Don't reread the APIC ID as it was already done from |
1802 | * check_x2apic() and the APIC driver still is a x2APIC variant, |
1803 | * which fails to do the read after x2APIC was disabled. |
1804 | */ |
1805 | apic_set_fixmap(read_apic: false); |
1806 | } |
1807 | |
1808 | static __init void x2apic_enable(void) |
1809 | { |
1810 | if (x2apic_state != X2APIC_OFF) |
1811 | return; |
1812 | |
1813 | x2apic_mode = 1; |
1814 | x2apic_state = X2APIC_ON; |
1815 | __x2apic_enable(); |
1816 | } |
1817 | |
1818 | static __init void try_to_enable_x2apic(int remap_mode) |
1819 | { |
1820 | if (x2apic_state == X2APIC_DISABLED) |
1821 | return; |
1822 | |
1823 | if (remap_mode != IRQ_REMAP_X2APIC_MODE) { |
1824 | u32 apic_limit = 255; |
1825 | |
1826 | /* |
1827 | * Using X2APIC without IR is not architecturally supported |
1828 | * on bare metal but may be supported in guests. |
1829 | */ |
1830 | if (!x86_init.hyper.x2apic_available()) { |
1831 | pr_info("x2apic: IRQ remapping doesn't support X2APIC mode\n"); |
1832 | x2apic_disable(); |
1833 | return; |
1834 | } |
1835 | |
1836 | /* |
1837 | * If the hypervisor supports extended destination ID in |
1838 | * MSI, that increases the maximum APIC ID that can be |
1839 | * used for non-remapped IRQ domains. |
1840 | */ |
1841 | if (x86_init.hyper.msi_ext_dest_id()) { |
1842 | virt_ext_dest_id = 1; |
1843 | apic_limit = 32767; |
1844 | } |
1845 | |
1846 | /* |
1847 | * Without IR, all CPUs can be addressed by IOAPIC/MSI only |
1848 | * in physical mode, and CPUs with an APIC ID that cannot |
1849 | * be addressed must not be brought online. |
1850 | */ |
1851 | x2apic_set_max_apicid(apicid: apic_limit); |
1852 | x2apic_phys = 1; |
1853 | } |
1854 | x2apic_enable(); |
1855 | } |
1856 | |
1857 | void __init check_x2apic(void) |
1858 | { |
1859 | if (x2apic_enabled()) { |
1860 | pr_info("x2apic: enabled by BIOS, switching to x2apic ops\n"); |
1861 | x2apic_mode = 1; |
1862 | if (x2apic_hw_locked()) |
1863 | x2apic_state = X2APIC_ON_LOCKED; |
1864 | else |
1865 | x2apic_state = X2APIC_ON; |
1866 | apic_read_boot_cpu_id(x2apic: true); |
1867 | } else if (!boot_cpu_has(X86_FEATURE_X2APIC)) { |
1868 | x2apic_state = X2APIC_DISABLED; |
1869 | } |
1870 | } |
1871 | #else /* CONFIG_X86_X2APIC */ |
1872 | void __init check_x2apic(void) |
1873 | { |
1874 | if (!apic_is_x2apic_enabled()) |
1875 | return; |
1876 | /* |
1877 | * Checkme: Can we simply turn off x2APIC here instead of disabling the APIC? |
1878 | */ |
1879 | pr_err("Kernel does not support x2APIC, please recompile with CONFIG_X86_X2APIC.\n"); |
1880 | pr_err("Disabling APIC, expect reduced performance and functionality.\n"); |
1881 | |
1882 | apic_is_disabled = true; |
1883 | setup_clear_cpu_cap(X86_FEATURE_APIC); |
1884 | } |
1885 | |
1886 | static inline void try_to_enable_x2apic(int remap_mode) { } |
1887 | static inline void __x2apic_enable(void) { } |
1888 | #endif /* !CONFIG_X86_X2APIC */ |
1889 | |
1890 | void __init enable_IR_x2apic(void) |
1891 | { |
1892 | unsigned long flags; |
1893 | int ret, ir_stat; |
1894 | |
1895 | if (ioapic_is_disabled) { |
1896 | pr_info("Not enabling interrupt remapping due to skipped IO-APIC setup\n"); |
1897 | return; |
1898 | } |
1899 | |
1900 | ir_stat = irq_remapping_prepare(); |
1901 | if (ir_stat < 0 && !x2apic_supported()) |
1902 | return; |
1903 | |
1904 | ret = save_ioapic_entries(); |
1905 | if (ret) { |
1906 | pr_info("Saving IO-APIC state failed: %d\n", ret); |
1907 | return; |
1908 | } |
1909 | |
1910 | local_irq_save(flags); |
1911 | legacy_pic->mask_all(); |
1912 | mask_ioapic_entries(); |
1913 | |
1914 | /* If irq_remapping_prepare() succeeded, try to enable it */ |
1915 | if (ir_stat >= 0) |
1916 | ir_stat = irq_remapping_enable(); |
1917 | /* ir_stat contains the remap mode or an error code */ |
1918 | try_to_enable_x2apic(remap_mode: ir_stat); |
1919 | |
