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_test(void);
50 void lapic_timer_oneshot_intr_enable(void);
53 TUNABLE_INT("hw.lapic_timer_test", &lapic_timer_test
);
56 * pointers to pmapped apic hardware.
59 volatile ioapic_t
**ioapic
;
61 static sysclock_t lapic_timer_freq
;
62 static int lapic_timer_divisor_idx
= -1;
63 static const uint32_t lapic_timer_divisors
[] = {
64 APIC_TDCR_2
, APIC_TDCR_4
, APIC_TDCR_8
, APIC_TDCR_16
,
65 APIC_TDCR_32
, APIC_TDCR_64
, APIC_TDCR_128
, APIC_TDCR_1
67 #define APIC_TIMER_NDIVISORS \
68 (int)(sizeof(lapic_timer_divisors) / sizeof(lapic_timer_divisors[0]))
72 * Enable APIC, configure interrupts.
75 apic_initialize(boolean_t bsp
)
81 * setup LVT1 as ExtINT on the BSP. This is theoretically an
82 * aggregate interrupt input from the 8259. The INTA cycle
83 * will be routed to the external controller (the 8259) which
84 * is expected to supply the vector.
86 * Must be setup edge triggered, active high.
88 * Disable LVT1 on the APs. It doesn't matter what delivery
89 * mode we use because we leave it masked.
91 temp
= lapic
.lvt_lint0
;
92 temp
&= ~(APIC_LVT_MASKED
| APIC_LVT_TRIG_MASK
|
93 APIC_LVT_POLARITY_MASK
| APIC_LVT_DM_MASK
);
94 if (mycpu
->gd_cpuid
== 0)
95 temp
|= APIC_LVT_DM_EXTINT
;
97 temp
|= APIC_LVT_DM_FIXED
| APIC_LVT_MASKED
;
98 lapic
.lvt_lint0
= temp
;
101 * setup LVT2 as NMI, masked till later. Edge trigger, active high.
103 temp
= lapic
.lvt_lint1
;
104 temp
&= ~(APIC_LVT_MASKED
| APIC_LVT_TRIG_MASK
|
105 APIC_LVT_POLARITY_MASK
| APIC_LVT_DM_MASK
);
106 temp
|= APIC_LVT_MASKED
| APIC_LVT_DM_NMI
;
107 lapic
.lvt_lint1
= temp
;
110 * Mask the apic error interrupt, apic performance counter
113 lapic
.lvt_error
= lapic
.lvt_error
| APIC_LVT_MASKED
;
114 lapic
.lvt_pcint
= lapic
.lvt_pcint
| APIC_LVT_MASKED
;
116 /* Set apic timer vector and mask the apic timer interrupt. */
117 timer
= lapic
.lvt_timer
;
118 timer
&= ~APIC_LVTT_VECTOR
;
119 timer
|= XTIMER_OFFSET
;
120 timer
|= APIC_LVTT_MASKED
;
121 lapic
.lvt_timer
= timer
;
124 * Set the Task Priority Register as needed. At the moment allow
125 * interrupts on all cpus (the APs will remain CLId until they are
126 * ready to deal). We could disable all but IPIs by setting
127 * temp |= TPR_IPI_ONLY for cpu != 0.
130 temp
&= ~APIC_TPR_PRIO
; /* clear priority field */
133 * If we are NOT running the IO APICs, the LAPIC will only be used
134 * for IPIs. Set the TPR to prevent any unintentional interrupts.
136 temp
|= TPR_IPI_ONLY
;
142 * enable the local APIC
145 temp
|= APIC_SVR_ENABLE
; /* enable the APIC */
146 temp
&= ~APIC_SVR_FOCUS_DISABLE
; /* enable lopri focus processor */
149 * Set the spurious interrupt vector. The low 4 bits of the vector
152 if ((XSPURIOUSINT_OFFSET
& 0x0F) != 0x0F)
153 panic("bad XSPURIOUSINT_OFFSET: 0x%08x", XSPURIOUSINT_OFFSET
);
154 temp
&= ~APIC_SVR_VECTOR
;
155 temp
|= XSPURIOUSINT_OFFSET
;
160 * Pump out a few EOIs to clean out interrupts that got through
161 * before we were able to set the TPR.
