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_base/apic/lapic.h>
38 #include <machine_base/apic/ioapic_abi.h>
39 #include <machine/segments.h>
40 #include <machine/specialreg.h>
41 #include <sys/thread2.h>
43 #include <machine/intr_machdep.h>
47 volatile lapic_t *lapic;
49 static void lapic_timer_calibrate(void);
50 static void lapic_timer_set_divisor(int);
51 static void lapic_timer_fixup_handler(void *);
52 static void lapic_timer_restart_handler(void *);
54 void lapic_timer_process(void);
55 void lapic_timer_process_frame(struct intrframe *);
56 void lapic_timer_always(struct intrframe *);
58 static int lapic_timer_enable = 1;
59 TUNABLE_INT("hw.lapic_timer_enable", &lapic_timer_enable);
61 static void lapic_timer_intr_reload(struct cputimer_intr *, sysclock_t);
62 static void lapic_timer_intr_enable(struct cputimer_intr *);
63 static void lapic_timer_intr_restart(struct cputimer_intr *);
64 static void lapic_timer_intr_pmfixup(struct cputimer_intr *);
66 static struct cputimer_intr lapic_cputimer_intr = {
68 .reload = lapic_timer_intr_reload,
69 .enable = lapic_timer_intr_enable,
70 .config = cputimer_intr_default_config,
71 .restart = lapic_timer_intr_restart,
72 .pmfixup = lapic_timer_intr_pmfixup,
73 .initclock = cputimer_intr_default_initclock,
74 .next = SLIST_ENTRY_INITIALIZER,
76 .type = CPUTIMER_INTR_LAPIC,
77 .prio = CPUTIMER_INTR_PRIO_LAPIC,
78 .caps = CPUTIMER_INTR_CAP_NONE
81 static int lapic_timer_divisor_idx = -1;
82 static const uint32_t lapic_timer_divisors[] = {
83 APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
84 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128, APIC_TDCR_1
86 #define APIC_TIMER_NDIVISORS (int)(NELEM(lapic_timer_divisors))
89 * APIC ID <-> CPU ID mapping structures.
91 int cpu_id_to_apic_id[NAPICID];
92 int apic_id_to_cpu_id[NAPICID];
102 * Enable LAPIC, configure interrupts.
105 lapic_init(boolean_t bsp)
113 * Since IDT is shared between BSP and APs, these vectors
114 * only need to be installed once; we do it on BSP.
117 /* Install a 'Spurious INTerrupt' vector */
118 setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
119 SDT_SYSIGT, SEL_KPL, 0);
121 /* Install an inter-CPU IPI for TLB invalidation */
122 setidt(XINVLTLB_OFFSET, Xinvltlb,
123 SDT_SYSIGT, SEL_KPL, 0);
125 /* Install an inter-CPU IPI for IPIQ messaging */
126 setidt(XIPIQ_OFFSET, Xipiq,
127 SDT_SYSIGT, SEL_KPL, 0);
129 /* Install a timer vector */
130 setidt(XTIMER_OFFSET, Xtimer,
131 SDT_SYSIGT, SEL_KPL, 0);
133 /* Install an inter-CPU IPI for CPU stop/restart */
134 setidt(XCPUSTOP_OFFSET, Xcpustop,
135 SDT_SYSIGT, SEL_KPL, 0);
139 * Setup LINT0 as ExtINT on the BSP. This is theoretically an
140 * aggregate interrupt input from the 8259. The INTA cycle
141 * will be routed to the external controller (the 8259) which
142 * is expected to supply the vector.
144 * Must be setup edge triggered, active high.
146 * Disable LINT0 on BSP, if I/O APIC is enabled.
148 * Disable LINT0 on the APs. It doesn't matter what delivery
149 * mode we use because we leave it masked.
151 temp = lapic->lvt_lint0;
152 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
153 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
155 temp |= APIC_LVT_DM_EXTINT;
157 temp |= APIC_LVT_MASKED;
159 temp |= APIC_LVT_DM_FIXED | APIC_LVT_MASKED;
161 lapic->lvt_lint0 = temp;
164 * Setup LINT1 as NMI.
166 * Must be setup edge trigger, active high.
168 * Enable LINT1 on BSP, if I/O APIC is enabled.
170 * Disable LINT1 on the APs.
172 temp = lapic->lvt_lint1;
173 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
174 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
175 temp |= APIC_LVT_MASKED | APIC_LVT_DM_NMI;
176 if (bsp && apic_io_enable)
177 temp &= ~APIC_LVT_MASKED;
178 lapic->lvt_lint1 = temp;
181 * Mask the LAPIC error interrupt, LAPIC performance counter
184 lapic->lvt_error = lapic->lvt_error | APIC_LVT_MASKED;
185 lapic->lvt_pcint = lapic->lvt_pcint | APIC_LVT_MASKED;
188 * Set LAPIC timer vector and mask the LAPIC timer interrupt.
