2 * Copyright (c) 1996, by Steve Passe
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. The name of the developer may NOT be used to endorse or promote products
11 * derived from this software without specific prior written permission.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * $FreeBSD: src/sys/i386/i386/mpapic.c,v 1.37.2.7 2003/01/25 02:31:47 peter Exp $
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/kernel.h>
32 #include <sys/machintr.h>
33 #include <machine/globaldata.h>
34 #include <machine/smp.h>
35 #include <machine/md_var.h>
36 #include <machine/pmap.h>
37 #include <machine/specialreg.h>
38 #include <machine_base/apic/lapic.h>
39 #include <machine_base/apic/ioapic.h>
40 #include <machine_base/apic/ioapic_abi.h>
41 #include <machine_base/apic/apicvar.h>
42 #include <machine_base/icu/icu_var.h>
43 #include <machine/segments.h>
44 #include <sys/thread2.h>
46 #include <machine/cputypes.h>
47 #include <machine/intr_machdep.h>
51 volatile lapic_t *lapic;
53 static void lapic_timer_calibrate(void);
54 static void lapic_timer_set_divisor(int);
55 static void lapic_timer_fixup_handler(void *);
56 static void lapic_timer_restart_handler(void *);
58 void lapic_timer_process(void);
59 void lapic_timer_process_frame(struct intrframe *);
60 void lapic_timer_always(struct intrframe *);
62 static int lapic_timer_enable = 1;
63 TUNABLE_INT("hw.lapic_timer_enable", &lapic_timer_enable);
65 static void lapic_timer_intr_reload(struct cputimer_intr *, sysclock_t);
66 static void lapic_timer_intr_enable(struct cputimer_intr *);
67 static void lapic_timer_intr_restart(struct cputimer_intr *);
68 static void lapic_timer_intr_pmfixup(struct cputimer_intr *);
70 static struct cputimer_intr lapic_cputimer_intr = {
72 .reload = lapic_timer_intr_reload,
73 .enable = lapic_timer_intr_enable,
74 .config = cputimer_intr_default_config,
75 .restart = lapic_timer_intr_restart,
76 .pmfixup = lapic_timer_intr_pmfixup,
77 .initclock = cputimer_intr_default_initclock,
78 .next = SLIST_ENTRY_INITIALIZER,
80 .type = CPUTIMER_INTR_LAPIC,
81 .prio = CPUTIMER_INTR_PRIO_LAPIC,
82 .caps = CPUTIMER_INTR_CAP_NONE
85 static int lapic_timer_divisor_idx = -1;
86 static const uint32_t lapic_timer_divisors[] = {
87 APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
88 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128, APIC_TDCR_1
90 #define APIC_TIMER_NDIVISORS (int)(NELEM(lapic_timer_divisors))
93 * APIC ID <-> CPU ID mapping structures.
95 int cpu_id_to_apic_id[NAPICID];
96 int apic_id_to_cpu_id[NAPICID];
100 * Enable LAPIC, configure interrupts.
103 lapic_init(boolean_t bsp)
111 * Since IDT is shared between BSP and APs, these vectors
112 * only need to be installed once; we do it on BSP.
115 if (cpu_vendor_id == CPU_VENDOR_AMD &&
116 CPUID_TO_FAMILY(cpu_id) >= 0xf) {
120 * Set the LINTEN bit in the HyperTransport
121 * Transaction Control Register.
123 * This will cause EXTINT and NMI interrupts
124 * routed over the hypertransport bus to be
125 * fed into the LAPIC LINT0/LINT1. If the bit
126 * isn't set, the interrupts will go to the
127 * general cpu INTR/NMI pins. On a dual-core
128 * cpu the interrupt winds up going to BOTH cpus.
129 * The first cpu that does the interrupt ack
130 * cycle will get the correct interrupt. The
131 * second cpu that does it will get a spurious
132 * interrupt vector (typically IRQ 7).