1920 | if (ir_stat < 0) |
1921 | restore_ioapic_entries(); |
1922 | legacy_pic->restore_mask(); |
1923 | local_irq_restore(flags); |
1924 | } |
1925 | |
1926 | #ifdef CONFIG_X86_64 |
1927 | /* |
1928 | * Detect and enable local APICs on non-SMP boards. |
1929 | * Original code written by Keir Fraser. |
1930 | * On AMD64 we trust the BIOS - if it says no APIC it is likely |
1931 | * not correctly set up (usually the APIC timer won't work etc.) |
1932 | */ |
1933 | static bool __init detect_init_APIC(void) |
1934 | { |
1935 | if (!boot_cpu_has(X86_FEATURE_APIC)) { |
1936 | pr_info("No local APIC present\n"); |
1937 | return false; |
1938 | } |
1939 | |
1940 | register_lapic_address(APIC_DEFAULT_PHYS_BASE); |
1941 | return true; |
1942 | } |
1943 | #else |
1944 | |
1945 | static bool __init apic_verify(unsigned long addr) |
1946 | { |
1947 | u32 features, h, l; |
1948 | |
1949 | /* |
1950 | * The APIC feature bit should now be enabled |
1951 | * in `cpuid' |
1952 | */ |
1953 | features = cpuid_edx(1); |
1954 | if (!(features & (1 << X86_FEATURE_APIC))) { |
1955 | pr_warn("Could not enable APIC!\n"); |
1956 | return false; |
1957 | } |
1958 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC); |
1959 | |
1960 | /* The BIOS may have set up the APIC at some other address */ |
1961 | if (boot_cpu_data.x86 >= 6) { |
1962 | rdmsr(MSR_IA32_APICBASE, l, h); |
1963 | if (l & MSR_IA32_APICBASE_ENABLE) |
1964 | addr = l & MSR_IA32_APICBASE_BASE; |
1965 | } |
1966 | |
1967 | register_lapic_address(addr); |
1968 | pr_info("Found and enabled local APIC!\n"); |
1969 | return true; |
1970 | } |
1971 | |
1972 | bool __init apic_force_enable(unsigned long addr) |
1973 | { |
1974 | u32 h, l; |
1975 | |
1976 | if (apic_is_disabled) |
1977 | return false; |
1978 | |
1979 | /* |
1980 | * Some BIOSes disable the local APIC in the APIC_BASE |
1981 | * MSR. This can only be done in software for Intel P6 or later |
1982 | * and AMD K7 (Model > 1) or later. |
1983 | */ |
1984 | if (boot_cpu_data.x86 >= 6) { |
1985 | rdmsr(MSR_IA32_APICBASE, l, h); |
1986 | if (!(l & MSR_IA32_APICBASE_ENABLE)) { |
1987 | pr_info("Local APIC disabled by BIOS -- reenabling.\n"); |
1988 | l &= ~MSR_IA32_APICBASE_BASE; |
1989 | l |= MSR_IA32_APICBASE_ENABLE | addr; |
1990 | wrmsr(MSR_IA32_APICBASE, l, h); |
1991 | enabled_via_apicbase = 1; |
1992 | } |
1993 | } |
1994 | return apic_verify(addr); |
1995 | } |
1996 | |
1997 | /* |
1998 | * Detect and initialize APIC |
1999 | */ |
2000 | static bool __init detect_init_APIC(void) |
2001 | { |
2002 | /* Disabled by kernel option? */ |
2003 | if (apic_is_disabled) |
2004 | return false; |
2005 | |
2006 | switch (boot_cpu_data.x86_vendor) { |
2007 | case X86_VENDOR_AMD: |
2008 | if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) || |
2009 | (boot_cpu_data.x86 >= 15)) |
2010 | break; |
2011 | goto no_apic; |
2012 | case X86_VENDOR_HYGON: |
2013 | break; |
2014 | case X86_VENDOR_INTEL: |
2015 | if ((boot_cpu_data.x86 == 5 && boot_cpu_has(X86_FEATURE_APIC)) || |
2016 | boot_cpu_data.x86_vfm >= INTEL_PENTIUM_PRO) |
2017 | break; |
2018 | goto no_apic; |
2019 | default: |
2020 | goto no_apic; |
2021 | } |
2022 | |
2023 | if (!boot_cpu_has(X86_FEATURE_APIC)) { |
2024 | /* |
2025 | * Over-ride BIOS and try to enable the local APIC only if |
2026 | * "lapic" specified. |
2027 | */ |
2028 | if (!force_enable_local_apic) { |
2029 | pr_info("Local APIC disabled by BIOS -- " |
2030 | "you can enable it with \"lapic\"\n"); |
2031 | return false; |
2032 | } |
2033 | if (!apic_force_enable(APIC_DEFAULT_PHYS_BASE)) |
2034 | return false; |
2035 | } else { |
2036 | if (!apic_verify(APIC_DEFAULT_PHYS_BASE)) |
2037 | return false; |
2038 | } |
2039 | |
2040 | apic_pm_activate(); |
2041 | |
2042 | return true; |
2043 | |
2044 | no_apic: |
2045 | pr_info("No local APIC present or hardware disabled\n"); |
2046 | return false; |
