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 <sys/thread2.h>
42 #include <machine/intr_machdep.h>
46 volatile lapic_t *lapic;
48 static void lapic_timer_calibrate(void);
49 static void lapic_timer_set_divisor(int);
50 static void lapic_timer_fixup_handler(void *);
51 static void lapic_timer_restart_handler(void *);
53 void lapic_timer_process(void);
54 void lapic_timer_process_frame(struct intrframe *);
55 void lapic_timer_always(struct intrframe *);
57 static int lapic_timer_enable = 1;
58 TUNABLE_INT("hw.lapic_timer_enable", &lapic_timer_enable);
60 static void lapic_timer_intr_reload(struct cputimer_intr *, sysclock_t);
61 static void lapic_timer_intr_enable(struct cputimer_intr *);
62 static void lapic_timer_intr_restart(struct cputimer_intr *);
63 static void lapic_timer_intr_pmfixup(struct cputimer_intr *);
65 static struct cputimer_intr lapic_cputimer_intr = {
67 .reload = lapic_timer_intr_reload,
68 .enable = lapic_timer_intr_enable,
69 .config = cputimer_intr_default_config,
70 .restart = lapic_timer_intr_restart,
71 .pmfixup = lapic_timer_intr_pmfixup,
72 .initclock = cputimer_intr_default_initclock,
73 .next = SLIST_ENTRY_INITIALIZER,
75 .type = CPUTIMER_INTR_LAPIC,
76 .prio = CPUTIMER_INTR_PRIO_LAPIC,
77 .caps = CPUTIMER_INTR_CAP_NONE
80 static int lapic_timer_divisor_idx = -1;
81 static const uint32_t lapic_timer_divisors[] = {
82 APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
83 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128, APIC_TDCR_1
85 #define APIC_TIMER_NDIVISORS (int)(NELEM(lapic_timer_divisors))
88 * APIC ID <-> CPU ID mapping structures.
90 int cpu_id_to_apic_id[NAPICID];
91 int apic_id_to_cpu_id[NAPICID];
101 * Enable LAPIC, configure interrupts.
104 lapic_init(boolean_t bsp)
112 * Since IDT is shared between BSP and APs, these vectors
113 * only need to be installed once; we do it on BSP.
116 /* Install a 'Spurious INTerrupt' vector */
117 setidt(XSPURIOUSINT_OFFSET, Xspuriousint,
118 SDT_SYSIGT, SEL_KPL, 0);
120 /* Install an inter-CPU IPI for TLB invalidation */
121 setidt(XINVLTLB_OFFSET, Xinvltlb,
122 SDT_SYSIGT, SEL_KPL, 0);
124 /* Install an inter-CPU IPI for IPIQ messaging */
125 setidt(XIPIQ_OFFSET, Xipiq,
126 SDT_SYSIGT, SEL_KPL, 0);
128 /* Install a timer vector */
129 setidt(XTIMER_OFFSET, Xtimer,
130 SDT_SYSIGT, SEL_KPL, 0);
132 /* Install an inter-CPU IPI for CPU stop/restart */
133 setidt(XCPUSTOP_OFFSET, Xcpustop,
134 SDT_SYSIGT, SEL_KPL, 0);
138 * Setup LINT0 as ExtINT on the BSP. This is theoretically an
139 * aggregate interrupt input from the 8259. The INTA cycle
140 * will be routed to the external controller (the 8259) which
141 * is expected to supply the vector.
143 * Must be setup edge triggered, active high.
145 * Disable LINT0 on BSP, if I/O APIC is enabled.
147 * Disable LINT0 on the APs. It doesn't matter what delivery
148 * mode we use because we leave it masked.
150 temp = lapic->lvt_lint0;
151 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
152 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
154 temp |= APIC_LVT_DM_EXTINT;
156 temp |= APIC_LVT_MASKED;
158 temp |= APIC_LVT_DM_FIXED | APIC_LVT_MASKED;
160 lapic->lvt_lint0 = temp;
163 * Setup LINT1 as NMI.
165 * Must be setup edge trigger, active high.
167 * Enable LINT1 on BSP, if I/O APIC is enabled.
169 * Disable LINT1 on the APs.
171 temp = lapic->lvt_lint1;
172 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
173 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
174 temp |= APIC_LVT_MASKED | APIC_LVT_DM_NMI;
175 if (bsp && apic_io_enable)
176 temp &= ~APIC_LVT_MASKED;
177 lapic->lvt_lint1 = temp;
180 * Mask the LAPIC error interrupt, LAPIC performance counter
183 lapic->lvt_error = lapic->lvt_error | APIC_LVT_MASKED;
184 lapic->lvt_pcint = lapic->lvt_pcint | APIC_LVT_MASKED;
187 * Set LAPIC timer vector and mask the LAPIC timer interrupt.
