Sysace: Use the established platform bus api
[wrt350n-kernel.git] / arch / x86_64 / kernel / apic.c
blob925758dbca0c6016b52e91ab90e7503c90ea08c0
1 /*
2 * Local APIC handling, local APIC timers
4 * (c) 1999, 2000 Ingo Molnar <mingo@redhat.com>
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.
17 #include <linux/init.h>
19 #include <linux/mm.h>
20 #include <linux/delay.h>
21 #include <linux/bootmem.h>
22 #include <linux/interrupt.h>
23 #include <linux/mc146818rtc.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/sysdev.h>
26 #include <linux/module.h>
27 #include <linux/ioport.h>
29 #include <asm/atomic.h>
30 #include <asm/smp.h>
31 #include <asm/mtrr.h>
32 #include <asm/mpspec.h>
33 #include <asm/pgalloc.h>
34 #include <asm/mach_apic.h>
35 #include <asm/nmi.h>
36 #include <asm/idle.h>
37 #include <asm/proto.h>
38 #include <asm/timex.h>
39 #include <asm/hpet.h>
40 #include <asm/apic.h>
42 int apic_mapped;
43 int apic_verbosity;
44 int apic_runs_main_timer;
45 int apic_calibrate_pmtmr __initdata;
47 int disable_apic_timer __initdata;
49 /* Local APIC timer works in C2? */
50 int local_apic_timer_c2_ok;
51 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
53 static struct resource *ioapic_resources;
54 static struct resource lapic_resource = {
55 .name = "Local APIC",
56 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
60 * cpu_mask that denotes the CPUs that needs timer interrupt coming in as
61 * IPIs in place of local APIC timers
63 static cpumask_t timer_interrupt_broadcast_ipi_mask;
65 /* Using APIC to generate smp_local_timer_interrupt? */
66 int using_apic_timer __read_mostly = 0;
68 static void apic_pm_activate(void);
70 void apic_wait_icr_idle(void)
72 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
73 cpu_relax();
76 unsigned int safe_apic_wait_icr_idle(void)
78 unsigned int send_status;
79 int timeout;
81 timeout = 0;
82 do {
83 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
84 if (!send_status)
85 break;
86 udelay(100);
87 } while (timeout++ < 1000);
89 return send_status;
92 void enable_NMI_through_LVT0 (void * dummy)
94 unsigned int v;
96 /* unmask and set to NMI */
97 v = APIC_DM_NMI;
98 apic_write(APIC_LVT0, v);
101 int get_maxlvt(void)
103 unsigned int v, maxlvt;
105 v = apic_read(APIC_LVR);
106 maxlvt = GET_APIC_MAXLVT(v);
107 return maxlvt;
111 * 'what should we do if we get a hw irq event on an illegal vector'.
112 * each architecture has to answer this themselves.
114 void ack_bad_irq(unsigned int irq)
116 printk("unexpected IRQ trap at vector %02x\n", irq);
118 * Currently unexpected vectors happen only on SMP and APIC.
119 * We _must_ ack these because every local APIC has only N
120 * irq slots per priority level, and a 'hanging, unacked' IRQ
121 * holds up an irq slot - in excessive cases (when multiple
122 * unexpected vectors occur) that might lock up the APIC
123 * completely.
124 * But don't ack when the APIC is disabled. -AK
126 if (!disable_apic)
127 ack_APIC_irq();
130 void clear_local_APIC(void)
132 int maxlvt;
133 unsigned int v;
135 maxlvt = get_maxlvt();
138 * Masking an LVT entry can trigger a local APIC error
139 * if the vector is zero. Mask LVTERR first to prevent this.
141 if (maxlvt >= 3) {
142 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
143 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
146 * Careful: we have to set masks only first to deassert
147 * any level-triggered sources.