2047 | } |
2048 | #endif |
2049 | |
2050 | /** |
2051 | * init_apic_mappings - initialize APIC mappings |
2052 | */ |
2053 | void __init init_apic_mappings(void) |
2054 | { |
2055 | if (apic_validate_deadline_timer()) |
2056 | pr_info("TSC deadline timer available\n"); |
2057 | |
2058 | if (x2apic_mode) |
2059 | return; |
2060 | |
2061 | if (!smp_found_config) { |
2062 | if (!detect_init_APIC()) { |
2063 | pr_info("APIC: disable apic facility\n"); |
2064 | apic_disable(); |
2065 | } |
2066 | } |
2067 | } |
2068 | |
2069 | static __init void apic_set_fixmap(bool read_apic) |
2070 | { |
2071 | set_fixmap_nocache(FIX_APIC_BASE, mp_lapic_addr); |
2072 | apic_mmio_base = APIC_BASE; |
2073 | apic_pr_verbose("Mapped APIC to %16lx (%16lx)\n", apic_mmio_base, mp_lapic_addr); |
2074 | if (read_apic) |
2075 | apic_read_boot_cpu_id(x2apic: false); |
2076 | } |
2077 | |
2078 | void __init register_lapic_address(unsigned long address) |
2079 | { |
2080 | /* This should only happen once */ |
2081 | WARN_ON_ONCE(mp_lapic_addr); |
2082 | mp_lapic_addr = address; |
2083 | |
2084 | if (!x2apic_mode) |
2085 | apic_set_fixmap(read_apic: true); |
2086 | } |
2087 | |
2088 | /* |
2089 | * Local APIC interrupts |
2090 | */ |
2091 | |
2092 | /* |
2093 | * Common handling code for spurious_interrupt and spurious_vector entry |
2094 | * points below. No point in allowing the compiler to inline it twice. |
2095 | */ |
2096 | static noinline void handle_spurious_interrupt(u8 vector) |
2097 | { |
2098 | u32 v; |
2099 | |
2100 | trace_spurious_apic_entry(vector); |
2101 | |
2102 | inc_irq_stat(irq_spurious_count); |
2103 | |
2104 | /* |
2105 | * If this is a spurious interrupt then do not acknowledge |
2106 | */ |
2107 | if (vector == SPURIOUS_APIC_VECTOR) { |
2108 | /* See SDM vol 3 */ |
2109 | pr_info("Spurious APIC interrupt (vector 0xFF) on CPU#%d, should never happen.\n", |
2110 | smp_processor_id()); |
2111 | goto out; |
2112 | } |
2113 | |
2114 | /* |
2115 | * If it is a vectored one, verify it's set in the ISR. If set, |
2116 | * acknowledge it. |
2117 | */ |
2118 | v = apic_read(APIC_ISR + ((vector & ~0x1f) >> 1)); |
2119 | if (v & (1 << (vector & 0x1f))) { |
2120 | pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Acked\n", |
2121 | vector, smp_processor_id()); |
2122 | apic_eoi(); |
2123 | } else { |
2124 | pr_info("Spurious interrupt (vector 0x%02x) on CPU#%d. Not pending!\n", |
2125 | vector, smp_processor_id()); |
2126 | } |
2127 | out: |
2128 | trace_spurious_apic_exit(vector); |
2129 | } |
2130 | |
2131 | /** |
2132 | * spurious_interrupt - Catch all for interrupts raised on unused vectors |
2133 | * @regs: Pointer to pt_regs on stack |
2134 | * @vector: The vector number |
2135 | * |
2136 | * This is invoked from ASM entry code to catch all interrupts which |
2137 | * trigger on an entry which is routed to the common_spurious idtentry |
2138 | * point. |
2139 | */ |
2140 | DEFINE_IDTENTRY_IRQ(spurious_interrupt) |
2141 | { |
2142 | handle_spurious_interrupt(vector); |
2143 | } |
2144 | |
2145 | DEFINE_IDTENTRY_SYSVEC(sysvec_spurious_apic_interrupt) |
2146 | { |
2147 | handle_spurious_interrupt(SPURIOUS_APIC_VECTOR); |
2148 | } |
2149 | |
2150 | /* |
2151 | * This interrupt should never happen with our APIC/SMP architecture |
2152 | */ |
2153 | DEFINE_IDTENTRY_SYSVEC(sysvec_error_interrupt) |
2154 | { |
2155 | static const char * const error_interrupt_reason[] = { |
2156 | "Send CS error", /* APIC Error Bit 0 */ |
2157 | "Receive CS error", /* APIC Error Bit 1 */ |
2158 | "Send accept error", /* APIC Error Bit 2 */ |
2159 | "Receive accept error", /* APIC Error Bit 3 */ |
2160 | "Redirectable IPI", /* APIC Error Bit 4 */ |
2161 | "Send illegal vector", /* APIC Error Bit 5 */ |
2162 | "Received illegal vector", /* APIC Error Bit 6 */ |
2163 | "Illegal register address", /* APIC Error Bit 7 */ |