189 timer = lapic->lvt_timer;
190 timer &= ~APIC_LVTT_VECTOR;
191 timer |= XTIMER_OFFSET;
192 timer |= APIC_LVTT_MASKED;
193 lapic->lvt_timer = timer;
196 * Set the Task Priority Register as needed. At the moment allow
197 * interrupts on all cpus (the APs will remain CLId until they are
198 * ready to deal). We could disable all but IPIs by setting
199 * temp |= TPR_IPI for cpu != 0.
202 temp &= ~APIC_TPR_PRIO; /* clear priority field */
203 #ifdef SMP /* APIC-IO */
204 if (!apic_io_enable) {
207 * If we are NOT running the IO APICs, the LAPIC will only be used
208 * for IPIs. Set the TPR to prevent any unintentional interrupts.
211 #ifdef SMP /* APIC-IO */
220 temp |= APIC_SVR_ENABLE; /* enable the LAPIC */
221 temp &= ~APIC_SVR_FOCUS_DISABLE; /* enable lopri focus processor */
224 * Set the spurious interrupt vector. The low 4 bits of the vector
227 if ((XSPURIOUSINT_OFFSET & 0x0F) != 0x0F)
228 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET);
229 temp &= ~APIC_SVR_VECTOR;
230 temp |= XSPURIOUSINT_OFFSET;
235 * Pump out a few EOIs to clean out interrupts that got through
236 * before we were able to set the TPR.
243 lapic_timer_calibrate();
244 if (lapic_timer_enable) {
245 cputimer_intr_register(&lapic_cputimer_intr);
246 cputimer_intr_select(&lapic_cputimer_intr, 0);
249 lapic_timer_set_divisor(lapic_timer_divisor_idx);
253 apic_dump("apic_initialize()");
257 lapic_timer_set_divisor(int divisor_idx)
259 KKASSERT(divisor_idx >= 0 && divisor_idx < APIC_TIMER_NDIVISORS);
260 lapic->dcr_timer = lapic_timer_divisors[divisor_idx];
264 lapic_timer_oneshot(u_int count)
268 value = lapic->lvt_timer;
269 value &= ~APIC_LVTT_PERIODIC;
270 lapic->lvt_timer = value;
271 lapic->icr_timer = count;
275 lapic_timer_oneshot_quick(u_int count)
277 lapic->icr_timer = count;
281 lapic_timer_calibrate(void)
285 /* Try to calibrate the local APIC timer. */
286 for (lapic_timer_divisor_idx = 0;
287 lapic_timer_divisor_idx < APIC_TIMER_NDIVISORS;
288 lapic_timer_divisor_idx++) {
289 lapic_timer_set_divisor(lapic_timer_divisor_idx);
290 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
292 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
293 if (value != APIC_TIMER_MAX_COUNT)
296 if (lapic_timer_divisor_idx >= APIC_TIMER_NDIVISORS)
297 panic("lapic: no proper timer divisor?!\n");
298 lapic_cputimer_intr.freq = value / 2;
300 kprintf("lapic: divisor index %d, frequency %u Hz\n",
301 lapic_timer_divisor_idx, lapic_cputimer_intr.freq);
305 lapic_timer_process_oncpu(struct globaldata *gd, struct intrframe *frame)
309 gd->gd_timer_running = 0;
311 count = sys_cputimer->count();
312 if (TAILQ_FIRST(&gd->gd_systimerq) != NULL)
313 systimer_intr(&count, 0, frame);
317 lapic_timer_process(void)
319 lapic_timer_process_oncpu(mycpu, NULL);
323 lapic_timer_process_frame(struct intrframe *frame)
325 lapic_timer_process_oncpu(mycpu, frame);
329 * This manual debugging code is called unconditionally from Xtimer
330 * (the lapic timer interrupt) whether the current thread is in a
331 * critical section or not) and can be useful in tracking down lockups.