135 (1 << 31) | /* enable */
136 (0 << 16) | /* bus */
137 (0x18 << 11) | /* dev (cpu + 0x18) */
138 (0 << 8) | /* func */
142 if ((tcr & 0x00010000) == 0) {
143 kprintf("LAPIC: AMD LINTEN on\n");
144 outl(0xcfc, tcr|0x00010000);
149 /* Install a 'Spurious INTerrupt' vector */
150 setidt_global(XSPURIOUSINT_OFFSET, Xspuriousint,
151 SDT_SYSIGT, SEL_KPL, 0);
153 /* Install a timer vector */
154 setidt_global(XTIMER_OFFSET, Xtimer,
155 SDT_SYSIGT, SEL_KPL, 0);
157 /* Install an inter-CPU IPI for TLB invalidation */
158 setidt_global(XINVLTLB_OFFSET, Xinvltlb,
159 SDT_SYSIGT, SEL_KPL, 0);
161 /* Install an inter-CPU IPI for IPIQ messaging */
162 setidt_global(XIPIQ_OFFSET, Xipiq,
163 SDT_SYSIGT, SEL_KPL, 0);
165 /* Install an inter-CPU IPI for CPU stop/restart */
166 setidt_global(XCPUSTOP_OFFSET, Xcpustop,
167 SDT_SYSIGT, SEL_KPL, 0);
171 * Setup LINT0 as ExtINT on the BSP. This is theoretically an
172 * aggregate interrupt input from the 8259. The INTA cycle
173 * will be routed to the external controller (the 8259) which
174 * is expected to supply the vector.
176 * Must be setup edge triggered, active high.
178 * Disable LINT0 on BSP, if I/O APIC is enabled.
180 * Disable LINT0 on the APs. It doesn't matter what delivery
181 * mode we use because we leave it masked.
183 temp = lapic->lvt_lint0;
184 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
185 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
187 temp |= APIC_LVT_DM_EXTINT;
189 temp |= APIC_LVT_MASKED;
191 temp |= APIC_LVT_DM_FIXED | APIC_LVT_MASKED;
193 lapic->lvt_lint0 = temp;
196 * Setup LINT1 as NMI.
198 * Must be setup edge trigger, active high.
200 * Enable LINT1 on BSP, if I/O APIC is enabled.
202 * Disable LINT1 on the APs.
204 temp = lapic->lvt_lint1;
205 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
206 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
207 temp |= APIC_LVT_MASKED | APIC_LVT_DM_NMI;
208 if (bsp && ioapic_enable)
209 temp &= ~APIC_LVT_MASKED;
210 lapic->lvt_lint1 = temp;
213 * Mask the LAPIC error interrupt, LAPIC performance counter
216 lapic->lvt_error = lapic->lvt_error | APIC_LVT_MASKED;
217 lapic->lvt_pcint = lapic->lvt_pcint | APIC_LVT_MASKED;
220 * Set LAPIC timer vector and mask the LAPIC timer interrupt.
222 timer = lapic->lvt_timer;
223 timer &= ~APIC_LVTT_VECTOR;
224 timer |= XTIMER_OFFSET;
225 timer |= APIC_LVTT_MASKED;
226 lapic->lvt_timer = timer;
229 * Set the Task Priority Register as needed. At the moment allow
230 * interrupts on all cpus (the APs will remain CLId until they are
234 temp &= ~APIC_TPR_PRIO; /* clear priority field */
241 temp |= APIC_SVR_ENABLE; /* enable the LAPIC */
242 temp &= ~APIC_SVR_FOCUS_DISABLE; /* enable lopri focus processor */
245 * Set the spurious interrupt vector. The low 4 bits of the vector
248 if ((XSPURIOUSINT_OFFSET & 0x0F) != 0x0F)
249 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET);
250 temp &= ~APIC_SVR_VECTOR;
251 temp |= XSPURIOUSINT_OFFSET;
256 * Pump out a few EOIs to clean out interrupts that got through
257 * before we were able to set the TPR.