149 v = apic_read(APIC_LVTT);
150 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
151 v = apic_read(APIC_LVT0);
152 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
153 v = apic_read(APIC_LVT1);
154 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
155 if (maxlvt >= 4) {
156 v = apic_read(APIC_LVTPC);
157 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
161 * Clean APIC state for other OSs:
163 apic_write(APIC_LVTT, APIC_LVT_MASKED);
164 apic_write(APIC_LVT0, APIC_LVT_MASKED);
165 apic_write(APIC_LVT1, APIC_LVT_MASKED);
166 if (maxlvt >= 3)
167 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
168 if (maxlvt >= 4)
169 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
170 apic_write(APIC_ESR, 0);
171 apic_read(APIC_ESR);
174 void disconnect_bsp_APIC(int virt_wire_setup)
176 /* Go back to Virtual Wire compatibility mode */
177 unsigned long value;
179 /* For the spurious interrupt use vector F, and enable it */
180 value = apic_read(APIC_SPIV);
181 value &= ~APIC_VECTOR_MASK;
182 value |= APIC_SPIV_APIC_ENABLED;
183 value |= 0xf;
184 apic_write(APIC_SPIV, value);
186 if (!virt_wire_setup) {
187 /* For LVT0 make it edge triggered, active high, external and enabled */
188 value = apic_read(APIC_LVT0);
189 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
190 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
191 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED );
192 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
193 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
194 apic_write(APIC_LVT0, value);
195 } else {
196 /* Disable LVT0 */
197 apic_write(APIC_LVT0, APIC_LVT_MASKED);
200 /* For LVT1 make it edge triggered, active high, nmi and enabled */
201 value = apic_read(APIC_LVT1);
202 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
203 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
204 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
205 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
206 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
207 apic_write(APIC_LVT1, value);
210 void disable_local_APIC(void)
212 unsigned int value;
214 clear_local_APIC();
217 * Disable APIC (implies clearing of registers
218 * for 82489DX!).
220 value = apic_read(APIC_SPIV);
221 value &= ~APIC_SPIV_APIC_ENABLED;
222 apic_write(APIC_SPIV, value);
226 * This is to verify that we're looking at a real local APIC.
227 * Check these against your board if the CPUs aren't getting
228 * started for no apparent reason.
230 int __init verify_local_APIC(void)
232 unsigned int reg0, reg1;
235 * The version register is read-only in a real APIC.
237 reg0 = apic_read(APIC_LVR);
238 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
239 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
240 reg1 = apic_read(APIC_LVR);
241 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
244 * The two version reads above should print the same
245 * numbers. If the second one is different, then we
246 * poke at a non-APIC.
248 if (reg1 != reg0)
249 return 0;
252 * Check if the version looks reasonably.
254 reg1 = GET_APIC_VERSION(reg0);
255 if (reg1 == 0x00 || reg1 == 0xff)
256 return 0;
257 reg1 = get_maxlvt();
258 if (reg1 < 0x02 || reg1 == 0xff)
259 return 0;
262 * The ID register is read/write in a real APIC.
264 reg0 = apic_read(APIC_ID);
265 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
266 apic_write(APIC_ID, reg0 ^ APIC_ID_MASK);
267 reg1 = apic_read(APIC_ID);
268 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
269 apic_write(APIC_ID, reg0);
270 if (reg1 != (reg0 ^ APIC_ID_MASK))
271 return 0;
274 * The next two are just to see if we have sane values.
275 * They're only really relevant if we're in Virtual Wire
276 * compatibility mode, but most boxes are anymore.
278 reg0 = apic_read(APIC_LVT0);
279 apic_printk(APIC_DEBUG,"Getting LVT0: %x\n", reg0);
280 reg1 = apic_read(APIC_LVT1);
281 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
283 return 1;
286 void __init sync_Arb_IDs(void)
288 /* Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 */
289 unsigned int ver = GET_APIC_VERSION(apic_read(APIC_LVR));
290 if (ver >= 0x14) /* P4 or higher */
291 return;
294 * Wait for idle.
296 apic_wait_icr_idle();
298 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
299 apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG
300 | APIC_DM_INIT);
304 * An initial setup of the virtual wire mode.
306 void __init init_bsp_APIC(void)
308 unsigned int value;
311 * Don't do the setup now if we have a SMP BIOS as the
312 * through-I/O-APIC virtual wire mode might be active.
314 if (smp_found_config || !cpu_has_apic)
315 return;
317 value = apic_read(APIC_LVR);
320 * Do not trust the local APIC being empty at bootup.
322 clear_local_APIC();
325 * Enable APIC.
327 value = apic_read(APIC_SPIV);
328 value &= ~APIC_VECTOR_MASK;
329 value |= APIC_SPIV_APIC_ENABLED;
330 value |= APIC_SPIV_FOCUS_DISABLED;
331 value |= SPURIOUS_APIC_VECTOR;
332 apic_write(APIC_SPIV, value);
335 * Set up the virtual wire mode.
