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 $
26 * $DragonFly: src/sys/platform/pc32/apic/mpapic.c,v 1.22 2008/04/20 13:44:26 swildner Exp $
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/kernel.h>
32 #include <machine/globaldata.h>
33 #include <machine/smp.h>
34 #include <machine/md_var.h>
35 #include <machine_base/apic/mpapic.h>
36 #include <machine/segments.h>
37 #include <sys/thread2.h>
39 #include <machine_base/isa/intr_machdep.h> /* Xspuriousint() */
41 /* EISA Edge/Level trigger control registers */
42 #define ELCR0 0x4d0 /* eisa irq 0-7 */
43 #define ELCR1 0x4d1 /* eisa irq 8-15 */
45 static void lapic_timer_calibrate(void);
46 static void lapic_timer_set_divisor(int);
47 static void lapic_timer_intr_reload(sysclock_t);
49 void lapic_timer_process(void);
50 void lapic_timer_process_frame(struct intrframe *);
51 void lapic_timer_intr_test(void);
52 void lapic_timer_oneshot_intr_enable(void);
55 int lapic_timer_enable;
57 TUNABLE_INT("hw.lapic_timer_test", &lapic_timer_test);
58 TUNABLE_INT("hw.lapic_timer_enable", &lapic_timer_enable);
61 * pointers to pmapped apic hardware.
64 volatile ioapic_t **ioapic;
66 static sysclock_t lapic_timer_freq;
67 static int lapic_timer_divisor_idx = -1;
68 static const uint32_t lapic_timer_divisors[] = {
69 APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
70 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128, APIC_TDCR_1
72 #define APIC_TIMER_NDIVISORS \
73 (int)(sizeof(lapic_timer_divisors) / sizeof(lapic_timer_divisors[0]))
77 * Enable APIC, configure interrupts.
80 apic_initialize(boolean_t bsp)
86 * setup LVT1 as ExtINT on the BSP. This is theoretically an
87 * aggregate interrupt input from the 8259. The INTA cycle
88 * will be routed to the external controller (the 8259) which
89 * is expected to supply the vector.
91 * Must be setup edge triggered, active high.
93 * Disable LVT1 on the APs. It doesn't matter what delivery
94 * mode we use because we leave it masked.
96 temp = lapic.lvt_lint0;
97 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
98 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
99 if (mycpu->gd_cpuid == 0)
100 temp |= APIC_LVT_DM_EXTINT;
102 temp |= APIC_LVT_DM_FIXED | APIC_LVT_MASKED;
103 lapic.lvt_lint0 = temp;
106 * setup LVT2 as NMI, masked till later. Edge trigger, active high.
108 temp = lapic.lvt_lint1;
109 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
110 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
111 temp |= APIC_LVT_MASKED | APIC_LVT_DM_NMI;
112 lapic.lvt_lint1 = temp;
115 * Mask the apic error interrupt, apic performance counter
118 lapic.lvt_error = lapic.lvt_error | APIC_LVT_MASKED;
119 lapic.lvt_pcint = lapic.lvt_pcint | APIC_LVT_MASKED;
121 /* Set apic timer vector and mask the apic timer interrupt. */
122 timer = lapic.lvt_timer;
123 timer &= ~APIC_LVTT_VECTOR;
124 timer |= XTIMER_OFFSET;
125 timer |= APIC_LVTT_MASKED;
126 lapic.lvt_timer = timer;
129 * Set the Task Priority Register as needed. At the moment allow
130 * interrupts on all cpus (the APs will remain CLId until they are
131 * ready to deal). We could disable all but IPIs by setting
132 * temp |= TPR_IPI_ONLY for cpu != 0.
135 temp &= ~APIC_TPR_PRIO; /* clear priority field */
138 * If we are NOT running the IO APICs, the LAPIC will only be used
139 * for IPIs. Set the TPR to prevent any unintentional interrupts.