264 lapic_timer_calibrate();
265 if (lapic_timer_enable) {
266 cputimer_intr_register(&lapic_cputimer_intr);
267 cputimer_intr_select(&lapic_cputimer_intr, 0);
270 lapic_timer_set_divisor(lapic_timer_divisor_idx);
274 apic_dump("apic_initialize()");
278 lapic_timer_set_divisor(int divisor_idx)
280 KKASSERT(divisor_idx >= 0 && divisor_idx < APIC_TIMER_NDIVISORS);
281 lapic->dcr_timer = lapic_timer_divisors[divisor_idx];
285 lapic_timer_oneshot(u_int count)
289 value = lapic->lvt_timer;
290 value &= ~APIC_LVTT_PERIODIC;
291 lapic->lvt_timer = value;
292 lapic->icr_timer = count;
296 lapic_timer_oneshot_quick(u_int count)
298 lapic->icr_timer = count;
302 lapic_timer_calibrate(void)
306 /* Try to calibrate the local APIC timer. */
307 for (lapic_timer_divisor_idx = 0;
308 lapic_timer_divisor_idx < APIC_TIMER_NDIVISORS;
309 lapic_timer_divisor_idx++) {
310 lapic_timer_set_divisor(lapic_timer_divisor_idx);
311 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
313 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
314 if (value != APIC_TIMER_MAX_COUNT)
317 if (lapic_timer_divisor_idx >= APIC_TIMER_NDIVISORS)
318 panic("lapic: no proper timer divisor?!");
319 lapic_cputimer_intr.freq = value / 2;
321 kprintf("lapic: divisor index %d, frequency %u Hz\n",
322 lapic_timer_divisor_idx, lapic_cputimer_intr.freq);
326 lapic_timer_process_oncpu(struct globaldata *gd, struct intrframe *frame)
330 gd->gd_timer_running = 0;
332 count = sys_cputimer->count();
333 if (TAILQ_FIRST(&gd->gd_systimerq) != NULL)
334 systimer_intr(&count, 0, frame);
338 lapic_timer_process(void)
340 lapic_timer_process_oncpu(mycpu, NULL);
344 lapic_timer_process_frame(struct intrframe *frame)
346 lapic_timer_process_oncpu(mycpu, frame);
350 * This manual debugging code is called unconditionally from Xtimer
351 * (the lapic timer interrupt) whether the current thread is in a
352 * critical section or not) and can be useful in tracking down lockups.
354 * NOTE: MANUAL DEBUG CODE
357 static int saveticks[SMP_MAXCPU];
358 static int savecounts[SMP_MAXCPU];
362 lapic_timer_always(struct intrframe *frame)
365 globaldata_t gd = mycpu;
366 int cpu = gd->gd_cpuid;
372 gptr = (short *)0xFFFFFFFF800b8000 + 80 * cpu;
373 *gptr = ((*gptr + 1) & 0x00FF) | 0x0700;
376 ksnprintf(buf, sizeof(buf), " %p %16s %d %16s ",
377 (void *)frame->if_rip, gd->gd_curthread->td_comm, ticks,
379 for (i = 0; buf[i]; ++i) {
380 gptr[i] = 0x0700 | (unsigned char)buf[i];
384 if (saveticks[gd->gd_cpuid] != ticks) {
385 saveticks[gd->gd_cpuid] = ticks;
386 savecounts[gd->gd_cpuid] = 0;
388 ++savecounts[gd->gd_cpuid];
389 if (savecounts[gd->gd_cpuid] > 2000 && panicstr == NULL) {
390 panic("cpud %d panicing on ticks failure",
393 for (i = 0; i < ncpus; ++i) {
395 if (saveticks[i] && panicstr == NULL) {
396 delta = saveticks[i] - ticks;
397 if (delta < -10 || delta > 10) {
398 panic("cpu %d panicing on cpu %d watchdog",
408 lapic_timer_intr_reload(struct cputimer_intr *cti, sysclock_t reload)
410 struct globaldata *gd = mycpu;
412 reload = (int64_t)reload * cti->freq / sys_cputimer->freq;
416 if (gd->gd_timer_running) {
417 if (reload < lapic->ccr_timer)
418 lapic_timer_oneshot_quick(reload);
420 gd->gd_timer_running = 1;
421 lapic_timer_oneshot_quick(reload);
426 lapic_timer_intr_enable(struct cputimer_intr *cti __unused)
430 timer = lapic->lvt_timer;
431 timer &= ~(APIC_LVTT_MASKED | APIC_LVTT_PERIODIC);
432 lapic->lvt_timer = timer;
434 lapic_timer_fixup_handler(NULL);
438 lapic_timer_fixup_handler(void *arg)
445 if (cpu_vendor_id == CPU_VENDOR_AMD) {
447 * Detect the presence of C1E capability mostly on latest
448 * dual-cores (or future) k8 family. This feature renders
449 * the local APIC timer dead, so we disable it by reading