2164 | }; |
2165 | u32 v, i = 0; |
2166 | |
2167 | trace_error_apic_entry(ERROR_APIC_VECTOR); |
2168 | |
2169 | /* First tickle the hardware, only then report what went on. -- REW */ |
2170 | if (lapic_get_maxlvt() > 3) /* Due to the Pentium erratum 3AP. */ |
2171 | apic_write(APIC_ESR, val: 0); |
2172 | v = apic_read(APIC_ESR); |
2173 | apic_eoi(); |
2174 | atomic_inc(v: &irq_err_count); |
2175 | |
2176 | apic_pr_debug("APIC error on CPU%d: %02x", smp_processor_id(), v); |
2177 | |
2178 | v &= 0xff; |
2179 | while (v) { |
2180 | if (v & 0x1) |
2181 | apic_pr_debug_cont(" : %s", error_interrupt_reason[i]); |
2182 | i++; |
2183 | v >>= 1; |
2184 | } |
2185 | |
2186 | apic_pr_debug_cont("\n"); |
2187 | |
2188 | trace_error_apic_exit(ERROR_APIC_VECTOR); |
2189 | } |
2190 | |
2191 | /** |
2192 | * connect_bsp_APIC - attach the APIC to the interrupt system |
2193 | */ |
2194 | static void __init connect_bsp_APIC(void) |
2195 | { |
2196 | #ifdef CONFIG_X86_32 |
2197 | if (pic_mode) { |
2198 | /* |
2199 | * Do not trust the local APIC being empty at bootup. |
2200 | */ |
2201 | clear_local_APIC(); |
2202 | /* |
2203 | * PIC mode, enable APIC mode in the IMCR, i.e. connect BSP's |
2204 | * local APIC to INT and NMI lines. |
2205 | */ |
2206 | apic_pr_verbose("Leaving PIC mode, enabling APIC mode.\n"); |
2207 | imcr_pic_to_apic(); |
2208 | } |
2209 | #endif |
2210 | } |
2211 | |
2212 | /** |
2213 | * disconnect_bsp_APIC - detach the APIC from the interrupt system |
2214 | * @virt_wire_setup: indicates, whether virtual wire mode is selected |
2215 | * |
2216 | * Virtual wire mode is necessary to deliver legacy interrupts even when the |
2217 | * APIC is disabled. |
2218 | */ |
2219 | void disconnect_bsp_APIC(int virt_wire_setup) |
2220 | { |
2221 | unsigned int value; |
2222 | |
2223 | #ifdef CONFIG_X86_32 |
2224 | if (pic_mode) { |
2225 | /* |
2226 | * Put the board back into PIC mode (has an effect only on |
2227 | * certain older boards). Note that APIC interrupts, including |
2228 | * IPIs, won't work beyond this point! The only exception are |
2229 | * INIT IPIs. |
2230 | */ |
2231 | apic_pr_verbose("Disabling APIC mode, entering PIC mode.\n"); |
2232 | imcr_apic_to_pic(); |
2233 | return; |
2234 | } |
2235 | #endif |
2236 | |
2237 | /* Go back to Virtual Wire compatibility mode */ |
2238 | |
2239 | /* For the spurious interrupt use vector F, and enable it */ |
2240 | value = apic_read(APIC_SPIV); |
2241 | value &= ~APIC_VECTOR_MASK; |
2242 | value |= APIC_SPIV_APIC_ENABLED; |
2243 | value |= 0xf; |
2244 | apic_write(APIC_SPIV, val: value); |
2245 | |
2246 | if (!virt_wire_setup) { |
2247 | /* |
2248 | * For LVT0 make it edge triggered, active high, |
2249 | * external and enabled |
2250 | */ |
2251 | value = apic_read(APIC_LVT0); |
2252 | value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING | |
2253 | APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR | |
2254 | APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); |
2255 | value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; |
2256 | value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT); |
2257 | apic_write(APIC_LVT0, val: value); |
2258 | } else { |
2259 | /* Disable LVT0 */ |
2260 | apic_write(APIC_LVT0, APIC_LVT_MASKED); |
2261 | } |
2262 | |
2263 | /* |
2264 | * For LVT1 make it edge triggered, active high, |
2265 | * nmi and enabled |
2266 | */ |
2267 | value = apic_read(APIC_LVT1); |
2268 | value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING | |
2269 | APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR | |
2270 | APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED); |
2271 | value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING; |
2272 | value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI); |
2273 | apic_write(APIC_LVT1, val: value); |
2274 | } |
2275 | |