333 * NOTE: MANUAL DEBUG CODE
336 static int saveticks[SMP_MAXCPU];
337 static int savecounts[SMP_MAXCPU];
341 lapic_timer_always(struct intrframe *frame)
344 globaldata_t gd = mycpu;
345 int cpu = gd->gd_cpuid;
351 gptr = (short *)0xFFFFFFFF800b8000 + 80 * cpu;
352 *gptr = ((*gptr + 1) & 0x00FF) | 0x0700;
355 ksnprintf(buf, sizeof(buf), " %p %16s %d %16s ",
356 (void *)frame->if_rip, gd->gd_curthread->td_comm, ticks,
358 for (i = 0; buf[i]; ++i) {
359 gptr[i] = 0x0700 | (unsigned char)buf[i];
363 if (saveticks[gd->gd_cpuid] != ticks) {
364 saveticks[gd->gd_cpuid] = ticks;
365 savecounts[gd->gd_cpuid] = 0;
367 ++savecounts[gd->gd_cpuid];
368 if (savecounts[gd->gd_cpuid] > 2000 && panicstr == NULL) {
369 panic("cpud %d panicing on ticks failure",
372 for (i = 0; i < ncpus; ++i) {
374 if (saveticks[i] && panicstr == NULL) {
375 delta = saveticks[i] - ticks;
376 if (delta < -10 || delta > 10) {
377 panic("cpu %d panicing on cpu %d watchdog",
387 lapic_timer_intr_reload(struct cputimer_intr *cti, sysclock_t reload)
389 struct globaldata *gd = mycpu;
391 reload = (int64_t)reload * cti->freq / sys_cputimer->freq;
395 if (gd->gd_timer_running) {
396 if (reload < lapic->ccr_timer)
397 lapic_timer_oneshot_quick(reload);
399 gd->gd_timer_running = 1;
400 lapic_timer_oneshot_quick(reload);
405 lapic_timer_intr_enable(struct cputimer_intr *cti __unused)
409 timer = lapic->lvt_timer;
410 timer &= ~(APIC_LVTT_MASKED | APIC_LVTT_PERIODIC);
411 lapic->lvt_timer = timer;
413 lapic_timer_fixup_handler(NULL);
417 lapic_timer_fixup_handler(void *arg)
424 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
426 * Detect the presence of C1E capability mostly on latest
427 * dual-cores (or future) k8 family. This feature renders
428 * the local APIC timer dead, so we disable it by reading
429 * the Interrupt Pending Message register and clearing both
430 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
433 * "BIOS and Kernel Developer's Guide for AMD NPT
434 * Family 0Fh Processors"
435 * #32559 revision 3.00
437 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
438 (cpu_id & 0x0fff0000) >= 0x00040000) {
441 msr = rdmsr(0xc0010055);
442 if (msr & 0x18000000) {
443 struct globaldata *gd = mycpu;
445 kprintf("cpu%d: AMD C1E detected\n",
447 wrmsr(0xc0010055, msr & ~0x18000000ULL);
450 * We are kinda stalled;
453 gd->gd_timer_running = 1;
454 lapic_timer_oneshot_quick(2);
464 lapic_timer_restart_handler(void *dummy __unused)
468 lapic_timer_fixup_handler(&started);
470 struct globaldata *gd = mycpu;
472 gd->gd_timer_running = 1;
473 lapic_timer_oneshot_quick(2);
478 * This function is called only by ACPI-CA code currently:
479 * - AMD C1E fixup. AMD C1E only seems to happen after ACPI
480 * module controls PM. So once ACPI-CA is attached, we try
481 * to apply the fixup to prevent LAPIC timer from hanging.
484 lapic_timer_intr_pmfixup(struct cputimer_intr *cti __unused)
486 lwkt_send_ipiq_mask(smp_active_mask,
487 lapic_timer_fixup_handler, NULL);
491 lapic_timer_intr_restart(struct cputimer_intr *cti __unused)
493 lwkt_send_ipiq_mask(smp_active_mask, lapic_timer_restart_handler, NULL);
498 * dump contents of local APIC registers
503 kprintf("SMP: CPU%d %s:\n", mycpu->gd_cpuid, str);
504 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
505 lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
509 * Inter Processor Interrupt functions.
513 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
515 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
516 * vector is any valid SYSTEM INT vector
517 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
519 * A backlog of requests can create a deadlock between cpus. To avoid this
520 * we have to be able to accept IPIs at the same time we are trying to send
521 * them. The critical section prevents us from attempting to send additional
522 * IPIs reentrantly, but also prevents IPIQ processing so we have to call
523 * lwkt_process_ipiq() manually. It's rather messy and expensive for this
524 * to occur but fortunately it does not happen too often.