190 timer = lapic->lvt_timer;
191 timer &= ~APIC_LVTT_VECTOR;
192 timer |= XTIMER_OFFSET;
193 timer |= APIC_LVTT_MASKED;
194 lapic->lvt_timer = timer;
197 * Set the Task Priority Register as needed. At the moment allow
198 * interrupts on all cpus (the APs will remain CLId until they are
202 temp &= ~APIC_TPR_PRIO; /* clear priority field */
209 temp |= APIC_SVR_ENABLE; /* enable the LAPIC */
210 temp &= ~APIC_SVR_FOCUS_DISABLE; /* enable lopri focus processor */
213 * Set the spurious interrupt vector. The low 4 bits of the vector
216 if ((XSPURIOUSINT_OFFSET & 0x0F) != 0x0F)
217 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET);
218 temp &= ~APIC_SVR_VECTOR;
219 temp |= XSPURIOUSINT_OFFSET;
224 * Pump out a few EOIs to clean out interrupts that got through
225 * before we were able to set the TPR.
232 lapic_timer_calibrate();
233 if (lapic_timer_enable) {
234 cputimer_intr_register(&lapic_cputimer_intr);
235 cputimer_intr_select(&lapic_cputimer_intr, 0);
238 lapic_timer_set_divisor(lapic_timer_divisor_idx);
242 apic_dump("apic_initialize()");
246 lapic_timer_set_divisor(int divisor_idx)
248 KKASSERT(divisor_idx >= 0 && divisor_idx < APIC_TIMER_NDIVISORS);
249 lapic->dcr_timer = lapic_timer_divisors[divisor_idx];
253 lapic_timer_oneshot(u_int count)
257 value = lapic->lvt_timer;
258 value &= ~APIC_LVTT_PERIODIC;
259 lapic->lvt_timer = value;
260 lapic->icr_timer = count;
264 lapic_timer_oneshot_quick(u_int count)
266 lapic->icr_timer = count;
270 lapic_timer_calibrate(void)
274 /* Try to calibrate the local APIC timer. */
275 for (lapic_timer_divisor_idx = 0;
276 lapic_timer_divisor_idx < APIC_TIMER_NDIVISORS;
277 lapic_timer_divisor_idx++) {
278 lapic_timer_set_divisor(lapic_timer_divisor_idx);
279 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
281 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
282 if (value != APIC_TIMER_MAX_COUNT)
285 if (lapic_timer_divisor_idx >= APIC_TIMER_NDIVISORS)
286 panic("lapic: no proper timer divisor?!\n");
287 lapic_cputimer_intr.freq = value / 2;
289 kprintf("lapic: divisor index %d, frequency %u Hz\n",
290 lapic_timer_divisor_idx, lapic_cputimer_intr.freq);
294 lapic_timer_process_oncpu(struct globaldata *gd, struct intrframe *frame)
298 gd->gd_timer_running = 0;
300 count = sys_cputimer->count();
301 if (TAILQ_FIRST(&gd->gd_systimerq) != NULL)
302 systimer_intr(&count, 0, frame);
306 lapic_timer_process(void)
308 lapic_timer_process_oncpu(mycpu, NULL);
312 lapic_timer_process_frame(struct intrframe *frame)
314 lapic_timer_process_oncpu(mycpu, frame);
318 * This manual debugging code is called unconditionally from Xtimer
319 * (the lapic timer interrupt) whether the current thread is in a
320 * critical section or not) and can be useful in tracking down lockups.