337 apic_write(APIC_LVT0, APIC_DM_EXTINT);
338 value = APIC_DM_NMI;
339 apic_write(APIC_LVT1, value);
342 void __cpuinit setup_local_APIC (void)
344 unsigned int value, maxlvt;
345 int i, j;
347 value = apic_read(APIC_LVR);
349 BUILD_BUG_ON((SPURIOUS_APIC_VECTOR & 0x0f) != 0x0f);
352 * Double-check whether this APIC is really registered.
353 * This is meaningless in clustered apic mode, so we skip it.
355 if (!apic_id_registered())
356 BUG();
359 * Intel recommends to set DFR, LDR and TPR before enabling
360 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
361 * document number 292116). So here it goes...
363 init_apic_ldr();
366 * Set Task Priority to 'accept all'. We never change this
367 * later on.
369 value = apic_read(APIC_TASKPRI);
370 value &= ~APIC_TPRI_MASK;
371 apic_write(APIC_TASKPRI, value);
374 * After a crash, we no longer service the interrupts and a pending
375 * interrupt from previous kernel might still have ISR bit set.
377 * Most probably by now CPU has serviced that pending interrupt and
378 * it might not have done the ack_APIC_irq() because it thought,
379 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
380 * does not clear the ISR bit and cpu thinks it has already serivced
381 * the interrupt. Hence a vector might get locked. It was noticed
382 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
384 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
385 value = apic_read(APIC_ISR + i*0x10);
386 for (j = 31; j >= 0; j--) {
387 if (value & (1<<j))
388 ack_APIC_irq();
393 * Now that we are all set up, enable the APIC
395 value = apic_read(APIC_SPIV);
396 value &= ~APIC_VECTOR_MASK;
398 * Enable APIC
400 value |= APIC_SPIV_APIC_ENABLED;
402 /* We always use processor focus */
405 * Set spurious IRQ vector
407 value |= SPURIOUS_APIC_VECTOR;
408 apic_write(APIC_SPIV, value);
411 * Set up LVT0, LVT1:
413 * set up through-local-APIC on the BP's LINT0. This is not
414 * strictly necessary in pure symmetric-IO mode, but sometimes
415 * we delegate interrupts to the 8259A.
418 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
420 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
421 if (!smp_processor_id() && !value) {
422 value = APIC_DM_EXTINT;
423 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", smp_processor_id());
424 } else {
425 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
426 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", smp_processor_id());
428 apic_write(APIC_LVT0, value);
431 * only the BP should see the LINT1 NMI signal, obviously.
433 if (!smp_processor_id())
434 value = APIC_DM_NMI;
435 else
436 value = APIC_DM_NMI | APIC_LVT_MASKED;
437 apic_write(APIC_LVT1, value);
440 unsigned oldvalue;
441 maxlvt = get_maxlvt();
442 oldvalue = apic_read(APIC_ESR);
443 value = ERROR_APIC_VECTOR; // enables sending errors
444 apic_write(APIC_LVTERR, value);
446 * spec says clear errors after enabling vector.
448 if (maxlvt > 3)
449 apic_write(APIC_ESR, 0);
450 value = apic_read(APIC_ESR);
451 if (value != oldvalue)
452 apic_printk(APIC_VERBOSE,
453 "ESR value after enabling vector: %08x, after %08x\n",
454 oldvalue, value);
457 nmi_watchdog_default();
458 setup_apic_nmi_watchdog(NULL);
459 apic_pm_activate();
462 #ifdef CONFIG_PM
464 static struct {
465 /* 'active' is true if the local APIC was enabled by us and
466 not the BIOS; this signifies that we are also responsible
467 for disabling it before entering apm/acpi suspend */
468 int active;
469 /* r/w apic fields */
470 unsigned int apic_id;
471 unsigned int apic_taskpri;
472 unsigned int apic_ldr;
473 unsigned int apic_dfr;
474 unsigned int apic_spiv;
475 unsigned int apic_lvtt;
476 unsigned int apic_lvtpc;
477 unsigned int apic_lvt0;
478 unsigned int apic_lvt1;
479 unsigned int apic_lvterr;
480 unsigned int apic_tmict;
481 unsigned int apic_tdcr;
482 unsigned int apic_thmr;
483 } apic_pm_state;
485 static int lapic_suspend(struct sys_device *dev, pm_message_t state)
487 unsigned long flags;
488 int maxlvt;
490 if (!apic_pm_state.active)
491 return 0;
493 maxlvt = get_maxlvt();
495 apic_pm_state.apic_id = apic_read(APIC_ID);
496 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
497 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
498 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
499 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
500 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
501 if (maxlvt >= 4)
502 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
503 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
504 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
505 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
506 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
507 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