2276 | void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg, |
2277 | bool dmar) |
2278 | { |
2279 | memset(msg, 0, sizeof(*msg)); |
2280 | |
2281 | msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW; |
2282 | msg->arch_addr_lo.dest_mode_logical = apic->dest_mode_logical; |
2283 | msg->arch_addr_lo.destid_0_7 = cfg->dest_apicid & 0xFF; |
2284 | |
2285 | msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_FIXED; |
2286 | msg->arch_data.vector = cfg->vector; |
2287 | |
2288 | msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH; |
2289 | /* |
2290 | * Only the IOMMU itself can use the trick of putting destination |
2291 | * APIC ID into the high bits of the address. Anything else would |
2292 | * just be writing to memory if it tried that, and needs IR to |
2293 | * address APICs which can't be addressed in the normal 32-bit |
2294 | * address range at 0xFFExxxxx. That is typically just 8 bits, but |
2295 | * some hypervisors allow the extended destination ID field in bits |
2296 | * 5-11 to be used, giving support for 15 bits of APIC IDs in total. |
2297 | */ |
2298 | if (dmar) |
2299 | msg->arch_addr_hi.destid_8_31 = cfg->dest_apicid >> 8; |
2300 | else if (virt_ext_dest_id && cfg->dest_apicid < 0x8000) |
2301 | msg->arch_addr_lo.virt_destid_8_14 = cfg->dest_apicid >> 8; |
2302 | else |
2303 | WARN_ON_ONCE(cfg->dest_apicid > 0xFF); |
2304 | } |
2305 | |
2306 | u32 x86_msi_msg_get_destid(struct msi_msg *msg, bool extid) |
2307 | { |
2308 | u32 dest = msg->arch_addr_lo.destid_0_7; |
2309 | |
2310 | if (extid) |
2311 | dest |= msg->arch_addr_hi.destid_8_31 << 8; |
2312 | return dest; |
2313 | } |
2314 | EXPORT_SYMBOL_GPL(x86_msi_msg_get_destid); |
2315 | |
2316 | static void __init apic_bsp_up_setup(void) |
2317 | { |
2318 | reset_phys_cpu_present_map(apicid: boot_cpu_physical_apicid); |
2319 | } |
2320 | |
2321 | /** |
2322 | * apic_bsp_setup - Setup function for local apic and io-apic |
2323 | * @upmode: Force UP mode (for APIC_init_uniprocessor) |
2324 | */ |
2325 | static void __init apic_bsp_setup(bool upmode) |
2326 | { |
2327 | connect_bsp_APIC(); |
2328 | if (upmode) |
2329 | apic_bsp_up_setup(); |
2330 | setup_local_APIC(); |
2331 | |
2332 | enable_IO_APIC(); |
2333 | end_local_APIC_setup(); |
2334 | irq_remap_enable_fault_handling(); |
2335 | setup_IO_APIC(); |
2336 | lapic_update_legacy_vectors(); |
2337 | } |
2338 | |
2339 | #ifdef CONFIG_UP_LATE_INIT |
2340 | void __init up_late_init(void) |
2341 | { |
2342 | if (apic_intr_mode == APIC_PIC) |
2343 | return; |
2344 | |
2345 | /* Setup local timer */ |
2346 | x86_init.timers.setup_percpu_clockev(); |
2347 | } |
2348 | #endif |
2349 | |
2350 | /* |
2351 | * Power management |
2352 | */ |
2353 | #ifdef CONFIG_PM |
2354 | |
2355 | static struct { |
2356 | /* |
2357 | * 'active' is true if the local APIC was enabled by us and |
2358 | * not the BIOS; this signifies that we are also responsible |
2359 | * for disabling it before entering apm/acpi suspend |
2360 | */ |
2361 | int active; |
2362 | /* r/w apic fields */ |
2363 | u32 apic_id; |
2364 | unsigned int apic_taskpri; |
2365 | unsigned int apic_ldr; |
2366 | unsigned int apic_dfr; |
2367 | unsigned int apic_spiv; |
2368 | unsigned int apic_lvtt; |
2369 | unsigned int apic_lvtpc; |
2370 | unsigned int apic_lvt0; |
2371 | unsigned int apic_lvt1; |
2372 | unsigned int apic_lvterr; |
2373 | unsigned int apic_tmict; |
2374 | unsigned int apic_tdcr; |
2375 | unsigned int apic_thmr; |
2376 | unsigned int apic_cmci; |
2377 | } apic_pm_state; |
2378 | |
2379 | static int lapic_suspend(void) |
2380 | { |
2381 | unsigned long flags; |
2382 | int maxlvt; |
2383 | |
2384 | if (!apic_pm_state.active) |
2385 | return 0; |
2386 | |
2387 | maxlvt = lapic_get_maxlvt(); |
2388 | |
2389 | apic_pm_state.apic_id = apic_read(APIC_ID); |
2390 | apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI); |
