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 <machine/globaldata.h>
32 #include <machine/smp.h>
33 #include <machine/md_var.h>
34 #include <machine_base/apic/mpapic.h>
35 #include <machine/segments.h>
36 #include <sys/thread2.h>
38 #include <machine_base/isa/intr_machdep.h> /* Xspuriousint() */
40 /* EISA Edge/Level trigger control registers */
41 #define ELCR0 0x4d0 /* eisa irq 0-7 */
42 #define ELCR1 0x4d1 /* eisa irq 8-15 */
44 static void lapic_timer_calibrate(void);
45 static void lapic_timer_set_divisor(int);
46 void lapic_timer_process(void);
47 void lapic_timer_process_frame(struct intrframe *);
50 * pointers to pmapped apic hardware.
53 volatile ioapic_t **ioapic;
55 static sysclock_t lapic_timer_freq;
56 static int lapic_timer_divisor_idx = -1;
57 static const uint32_t lapic_timer_divisors[] = {
58 APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
59 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128, APIC_TDCR_1
61 #define APIC_TIMER_NDIVISORS \
62 (int)(sizeof(lapic_timer_divisors) / sizeof(lapic_timer_divisors[0]))
66 * Enable APIC, configure interrupts.
69 apic_initialize(boolean_t bsp)
75 * setup LVT1 as ExtINT on the BSP. This is theoretically an
76 * aggregate interrupt input from the 8259. The INTA cycle
77 * will be routed to the external controller (the 8259) which
78 * is expected to supply the vector.
80 * Must be setup edge triggered, active high.
82 * Disable LVT1 on the APs. It doesn't matter what delivery
83 * mode we use because we leave it masked.
85 temp = lapic.lvt_lint0;
86 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
87 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
88 if (mycpu->gd_cpuid == 0)
89 temp |= APIC_LVT_DM_EXTINT;
91 temp |= APIC_LVT_DM_FIXED | APIC_LVT_MASKED;
92 lapic.lvt_lint0 = temp;
95 * setup LVT2 as NMI, masked till later. Edge trigger, active high.
97 temp = lapic.lvt_lint1;
98 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
99 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
100 temp |= APIC_LVT_MASKED | APIC_LVT_DM_NMI;
101 lapic.lvt_lint1 = temp;
104 * Mask the apic error interrupt, apic performance counter
107 lapic.lvt_error = lapic.lvt_error | APIC_LVT_MASKED;
108 lapic.lvt_pcint = lapic.lvt_pcint | APIC_LVT_MASKED;
110 /* Set apic timer vector and mask the apic timer interrupt. */
111 timer = lapic.lvt_timer;
112 timer &= ~APIC_LVTT_VECTOR;
113 timer |= XTIMER_OFFSET;
114 timer |= APIC_LVTT_MASKED;
115 lapic.lvt_timer = timer;
118 * Set the Task Priority Register as needed. At the moment allow
119 * interrupts on all cpus (the APs will remain CLId until they are
120 * ready to deal). We could disable all but IPIs by setting
121 * temp |= TPR_IPI_ONLY for cpu != 0.
124 temp &= ~APIC_TPR_PRIO; /* clear priority field */
127 * If we are NOT running the IO APICs, the LAPIC will only be used
128 * for IPIs. Set the TPR to prevent any unintentional interrupts.
130 temp |= TPR_IPI_ONLY;
136 * enable the local APIC
139 temp |= APIC_SVR_ENABLE; /* enable the APIC */
140 temp &= ~APIC_SVR_FOCUS_DISABLE; /* enable lopri focus processor */
143 * Set the spurious interrupt vector. The low 4 bits of the vector
146 if ((XSPURIOUSINT_OFFSET & 0x0F) != 0x0F)
147 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET);
148 temp &= ~APIC_SVR_VECTOR;
149 temp |= XSPURIOUSINT_OFFSET;
154 * Pump out a few EOIs to clean out interrupts that got through
155 * before we were able to set the TPR.