508 #ifdef CONFIG_X86_MCE_INTEL
509 if (maxlvt >= 5)
510 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
511 #endif
512 local_irq_save(flags);
513 disable_local_APIC();
514 local_irq_restore(flags);
515 return 0;
518 static int lapic_resume(struct sys_device *dev)
520 unsigned int l, h;
521 unsigned long flags;
522 int maxlvt;
524 if (!apic_pm_state.active)
525 return 0;
527 maxlvt = get_maxlvt();
529 local_irq_save(flags);
530 rdmsr(MSR_IA32_APICBASE, l, h);
531 l &= ~MSR_IA32_APICBASE_BASE;
532 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
533 wrmsr(MSR_IA32_APICBASE, l, h);
534 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
535 apic_write(APIC_ID, apic_pm_state.apic_id);
536 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
537 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
538 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
539 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
540 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
541 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
542 #ifdef CONFIG_X86_MCE_INTEL
543 if (maxlvt >= 5)
544 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
545 #endif
546 if (maxlvt >= 4)
547 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
548 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
549 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
550 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
551 apic_write(APIC_ESR, 0);
552 apic_read(APIC_ESR);
553 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
554 apic_write(APIC_ESR, 0);
555 apic_read(APIC_ESR);
556 local_irq_restore(flags);
557 return 0;
560 static struct sysdev_class lapic_sysclass = {
561 set_kset_name("lapic"),
562 .resume = lapic_resume,
563 .suspend = lapic_suspend,
566 static struct sys_device device_lapic = {
567 .id = 0,
568 .cls = &lapic_sysclass,
571 static void __cpuinit apic_pm_activate(void)
573 apic_pm_state.active = 1;
576 static int __init init_lapic_sysfs(void)
578 int error;
579 if (!cpu_has_apic)
580 return 0;
581 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
582 error = sysdev_class_register(&lapic_sysclass);
583 if (!error)
584 error = sysdev_register(&device_lapic);
585 return error;
587 device_initcall(init_lapic_sysfs);
589 #else /* CONFIG_PM */
591 static void apic_pm_activate(void) { }
593 #endif /* CONFIG_PM */
595 static int __init apic_set_verbosity(char *str)
597 if (str == NULL) {
598 skip_ioapic_setup = 0;
599 ioapic_force = 1;
600 return 0;
602 if (strcmp("debug", str) == 0)
603 apic_verbosity = APIC_DEBUG;
604 else if (strcmp("verbose", str) == 0)
605 apic_verbosity = APIC_VERBOSE;
606 else {
607 printk(KERN_WARNING "APIC Verbosity level %s not recognised"
608 " use apic=verbose or apic=debug\n", str);
609 return -EINVAL;
612 return 0;
614 early_param("apic", apic_set_verbosity);
617 * Detect and enable local APICs on non-SMP boards.
618 * Original code written by Keir Fraser.
619 * On AMD64 we trust the BIOS - if it says no APIC it is likely
620 * not correctly set up (usually the APIC timer won't work etc.)
623 static int __init detect_init_APIC (void)
625 if (!cpu_has_apic) {
626 printk(KERN_INFO "No local APIC present\n");
627 return -1;
630 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
631 boot_cpu_id = 0;
632 return 0;
635 #ifdef CONFIG_X86_IO_APIC
636 static struct resource * __init ioapic_setup_resources(void)
638 #define IOAPIC_RESOURCE_NAME_SIZE 11
639 unsigned long n;
640 struct resource *res;
641 char *mem;
642 int i;
644 if (nr_ioapics <= 0)
645 return NULL;
647 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
648 n *= nr_ioapics;
650 mem = alloc_bootmem(n);
651 res = (void *)mem;
653 if (mem != NULL) {
654 memset(mem, 0, n);
655 mem += sizeof(struct resource) * nr_ioapics;
657 for (i = 0; i < nr_ioapics; i++) {
658 res[i].name = mem;
659 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
660 sprintf(mem, "IOAPIC %u", i);
661 mem += IOAPIC_RESOURCE_NAME_SIZE;
665 ioapic_resources = res;
667 return res;
670 static int __init ioapic_insert_resources(void)
672 int i;
673 struct resource *r = ioapic_resources;
675 if (!r) {
676 printk("IO APIC resources could be not be allocated.\n");
677 return -1;
680 for (i = 0; i < nr_ioapics; i++) {
681 insert_resource(&iomem_resource, r);
682 r++;
685 return 0;
688 /* Insert the IO APIC resources after PCI initialization has occured to handle
689 * IO APICS that are mapped in on a BAR in PCI space. */
690 late_initcall(ioapic_insert_resources);
691 #endif
693 void __init init_apic_mappings(void)
695 unsigned long apic_phys;
698 * If no local APIC can be found then set up a fake all
699 * zeroes page to simulate the local APIC and another
700 * one for the IO-APIC.