2391 | apic_pm_state.apic_ldr = apic_read(APIC_LDR); |
2392 | apic_pm_state.apic_dfr = apic_read(APIC_DFR); |
2393 | apic_pm_state.apic_spiv = apic_read(APIC_SPIV); |
2394 | apic_pm_state.apic_lvtt = apic_read(APIC_LVTT); |
2395 | if (maxlvt >= 4) |
2396 | apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC); |
2397 | apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0); |
2398 | apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1); |
2399 | apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR); |
2400 | apic_pm_state.apic_tmict = apic_read(APIC_TMICT); |
2401 | apic_pm_state.apic_tdcr = apic_read(APIC_TDCR); |
2402 | #ifdef CONFIG_X86_THERMAL_VECTOR |
2403 | if (maxlvt >= 5) |
2404 | apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR); |
2405 | #endif |
2406 | #ifdef CONFIG_X86_MCE_INTEL |
2407 | if (maxlvt >= 6) |
2408 | apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI); |
2409 | #endif |
2410 | |
2411 | local_irq_save(flags); |
2412 | |
2413 | /* |
2414 | * Mask IOAPIC before disabling the local APIC to prevent stale IRR |
2415 | * entries on some implementations. |
2416 | */ |
2417 | mask_ioapic_entries(); |
2418 | |
2419 | disable_local_APIC(); |
2420 | |
2421 | irq_remapping_disable(); |
2422 | |
2423 | local_irq_restore(flags); |
2424 | return 0; |
2425 | } |
2426 | |
2427 | static void lapic_resume(void) |
2428 | { |
2429 | unsigned int l, h; |
2430 | unsigned long flags; |
2431 | int maxlvt; |
2432 | |
2433 | if (!apic_pm_state.active) |
2434 | return; |
2435 | |
2436 | local_irq_save(flags); |
2437 | |
2438 | /* |
2439 | * IO-APIC and PIC have their own resume routines. |
2440 | * We just mask them here to make sure the interrupt |
2441 | * subsystem is completely quiet while we enable x2apic |
2442 | * and interrupt-remapping. |
2443 | */ |
2444 | mask_ioapic_entries(); |
2445 | legacy_pic->mask_all(); |
2446 | |
2447 | if (x2apic_mode) { |
2448 | __x2apic_enable(); |
2449 | } else { |
2450 | /* |
2451 | * Make sure the APICBASE points to the right address |
2452 | * |
2453 | * FIXME! This will be wrong if we ever support suspend on |
2454 | * SMP! We'll need to do this as part of the CPU restore! |
2455 | */ |
2456 | if (boot_cpu_data.x86 >= 6) { |
2457 | rdmsr(MSR_IA32_APICBASE, l, h); |
2458 | l &= ~MSR_IA32_APICBASE_BASE; |
2459 | l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr; |
2460 | wrmsr(MSR_IA32_APICBASE, low: l, high: h); |
2461 | } |
2462 | } |
2463 | |
2464 | maxlvt = lapic_get_maxlvt(); |
2465 | apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED); |
2466 | apic_write(APIC_ID, val: apic_pm_state.apic_id); |
2467 | apic_write(APIC_DFR, val: apic_pm_state.apic_dfr); |
2468 | apic_write(APIC_LDR, val: apic_pm_state.apic_ldr); |
2469 | apic_write(APIC_TASKPRI, val: apic_pm_state.apic_taskpri); |
2470 | apic_write(APIC_SPIV, val: apic_pm_state.apic_spiv); |
2471 | apic_write(APIC_LVT0, val: apic_pm_state.apic_lvt0); |
2472 | apic_write(APIC_LVT1, val: apic_pm_state.apic_lvt1); |
2473 | #ifdef CONFIG_X86_THERMAL_VECTOR |
2474 | if (maxlvt >= 5) |
2475 | apic_write(APIC_LVTTHMR, val: apic_pm_state.apic_thmr); |
2476 | #endif |
2477 | #ifdef CONFIG_X86_MCE_INTEL |
2478 | if (maxlvt >= 6) |
2479 | apic_write(APIC_LVTCMCI, val: apic_pm_state.apic_cmci); |
2480 | #endif |
2481 | if (maxlvt >= 4) |
2482 | apic_write(APIC_LVTPC, val: apic_pm_state.apic_lvtpc); |
2483 | apic_write(APIC_LVTT, val: apic_pm_state.apic_lvtt); |
2484 | apic_write(APIC_TDCR, val: apic_pm_state.apic_tdcr); |
2485 | apic_write(APIC_TMICT, val: apic_pm_state.apic_tmict); |
2486 | apic_write(APIC_ESR, val: 0); |
2487 | apic_read(APIC_ESR); |
2488 | apic_write(APIC_LVTERR, val: apic_pm_state.apic_lvterr); |
2489 | apic_write(APIC_ESR, val: 0); |
2490 | apic_read(APIC_ESR); |
2491 | |
2492 | irq_remapping_reenable(x2apic_mode); |
2493 | |
2494 | local_irq_restore(flags); |