161 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
163 * Detect the presence of C1E capability mostly on latest
164 * dual-cores (or future) k8 family. This feature renders
165 * the local APIC timer dead, so we disable it by reading
166 * the Interrupt Pending Message register and clearing both
167 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
170 * "BIOS and Kernel Developer's Guide for AMD NPT
171 * Family 0Fh Processors"
172 * #32559 revision 3.00
174 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
175 (cpu_id & 0x0fff0000) >= 0x00040000) {
178 msr = rdmsr(0xc0010055);
179 if (msr & 0x18000000)
180 wrmsr(0xc0010055, msr & ~0x18000000ULL);
185 lapic_timer_calibrate();
187 lapic_timer_set_divisor(lapic_timer_divisor_idx);
190 apic_dump("apic_initialize()");
195 lapic_timer_set_divisor(int divisor_idx)
197 KKASSERT(divisor_idx >= 0 && divisor_idx < APIC_TIMER_NDIVISORS);
198 lapic.dcr_timer = lapic_timer_divisors[divisor_idx];
202 lapic_timer_oneshot(u_int count)
206 value = lapic.lvt_timer;
207 value &= ~APIC_LVTT_PERIODIC;
208 lapic.lvt_timer = value;
209 lapic.icr_timer = count;
213 lapic_timer_calibrate(void)
217 /* Try to calibrate the local APIC timer. */
218 for (lapic_timer_divisor_idx = 0;
219 lapic_timer_divisor_idx < APIC_TIMER_NDIVISORS;
220 lapic_timer_divisor_idx++) {
221 lapic_timer_set_divisor(lapic_timer_divisor_idx);
222 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
224 value = APIC_TIMER_MAX_COUNT - lapic.ccr_timer;
225 if (value != APIC_TIMER_MAX_COUNT)
228 if (lapic_timer_divisor_idx >= APIC_TIMER_NDIVISORS)
229 panic("lapic: no proper timer divisor?!\n");
230 lapic_timer_freq = value / 2;
232 kprintf("lapic: divisor index %d, frequency %u Hz\n",
233 lapic_timer_divisor_idx, lapic_timer_freq);
237 lapic_timer_process(void)
242 lapic_timer_process_frame(struct intrframe *frame)
248 * dump contents of local APIC registers
253 kprintf("SMP: CPU%d %s:\n", mycpu->gd_cpuid, str);
254 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
255 lapic.lvt_lint0, lapic.lvt_lint1, lapic.tpr, lapic.svr);
265 #define IOAPIC_ISA_INTS 16
266 #define REDIRCNT_IOAPIC(A) \
267 ((int)((io_apic_versions[(A)] & IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) + 1)
269 static int trigger (int apic, int pin, u_int32_t * flags);
270 static void polarity (int apic, int pin, u_int32_t * flags, int level);
272 #define DEFAULT_FLAGS \
278 #define DEFAULT_ISA_FLAGS \
287 io_apic_set_id(int apic, int id)
291 ux = io_apic_read(apic, IOAPIC_ID); /* get current contents */
292 if (((ux & APIC_ID_MASK) >> 24) != id) {
293 kprintf("Changing APIC ID for IO APIC #%d"
294 " from %d to %d on chip\n",
295 apic, ((ux & APIC_ID_MASK) >> 24), id);
296 ux &= ~APIC_ID_MASK; /* clear the ID field */
298 io_apic_write(apic, IOAPIC_ID, ux); /* write new value */
299 ux = io_apic_read(apic, IOAPIC_ID); /* re-read && test */
300 if (((ux & APIC_ID_MASK) >> 24) != id)
301 panic("can't control IO APIC #%d ID, reg: 0x%08x",
308 io_apic_get_id(int apic)
310 return (io_apic_read(apic, IOAPIC_ID) & APIC_ID_MASK) >> 24;
319 extern int apic_pin_trigger; /* 'opaque' */
322 io_apic_setup_intpin(int apic, int pin)
324 int bus, bustype, irq;
325 u_char select; /* the select register is 8 bits */
326 u_int32_t flags; /* the window register is 32 bits */
327 u_int32_t target; /* the window register is 32 bits */
328 u_int32_t vector; /* the window register is 32 bits */
331 select = pin * 2 + IOAPIC_REDTBL0; /* register */
334 * Always clear an IO APIC pin before [re]programming it. This is
335 * particularly important if the pin is set up for a level interrupt
336 * as the IOART_REM_IRR bit might be set. When we reprogram the
337 * vector any EOI from pending ints on this pin could be lost and
338 * IRR might never get reset.
340 * To fix this problem, clear the vector and make sure it is
341 * programmed as an edge interrupt. This should theoretically
342 * clear IRR so we can later, safely program it as a level
347 flags = io_apic_read(apic, select) & IOART_RESV;
348 flags |= IOART_INTMSET | IOART_TRGREDG | IOART_INTAHI;
349 flags |= IOART_DESTPHY | IOART_DELFIXED;
351 target = io_apic_read(apic, select + 1) & IOART_HI_DEST_RESV;
352 target |= 0; /* fixed mode cpu mask of 0 - don't deliver anywhere */
356 io_apic_write(apic, select, flags | vector);
357 io_apic_write(apic, select + 1, target);
362 * We only deal with vectored interrupts here. ? documentation is
363 * lacking, I'm guessing an interrupt type of 0 is the 'INT' type,
366 * This test also catches unconfigured pins.