702 if (!smp_found_config && detect_init_APIC()) {
703 apic_phys = (unsigned long) alloc_bootmem_pages(PAGE_SIZE);
704 apic_phys = __pa(apic_phys);
705 } else
706 apic_phys = mp_lapic_addr;
708 set_fixmap_nocache(FIX_APIC_BASE, apic_phys);
709 apic_mapped = 1;
710 apic_printk(APIC_VERBOSE,"mapped APIC to %16lx (%16lx)\n", APIC_BASE, apic_phys);
712 /* Put local APIC into the resource map. */
713 lapic_resource.start = apic_phys;
714 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
715 insert_resource(&iomem_resource, &lapic_resource);
718 * Fetch the APIC ID of the BSP in case we have a
719 * default configuration (or the MP table is broken).
721 boot_cpu_id = GET_APIC_ID(apic_read(APIC_ID));
724 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
725 int i;
726 struct resource *ioapic_res;
728 ioapic_res = ioapic_setup_resources();
729 for (i = 0; i < nr_ioapics; i++) {
730 if (smp_found_config) {
731 ioapic_phys = mp_ioapics[i].mpc_apicaddr;
732 } else {
733 ioapic_phys = (unsigned long) alloc_bootmem_pages(PAGE_SIZE);
734 ioapic_phys = __pa(ioapic_phys);
736 set_fixmap_nocache(idx, ioapic_phys);
737 apic_printk(APIC_VERBOSE,"mapped IOAPIC to %016lx (%016lx)\n",
738 __fix_to_virt(idx), ioapic_phys);
739 idx++;
741 if (ioapic_res != NULL) {
742 ioapic_res->start = ioapic_phys;
743 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
744 ioapic_res++;
751 * This function sets up the local APIC timer, with a timeout of
752 * 'clocks' APIC bus clock. During calibration we actually call
753 * this function twice on the boot CPU, once with a bogus timeout
754 * value, second time for real. The other (noncalibrating) CPUs
755 * call this function only once, with the real, calibrated value.
757 * We do reads before writes even if unnecessary, to get around the
758 * P5 APIC double write bug.
761 #define APIC_DIVISOR 16
763 static void __setup_APIC_LVTT(unsigned int clocks)
765 unsigned int lvtt_value, tmp_value;
766 int cpu = smp_processor_id();
768 lvtt_value = APIC_LVT_TIMER_PERIODIC | LOCAL_TIMER_VECTOR;
770 if (cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask))
771 lvtt_value |= APIC_LVT_MASKED;
773 apic_write(APIC_LVTT, lvtt_value);
776 * Divide PICLK by 16
778 tmp_value = apic_read(APIC_TDCR);
779 apic_write(APIC_TDCR, (tmp_value
780 & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
781 | APIC_TDR_DIV_16);
783 apic_write(APIC_TMICT, clocks/APIC_DIVISOR);
786 static void setup_APIC_timer(unsigned int clocks)
788 unsigned long flags;
790 local_irq_save(flags);
792 /* wait for irq slice */
793 if (hpet_address && hpet_use_timer) {
794 u32 trigger = hpet_readl(HPET_T0_CMP);
795 while (hpet_readl(HPET_T0_CMP) == trigger)
796 /* do nothing */ ;
797 } else {
798 int c1, c2;
799 outb_p(0x00, 0x43);
800 c2 = inb_p(0x40);
801 c2 |= inb_p(0x40) << 8;
802 do {
803 c1 = c2;
804 outb_p(0x00, 0x43);
805 c2 = inb_p(0x40);
806 c2 |= inb_p(0x40) << 8;
807 } while (c2 - c1 < 300);
809 __setup_APIC_LVTT(clocks);
810 /* Turn off PIT interrupt if we use APIC timer as main timer.