141 temp |= TPR_IPI_ONLY;
147 * enable the local APIC
150 temp |= APIC_SVR_ENABLE; /* enable the APIC */
151 temp &= ~APIC_SVR_FOCUS_DISABLE; /* enable lopri focus processor */
154 * Set the spurious interrupt vector. The low 4 bits of the vector
157 if ((XSPURIOUSINT_OFFSET & 0x0F) != 0x0F)
158 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET);
159 temp &= ~APIC_SVR_VECTOR;
160 temp |= XSPURIOUSINT_OFFSET;
165 * Pump out a few EOIs to clean out interrupts that got through
166 * before we were able to set the TPR.
172 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
174 * Detect the presence of C1E capability mostly on latest
175 * dual-cores (or future) k8 family. This feature renders
176 * the local APIC timer dead, so we disable it by reading
177 * the Interrupt Pending Message register and clearing both
178 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
181 * "BIOS and Kernel Developer's Guide for AMD NPT
182 * Family 0Fh Processors"
183 * #32559 revision 3.00
185 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
186 (cpu_id & 0x0fff0000) >= 0x00040000) {
189 msr = rdmsr(0xc0010055);
190 if (msr & 0x18000000)
191 wrmsr(0xc0010055, msr & ~0x18000000ULL);
196 lapic_timer_calibrate();
197 if (lapic_timer_enable)
198 cputimer_intr_reload = lapic_timer_intr_reload;
200 lapic_timer_set_divisor(lapic_timer_divisor_idx);
204 apic_dump("apic_initialize()");
209 lapic_timer_set_divisor(int divisor_idx)
211 KKASSERT(divisor_idx >= 0 && divisor_idx < APIC_TIMER_NDIVISORS);
212 lapic.dcr_timer = lapic_timer_divisors[divisor_idx];
216 lapic_timer_oneshot(u_int count)
220 value = lapic.lvt_timer;
221 value &= ~APIC_LVTT_PERIODIC;
222 lapic.lvt_timer = value;
223 lapic.icr_timer = count;
227 lapic_timer_oneshot_quick(u_int count)
229 lapic.icr_timer = count;
233 lapic_timer_calibrate(void)
237 /* Try to calibrate the local APIC timer. */
238 for (lapic_timer_divisor_idx = 0;
239 lapic_timer_divisor_idx < APIC_TIMER_NDIVISORS;
240 lapic_timer_divisor_idx++) {
241 lapic_timer_set_divisor(lapic_timer_divisor_idx);
242 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
244 value = APIC_TIMER_MAX_COUNT - lapic.ccr_timer;
245 if (value != APIC_TIMER_MAX_COUNT)
248 if (lapic_timer_divisor_idx >= APIC_TIMER_NDIVISORS)
249 panic("lapic: no proper timer divisor?!\n");
250 lapic_timer_freq = value / 2;
252 kprintf("lapic: divisor index %d, frequency %u Hz\n",
253 lapic_timer_divisor_idx, lapic_timer_freq);
257 lapic_timer_process_oncpu(struct globaldata *gd, struct intrframe *frame)
261 gd->gd_timer_running = 0;
263 count = sys_cputimer->count();
264 if (TAILQ_FIRST(&gd->gd_systimerq) != NULL)
265 systimer_intr(&count, 0, frame);
269 lapic_timer_process(void)
271 struct globaldata *gd = mycpu;
273 if (__predict_false(lapic_timer_test)) {
274 gd->gd_timer_running = 0;
275 kprintf("%d proc\n", gd->gd_cpuid);
277 lapic_timer_process_oncpu(gd, NULL);
282 lapic_timer_process_frame(struct intrframe *frame)
284 struct globaldata *gd = mycpu;
286 if (__predict_false(lapic_timer_test)) {
287 gd->gd_timer_running = 0;
288 kprintf("%d proc frame\n", gd->gd_cpuid);
290 lapic_timer_process_oncpu(gd, frame);
295 lapic_timer_intr_test(void)
297 struct globaldata *gd = mycpu;
299 if (!gd->gd_timer_running) {
300 gd->gd_timer_running = 1;
301 KKASSERT(lapic_timer_freq != 0);
302 lapic_timer_oneshot_quick(lapic_timer_freq);
307 lapic_timer_intr_reload(sysclock_t reload)
309 struct globaldata *gd = mycpu;
311 reload = (int64_t)reload * lapic_timer_freq / sys_cputimer->freq;