167 if (strcmp(cpu_vendor
, "AuthenticAMD") == 0) {
169 * Detect the presence of C1E capability mostly on latest
170 * dual-cores (or future) k8 family. This feature renders
171 * the local APIC timer dead, so we disable it by reading
172 * the Interrupt Pending Message register and clearing both
173 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
176 * "BIOS and Kernel Developer's Guide for AMD NPT
177 * Family 0Fh Processors"
178 * #32559 revision 3.00
180 if ((cpu_id
& 0x00000f00) == 0x00000f00 &&
181 (cpu_id
& 0x0fff0000) >= 0x00040000) {
184 msr
= rdmsr(0xc0010055);
185 if (msr
& 0x18000000)
186 wrmsr(0xc0010055, msr
& ~0x18000000ULL
);
191 lapic_timer_calibrate();
193 lapic_timer_set_divisor(lapic_timer_divisor_idx
);
196 apic_dump("apic_initialize()");
201 lapic_timer_set_divisor(int divisor_idx
)
203 KKASSERT(divisor_idx
>= 0 && divisor_idx
< APIC_TIMER_NDIVISORS
);
204 lapic
.dcr_timer
= lapic_timer_divisors
[divisor_idx
];
208 lapic_timer_oneshot(u_int count
)
212 value
= lapic
.lvt_timer
;
213 value
&= ~APIC_LVTT_PERIODIC
;
214 lapic
.lvt_timer
= value
;
215 lapic
.icr_timer
= count
;
219 lapic_timer_oneshot_quick(u_int count
)
221 lapic
.icr_timer
= count
;
225 lapic_timer_calibrate(void)
229 /* Try to calibrate the local APIC timer. */
230 for (lapic_timer_divisor_idx
= 0;
231 lapic_timer_divisor_idx
< APIC_TIMER_NDIVISORS
;
232 lapic_timer_divisor_idx
++) {
233 lapic_timer_set_divisor(lapic_timer_divisor_idx
);
234 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT
);
236 value
= APIC_TIMER_MAX_COUNT
- lapic
.ccr_timer
;
237 if (value
!= APIC_TIMER_MAX_COUNT
)
240 if (lapic_timer_divisor_idx
>= APIC_TIMER_NDIVISORS
)
241 panic("lapic: no proper timer divisor?!\n");
242 lapic_timer_freq
= value
/ 2;
244 kprintf("lapic: divisor index %d, frequency %u Hz\n",
245 lapic_timer_divisor_idx
, lapic_timer_freq
);
249 lapic_timer_process(void)
251 struct globaldata
*gd
= mycpu
;
253 gd
->gd_timer_running
= 0;
255 if (lapic_timer_test
)
256 kprintf("%d proc\n", gd
->gd_cpuid
);
260 lapic_timer_process_frame(struct intrframe
*frame
)
262 struct globaldata
*gd
= mycpu
;
264 gd
->gd_timer_running
= 0;
266 if (lapic_timer_test
)
267 kprintf("%d proc frame\n", gd
->gd_cpuid
);
271 lapic_timer_intr_test(void)
273 struct globaldata
*gd
= mycpu
;
275 if (!gd
->gd_timer_running
) {
276 gd
->gd_timer_running
= 1;
277 KKASSERT(lapic_timer_freq
!= 0);
278 lapic_timer_oneshot_quick(lapic_timer_freq
);
283 lapic_timer_oneshot_intr_enable(void)
287 timer
= lapic
.lvt_timer
;
288 timer
&= ~(APIC_LVTT_MASKED
| APIC_LVTT_PERIODIC
);
289 lapic
.lvt_timer
= timer
;
294 * dump contents of local APIC registers
299 kprintf("SMP: CPU%d %s:\n", mycpu
->gd_cpuid
, str
);
300 kprintf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
301 lapic
.lvt_lint0
, lapic
.lvt_lint1
, lapic
.tpr
, lapic
.svr
);
311 #define IOAPIC_ISA_INTS 16
312 #define REDIRCNT_IOAPIC(A) \
313 ((int)((io_apic_versions[(A)] & IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) + 1)
315 static int trigger (int apic
, int pin
, u_int32_t
* flags
);
316 static void polarity (int apic
, int pin
, u_int32_t
* flags
, int level
);
318 #define DEFAULT_FLAGS \
324 #define DEFAULT_ISA_FLAGS \
333 io_apic_set_id(int apic