811 Only works with the PM timer right now
812 TBD fix it for HPET too. */
813 if ((pmtmr_ioport != 0) &&
814 smp_processor_id() == boot_cpu_id &&
815 apic_runs_main_timer == 1 &&
816 !cpu_isset(boot_cpu_id, timer_interrupt_broadcast_ipi_mask)) {
817 stop_timer_interrupt();
818 apic_runs_main_timer++;
820 local_irq_restore(flags);
824 * In this function we calibrate APIC bus clocks to the external
825 * timer. Unfortunately we cannot use jiffies and the timer irq
826 * to calibrate, since some later bootup code depends on getting
827 * the first irq? Ugh.
829 * We want to do the calibration only once since we
830 * want to have local timer irqs syncron. CPUs connected
831 * by the same APIC bus have the very same bus frequency.
832 * And we want to have irqs off anyways, no accidental
833 * APIC irq that way.
836 #define TICK_COUNT 100000000
838 static int __init calibrate_APIC_clock(void)
840 unsigned apic, apic_start;
841 unsigned long tsc, tsc_start;
842 int result;
844 * Put whatever arbitrary (but long enough) timeout
845 * value into the APIC clock, we just want to get the
846 * counter running for calibration.
848 __setup_APIC_LVTT(4000000000);
850 apic_start = apic_read(APIC_TMCCT);
851 #ifdef CONFIG_X86_PM_TIMER
852 if (apic_calibrate_pmtmr && pmtmr_ioport) {
853 pmtimer_wait(5000); /* 5ms wait */
854 apic = apic_read(APIC_TMCCT);
855 result = (apic_start - apic) * 1000L / 5;
856 } else
857 #endif
859 rdtscll(tsc_start);
861 do {
862 apic = apic_read(APIC_TMCCT);
863 rdtscll(tsc);
864 } while ((tsc - tsc_start) < TICK_COUNT &&
865 (apic_start - apic) < TICK_COUNT);
867 result = (apic_start - apic) * 1000L * tsc_khz /
868 (tsc - tsc_start);
870 printk("result %d\n", result);
873 printk(KERN_INFO "Detected %d.%03d MHz APIC timer.\n",
874 result / 1000 / 1000, result / 1000 % 1000);
876 return result * APIC_DIVISOR / HZ;
879 static unsigned int calibration_result;
881 void __init setup_boot_APIC_clock (void)
883 if (disable_apic_timer) {
884 printk(KERN_INFO "Disabling APIC timer\n");
885 return;
888 printk(KERN_INFO "Using local APIC timer interrupts.\n");
889 using_apic_timer = 1;
891 local_irq_disable();
893 calibration_result = calibrate_APIC_clock();
895 * Now set up the timer for real.
897 setup_APIC_timer(calibration_result);
899 local_irq_enable();
902 void __cpuinit setup_secondary_APIC_clock(void)
904 local_irq_disable(); /* FIXME: Do we need this? --RR */
905 setup_APIC_timer(calibration_result);
906 local_irq_enable();
909 void disable_APIC_timer(void)
911 if (using_apic_timer) {
912 unsigned long v;
914 v = apic_read(APIC_LVTT);
916 * When an illegal vector value (0-15) is written to an LVT
917 * entry and delivery mode is Fixed, the APIC may signal an
918 * illegal vector error, with out regard to whether the mask
919 * bit is set or whether an interrupt is actually seen on input.
921 * Boot sequence might call this function when the LVTT has
922 * '0' vector value. So make sure vector field is set to
923 * valid value.