315 if (gd->gd_timer_running) {
316 if (reload < lapic.ccr_timer)
317 lapic_timer_oneshot_quick(reload);
319 gd->gd_timer_running = 1;
320 lapic_timer_oneshot_quick(reload);
325 lapic_timer_oneshot_intr_enable(void)
329 timer = lapic.lvt_timer;
330 timer &= ~(APIC_LVTT_MASKED | APIC_LVTT_PERIODIC);
331 lapic.lvt_timer = timer;
336 * dump contents of local APIC registers
341 kprintf("SMP: CPU%d %s:\n", mycpu->gd_cpuid, str);
342 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
343 lapic.lvt_lint0, lapic.lvt_lint1, lapic.tpr, lapic.svr);
353 #define IOAPIC_ISA_INTS 16
354 #define REDIRCNT_IOAPIC(A) \
355 ((int)((io_apic_versions[(A)] & IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) + 1)
357 static int trigger (int apic, int pin, u_int32_t * flags);
358 static void polarity (int apic, int pin, u_int32_t * flags, int level);
360 #define DEFAULT_FLAGS \
366 #define DEFAULT_ISA_FLAGS \
375 io_apic_set_id(int apic, int id)
379 ux = io_apic_read(apic, IOAPIC_ID); /* get current contents */
380 if (((ux & APIC_ID_MASK) >> 24) != id) {
381 kprintf("Changing APIC ID for IO APIC #%d"
382 " from %d to %d on chip\n",
383 apic, ((ux & APIC_ID_MASK) >> 24), id);
384 ux &= ~APIC_ID_MASK; /* clear the ID field */
386 io_apic_write(apic, IOAPIC_ID, ux); /* write new value */
387 ux = io_apic_read(apic, IOAPIC_ID); /* re-read && test */
388 if (((ux & APIC_ID_MASK) >> 24) != id)
389 panic("can't control IO APIC #%d ID, reg: 0x%08x",
396 io_apic_get_id(int apic)
398 return (io_apic_read(apic, IOAPIC_ID) & APIC_ID_MASK) >> 24;
407 extern int apic_pin_trigger; /* 'opaque' */
410 io_apic_setup_intpin(int apic, int pin)
412 int bus, bustype, irq;
413 u_char select; /* the select register is 8 bits */
414 u_int32_t flags; /* the window register is 32 bits */
415 u_int32_t target; /* the window register is 32 bits */
416 u_int32_t vector; /* the window register is 32 bits */
419 select = pin * 2 + IOAPIC_REDTBL0; /* register */
422 * Always clear an IO APIC pin before [re]programming it. This is
423 * particularly important if the pin is set up for a level interrupt
424 * as the IOART_REM_IRR bit might be set. When we reprogram the
425 * vector any EOI from pending ints on this pin could be lost and
426 * IRR might never get reset.
428 * To fix this problem, clear the vector and make sure it is
429 * programmed as an edge interrupt. This should theoretically
430 * clear IRR so we can later, safely program it as a level
435 flags = io_apic_read(apic, select) & IOART_RESV;
436 flags |= IOART_INTMSET | IOART_TRGREDG | IOART_INTAHI;
437 flags |= IOART_DESTPHY | IOART_DELFIXED;
439 target = io_apic_read(apic, select + 1) & IOART_HI_DEST_RESV;
440 target |= 0; /* fixed mode cpu mask of 0 - don't deliver anywhere */
444 io_apic_write(apic, select, flags | vector);
445 io_apic_write(apic, select + 1, target);
450 * We only deal with vectored interrupts here. ? documentation is
451 * lacking, I'm guessing an interrupt type of 0 is the 'INT' type,
454 * This test also catches unconfigured pins.
456 if (apic_int_type(apic, pin) != 0)
460 * Leave the pin unprogrammed if it does not correspond to
463 irq = apic_irq(apic, pin);
467 /* determine the bus type for this pin */
468 bus = apic_src_bus_id(apic, pin);
471 bustype = apic_bus_type(bus);
473 if ((bustype == ISA) &&
474 (pin < IOAPIC_ISA_INTS) &&
476 (apic_polarity(apic, pin) == 0x1) &&
477 (apic_trigger(apic, pin) == 0x3)) {
479 * A broken BIOS might describe some ISA
480 * interrupts as active-high level-triggered.
