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 void lapic_timer_process(void);
48 void lapic_timer_process_frame(struct intrframe *);
49 void lapic_timer_intr_reload(sysclock_t);
52 TUNABLE_INT("hw.lapic_timer_test", &lapic_timer_test);
55 * pointers to pmapped apic hardware.
58 volatile ioapic_t **ioapic;
60 static sysclock_t lapic_timer_freq;
61 static int lapic_timer_divisor_idx = -1;
62 static const uint32_t lapic_timer_divisors[] = {
63 APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
64 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128, APIC_TDCR_1
66 #define APIC_TIMER_NDIVISORS \
67 (int)(sizeof(lapic_timer_divisors) / sizeof(lapic_timer_divisors[0]))
71 * Enable APIC, configure interrupts.
74 apic_initialize(boolean_t bsp)
80 * setup LVT1 as ExtINT on the BSP. This is theoretically an
81 * aggregate interrupt input from the 8259. The INTA cycle
82 * will be routed to the external controller (the 8259) which
83 * is expected to supply the vector.
85 * Must be setup edge triggered, active high.
87 * Disable LVT1 on the APs. It doesn't matter what delivery
88 * mode we use because we leave it masked.
90 temp = lapic.lvt_lint0;
91 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
92 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
93 if (mycpu->gd_cpuid == 0)
94 temp |= APIC_LVT_DM_EXTINT;
96 temp |= APIC_LVT_DM_FIXED | APIC_LVT_MASKED;
97 lapic.lvt_lint0 = temp;
100 * setup LVT2 as NMI, masked till later. Edge trigger, active high.
102 temp = lapic.lvt_lint1;
103 temp &= ~(APIC_LVT_MASKED | APIC_LVT_TRIG_MASK |
104 APIC_LVT_POLARITY_MASK | APIC_LVT_DM_MASK);
105 temp |= APIC_LVT_MASKED | APIC_LVT_DM_NMI;
106 lapic.lvt_lint1 = temp;
109 * Mask the apic error interrupt, apic performance counter
112 lapic.lvt_error = lapic.lvt_error | APIC_LVT_MASKED;
113 lapic.lvt_pcint = lapic.lvt_pcint | APIC_LVT_MASKED;
115 /* Set apic timer vector and mask the apic timer interrupt. */
116 timer = lapic.lvt_timer;
117 timer &= ~APIC_LVTT_VECTOR;
118 timer |= XTIMER_OFFSET;
119 timer |= APIC_LVTT_MASKED;
120 lapic.lvt_timer = timer;
123 * Set the Task Priority Register as needed. At the moment allow
124 * interrupts on all cpus (the APs will remain CLId until they are
125 * ready to deal). We could disable all but IPIs by setting
126 * temp |= TPR_IPI_ONLY for cpu != 0.
129 temp &= ~APIC_TPR_PRIO; /* clear priority field */
132 * If we are NOT running the IO APICs, the LAPIC will only be used
133 * for IPIs. Set the TPR to prevent any unintentional interrupts.
135 temp |= TPR_IPI_ONLY;
141 * enable the local APIC
144 temp |= APIC_SVR_ENABLE; /* enable the APIC */
145 temp &= ~APIC_SVR_FOCUS_DISABLE; /* enable lopri focus processor */
148 * Set the spurious interrupt vector. The low 4 bits of the vector
151 if ((XSPURIOUSINT_OFFSET & 0x0F) != 0x0F)
152 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET);
153 temp &= ~APIC_SVR_VECTOR;
154 temp |= XSPURIOUSINT_OFFSET;
159 * Pump out a few EOIs to clean out interrupts that got through
160 * before we were able to set the TPR.