925 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
926 apic_write(APIC_LVTT, v);
930 void enable_APIC_timer(void)
932 int cpu = smp_processor_id();
934 if (using_apic_timer &&
935 !cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask)) {
936 unsigned long v;
938 v = apic_read(APIC_LVTT);
939 apic_write(APIC_LVTT, v & ~APIC_LVT_MASKED);
943 void switch_APIC_timer_to_ipi(void *cpumask)
945 cpumask_t mask = *(cpumask_t *)cpumask;
946 int cpu = smp_processor_id();
948 if (cpu_isset(cpu, mask) &&
949 !cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask)) {
950 disable_APIC_timer();
951 cpu_set(cpu, timer_interrupt_broadcast_ipi_mask);
954 EXPORT_SYMBOL(switch_APIC_timer_to_ipi);
956 void smp_send_timer_broadcast_ipi(void)
958 int cpu = smp_processor_id();
959 cpumask_t mask;
961 cpus_and(mask, cpu_online_map, timer_interrupt_broadcast_ipi_mask);
963 if (cpu_isset(cpu, mask)) {
964 cpu_clear(cpu, mask);
965 add_pda(apic_timer_irqs, 1);
966 smp_local_timer_interrupt();
969 if (!cpus_empty(mask)) {
970 send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
974 void switch_ipi_to_APIC_timer(void *cpumask)
976 cpumask_t mask = *(cpumask_t *)cpumask;
977 int cpu = smp_processor_id();
979 if (cpu_isset(cpu, mask) &&
980 cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask)) {
981 cpu_clear(cpu, timer_interrupt_broadcast_ipi_mask);
982 enable_APIC_timer();
985 EXPORT_SYMBOL(switch_ipi_to_APIC_timer);
987 int setup_profiling_timer(unsigned int multiplier)
989 return -EINVAL;
992 void setup_APIC_extended_lvt(unsigned char lvt_off, unsigned char vector,
993 unsigned char msg_type, unsigned char mask)
995 unsigned long reg = (lvt_off << 4) + K8_APIC_EXT_LVT_BASE;
996 unsigned int v = (mask << 16) | (msg_type << 8) | vector;
997 apic_write(reg, v);
1000 #undef APIC_DIVISOR
1003 * Local timer interrupt handler. It does both profiling and
1004 * process statistics/rescheduling.
1006 * We do profiling in every local tick, statistics/rescheduling
1007 * happen only every 'profiling multiplier' ticks. The default
1008 * multiplier is 1 and it can be changed by writing the new multiplier
1009 * value into /proc/profile.
1012 void smp_local_timer_interrupt(void)
1014 profile_tick(CPU_PROFILING);
1015 #ifdef CONFIG_SMP
1016 update_process_times(user_mode(get_irq_regs()));
1017 #endif
1018 if (apic_runs_main_timer > 1 && smp_processor_id() == boot_cpu_id)
1019 main_timer_handler();
1021 * We take the 'long' return path, and there every subsystem
1022 * grabs the appropriate locks (kernel lock/ irq lock).
1024 * We might want to decouple profiling from the 'long path',
1025 * and do the profiling totally in assembly.
1027 * Currently this isn't too much of an issue (performance wise),
1028 * we can take more than 100K local irqs per second on a 100 MHz P5.
1033 * Local APIC timer interrupt. This is the most natural way for doing
1034 * local interrupts, but local timer interrupts can be emulated by
1035 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
1037 * [ if a single-CPU system runs an SMP kernel then we call the local
1038 * interrupt as well. Thus we cannot inline the local irq ... ]
1040 void smp_apic_timer_interrupt(struct pt_regs *regs)
1042 struct pt_regs *old_regs = set_irq_regs(regs);
1045 * the NMI deadlock-detector uses this.
1047 add_pda(apic_timer_irqs, 1);
1050 * NOTE! We'd better ACK the irq immediately,
1051 * because timer handling can be slow.
1053 ack_APIC_irq();
1055 * update_process_times() expects us to have done irq_enter().
1056 * Besides, if we don't timer interrupts ignore the global
1057 * interrupt lock, which is the WrongThing (tm) to do.
1059 exit_idle();
1060 irq_enter();
1061 smp_local_timer_interrupt();
1062 irq_exit();
1063 set_irq_regs(old_regs);
1067 * apic_is_clustered_box() -- Check if we can expect good TSC
1069 * Thus far, the major user of this is IBM's Summit2 series:
1071 * Clustered boxes may have unsynced TSC problems if they are
1072 * multi-chassis. Use available data to take a good guess.
1073 * If in doubt, go HPET.
1075 __cpuinit int apic_is_clustered_box(void)
1077 int i, clusters, zeros;
1078 unsigned id;
1079 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
1081 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
1083 for (i = 0; i < NR_CPUS; i++) {
1084 id = bios_cpu_apicid[i];
1085 if (id != BAD_APICID)
1086 __set_bit(APIC_CLUSTERID(id), clustermap);
1089 /* Problem: Partially populated chassis may not have CPUs in some of
1090 * the APIC clusters they have been allocated. Only present CPUs have
1091 * bios_cpu_apicid entries, thus causing zeroes in the bitmap. Since
1092 * clusters are allocated sequentially, count zeros only if they are
1093 * bounded by ones.