, int id
)
337 ux
= io_apic_read(apic
, IOAPIC_ID
); /* get current contents */
338 if (((ux
& APIC_ID_MASK
) >> 24) != id
) {
339 kprintf("Changing APIC ID for IO APIC #%d"
340 " from %d to %d on chip\n",
341 apic
, ((ux
& APIC_ID_MASK
) >> 24), id
);
342 ux
&= ~APIC_ID_MASK
; /* clear the ID field */
344 io_apic_write(apic
, IOAPIC_ID
, ux
); /* write new value */
345 ux
= io_apic_read(apic
, IOAPIC_ID
); /* re-read && test */
346 if (((ux
& APIC_ID_MASK
) >> 24) != id
)
347 panic("can't control IO APIC #%d ID, reg: 0x%08x",
354 io_apic_get_id(int apic
)
356 return (io_apic_read(apic
, IOAPIC_ID
) & APIC_ID_MASK
) >> 24;
365 extern int apic_pin_trigger
; /* 'opaque' */
368 io_apic_setup_intpin(int apic
, int pin
)
370 int bus
, bustype
, irq
;
371 u_char select
; /* the select register is 8 bits */
372 u_int32_t flags
; /* the window register is 32 bits */
373 u_int32_t target
; /* the window register is 32 bits */
374 u_int32_t vector
; /* the window register is 32 bits */
377 select
= pin
* 2 + IOAPIC_REDTBL0
; /* register */
380 * Always clear an IO APIC pin before [re]programming it. This is
381 * particularly important if the pin is set up for a level interrupt
382 * as the IOART_REM_IRR bit might be set. When we reprogram the
383 * vector any EOI from pending ints on this pin could be lost and
384 * IRR might never get reset.
386 * To fix this problem, clear the vector and make sure it is
387 * programmed as an edge interrupt. This should theoretically
388 * clear IRR so we can later, safely program it as a level
393 flags
= io_apic_read(apic
, select
) & IOART_RESV
;
394 flags
|= IOART_INTMSET
| IOART_TRGREDG
| IOART_INTAHI
;
395 flags
|= IOART_DESTPHY
| IOART_DELFIXED
;
397 target
= io_apic_read(apic
, select
+ 1) & IOART_HI_DEST_RESV
;
398 target
|= 0; /* fixed mode cpu mask of 0 - don't deliver anywhere */
402 io_apic_write(apic
, select
, flags
| vector
);
403 io_apic_write(apic
, select
+ 1, target
);
408 * We only deal with vectored interrupts here. ? documentation is
409 * lacking, I'm guessing an interrupt type of 0 is the 'INT' type,
412 * This test also catches unconfigured pins.
414 if (apic_int_type(apic
, pin
) != 0)
418 * Leave the pin unprogrammed if it does not correspond to
421 irq
= apic_irq(apic
, pin
);
425 /* determine the bus type for this pin */
426 bus
= apic_src_bus_id(apic
, pin
);
429 bustype
= apic_bus_type(bus
);
431 if ((bustype
== ISA
) &&
432 (pin
< IOAPIC_ISA_INTS
) &&
434 (apic_polarity(apic
, pin
) == 0x1) &&
435 (apic_trigger(apic
, pin
) == 0x3)) {
437 * A broken BIOS might describe some ISA
438 * interrupts as active-high level-triggered.
439 * Use default ISA flags for those interrupts.
441 flags
= DEFAULT_ISA_FLAGS
;
444 * Program polarity and trigger mode according to
447 flags
= DEFAULT_FLAGS
;
448 level
= trigger(apic
, pin
, &flags
);
450 apic_pin_trigger
|= (1 << irq
);
451 polarity(apic
, pin
, &flags
, level
);
455 kprintf("IOAPIC #%d intpin %d -> irq %d\n",
460 * Program the appropriate registers. This routing may be
461 * overridden when an interrupt handler for a device is
462 * actually added (see register_int(), which calls through
463 * the MACHINTR ABI to set up an interrupt handler/vector).