166 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
168 * Detect the presence of C1E capability mostly on latest
169 * dual-cores (or future) k8 family. This feature renders
170 * the local APIC timer dead, so we disable it by reading
171 * the Interrupt Pending Message register and clearing both
172 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
175 * "BIOS and Kernel Developer's Guide for AMD NPT
176 * Family 0Fh Processors"
177 * #32559 revision 3.00
179 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
180 (cpu_id & 0x0fff0000) >= 0x00040000) {
183 msr = rdmsr(0xc0010055);
184 if (msr & 0x18000000)
185 wrmsr(0xc0010055, msr & ~0x18000000ULL);
190 lapic_timer_calibrate();
192 lapic_timer_set_divisor(lapic_timer_divisor_idx);
195 apic_dump("apic_initialize()");
200 lapic_timer_set_divisor(int divisor_idx)
202 KKASSERT(divisor_idx >= 0 && divisor_idx < APIC_TIMER_NDIVISORS);
203 lapic.dcr_timer = lapic_timer_divisors[divisor_idx];
207 lapic_timer_oneshot(u_int count)
211 value = lapic.lvt_timer;
212 value &= ~APIC_LVTT_PERIODIC;
213 lapic.lvt_timer = value;
214 lapic.icr_timer = count;
218 lapic_timer_calibrate(void)
222 /* Try to calibrate the local APIC timer. */
223 for (lapic_timer_divisor_idx = 0;
224 lapic_timer_divisor_idx < APIC_TIMER_NDIVISORS;
225 lapic_timer_divisor_idx++) {
226 lapic_timer_set_divisor(lapic_timer_divisor_idx);
227 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
229 value = APIC_TIMER_MAX_COUNT - lapic.ccr_timer;
230 if (value != APIC_TIMER_MAX_COUNT)
233 if (lapic_timer_divisor_idx >= APIC_TIMER_NDIVISORS)
234 panic("lapic: no proper timer divisor?!\n");
235 lapic_timer_freq = value / 2;
237 kprintf("lapic: divisor index %d, frequency %u Hz\n",
238 lapic_timer_divisor_idx, lapic_timer_freq);
242 lapic_timer_process(void)
244 struct globaldata *gd = mycpu;
246 gd->gd_timer_running = 1;
248 if (lapic_timer_test)
249 kprintf("%d proc\n", gd->gd_cpuid);
253 lapic_timer_process_frame(struct intrframe *frame)
255 struct globaldata *gd = mycpu;
257 gd->gd_timer_running = 1;
259 if (lapic_timer_test)
260 kprintf("%d proc frame\n", gd->gd_cpuid);
264 lapic_timer_intr_reload(sysclock_t reload)
266 struct globaldata *gd = mycpu;
268 if (lapic_timer_test) {
269 if (gd->gd_timer_running == 2) {
271 } else if (gd->gd_timer_running == 1) {
272 gd->gd_timer_running = 2;
273 KKASSERT(lapic_timer_freq != 0);
274 lapic_timer_oneshot(lapic_timer_freq);
275 } else if (gd->gd_timer_running == 0) {
278 timer = lapic.lvt_timer;
279 timer &= ~APIC_LVTT_MASKED;
280 lapic.lvt_timer = timer;
282 gd->gd_timer_running = 2;
283 KKASSERT(lapic_timer_freq != 0);
284 lapic_timer_oneshot(lapic_timer_freq);
291 * dump contents of local APIC registers
296 kprintf("SMP: CPU%d %s:\n", mycpu->gd_cpuid, str);
297 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
298 lapic.lvt_lint0, lapic.lvt_lint1, lapic.tpr, lapic.svr);
308 #define IOAPIC_ISA_INTS 16
309 #define REDIRCNT_IOAPIC(A) \
310 ((int)((io_apic_versions[(A)] & IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) + 1)
312 static int trigger (int apic, int pin, u_int32_t * flags);
313 static void polarity (int apic, int pin, u_int32_t * flags, int level);
315 #define DEFAULT_FLAGS \
321 #define DEFAULT_ISA_FLAGS \
330 io_apic_set_id(int apic, int id)
334 ux = io_apic_read(apic, IOAPIC_ID); /* get current contents */
335 if (((ux & APIC_ID_MASK) >> 24) != id) {
336 kprintf("Changing APIC ID for IO APIC #%d"
337 " from %d to %d on chip\n",
338 apic, ((ux & APIC_ID_MASK) >> 24), id);
339 ux &= ~APIC_ID_MASK; /* clear the ID field */
341 io_apic_write(apic, IOAPIC_ID, ux); /* write new value */
342 ux = io_apic_read(apic, IOAPIC_ID); /* re-read && test */
343 if (((ux & APIC_ID_MASK) >> 24) != id)
344 panic("can't control IO APIC #%d ID, reg: 0x%08x",
351 io_apic_get_id(int apic)
353 return (io_apic_read(apic, IOAPIC_ID) & APIC_ID_MASK) >> 24;
362 extern int apic_pin_trigger; /* 'opaque' */
365 io_apic_setup_intpin(int apic, int pin)
367 int bus, bustype, irq;
368 u_char select; /* the select register is 8 bits */
369 u_int32_t flags; /* the window register is 32 bits */
370 u_int32_t target; /* the window register is 32 bits */
371 u_int32_t vector; /* the window register is 32 bits */
374 select = pin * 2 + IOAPIC_REDTBL0; /* register */
377 * Always clear an IO APIC pin before [re]programming it. This is
378 * particularly important if the pin is set up for a level interrupt
379 * as the IOART_REM_IRR bit might be set. When we reprogram the
380 * vector any EOI from pending ints on this pin could be lost and
381 * IRR might never get reset.
