2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/acpi.h>
31 #include <linux/sysdev.h>
32 #include <linux/msi.h>
33 #include <linux/htirq.h>
34 #include <linux/dmar.h>
35 #include <linux/jiffies.h>
37 #include <acpi/acpi_bus.h>
39 #include <linux/bootmem.h>
40 #include <linux/dmar.h>
46 #include <asm/proto.h>
49 #include <asm/i8259.h>
51 #include <asm/msidef.h>
52 #include <asm/hypertransport.h>
53 #include <asm/irq_remapping.h>
56 #include <mach_apic.h>
58 #define __apicdebuginit(type) static type __init
65 struct irq_pin_list
*irq_2_pin
;
68 unsigned move_cleanup_count
;
70 u8 move_in_progress
: 1;
73 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
74 static struct irq_cfg irq_cfg_legacy
[] __initdata
= {
75 [0] = { .irq
= 0, .domain
= CPU_MASK_ALL
, .vector
= IRQ0_VECTOR
, },
76 [1] = { .irq
= 1, .domain
= CPU_MASK_ALL
, .vector
= IRQ1_VECTOR
, },
77 [2] = { .irq
= 2, .domain
= CPU_MASK_ALL
, .vector
= IRQ2_VECTOR
, },
78 [3] = { .irq
= 3, .domain
= CPU_MASK_ALL
, .vector
= IRQ3_VECTOR
, },
79 [4] = { .irq
= 4, .domain
= CPU_MASK_ALL
, .vector
= IRQ4_VECTOR
, },
80 [5] = { .irq
= 5, .domain
= CPU_MASK_ALL
, .vector
= IRQ5_VECTOR
, },
81 [6] = { .irq
= 6, .domain
= CPU_MASK_ALL
, .vector
= IRQ6_VECTOR
, },
82 [7] = { .irq
= 7, .domain
= CPU_MASK_ALL
, .vector
= IRQ7_VECTOR
, },
83 [8] = { .irq
= 8, .domain
= CPU_MASK_ALL
, .vector
= IRQ8_VECTOR
, },
84 [9] = { .irq
= 9, .domain
= CPU_MASK_ALL
, .vector
= IRQ9_VECTOR
, },
85 [10] = { .irq
= 10, .domain
= CPU_MASK_ALL
, .vector
= IRQ10_VECTOR
, },
86 [11] = { .irq
= 11, .domain
= CPU_MASK_ALL
, .vector
= IRQ11_VECTOR
, },
87 [12] = { .irq
= 12, .domain
= CPU_MASK_ALL
, .vector
= IRQ12_VECTOR
, },
88 [13] = { .irq
= 13, .domain
= CPU_MASK_ALL
, .vector
= IRQ13_VECTOR
, },
89 [14] = { .irq
= 14, .domain
= CPU_MASK_ALL
, .vector
= IRQ14_VECTOR
, },
90 [15] = { .irq
= 15, .domain
= CPU_MASK_ALL
, .vector
= IRQ15_VECTOR
, },
93 static struct irq_cfg irq_cfg_init
= { .irq
= -1U, };
94 /* need to be biger than size of irq_cfg_legacy */
95 static int nr_irq_cfg
= 32;
97 static int __init
parse_nr_irq_cfg(char *arg
)
100 nr_irq_cfg
= simple_strtoul(arg
, NULL
, 0);
107 early_param("nr_irq_cfg", parse_nr_irq_cfg
);
109 static void init_one_irq_cfg(struct irq_cfg
*cfg
)
111 memcpy(cfg
, &irq_cfg_init
, sizeof(struct irq_cfg
));
114 static void __init
init_work(void *data
)
116 struct dyn_array
*da
= data
;
122 memcpy(cfg
, irq_cfg_legacy
, sizeof(irq_cfg_legacy
));
124 i
= sizeof(irq_cfg_legacy
)/sizeof(irq_cfg_legacy
[0]);
125 for (; i
< *da
->nr
; i
++)
126 init_one_irq_cfg(&cfg
[i
]);
128 for (i
= 1; i
< *da
->nr
; i
++)
129 cfg
[i
-1].next
= &cfg
[i
];
132 static struct irq_cfg
*irq_cfgx
;
133 DEFINE_DYN_ARRAY(irq_cfgx
, sizeof(struct irq_cfg
), nr_irq_cfg
, PAGE_SIZE
, init_work
);
135 static struct irq_cfg
*irq_cfg(unsigned int irq
)
155 static struct irq_cfg
*irq_cfg_alloc(unsigned int irq
)
157 struct irq_cfg
*cfg
, *cfg_pri
;
163 cfg_pri
= cfg
= &irq_cfgx
[0];
168 if (cfg
->irq
== -1U) {
178 * we run out of pre-allocate ones, allocate more
180 printk(KERN_DEBUG
"try to get more irq_cfg %d\n", nr_irq_cfg
);
183 cfg
= kzalloc(sizeof(struct irq_cfg
)*nr_irq_cfg
, GFP_ATOMIC
);
185 cfg
= __alloc_bootmem_nopanic(sizeof(struct irq_cfg
)*nr_irq_cfg
, PAGE_SIZE
, 0);
188 panic("please boot with nr_irq_cfg= %d\n", count
* 2);