322 * NOTE: MANUAL DEBUG CODE
325 static int saveticks[SMP_MAXCPU];
326 static int savecounts[SMP_MAXCPU];
330 lapic_timer_always(struct intrframe *frame)
333 globaldata_t gd = mycpu;
334 int cpu = gd->gd_cpuid;
340 gptr = (short *)0xFFFFFFFF800b8000 + 80 * cpu;
341 *gptr = ((*gptr + 1) & 0x00FF) | 0x0700;
344 ksnprintf(buf, sizeof(buf), " %p %16s %d %16s ",
345 (void *)frame->if_rip, gd->gd_curthread->td_comm, ticks,
347 for (i = 0; buf[i]; ++i) {
348 gptr[i] = 0x0700 | (unsigned char)buf[i];
352 if (saveticks[gd->gd_cpuid] != ticks) {
353 saveticks[gd->gd_cpuid] = ticks;
354 savecounts[gd->gd_cpuid] = 0;
356 ++savecounts[gd->gd_cpuid];
357 if (savecounts[gd->gd_cpuid] > 2000 && panicstr == NULL) {
358 panic("cpud %d panicing on ticks failure",
361 for (i = 0; i < ncpus; ++i) {
363 if (saveticks[i] && panicstr == NULL) {
364 delta = saveticks[i] - ticks;
365 if (delta < -10 || delta > 10) {
366 panic("cpu %d panicing on cpu %d watchdog",
376 lapic_timer_intr_reload(struct cputimer_intr *cti, sysclock_t reload)
378 struct globaldata *gd = mycpu;
380 reload = (int64_t)reload * cti->freq / sys_cputimer->freq;
384 if (gd->gd_timer_running) {
385 if (reload < lapic->ccr_timer)
386 lapic_timer_oneshot_quick(reload);
388 gd->gd_timer_running = 1;
389 lapic_timer_oneshot_quick(reload);
394 lapic_timer_intr_enable(struct cputimer_intr *cti __unused)
398 timer = lapic->lvt_timer;
399 timer &= ~(APIC_LVTT_MASKED | APIC_LVTT_PERIODIC);
400 lapic->lvt_timer = timer;
402 lapic_timer_fixup_handler(NULL);
406 lapic_timer_fixup_handler(void *arg)
413 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
415 * Detect the presence of C1E capability mostly on latest
416 * dual-cores (or future) k8 family. This feature renders
417 * the local APIC timer dead, so we disable it by reading
418 * the Interrupt Pending Message register and clearing both
419 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
422 * "BIOS and Kernel Developer's Guide for AMD NPT
423 * Family 0Fh Processors"
424 * #32559 revision 3.00
426 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
427 (cpu_id & 0x0fff0000) >= 0x00040000) {
430 msr = rdmsr(0xc0010055);
431 if (msr & 0x18000000) {
432 struct globaldata *gd = mycpu;
434 kprintf("cpu%d: AMD C1E detected\n",
436 wrmsr(0xc0010055, msr & ~0x18000000ULL);
439 * We are kinda stalled;
442 gd->gd_timer_running = 1;
443 lapic_timer_oneshot_quick(2);
453 lapic_timer_restart_handler(void *dummy __unused)
457 lapic_timer_fixup_handler(&started);
459 struct globaldata *gd = mycpu;
461 gd->gd_timer_running = 1;
462 lapic_timer_oneshot_quick(2);
467 * This function is called only by ACPI-CA code currently:
468 * - AMD C1E fixup. AMD C1E only seems to happen after ACPI
469 * module controls PM. So once ACPI-CA is attached, we try
470 * to apply the fixup to prevent LAPIC timer from hanging.
473 lapic_timer_intr_pmfixup(struct cputimer_intr *cti __unused)
475 lwkt_send_ipiq_mask(smp_active_mask,
476 lapic_timer_fixup_handler, NULL);
480 lapic_timer_intr_restart(struct cputimer_intr *cti __unused)
482 lwkt_send_ipiq_mask(smp_active_mask, lapic_timer_restart_handler, NULL);
487 * dump contents of local APIC registers
492 kprintf("SMP: CPU%d %s:\n", mycpu->gd_cpuid, str);
493 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
494 lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
498 * Inter Processor Interrupt functions.
502 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
504 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
505 * vector is any valid SYSTEM INT vector
506 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
508 * A backlog of requests can create a deadlock between cpus. To avoid this
509 * we have to be able to accept IPIs at the same time we are trying to send
510 * them. The critical section prevents us from attempting to send additional
511 * IPIs reentrantly, but also prevents IPIQ processing so we have to call
512 * lwkt_process_ipiq() manually. It's rather messy and expensive for this
513 * to occur but fortunately it does not happen too often.
516 apic_ipi(int dest_type, int vector, int delivery_mode)
521 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
522 unsigned long rflags = read_rflags();
524 DEBUG_PUSH_INFO("apic_ipi");
525 while ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
529 write_rflags(rflags);
532 icr_lo = (lapic->icr_lo & APIC_ICRLO_RESV_MASK) | dest_type |
533 delivery_mode | vector;
534 lapic->icr_lo = icr_lo;
540 single_apic_ipi(int cpu, int vector, int delivery_mode)
546 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
547 unsigned long rflags = read_rflags();
549 DEBUG_PUSH_INFO("single_apic_ipi");
550 while ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
554 write_rflags(rflags);
556 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
557 icr_hi |= (CPUID_TO_APICID(cpu) << 24);
558 lapic->icr_hi = icr_hi;
561 icr_lo = (lapic->icr_lo & APIC_ICRLO_RESV_MASK)
562 | APIC_DEST_DESTFLD | delivery_mode | vector;
565 lapic->icr_lo = icr_lo;
572 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
574 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
575 * to the target, and the scheduler does not 'poll' for IPI messages.