383 * To fix this problem, clear the vector and make sure it is
384 * programmed as an edge interrupt. This should theoretically
385 * clear IRR so we can later, safely program it as a level
390 flags = io_apic_read(apic, select) & IOART_RESV;
391 flags |= IOART_INTMSET | IOART_TRGREDG | IOART_INTAHI;
392 flags |= IOART_DESTPHY | IOART_DELFIXED;
394 target = io_apic_read(apic, select + 1) & IOART_HI_DEST_RESV;
395 target |= 0; /* fixed mode cpu mask of 0 - don't deliver anywhere */
399 io_apic_write(apic, select, flags | vector);
400 io_apic_write(apic, select + 1, target);
405 * We only deal with vectored interrupts here. ? documentation is
406 * lacking, I'm guessing an interrupt type of 0 is the 'INT' type,
409 * This test also catches unconfigured pins.
411 if (apic_int_type(apic, pin) != 0)
415 * Leave the pin unprogrammed if it does not correspond to
418 irq = apic_irq(apic, pin);
422 /* determine the bus type for this pin */
423 bus = apic_src_bus_id(apic, pin);
426 bustype = apic_bus_type(bus);
428 if ((bustype == ISA) &&
429 (pin < IOAPIC_ISA_INTS) &&
431 (apic_polarity(apic, pin) == 0x1) &&
432 (apic_trigger(apic, pin) == 0x3)) {
434 * A broken BIOS might describe some ISA
435 * interrupts as active-high level-triggered.
436 * Use default ISA flags for those interrupts.
438 flags = DEFAULT_ISA_FLAGS;
441 * Program polarity and trigger mode according to
444 flags = DEFAULT_FLAGS;
445 level = trigger(apic, pin, &flags);
447 apic_pin_trigger |= (1 << irq);
448 polarity(apic, pin, &flags, level);
452 kprintf("IOAPIC #%d intpin %d -> irq %d\n",
457 * Program the appropriate registers. This routing may be
458 * overridden when an interrupt handler for a device is
459 * actually added (see register_int(), which calls through
460 * the MACHINTR ABI to set up an interrupt handler/vector).
462 * The order in which we must program the two registers for
463 * safety is unclear! XXX
467 vector = IDT_OFFSET + irq; /* IDT vec */
468 target = io_apic_read(apic, select + 1) & IOART_HI_DEST_RESV;
469 target |= IOART_HI_DEST_BROADCAST;
470 flags |= io_apic_read(apic, select) & IOART_RESV;
471 io_apic_write(apic, select, flags | vector);
472 io_apic_write(apic, select + 1, target);
478 io_apic_setup(int apic)
484 apic_pin_trigger = 0; /* default to edge-triggered */
486 maxpin = REDIRCNT_IOAPIC(apic); /* pins in APIC */
487 kprintf("Programming %d pins in IOAPIC #%d\n", maxpin, apic);
489 for (pin = 0; pin < maxpin; ++pin) {
490 io_apic_setup_intpin(apic, pin);
493 if (apic_int_type(apic, pin) >= 0) {
494 kprintf("Warning: IOAPIC #%d pin %d does not exist,"
495 " cannot program!\n", apic, pin);
500 /* return GOOD status */
503 #undef DEFAULT_ISA_FLAGS
507 #define DEFAULT_EXTINT_FLAGS \
516 * Setup the source of External INTerrupts.
519 ext_int_setup(int apic, int intr)
521 u_char select; /* the select register is 8 bits */
522 u_int32_t flags; /* the window register is 32 bits */
523 u_int32_t target; /* the window register is 32 bits */
524 u_int32_t vector; /* the window register is 32 bits */
526 if (apic_int_type(apic, intr) != 3)
529 target = IOART_HI_DEST_BROADCAST;
530 select = IOAPIC_REDTBL0 + (2 * intr);
531 vector = IDT_OFFSET + intr;
532 flags = DEFAULT_EXTINT_FLAGS;
534 io_apic_write(apic, select, flags | vector);
535 io_apic_write(apic, select + 1, target);
539 #undef DEFAULT_EXTINT_FLAGS
543 * Set the trigger level for an IO APIC pin.
546 trigger(int apic, int pin, u_int32_t * flags)
551 static int intcontrol = -1;
553 switch (apic_trigger(apic, pin)) {
559 *flags &= ~IOART_TRGRLVL; /* *flags |= IOART_TRGREDG */
563 *flags |= IOART_TRGRLVL;
571 if ((id = apic_src_bus_id(apic, pin)) == -1)
574 switch (apic_bus_type(id)) {
576 *flags &= ~IOART_TRGRLVL; /* *flags |= IOART_TRGREDG; */
580 eirq = apic_src_bus_irq(apic, pin);
582 if (eirq < 0 || eirq > 15) {
583 kprintf("EISA IRQ %d?!?!\n", eirq);
587 if (intcontrol == -1) {
588 intcontrol = inb(ELCR1) << 8;
589 intcontrol |= inb(ELCR0);
590 kprintf("EISA INTCONTROL = %08x\n", intcontrol);
593 /* Use ELCR settings to determine level or edge mode */
594 level = (intcontrol >> eirq) & 1;
597 * Note that on older Neptune chipset based systems, any
598 * pci interrupts often show up here and in the ELCR as well
599 * as level sensitive interrupts attributed to the EISA bus.