190 for (i
= 0; i
< nr_irq_cfg
; i
++)
191 init_one_irq_cfg(&cfg
[i
]);
193 for (i
= 1; i
< nr_irq_cfg
; i
++)
194 cfg
[i
-1].next
= &cfg
[i
];
202 static int assign_irq_vector(int irq
, cpumask_t mask
);
204 int first_system_vector
= 0xfe;
206 char system_vectors
[NR_VECTORS
] = { [0 ... NR_VECTORS
-1] = SYS_VECTOR_FREE
};
208 int sis_apic_bug
; /* not actually supported, dummy for compile */
210 static int no_timer_check
;
212 static int disable_timer_pin_1 __initdata
;
214 int timer_through_8259 __initdata
;
216 /* Where if anywhere is the i8259 connect in external int mode */
217 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
219 static DEFINE_SPINLOCK(ioapic_lock
);
220 static DEFINE_SPINLOCK(vector_lock
);
223 * # of IRQ routing registers
225 int nr_ioapic_registers
[MAX_IO_APICS
];
227 /* I/O APIC RTE contents at the OS boot up */
228 struct IO_APIC_route_entry
*early_ioapic_entries
[MAX_IO_APICS
];
230 /* I/O APIC entries */
231 struct mp_config_ioapic mp_ioapics
[MAX_IO_APICS
];
234 /* MP IRQ source entries */
235 struct mp_config_intsrc mp_irqs
[MAX_IRQ_SOURCES
];
237 /* # of MP IRQ source entries */
240 DECLARE_BITMAP(mp_bus_not_pci
, MAX_MP_BUSSES
);
243 * Rough estimation of how many shared IRQs there are, can
244 * be changed anytime.
250 * This is performance-critical, we want to do it O(1)
252 * the indexing order of this array favors 1:1 mappings
253 * between pins and IRQs.
256 struct irq_pin_list
{
258 struct irq_pin_list
*next
;
261 static struct irq_pin_list
*irq_2_pin_head
;
262 /* fill one page ? */
263 static int nr_irq_2_pin
= 0x100;
264 static struct irq_pin_list
*irq_2_pin_ptr
;
265 static void __init
irq_2_pin_init_work(void *data
)
267 struct dyn_array
*da
= data
;
268 struct irq_pin_list
*pin
;
273 for (i
= 1; i
< *da
->nr
; i
++)
274 pin
[i
-1].next
= &pin
[i
];
276 irq_2_pin_ptr
= &pin
[0];
278 DEFINE_DYN_ARRAY(irq_2_pin_head
, sizeof(struct irq_pin_list
), nr_irq_2_pin
, PAGE_SIZE
, irq_2_pin_init_work
);
280 static struct irq_pin_list
*get_one_free_irq_2_pin(void)
282 struct irq_pin_list
*pin
;
288 irq_2_pin_ptr
= pin
->next
;
294 * we run out of pre-allocate ones, allocate more
296 printk(KERN_DEBUG
"try to get more irq_2_pin %d\n", nr_irq_2_pin
);
299 pin
= kzalloc(sizeof(struct irq_pin_list
)*nr_irq_2_pin
,
302 pin
= __alloc_bootmem_nopanic(sizeof(struct irq_pin_list
) *
303 nr_irq_2_pin
, PAGE_SIZE
, 0);
306 panic("can not get more irq_2_pin\n");
308 for (i
= 1; i
< nr_irq_2_pin
; i
++)
309 pin
[i
-1].next
= &pin
[i
];
311 irq_2_pin_ptr
= pin
->next
;
319 unsigned int unused
[3];
323 static __attribute_const__
struct io_apic __iomem
*io_apic_base(int idx
)
325 return (void __iomem
*) __fix_to_virt(FIX_IO_APIC_BASE_0
+ idx
)
326 + (mp_ioapics
[idx
].mp_apicaddr
& ~PAGE_MASK
);
329 static inline unsigned int io_apic_read(unsigned int apic
, unsigned int reg
)
331 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
332 writel(reg
, &io_apic
->index
);
333 return readl(&io_apic
->data
);
336 static inline void io_apic_write(unsigned int apic
, unsigned int reg
, unsigned int value
)
338 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
339 writel(reg
, &io_apic
->index
);
340 writel(value
, &io_apic
->data
);
344 * Re-write a value: to be used for read-modify-write
345 * cycles where the read already set up the index register.