368 if (apic_int_type(apic, pin) != 0)
372 * Leave the pin unprogrammed if it does not correspond to
375 irq = apic_irq(apic, pin);
379 /* determine the bus type for this pin */
380 bus = apic_src_bus_id(apic, pin);
383 bustype = apic_bus_type(bus);
385 if ((bustype == ISA) &&
386 (pin < IOAPIC_ISA_INTS) &&
388 (apic_polarity(apic, pin) == 0x1) &&
389 (apic_trigger(apic, pin) == 0x3)) {
391 * A broken BIOS might describe some ISA
392 * interrupts as active-high level-triggered.
393 * Use default ISA flags for those interrupts.
395 flags = DEFAULT_ISA_FLAGS;
398 * Program polarity and trigger mode according to
401 flags = DEFAULT_FLAGS;
402 level = trigger(apic, pin, &flags);
404 apic_pin_trigger |= (1 << irq);
405 polarity(apic, pin, &flags, level);
409 kprintf("IOAPIC #%d intpin %d -> irq %d\n",
414 * Program the appropriate registers. This routing may be
415 * overridden when an interrupt handler for a device is
416 * actually added (see register_int(), which calls through
417 * the MACHINTR ABI to set up an interrupt handler/vector).
419 * The order in which we must program the two registers for
420 * safety is unclear! XXX
424 vector = IDT_OFFSET + irq; /* IDT vec */
425 target = io_apic_read(apic, select + 1) & IOART_HI_DEST_RESV;
426 target |= IOART_HI_DEST_BROADCAST;
427 flags |= io_apic_read(apic, select) & IOART_RESV;
428 io_apic_write(apic, select, flags | vector);
429 io_apic_write(apic, select + 1, target);
435 io_apic_setup(int apic)
441 apic_pin_trigger = 0; /* default to edge-triggered */
443 maxpin = REDIRCNT_IOAPIC(apic); /* pins in APIC */
444 kprintf("Programming %d pins in IOAPIC #%d\n", maxpin, apic);
446 for (pin = 0; pin < maxpin; ++pin) {
447 io_apic_setup_intpin(apic, pin);
450 if (apic_int_type(apic, pin) >= 0) {
451 kprintf("Warning: IOAPIC #%d pin %d does not exist,"
452 " cannot program!\n", apic, pin);
457 /* return GOOD status */
460 #undef DEFAULT_ISA_FLAGS
464 #define DEFAULT_EXTINT_FLAGS \
473 * Setup the source of External INTerrupts.
476 ext_int_setup(int apic, int intr)
478 u_char select; /* the select register is 8 bits */
479 u_int32_t flags; /* the window register is 32 bits */
480 u_int32_t target; /* the window register is 32 bits */
481 u_int32_t vector; /* the window register is 32 bits */
483 if (apic_int_type(apic, intr) != 3)
486 target = IOART_HI_DEST_BROADCAST;
487 select = IOAPIC_REDTBL0 + (2 * intr);
488 vector = IDT_OFFSET + intr;
489 flags = DEFAULT_EXTINT_FLAGS;
491 io_apic_write(apic, select, flags | vector);
492 io_apic_write(apic, select + 1, target);
496 #undef DEFAULT_EXTINT_FLAGS
500 * Set the trigger level for an IO APIC pin.
503 trigger(int apic, int pin, u_int32_t * flags)
508 static int intcontrol = -1;
510 switch (apic_trigger(apic, pin)) {
516 *flags &= ~IOART_TRGRLVL; /* *flags |= IOART_TRGREDG */
520 *flags |= IOART_TRGRLVL;
528 if ((id = apic_src_bus_id(apic, pin)) == -1)
531 switch (apic_bus_type(id)) {
533 *flags &= ~IOART_TRGRLVL; /* *flags |= IOART_TRGREDG; */
537 eirq = apic_src_bus_irq(apic, pin);
539 if (eirq < 0 || eirq > 15) {
540 kprintf("EISA IRQ %d?!?!\n", eirq);
544 if (intcontrol == -1) {
545 intcontrol = inb(ELCR1) << 8;
546 intcontrol |= inb(ELCR0);
547 kprintf("EISA INTCONTROL = %08x\n", intcontrol);
550 /* Use ELCR settings to determine level or edge mode */
551 level = (intcontrol >> eirq) & 1;
554 * Note that on older Neptune chipset based systems, any
555 * pci interrupts often show up here and in the ELCR as well
556 * as level sensitive interrupts attributed to the EISA bus.