527 apic_ipi(int dest_type, int vector, int delivery_mode)
532 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
533 unsigned long rflags = read_rflags();
535 DEBUG_PUSH_INFO("apic_ipi");
536 while ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
540 write_rflags(rflags);
543 icr_lo = (lapic->icr_lo & APIC_ICRLO_RESV_MASK) | dest_type |
544 delivery_mode | vector;
545 lapic->icr_lo = icr_lo;
551 single_apic_ipi(int cpu, int vector, int delivery_mode)
557 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
558 unsigned long rflags = read_rflags();
560 DEBUG_PUSH_INFO("single_apic_ipi");
561 while ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
565 write_rflags(rflags);
567 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
568 icr_hi |= (CPUID_TO_APICID(cpu) << 24);
569 lapic->icr_hi = icr_hi;
572 icr_lo = (lapic->icr_lo & APIC_ICRLO_RESV_MASK)
573 | APIC_DEST_DESTFLD | delivery_mode | vector;
576 lapic->icr_lo = icr_lo;
583 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
585 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
586 * to the target, and the scheduler does not 'poll' for IPI messages.
589 single_apic_ipi_passive(int cpu, int vector, int delivery_mode)
595 if ((lapic->icr_lo & APIC_DELSTAT_MASK) != 0) {
599 icr_hi = lapic->icr_hi & ~APIC_ID_MASK;
600 icr_hi |= (CPUID_TO_APICID(cpu) << 24);
601 lapic->icr_hi = icr_hi;
604 icr_lo = (lapic->icr_lo & APIC_RESV2_MASK)
605 | APIC_DEST_DESTFLD | delivery_mode | vector;
608 lapic->icr_lo = icr_lo;
616 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
618 * target is a bitmask of destination cpus. Vector is any
619 * valid system INT vector. Delivery mode may be either
620 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
623 selected_apic_ipi(cpumask_t target, int vector, int delivery_mode)
627 int n = BSFCPUMASK(target);
628 target &= ~CPUMASK(n);
629 single_apic_ipi(n, vector, delivery_mode);
635 * Timer code, in development...
636 * - suggested by rgrimes@gndrsh.aac.dev.com
639 get_apic_timer_frequency(void)
641 return(lapic_cputimer_intr.freq);
645 * Load a 'downcount time' in uSeconds.
648 set_apic_timer(int us)
653 * When we reach here, lapic timer's frequency
654 * must have been calculated as well as the
655 * divisor (lapic->dcr_timer is setup during the
656 * divisor calculation).
658 KKASSERT(lapic_cputimer_intr.freq != 0 &&
659 lapic_timer_divisor_idx >= 0);
661 count = ((us * (int64_t)lapic_cputimer_intr.freq) + 999999) / 1000000;
662 lapic_timer_oneshot(count);
667 * Read remaining time in timer.
670 read_apic_timer(void)
673 /** XXX FIXME: we need to return the actual remaining time,
674 * for now we just return the remaining count.
677 return lapic->ccr_timer;
683 * Spin-style delay, set delay time in uS, spin till it drains.
688 set_apic_timer(count);
689 while (read_apic_timer())
694 lapic_unused_apic_id(int start)
698 for (i = start; i < NAPICID; ++i) {
699 if (APICID_TO_CPUID(i) == -1)
706 lapic_map(vm_offset_t lapic_addr)
708 lapic = pmap_mapdev_uncacheable(lapic_addr, sizeof(struct LAPIC));
710 kprintf("lapic: at 0x%08lx\n", lapic_addr);
713 static TAILQ_HEAD(, lapic_enumerator) lapic_enumerators =
714 TAILQ_HEAD_INITIALIZER(lapic_enumerators);
719 struct lapic_enumerator *e;
722 for (i = 0; i < NAPICID; ++i)
723 APICID_TO_CPUID(i) = -1;
725 TAILQ_FOREACH(e, &lapic_enumerators, lapic_link) {
726 error = e->lapic_probe(e);
731 panic("can't config lapic\n");
733 e->lapic_enumerate(e);
737 lapic_enumerator_register(struct lapic_enumerator *ne)
739 struct lapic_enumerator *e;
741 TAILQ_FOREACH(e, &lapic_enumerators, lapic_link) {
742 if (e->lapic_prio < ne->lapic_prio) {
743 TAILQ_INSERT_BEFORE(e, ne, lapic_link);
747 TAILQ_INSERT_TAIL(&lapic_enumerators, ne, lapic_link);
751 lapic_set_cpuid(int cpu_id, int apic_id)
753 CPUID_TO_APICID(cpu_id) = apic_id;
754 APICID_TO_CPUID(apic_id) = cpu_id;