1095 clusters = 0;
1096 zeros = 0;
1097 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
1098 if (test_bit(i, clustermap)) {
1099 clusters += 1 + zeros;
1100 zeros = 0;
1101 } else
1102 ++zeros;
1106 * If clusters > 2, then should be multi-chassis.
1107 * May have to revisit this when multi-core + hyperthreaded CPUs come
1108 * out, but AFAIK this will work even for them.
1110 return (clusters > 2);
1114 * This interrupt should _never_ happen with our APIC/SMP architecture
1116 asmlinkage void smp_spurious_interrupt(void)
1118 unsigned int v;
1119 exit_idle();
1120 irq_enter();
1122 * Check if this really is a spurious interrupt and ACK it
1123 * if it is a vectored one. Just in case...
1124 * Spurious interrupts should not be ACKed.
1126 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
1127 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
1128 ack_APIC_irq();
1130 irq_exit();
1134 * This interrupt should never happen with our APIC/SMP architecture
1137 asmlinkage void smp_error_interrupt(void)
1139 unsigned int v, v1;
1141 exit_idle();
1142 irq_enter();
1143 /* First tickle the hardware, only then report what went on. -- REW */
1144 v = apic_read(APIC_ESR);
1145 apic_write(APIC_ESR, 0);
1146 v1 = apic_read(APIC_ESR);
1147 ack_APIC_irq();
1148 atomic_inc(&irq_err_count);
1150 /* Here is what the APIC error bits mean:
1151 0: Send CS error
1152 1: Receive CS error
1153 2: Send accept error
1154 3: Receive accept error
1155 4: Reserved
1156 5: Send illegal vector
1157 6: Received illegal vector
1158 7: Illegal register address
1160 printk (KERN_DEBUG "APIC error on CPU%d: %02x(%02x)\n",
1161 smp_processor_id(), v , v1);
1162 irq_exit();
1165 int disable_apic;
1168 * This initializes the IO-APIC and APIC hardware if this is
1169 * a UP kernel.
1171 int __init APIC_init_uniprocessor (void)
1173 if (disable_apic) {
1174 printk(KERN_INFO "Apic disabled\n");
1175 return -1;
1177 if (!cpu_has_apic) {
1178 disable_apic = 1;
1179 printk(KERN_INFO "Apic disabled by BIOS\n");
1180 return -1;
1183 verify_local_APIC();
1185 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_id);
1186 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_id));
1188 setup_local_APIC();
1190 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
1191 setup_IO_APIC();
1192 else
1193 nr_ioapics = 0;
1194 setup_boot_APIC_clock();
1195 check_nmi_watchdog();
1196 return 0;
1199 static __init int setup_disableapic(char *str)
1201 disable_apic = 1;
1202 clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
1203 return 0;
1205 early_param("disableapic", setup_disableapic);
1207 /* same as disableapic, for compatibility */
1208 static __init int setup_nolapic(char *str)
1210 return setup_disableapic(str);
1212 early_param("nolapic", setup_nolapic);
1214 static int __init parse_lapic_timer_c2_ok(char *arg)
1216 local_apic_timer_c2_ok = 1;
1217 return 0;
1219 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
1221 static __init int setup_noapictimer(char *str)
1223 if (str[0] != ' ' && str[0] != 0)
1224 return 0;
1225 disable_apic_timer = 1;
1226 return 1;
1229 static __init int setup_apicmaintimer(char *str)
1231 apic_runs_main_timer = 1;
1232 nohpet = 1;
1233 return 1;
1235 __setup("apicmaintimer", setup_apicmaintimer);
1237 static __init int setup_noapicmaintimer(char *str)
1239 apic_runs_main_timer = -1;
1240 return 1;
1242 __setup("noapicmaintimer", setup_noapicmaintimer);
1244 static __init int setup_apicpmtimer(char *s)
1246 apic_calibrate_pmtmr = 1;
1247 notsc_setup(NULL);
1248 return setup_apicmaintimer(NULL);
1250 __setup("apicpmtimer", setup_apicpmtimer);
1252 __setup("noapictimer", setup_noapictimer);