578 single_apic_ipi_passive(int cpu, int vector, int delivery_mode)
584 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
588 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
589 icr_hi |= (CPUID_TO_APICID(cpu) << 24);
590 lapic->icr_hi = icr_hi;
593 icr_lo = (lapic->icr_lo & APIC_RESV2_MASK)
594 | APIC_DEST_DESTFLD | delivery_mode | vector;
597 lapic->icr_lo = icr_lo;
605 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
607 * target is a bitmask of destination cpus. Vector is any
608 * valid system INT vector. Delivery mode may be either
609 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
612 selected_apic_ipi(cpumask_t target, int vector, int delivery_mode)
616 int n = BSFCPUMASK(target);
617 target &= ~CPUMASK(n);
618 single_apic_ipi(n, vector, delivery_mode);
624 * Timer code, in development...
625 * - suggested by rgrimes@gndrsh.aac.dev.com
628 get_apic_timer_frequency(void)
630 return(lapic_cputimer_intr.freq);
634 * Load a 'downcount time' in uSeconds.
637 set_apic_timer(int us)
642 * When we reach here, lapic timer's frequency
643 * must have been calculated as well as the
644 * divisor (lapic->dcr_timer is setup during the
645 * divisor calculation).
647 KKASSERT(lapic_cputimer_intr.freq != 0 &&
648 lapic_timer_divisor_idx >= 0);
650 count = ((us * (int64_t)lapic_cputimer_intr.freq) + 999999) / 1000000;
651 lapic_timer_oneshot(count);
656 * Read remaining time in timer.
659 read_apic_timer(void)
662 /** XXX FIXME: we need to return the actual remaining time,
663 * for now we just return the remaining count.
666 return lapic->ccr_timer;
672 * Spin-style delay, set delay time in uS, spin till it drains.
677 set_apic_timer(count);
678 while (read_apic_timer())
683 lapic_unused_apic_id(int start)
687 for (i = start; i < NAPICID; ++i) {
688 if (APICID_TO_CPUID(i) == -1)
695 lapic_map(vm_offset_t lapic_addr)
697 lapic = pmap_mapdev_uncacheable(lapic_addr, sizeof(struct LAPIC));
699 kprintf("lapic: at 0x%08lx\n", lapic_addr);
702 static TAILQ_HEAD(, lapic_enumerator) lapic_enumerators =
703 TAILQ_HEAD_INITIALIZER(lapic_enumerators);
708 struct lapic_enumerator *e;
709 int error, i, enable, ap_max;
711 for (i = 0; i < NAPICID; ++i)
712 APICID_TO_CPUID(i) = -1;
715 TUNABLE_INT_FETCH("hw.lapic_enable", &enable);
717 kprintf("LAPIC: Warning LAPIC is disabled\n");
721 TAILQ_FOREACH(e, &lapic_enumerators, lapic_link) {
722 error = e->lapic_probe(e);
727 kprintf("LAPIC: Can't find LAPIC\n");
731 e->lapic_enumerate(e);
734 TUNABLE_INT_FETCH("hw.ap_max", &ap_max);
735 if (ap_max > MAXCPU - 1)
738 if (mp_naps > ap_max) {
739 kprintf("LAPIC: Warning use only %d out of %d "
745 if ((cpu_feature2 & CPUID2_VMM) && mp_naps == 0) {
748 * Special hack for vmware. It looks like that
749 * if only one CPU is configured (mp_naps == 0)
750 * in vmware (cpu_feature2 & CPUID2_VMM),
751 * then LAPIC will not work at all.
753 kprintf("LAPIC: single CPU virtual machine detected, "
761 lapic_enumerator_register(struct lapic_enumerator *ne)
763 struct lapic_enumerator *e;
765 TAILQ_FOREACH(e, &lapic_enumerators, lapic_link) {
766 if (e->lapic_prio < ne->lapic_prio) {
767 TAILQ_INSERT_BEFORE(e, ne, lapic_link);
771 TAILQ_INSERT_TAIL(&lapic_enumerators, ne, lapic_link);
775 lapic_set_cpuid(int cpu_id, int apic_id)
777 CPUID_TO_APICID(cpu_id) = apic_id;
778 APICID_TO_CPUID(apic_id) = cpu_id;
782 lapic_fixup_noioapic(void)
786 /* Only allowed on BSP */
787 KKASSERT(mycpuid == 0);
788 KKASSERT(!apic_io_enable);
790 temp = lapic->lvt_lint0;
791 temp &= ~APIC_LVT_MASKED;
792 lapic->lvt_lint0 = temp;
794 temp = lapic->lvt_lint1;
795 temp |= APIC_LVT_MASKED;
796 lapic->lvt_lint1 = temp;