465 * The order in which we must program the two registers for
466 * safety is unclear! XXX
470 vector
= IDT_OFFSET
+ irq
; /* IDT vec */
471 target
= io_apic_read(apic
, select
+ 1) & IOART_HI_DEST_RESV
;
472 target
|= IOART_HI_DEST_BROADCAST
;
473 flags
|= io_apic_read(apic
, select
) & IOART_RESV
;
474 io_apic_write(apic
, select
, flags
| vector
);
475 io_apic_write(apic
, select
+ 1, target
);
481 io_apic_setup(int apic
)
487 apic_pin_trigger
= 0; /* default to edge-triggered */
489 maxpin
= REDIRCNT_IOAPIC(apic
); /* pins in APIC */
490 kprintf("Programming %d pins in IOAPIC #%d\n", maxpin
, apic
);
492 for (pin
= 0; pin
< maxpin
; ++pin
) {
493 io_apic_setup_intpin(apic
, pin
);
496 if (apic_int_type(apic
, pin
) >= 0) {
497 kprintf("Warning: IOAPIC #%d pin %d does not exist,"
498 " cannot program!\n", apic
, pin
);
503 /* return GOOD status */
506 #undef DEFAULT_ISA_FLAGS
510 #define DEFAULT_EXTINT_FLAGS \
519 * Setup the source of External INTerrupts.
522 ext_int_setup(int apic
, int intr
)
524 u_char select
; /* the select register is 8 bits */
525 u_int32_t flags
; /* the window register is 32 bits */
526 u_int32_t target
; /* the window register is 32 bits */
527 u_int32_t vector
; /* the window register is 32 bits */
529 if (apic_int_type(apic
, intr
) != 3)
532 target
= IOART_HI_DEST_BROADCAST
;
533 select
= IOAPIC_REDTBL0
+ (2 * intr
);
534 vector
= IDT_OFFSET
+ intr
;
535 flags
= DEFAULT_EXTINT_FLAGS
;
537 io_apic_write(apic
, select
, flags
| vector
);
538 io_apic_write(apic
, select
+ 1, target
);
542 #undef DEFAULT_EXTINT_FLAGS
546 * Set the trigger level for an IO APIC pin.
549 trigger(int apic
, int pin
, u_int32_t
* flags
)
554 static int intcontrol
= -1;
556 switch (apic_trigger(apic
, pin
)) {
562 *flags
&= ~IOART_TRGRLVL
; /* *flags |= IOART_TRGREDG */
566 *flags
|= IOART_TRGRLVL
;
574 if ((id
= apic_src_bus_id(apic
, pin
)) == -1)
577 switch (apic_bus_type(id
)) {
579 *flags
&= ~IOART_TRGRLVL
; /* *flags |= IOART_TRGREDG; */
583 eirq
= apic_src_bus_irq(apic
, pin
);
585 if (eirq
< 0 || eirq
> 15) {
586 kprintf("EISA IRQ %d?!?!\n", eirq
);
590 if (intcontrol
== -1) {
591 intcontrol
= inb(ELCR1
) << 8;
592 intcontrol
|= inb(ELCR0
);
593 kprintf("EISA INTCONTROL = %08x\n", intcontrol
);
596 /* Use ELCR settings to determine level or edge mode */
597 level
= (intcontrol
>> eirq
) & 1;
600 * Note that on older Neptune chipset based systems, any
601 * pci interrupts often show up here and in the ELCR as well
602 * as level sensitive interrupts attributed to the EISA bus.
606 *flags
|= IOART_TRGRLVL
;
608 *flags
&= ~IOART_TRGRLVL
;
613 *flags
|= IOART_TRGRLVL
;
622 panic("bad APIC IO INT flags");
627 * Set the polarity value for an IO APIC pin.
630 polarity(int apic
, int pin
, u_int32_t
* flags
, int level
)
634 switch (apic_polarity(apic
, pin
)) {
640 *flags
&= ~IOART_INTALO
; /* *flags |= IOART_INTAHI */
644 *flags
|= IOART_INTALO
;
652 if ((id
= apic_src_bus_id(apic
, pin
)) == -1)
655 switch (apic_bus_type(id
)) {
657 *flags
&= ~IOART_INTALO
; /* *flags |= IOART_INTAHI */
661 /* polarity converter always gives active high */
662 *flags
&= ~IOART_INTALO
;
666 *flags
|= IOART_INTALO
;
675 panic("bad APIC IO INT flags");
680 * Print contents of apic_imen.
682 extern u_int apic_imen
; /* keep apic_imen 'opaque' */
688 kprintf("SMP: enabled INTs: ");
689 for (x
= 0; x
< 24; ++x
)
690 if ((apic_imen
& (1 << x
)) == 0)
692 kprintf("apic_imen: 0x%08x\n", apic_imen
);
697 * Inter Processor Interrupt functions.
703 * Send APIC IPI 'vector' to 'destType' via 'deliveryMode'.