2495 | } |
2496 | |
2497 | /* |
2498 | * This device has no shutdown method - fully functioning local APICs |
2499 | * are needed on every CPU up until machine_halt/restart/poweroff. |
2500 | */ |
2501 | |
2502 | static struct syscore_ops lapic_syscore_ops = { |
2503 | .resume = lapic_resume, |
2504 | .suspend = lapic_suspend, |
2505 | }; |
2506 | |
2507 | static void apic_pm_activate(void) |
2508 | { |
2509 | apic_pm_state.active = 1; |
2510 | } |
2511 | |
2512 | static int __init init_lapic_sysfs(void) |
2513 | { |
2514 | /* XXX: remove suspend/resume procs if !apic_pm_state.active? */ |
2515 | if (boot_cpu_has(X86_FEATURE_APIC)) |
2516 | register_syscore_ops(ops: &lapic_syscore_ops); |
2517 | |
2518 | return 0; |
2519 | } |
2520 | |
2521 | /* local apic needs to resume before other devices access its registers. */ |
2522 | core_initcall(init_lapic_sysfs); |
2523 | |
2524 | #else /* CONFIG_PM */ |
2525 | |
2526 | static void apic_pm_activate(void) { } |
2527 | |
2528 | #endif /* CONFIG_PM */ |
2529 | |
2530 | #ifdef CONFIG_X86_64 |
2531 | |
2532 | static int multi_checked; |
2533 | static int multi; |
2534 | |
2535 | static int set_multi(const struct dmi_system_id *d) |
2536 | { |
2537 | if (multi) |
2538 | return 0; |
2539 | pr_info("APIC: %s detected, Multi Chassis\n", d->ident); |
2540 | multi = 1; |
2541 | return 0; |
2542 | } |
2543 | |
2544 | static const struct dmi_system_id multi_dmi_table[] = { |
2545 | { |
2546 | .callback = set_multi, |
2547 | .ident = "IBM System Summit2", |
2548 | .matches = { |
2549 | DMI_MATCH(DMI_SYS_VENDOR, "IBM"), |
2550 | DMI_MATCH(DMI_PRODUCT_NAME, "Summit2"), |
2551 | }, |
2552 | }, |
2553 | {} |
2554 | }; |
2555 | |
2556 | static void dmi_check_multi(void) |
2557 | { |
2558 | if (multi_checked) |
2559 | return; |
2560 | |
2561 | dmi_check_system(list: multi_dmi_table); |
2562 | multi_checked = 1; |
2563 | } |
2564 | |
2565 | /* |
2566 | * apic_is_clustered_box() -- Check if we can expect good TSC |
2567 | * |
2568 | * Thus far, the major user of this is IBM's Summit2 series: |
2569 | * Clustered boxes may have unsynced TSC problems if they are |
2570 | * multi-chassis. |
2571 | * Use DMI to check them |
2572 | */ |
2573 | int apic_is_clustered_box(void) |
2574 | { |
2575 | dmi_check_multi(); |
2576 | return multi; |
2577 | } |
2578 | #endif |
2579 | |
2580 | /* |
2581 | * APIC command line parameters |
2582 | */ |
2583 | static int __init setup_nolapic(char *arg) |
2584 | { |
2585 | apic_is_disabled = true; |
2586 | setup_clear_cpu_cap(X86_FEATURE_APIC); |
2587 | return 0; |
2588 | } |
2589 | early_param("nolapic", setup_nolapic); |
2590 | |
2591 | static int __init parse_lapic_timer_c2_ok(char *arg) |
2592 | { |
2593 | local_apic_timer_c2_ok = 1; |
2594 | return 0; |
2595 | } |
2596 | early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok); |
2597 | |
2598 | static int __init parse_disable_apic_timer(char *arg) |
2599 | { |
2600 | disable_apic_timer = 1; |
2601 | return 0; |
2602 | } |
2603 | early_param("noapictimer", parse_disable_apic_timer); |
2604 | |
2605 | static int __init parse_nolapic_timer(char *arg) |
2606 | { |
2607 | disable_apic_timer = 1; |
2608 | return 0; |
2609 | } |
2610 | early_param("nolapic_timer", parse_nolapic_timer); |
2611 | |
2612 | static int __init apic_set_verbosity(char *arg) |
2613 | { |
2614 | if (!arg) { |
2615 | if (IS_ENABLED(CONFIG_X86_32)) |
2616 | return -EINVAL; |
2617 | |
2618 | ioapic_is_disabled = false; |
2619 | return 0; |
2620 | } |
2621 | |
2622 | if (strcmp("debug", arg) == 0) |
2623 | apic_verbosity = APIC_DEBUG; |
2624 | else if (strcmp("verbose", arg) == 0) |
2625 | apic_verbosity = APIC_VERBOSE; |
2626 | #ifdef CONFIG_X86_64 |
2627 | else { |
2628 | pr_warn("APIC Verbosity level %s not recognised" |
2629 | " use apic=verbose or apic=debug\n", arg); |
2630 | return -EINVAL; |
2631 | } |
2632 | #endif |