450 * the Interrupt Pending Message register and clearing both
451 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
454 * "BIOS and Kernel Developer's Guide for AMD NPT
455 * Family 0Fh Processors"
456 * #32559 revision 3.00
458 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
459 (cpu_id & 0x0fff0000) >= 0x00040000) {
462 msr = rdmsr(0xc0010055);
463 if (msr & 0x18000000) {
464 struct globaldata *gd = mycpu;
466 kprintf("cpu%d: AMD C1E detected\n",
468 wrmsr(0xc0010055, msr & ~0x18000000ULL);
471 * We are kinda stalled;
474 gd->gd_timer_running = 1;
475 lapic_timer_oneshot_quick(2);
485 lapic_timer_restart_handler(void *dummy __unused)
489 lapic_timer_fixup_handler(&started);
491 struct globaldata *gd = mycpu;
493 gd->gd_timer_running = 1;
494 lapic_timer_oneshot_quick(2);
499 * This function is called only by ACPI-CA code currently:
500 * - AMD C1E fixup. AMD C1E only seems to happen after ACPI
501 * module controls PM. So once ACPI-CA is attached, we try
502 * to apply the fixup to prevent LAPIC timer from hanging.
505 lapic_timer_intr_pmfixup(struct cputimer_intr *cti __unused)
507 lwkt_send_ipiq_mask(smp_active_mask,
508 lapic_timer_fixup_handler, NULL);
512 lapic_timer_intr_restart(struct cputimer_intr *cti __unused)
514 lwkt_send_ipiq_mask(smp_active_mask, lapic_timer_restart_handler, NULL);
519 * dump contents of local APIC registers
524 kprintf("SMP: CPU%d %s:\n", mycpu->gd_cpuid, str);
525 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
526 lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
530 * Inter Processor Interrupt functions.
534 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
536 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
537 * vector is any valid SYSTEM INT vector
538 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
542 * We now implement a per-cpu interlock (gd->gd_npoll) to prevent more than
543 * one IPI from being sent to any given cpu at a time. Thus we no longer
544 * have to process incoming IPIs while waiting for the status to clear.
545 * No deadlock should be possible.
547 * We now physically disable interrupts for the lapic ICR operation. If
548 * we do not do this then it looks like an EOI sent to the lapic (which
549 * occurs even with a critical section) can interfere with the command
550 * register ready status and cause an IPI to be lost.
552 * e.g. an interrupt can occur, issue the EOI, IRET, and cause the command
553 * register to busy just before we write to icr_lo, resulting in a lost
554 * issuance. This only appears to occur on Intel cpus and is not
555 * documented. It could simply be that cpus are so fast these days that
556 * it was always an issue, but is only now rearing its ugly head. This
560 apic_ipi(int dest_type, int vector, int delivery_mode)
562 unsigned long rflags;
565 rflags = read_rflags();
567 while ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
570 icr_lo = (lapic->icr_lo & APIC_ICRLO_RESV_MASK) | dest_type |
571 delivery_mode | vector;
572 lapic->icr_lo = icr_lo;
573 write_rflags(rflags);
579 single_apic_ipi(int cpu, int vector, int delivery_mode)
581 unsigned long rflags;
585 rflags = read_rflags();
587 while ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
590 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
591 icr_hi |= (CPUID_TO_APICID(cpu) << 24);
592 lapic->icr_hi = icr_hi;
595 icr_lo = (lapic->icr_lo & APIC_ICRLO_RESV_MASK) |
596 APIC_DEST_DESTFLD | delivery_mode | vector;
599 lapic->icr_lo = icr_lo;
600 write_rflags(rflags);
606 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
608 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
609 * to the target, and the scheduler does not 'poll' for IPI messages.