603 *flags |= IOART_TRGRLVL;
605 *flags &= ~IOART_TRGRLVL;
610 *flags |= IOART_TRGRLVL;
619 panic("bad APIC IO INT flags");
624 * Set the polarity value for an IO APIC pin.
627 polarity(int apic, int pin, u_int32_t * flags, int level)
631 switch (apic_polarity(apic, pin)) {
637 *flags &= ~IOART_INTALO; /* *flags |= IOART_INTAHI */
641 *flags |= IOART_INTALO;
649 if ((id = apic_src_bus_id(apic, pin)) == -1)
652 switch (apic_bus_type(id)) {
654 *flags &= ~IOART_INTALO; /* *flags |= IOART_INTAHI */
658 /* polarity converter always gives active high */
659 *flags &= ~IOART_INTALO;
663 *flags |= IOART_INTALO;
672 panic("bad APIC IO INT flags");
677 * Print contents of apic_imen.
679 extern u_int apic_imen; /* keep apic_imen 'opaque' */
685 kprintf("SMP: enabled INTs: ");
686 for (x = 0; x < 24; ++x)
687 if ((apic_imen & (1 << x)) == 0)
689 kprintf("apic_imen: 0x%08x\n", apic_imen);
694 * Inter Processor Interrupt functions.
700 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
702 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
703 * vector is any valid SYSTEM INT vector
704 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
706 * A backlog of requests can create a deadlock between cpus. To avoid this
707 * we have to be able to accept IPIs at the same time we are trying to send
708 * them. The critical section prevents us from attempting to send additional
709 * IPIs reentrantly, but also prevents IPIQ processing so we have to call
710 * lwkt_process_ipiq() manually. It's rather messy and expensive for this
711 * to occur but fortunately it does not happen too often.
714 apic_ipi(int dest_type, int vector, int delivery_mode)
719 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
720 unsigned int eflags = read_eflags();
722 while ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
725 write_eflags(eflags);
728 icr_lo = (lapic.icr_lo & APIC_ICRLO_RESV_MASK) | dest_type |
729 delivery_mode | vector;
730 lapic.icr_lo = icr_lo;
736 single_apic_ipi(int cpu, int vector, int delivery_mode)
742 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
743 unsigned int eflags = read_eflags();
745 while ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
748 write_eflags(eflags);
750 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
751 icr_hi |= (CPU_TO_ID(cpu) << 24);
752 lapic.icr_hi = icr_hi;
755 icr_lo = (lapic.icr_lo & APIC_ICRLO_RESV_MASK)
756 | APIC_DEST_DESTFLD | delivery_mode | vector;
759 lapic.icr_lo = icr_lo;
766 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
768 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
769 * to the target, and the scheduler does not 'poll' for IPI messages.
772 single_apic_ipi_passive(int cpu, int vector, int delivery_mode)
778 if ((lapic.icr_lo & APIC_DELSTAT_MASK) != 0) {
782 icr_hi = lapic.icr_hi & ~APIC_ID_MASK;
783 icr_hi |= (CPU_TO_ID(cpu) << 24);
784 lapic.icr_hi = icr_hi;
787 icr_lo = (lapic.icr_lo & APIC_RESV2_MASK)
788 | APIC_DEST_DESTFLD | delivery_mode | vector;
791 lapic.icr_lo = icr_lo;
799 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
801 * target is a bitmask of destination cpus. Vector is any
802 * valid system INT vector. Delivery mode may be either
803 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
806 selected_apic_ipi(u_int target, int vector, int delivery_mode)
810 int n = bsfl(target);
812 single_apic_ipi(n, vector, delivery_mode);
818 * Timer code, in development...
819 * - suggested by rgrimes@gndrsh.aac.dev.com
823 * Load a 'downcount time' in uSeconds.
826 set_apic_timer(int us)
831 * When we reach here, lapic timer's frequency
832 * must have been calculated as well as the
833 * divisor (lapic.dcr_timer is setup during the
834 * divisor calculation).
836 KKASSERT(lapic_timer_freq != 0 &&
837 lapic_timer_divisor_idx >= 0);
839 count = ((us * (int64_t)lapic_timer_freq) + 999999) / 1000000;
840 lapic_timer_oneshot(count);
845 * Read remaining time in timer.
848 read_apic_timer(void)
851 /** XXX FIXME: we need to return the actual remaining time,
852 * for now we just return the remaining count.
855 return lapic.ccr_timer;
861 * Spin-style delay, set delay time in uS, spin till it drains.
866 set_apic_timer(count);
867 while (read_apic_timer())