481 * Use default ISA flags for those interrupts.
483 flags = DEFAULT_ISA_FLAGS;
486 * Program polarity and trigger mode according to
489 flags = DEFAULT_FLAGS;
490 level = trigger(apic, pin, &flags);
492 apic_pin_trigger |= (1 << irq);
493 polarity(apic, pin, &flags, level);
497 kprintf("IOAPIC #%d intpin %d -> irq %d\n",
502 * Program the appropriate registers. This routing may be
503 * overridden when an interrupt handler for a device is
504 * actually added (see register_int(), which calls through
505 * the MACHINTR ABI to set up an interrupt handler/vector).
507 * The order in which we must program the two registers for
508 * safety is unclear! XXX
512 vector = IDT_OFFSET + irq; /* IDT vec */
513 target = io_apic_read(apic, select + 1) & IOART_HI_DEST_RESV;
514 target |= IOART_HI_DEST_BROADCAST;
515 flags |= io_apic_read(apic, select) & IOART_RESV;
516 io_apic_write(apic, select, flags | vector);
517 io_apic_write(apic, select + 1, target);
523 io_apic_setup(int apic)
529 apic_pin_trigger = 0; /* default to edge-triggered */
531 maxpin = REDIRCNT_IOAPIC(apic); /* pins in APIC */
532 kprintf("Programming %d pins in IOAPIC #%d\n", maxpin, apic);
534 for (pin = 0; pin < maxpin; ++pin) {
535 io_apic_setup_intpin(apic, pin);
538 if (apic_int_type(apic, pin) >= 0) {
539 kprintf("Warning: IOAPIC #%d pin %d does not exist,"
540 " cannot program!\n", apic, pin);
545 /* return GOOD status */
548 #undef DEFAULT_ISA_FLAGS
552 #define DEFAULT_EXTINT_FLAGS \
561 * Setup the source of External INTerrupts.
564 ext_int_setup(int apic, int intr)
566 u_char select; /* the select register is 8 bits */
567 u_int32_t flags; /* the window register is 32 bits */
568 u_int32_t target; /* the window register is 32 bits */
569 u_int32_t vector; /* the window register is 32 bits */
571 if (apic_int_type(apic, intr) != 3)
574 target = IOART_HI_DEST_BROADCAST;
575 select = IOAPIC_REDTBL0 + (2 * intr);
576 vector = IDT_OFFSET + intr;
577 flags = DEFAULT_EXTINT_FLAGS;
579 io_apic_write(apic, select, flags | vector);
580 io_apic_write(apic, select + 1, target);
584 #undef DEFAULT_EXTINT_FLAGS
588 * Set the trigger level for an IO APIC pin.
591 trigger(int apic, int pin, u_int32_t * flags)
596 static int intcontrol = -1;
598 switch (apic_trigger(apic, pin)) {
604 *flags &= ~IOART_TRGRLVL; /* *flags |= IOART_TRGREDG */
608 *flags |= IOART_TRGRLVL;
616 if ((id = apic_src_bus_id(apic, pin)) == -1)
619 switch (apic_bus_type(id)) {
621 *flags &= ~IOART_TRGRLVL; /* *flags |= IOART_TRGREDG; */
625 eirq = apic_src_bus_irq(apic, pin);
627 if (eirq < 0 || eirq > 15) {
628 kprintf("EISA IRQ %d?!?!\n", eirq);
632 if (intcontrol == -1) {
633 intcontrol = inb(ELCR1) << 8;
634 intcontrol |= inb(ELCR0);
635 kprintf("EISA INTCONTROL = %08x\n", intcontrol);
638 /* Use ELCR settings to determine level or edge mode */
639 level = (intcontrol >> eirq) & 1;
642 * Note that on older Neptune chipset based systems, any
643 * pci interrupts often show up here and in the ELCR as well
644 * as level sensitive interrupts attributed to the EISA bus.
648 *flags |= IOART_TRGRLVL;
650 *flags &= ~IOART_TRGRLVL;
655 *flags |= IOART_TRGRLVL;
664 panic("bad APIC IO INT flags");
669 * Set the polarity value for an IO APIC pin.