612 single_apic_ipi_passive(int cpu, int vector, int delivery_mode)
618 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
622 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
623 icr_hi |= (CPUID_TO_APICID(cpu) << 24);
624 lapic->icr_hi = icr_hi;
627 icr_lo = (lapic->icr_lo & APIC_RESV2_MASK)
628 | APIC_DEST_DESTFLD | delivery_mode | vector;
631 lapic->icr_lo = icr_lo;
639 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
641 * target is a bitmask of destination cpus. Vector is any
642 * valid system INT vector. Delivery mode may be either
643 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
646 selected_apic_ipi(cpumask_t target, int vector, int delivery_mode)
650 int n = BSFCPUMASK(target);
651 target &= ~CPUMASK(n);
652 single_apic_ipi(n, vector, delivery_mode);
658 * Timer code, in development...
659 * - suggested by rgrimes@gndrsh.aac.dev.com
662 get_apic_timer_frequency(void)
664 return(lapic_cputimer_intr.freq);
668 * Load a 'downcount time' in uSeconds.
671 set_apic_timer(int us)
676 * When we reach here, lapic timer's frequency
677 * must have been calculated as well as the
678 * divisor (lapic->dcr_timer is setup during the
679 * divisor calculation).
681 KKASSERT(lapic_cputimer_intr.freq != 0 &&
682 lapic_timer_divisor_idx >= 0);
684 count = ((us * (int64_t)lapic_cputimer_intr.freq) + 999999) / 1000000;
685 lapic_timer_oneshot(count);
690 * Read remaining time in timer.
693 read_apic_timer(void)
696 /** XXX FIXME: we need to return the actual remaining time,
697 * for now we just return the remaining count.
700 return lapic->ccr_timer;
706 * Spin-style delay, set delay time in uS, spin till it drains.
711 set_apic_timer(count);
712 while (read_apic_timer())
717 lapic_unused_apic_id(int start)
721 for (i = start; i < APICID_MAX; ++i) {
722 if (APICID_TO_CPUID(i) == -1)
729 lapic_map(vm_paddr_t lapic_addr)
731 lapic = pmap_mapdev_uncacheable(lapic_addr, sizeof(struct LAPIC));
734 static TAILQ_HEAD(, lapic_enumerator) lapic_enumerators =
735 TAILQ_HEAD_INITIALIZER(lapic_enumerators);
740 struct lapic_enumerator *e;
741 int error, i, ap_max;
743 KKASSERT(lapic_enable);
745 for (i = 0; i < NAPICID; ++i)
746 APICID_TO_CPUID(i) = -1;
748 TAILQ_FOREACH(e, &lapic_enumerators, lapic_link) {
749 error = e->lapic_probe(e);
754 kprintf("LAPIC: Can't find LAPIC\n");
758 error = e->lapic_enumerate(e);
760 kprintf("LAPIC: enumeration failed\n");
765 TUNABLE_INT_FETCH("hw.ap_max", &ap_max);
766 if (ap_max > MAXCPU - 1)
770 kprintf("LAPIC: Warning use only %d out of %d "
780 lapic_enumerator_register(struct lapic_enumerator *ne)
782 struct lapic_enumerator *e;
784 TAILQ_FOREACH(e, &lapic_enumerators, lapic_link) {
785 if (e->lapic_prio < ne->lapic_prio) {
786 TAILQ_INSERT_BEFORE(e, ne, lapic_link);
790 TAILQ_INSERT_TAIL(&lapic_enumerators, ne, lapic_link);
794 lapic_set_cpuid(int cpu_id, int apic_id)
796 CPUID_TO_APICID(cpu_id) = apic_id;
797 APICID_TO_CPUID(apic_id) = cpu_id;
801 lapic_fixup_noioapic(void)
805 /* Only allowed on BSP */
806 KKASSERT(mycpuid == 0);
807 KKASSERT(!ioapic_enable);
809 temp = lapic->lvt_lint0;
810 temp &= ~APIC_LVT_MASKED;
811 lapic->lvt_lint0 = temp;
813 temp = lapic->lvt_lint1;
814 temp |= APIC_LVT_MASKED;
815 lapic->lvt_lint1 = temp;
819 lapic_sysinit(void *dummy __unused)
824 error = lapic_config();
830 /* Initialize BSP's local APIC */
832 } else if (ioapic_enable) {
834 icu_reinit_noioapic();
837 SYSINIT(lapic, SI_BOOT2_LAPIC, SI_ORDER_FIRST, lapic_sysinit, NULL)