560 *flags |= IOART_TRGRLVL;
562 *flags &= ~IOART_TRGRLVL;
567 *flags |= IOART_TRGRLVL;
576 panic("bad APIC IO INT flags");
581 * Set the polarity value for an IO APIC pin.
584 polarity(int apic, int pin, u_int32_t * flags, int level)
588 switch (apic_polarity(apic, pin)) {
594 *flags &= ~IOART_INTALO; /* *flags |= IOART_INTAHI */
598 *flags |= IOART_INTALO;
606 if ((id = apic_src_bus_id(apic, pin)) == -1)
609 switch (apic_bus_type(id)) {
611 *flags &= ~IOART_INTALO; /* *flags |= IOART_INTAHI */
615 /* polarity converter always gives active high */
616 *flags &= ~IOART_INTALO;
620 *flags |= IOART_INTALO;
629 panic("bad APIC IO INT flags");
634 * Print contents of apic_imen.
636 extern u_int apic_imen; /* keep apic_imen 'opaque' */
642 kprintf("SMP: enabled INTs: ");
643 for (x = 0; x < 24; ++x)
644 if ((apic_imen & (1 << x)) == 0)
646 kprintf("apic_imen: 0x%08x\n", apic_imen);
651 * Inter Processor Interrupt functions.
657 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
659 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
660 * vector is any valid SYSTEM INT vector
661 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
663 * A backlog of requests can create a deadlock between cpus. To avoid this
664 * we have to be able to accept IPIs at the same time we are trying to send
665 * them. The critical section prevents us from attempting to send additional
666 * IPIs reentrantly, but also prevents IPIQ processing so we have to call
667 * lwkt_process_ipiq() manually. It's rather messy and expensive for this
668 * to occur but fortunately it does not happen too often.
671 apic_ipi(int dest_type, int vector, int delivery_mode)
676 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
677 unsigned int eflags = read_eflags();
679 while ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
682 write_eflags(eflags);
685 icr_lo = (lapic.icr_lo & APIC_ICRLO_RESV_MASK) | dest_type |
686 delivery_mode | vector;
687 lapic.icr_lo = icr_lo;
693 single_apic_ipi(int cpu, int vector, int delivery_mode)
699 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
700 unsigned int eflags = read_eflags();
702 while ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
705 write_eflags(eflags);
707 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
708 icr_hi |= (CPU_TO_ID(cpu) << 24);
709 lapic.icr_hi = icr_hi;
712 icr_lo = (lapic.icr_lo & APIC_ICRLO_RESV_MASK)
713 | APIC_DEST_DESTFLD | delivery_mode | vector;
716 lapic.icr_lo = icr_lo;
723 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
725 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
726 * to the target, and the scheduler does not 'poll' for IPI messages.
729 single_apic_ipi_passive(int cpu, int vector, int delivery_mode)
735 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
739 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
740 icr_hi |= (CPU_TO_ID(cpu) << 24);
741 lapic.icr_hi = icr_hi;
744 icr_lo = (lapic.icr_lo & APIC_RESV2_MASK)
745 | APIC_DEST_DESTFLD | delivery_mode | vector;
748 lapic.icr_lo = icr_lo;
756 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
758 * target is a bitmask of destination cpus. Vector is any
759 * valid system INT vector. Delivery mode may be either
760 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
763 selected_apic_ipi(u_int target, int vector, int delivery_mode)
767 int n = bsfl(target);
769 single_apic_ipi(n, vector, delivery_mode);
775 * Timer code, in development...
776 * - suggested by rgrimes@gndrsh.aac.dev.com
780 * Load a 'downcount time' in uSeconds.
783 set_apic_timer(int us)
788 * When we reach here, lapic timer's frequency
789 * must have been calculated as well as the
790 * divisor (lapic.dcr_timer is setup during the
791 * divisor calculation).
793 KKASSERT(lapic_timer_freq != 0 &&
794 lapic_timer_divisor_idx >= 0);
796 count = ((us * (int64_t)lapic_timer_freq) + 999999) / 1000000;
797 lapic_timer_oneshot(count);
802 * Read remaining time in timer.
805 read_apic_timer(void)
808 /** XXX FIXME: we need to return the actual remaining time,
809 * for now we just return the remaining count.
812 return lapic.ccr_timer;
818 * Spin-style delay, set delay time in uS, spin till it drains.
823 set_apic_timer(count);
824 while (read_apic_timer())