672 polarity(int apic, int pin, u_int32_t * flags, int level)
676 switch (apic_polarity(apic, pin)) {
682 *flags &= ~IOART_INTALO; /* *flags |= IOART_INTAHI */
686 *flags |= IOART_INTALO;
694 if ((id = apic_src_bus_id(apic, pin)) == -1)
697 switch (apic_bus_type(id)) {
699 *flags &= ~IOART_INTALO; /* *flags |= IOART_INTAHI */
703 /* polarity converter always gives active high */
704 *flags &= ~IOART_INTALO;
708 *flags |= IOART_INTALO;
717 panic("bad APIC IO INT flags");
722 * Print contents of apic_imen.
724 extern u_int apic_imen; /* keep apic_imen 'opaque' */
730 kprintf("SMP: enabled INTs: ");
731 for (x = 0; x < 24; ++x)
732 if ((apic_imen & (1 << x)) == 0)
734 kprintf("apic_imen: 0x%08x\n", apic_imen);
739 * Inter Processor Interrupt functions.
745 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
747 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
748 * vector is any valid SYSTEM INT vector
749 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
751 * A backlog of requests can create a deadlock between cpus. To avoid this
752 * we have to be able to accept IPIs at the same time we are trying to send
753 * them. The critical section prevents us from attempting to send additional
754 * IPIs reentrantly, but also prevents IPIQ processing so we have to call
755 * lwkt_process_ipiq() manually. It's rather messy and expensive for this
756 * to occur but fortunately it does not happen too often.
759 apic_ipi(int dest_type, int vector, int delivery_mode)
764 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
765 unsigned int eflags = read_eflags();
767 while ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
770 write_eflags(eflags);
773 icr_lo = (lapic.icr_lo & APIC_ICRLO_RESV_MASK) | dest_type |
774 delivery_mode | vector;
775 lapic.icr_lo = icr_lo;
781 single_apic_ipi(int cpu, int vector, int delivery_mode)
787 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
788 unsigned int eflags = read_eflags();
790 while ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
793 write_eflags(eflags);
795 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
796 icr_hi |= (CPU_TO_ID(cpu) << 24);
797 lapic.icr_hi = icr_hi;
800 icr_lo = (lapic.icr_lo & APIC_ICRLO_RESV_MASK)
801 | APIC_DEST_DESTFLD | delivery_mode | vector;
804 lapic.icr_lo = icr_lo;
811 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
813 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
814 * to the target, and the scheduler does not 'poll' for IPI messages.
817 single_apic_ipi_passive(int cpu, int vector, int delivery_mode)
823 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
827 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
828 icr_hi |= (CPU_TO_ID(cpu) << 24);
829 lapic.icr_hi = icr_hi;
832 icr_lo = (lapic.icr_lo & APIC_RESV2_MASK)
833 | APIC_DEST_DESTFLD | delivery_mode | vector;
836 lapic.icr_lo = icr_lo;
844 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
846 * target is a bitmask of destination cpus. Vector is any
847 * valid system INT vector. Delivery mode may be either
848 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
851 selected_apic_ipi(u_int target, int vector, int delivery_mode)
855 int n = bsfl(target);
857 single_apic_ipi(n, vector, delivery_mode);
863 * Timer code, in development...
864 * - suggested by rgrimes@gndrsh.aac.dev.com
868 * Load a 'downcount time' in uSeconds.
871 set_apic_timer(int us)
876 * When we reach here, lapic timer's frequency
877 * must have been calculated as well as the
878 * divisor (lapic.dcr_timer is setup during the
879 * divisor calculation).
881 KKASSERT(lapic_timer_freq != 0 &&
882 lapic_timer_divisor_idx >= 0);
884 count = ((us * (int64_t)lapic_timer_freq) + 999999) / 1000000;
885 lapic_timer_oneshot(count);
890 * Read remaining time in timer.
893 read_apic_timer(void)
896 /** XXX FIXME: we need to return the actual remaining time,
897 * for now we just return the remaining count.
900 return lapic.ccr_timer;
906 * Spin-style delay, set delay time in uS, spin till it drains.
911 set_apic_timer(count);
912 while (read_apic_timer())