705 * destType is 1 of: APIC_DEST_SELF, APIC_DEST_ALLISELF, APIC_DEST_ALLESELF
706 * vector is any valid SYSTEM INT vector
707 * delivery_mode is 1 of: APIC_DELMODE_FIXED, APIC_DELMODE_LOWPRIO
709 * A backlog of requests can create a deadlock between cpus. To avoid this
710 * we have to be able to accept IPIs at the same time we are trying to send
711 * them. The critical section prevents us from attempting to send additional
712 * IPIs reentrantly, but also prevents IPIQ processing so we have to call
713 * lwkt_process_ipiq() manually. It's rather messy and expensive for this
714 * to occur but fortunately it does not happen too often.
717 apic_ipi(int dest_type
, int vector
, int delivery_mode
)
722 if ((lapic
.icr_lo
& APIC_DELSTAT_MASK
) != 0) {
723 unsigned int eflags
= read_eflags();
725 while ((lapic
.icr_lo
& APIC_DELSTAT_MASK
) != 0) {
728 write_eflags(eflags
);
731 icr_lo
= (lapic
.icr_lo
& APIC_ICRLO_RESV_MASK
) | dest_type
|
732 delivery_mode
| vector
;
733 lapic
.icr_lo
= icr_lo
;
739 single_apic_ipi(int cpu
, int vector
, int delivery_mode
)
745 if ((lapic
.icr_lo
& APIC_DELSTAT_MASK
) != 0) {
746 unsigned int eflags
= read_eflags();
748 while ((lapic
.icr_lo
& APIC_DELSTAT_MASK
) != 0) {
751 write_eflags(eflags
);
753 icr_hi
= lapic
.icr_hi
& ~APIC_ID_MASK
;
754 icr_hi
|= (CPU_TO_ID(cpu
) << 24);
755 lapic
.icr_hi
= icr_hi
;
758 icr_lo
= (lapic
.icr_lo
& APIC_ICRLO_RESV_MASK
)
759 | APIC_DEST_DESTFLD
| delivery_mode
| vector
;
762 lapic
.icr_lo
= icr_lo
;
769 * Returns 0 if the apic is busy, 1 if we were able to queue the request.
771 * NOT WORKING YET! The code as-is may end up not queueing an IPI at all
772 * to the target, and the scheduler does not 'poll' for IPI messages.
775 single_apic_ipi_passive(int cpu
, int vector
, int delivery_mode
)
781 if ((lapic
.icr_lo
& APIC_DELSTAT_MASK
) != 0) {
785 icr_hi
= lapic
.icr_hi
& ~APIC_ID_MASK
;
786 icr_hi
|= (CPU_TO_ID(cpu
) << 24);
787 lapic
.icr_hi
= icr_hi
;
790 icr_lo
= (lapic
.icr_lo
& APIC_RESV2_MASK
)
791 | APIC_DEST_DESTFLD
| delivery_mode
| vector
;
794 lapic
.icr_lo
= icr_lo
;
802 * Send APIC IPI 'vector' to 'target's via 'delivery_mode'.
804 * target is a bitmask of destination cpus. Vector is any
805 * valid system INT vector. Delivery mode may be either
806 * APIC_DELMODE_FIXED or APIC_DELMODE_LOWPRIO.
809 selected_apic_ipi(u_int target
, int vector
, int delivery_mode
)
813 int n
= bsfl(target
);
815 single_apic_ipi(n
, vector
, delivery_mode
);
821 * Timer code, in development...
822 * - suggested by rgrimes@gndrsh.aac.dev.com
826 * Load a 'downcount time' in uSeconds.
829 set_apic_timer(int us
)
834 * When we reach here, lapic timer's frequency
835 * must have been calculated as well as the
836 * divisor (lapic.dcr_timer is setup during the
837 * divisor calculation).
839 KKASSERT(lapic_timer_freq
!= 0 &&
840 lapic_timer_divisor_idx
>= 0);
842 count
= ((us
* (int64_t)lapic_timer_freq
) + 999999) / 1000000;
843 lapic_timer_oneshot(count
);
848 * Read remaining time in timer.
851 read_apic_timer(void)
854 /** XXX FIXME: we need to return the actual remaining time,
855 * for now we just return the remaining count.
858 return lapic
.ccr_timer
;
864 * Spin-style delay, set delay time in uS, spin till it drains.
869 set_apic_timer(count
);
870 while (read_apic_timer())