2633 | |
2634 | return 0; |
2635 | } |
2636 | early_param("apic", apic_set_verbosity); |
2637 | |
2638 | static int __init lapic_insert_resource(void) |
2639 | { |
2640 | if (!apic_mmio_base) |
2641 | return -1; |
2642 | |
2643 | /* Put local APIC into the resource map. */ |
2644 | lapic_resource.start = apic_mmio_base; |
2645 | lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1; |
2646 | insert_resource(parent: &iomem_resource, new: &lapic_resource); |
2647 | |
2648 | return 0; |
2649 | } |
2650 | |
2651 | /* |
2652 | * need call insert after e820__reserve_resources() |
2653 | * that is using request_resource |
2654 | */ |
2655 | late_initcall(lapic_insert_resource); |
2656 | |
2657 | static int __init apic_set_extnmi(char *arg) |
2658 | { |
2659 | if (!arg) |
2660 | return -EINVAL; |
2661 | |
2662 | if (!strncmp("all", arg, 3)) |
2663 | apic_extnmi = APIC_EXTNMI_ALL; |
2664 | else if (!strncmp("none", arg, 4)) |
2665 | apic_extnmi = APIC_EXTNMI_NONE; |
2666 | else if (!strncmp("bsp", arg, 3)) |
2667 | apic_extnmi = APIC_EXTNMI_BSP; |
2668 | else { |
2669 | pr_warn("Unknown external NMI delivery mode `%s' ignored\n", arg); |
2670 | return -EINVAL; |
2671 | } |
2672 | |
2673 | return 0; |
2674 | } |
2675 | early_param("apic_extnmi", apic_set_extnmi); |
2676 |
Definitions
- boot_cpu_physical_apicid
- boot_cpu_apic_version
- apic_extnmi
- virt_ext_dest_id
- apic_mmio_base
- apic_accessible
- force_enable_local_apic
- parse_lapic
- apic_calibrate_pmtmr
- setup_apicpmtimer
- mp_lapic_addr
- apic_is_disabled
- disable_apic_timer
- local_apic_timer_c2_ok
- apic_verbosity
- pic_mode
- smp_found_config
- lapic_resource
- lapic_timer_period
- lapic_get_version
- lapic_is_integrated
- modern_apic
- apic_disable
- native_apic_icr_write
- native_apic_icr_read
- lapic_get_maxlvt
- __setup_APIC_LVTT
- eilvt_offsets
- eilvt_entry_is_changeable
- reserve_eilvt_offset
- setup_APIC_eilvt
- lapic_next_event
- lapic_next_deadline
- lapic_timer_shutdown
- lapic_timer_set_periodic_oneshot
- lapic_timer_set_periodic
- lapic_timer_set_oneshot
- lapic_timer_broadcast
- lapic_clockevent
- lapic_events
- deadline_match
- apic_validate_deadline_timer
- setup_APIC_timer
- __lapic_update_tsc_freq
- lapic_update_tsc_freq
- lapic_cal_loops
- lapic_cal_t1
- lapic_cal_t2
- lapic_cal_tsc1
- lapic_cal_tsc2
- lapic_cal_pm1
- lapic_cal_pm2
- lapic_cal_j1
- lapic_cal_j2
- lapic_cal_handler
- calibrate_by_pmtimer
- lapic_init_clockevent
- apic_needs_pit
- calibrate_APIC_clock
- setup_boot_APIC_clock
- setup_secondary_APIC_clock
- local_apic_timer_interrupt
- sysvec_apic_timer_interrupt
- clear_local_APIC
- apic_soft_disable
- disable_local_APIC
- lapic_shutdown
- sync_Arb_IDs
- apic_intr_mode
- __apic_intr_mode_select
- apic_intr_mode_select
- init_bsp_APIC
- apic_intr_mode_init
- lapic_setup_esr
- apic_ir
- apic_check_and_ack
- apic_pending_intr_clear
- setup_local_APIC
- end_local_APIC_setup
- apic_ap_setup
- apic_read_boot_cpu_id
- x2apic_mode
- x2apic_state
- x2apic_hw_locked
- __x2apic_disable
- __x2apic_enable
- setup_nox2apic
- x2apic_setup
- x2apic_disable
- x2apic_enable
- try_to_enable_x2apic
- check_x2apic
- enable_IR_x2apic
- detect_init_APIC
- init_apic_mappings
- apic_set_fixmap
- register_lapic_address
- handle_spurious_interrupt
- spurious_interrupt
- sysvec_spurious_apic_interrupt
- sysvec_error_interrupt
- connect_bsp_APIC
- disconnect_bsp_APIC
- __irq_msi_compose_msg
- x86_msi_msg_get_destid
- apic_bsp_up_setup
- apic_bsp_setup
- apic_pm_state
- lapic_suspend
- lapic_resume
- lapic_syscore_ops
- apic_pm_activate
- init_lapic_sysfs
- multi_checked
- multi
- set_multi
- multi_dmi_table
- dmi_check_multi
- apic_is_clustered_box
- setup_nolapic
- parse_lapic_timer_c2_ok
- parse_disable_apic_timer
- parse_nolapic_timer
- apic_set_verbosity
- lapic_insert_resource
Improve your Profiling and Debugging skills
Find out more