347 static inline void io_apic_modify(unsigned int apic
, unsigned int value
)
349 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
350 writel(value
, &io_apic
->data
);
353 static bool io_apic_level_ack_pending(unsigned int irq
)
355 struct irq_pin_list
*entry
;
357 struct irq_cfg
*cfg
= irq_cfg(irq
);
359 spin_lock_irqsave(&ioapic_lock
, flags
);
360 entry
= cfg
->irq_2_pin
;
368 reg
= io_apic_read(entry
->apic
, 0x10 + pin
*2);
369 /* Is the remote IRR bit set? */
370 if (reg
& IO_APIC_REDIR_REMOTE_IRR
) {
371 spin_unlock_irqrestore(&ioapic_lock
, flags
);
378 spin_unlock_irqrestore(&ioapic_lock
, flags
);
384 * Synchronize the IO-APIC and the CPU by doing
385 * a dummy read from the IO-APIC
387 static inline void io_apic_sync(unsigned int apic
)
389 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
390 readl(&io_apic
->data
);
393 #define __DO_ACTION(R, ACTION, FINAL) \
397 struct irq_cfg *cfg; \
398 struct irq_pin_list *entry; \
400 BUG_ON(irq >= nr_irqs); \
401 cfg = irq_cfg(irq); \
402 entry = cfg->irq_2_pin; \
408 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
410 io_apic_modify(entry->apic, reg); \
414 entry = entry->next; \
419 struct { u32 w1
, w2
; };
420 struct IO_APIC_route_entry entry
;
423 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
425 union entry_union eu
;
427 spin_lock_irqsave(&ioapic_lock
, flags
);
428 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
429 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
430 spin_unlock_irqrestore(&ioapic_lock
, flags
);
435 * When we write a new IO APIC routing entry, we need to write the high
436 * word first! If the mask bit in the low word is clear, we will enable
437 * the interrupt, and we need to make sure the entry is fully populated
438 * before that happens.
441 __ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
443 union entry_union eu
;
445 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
446 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
449 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
452 spin_lock_irqsave(&ioapic_lock
, flags
);
453 __ioapic_write_entry(apic
, pin
, e
);
454 spin_unlock_irqrestore(&ioapic_lock
, flags
);
458 * When we mask an IO APIC routing entry, we need to write the low
459 * word first, in order to set the mask bit before we change the
462 static void ioapic_mask_entry(int apic
, int pin
)
465 union entry_union eu
= { .entry
.mask
= 1 };
467 spin_lock_irqsave(&ioapic_lock
, flags
);
468 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
469 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
470 spin_unlock_irqrestore(&ioapic_lock
, flags
);
474 static void __target_IO_APIC_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
478 struct irq_pin_list
*entry
;
480 BUG_ON(irq
>= nr_irqs
);
482 entry
= cfg
->irq_2_pin
;
492 * With interrupt-remapping, destination information comes
493 * from interrupt-remapping table entry.
495 if (!irq_remapped(irq
))
496 io_apic_write(apic
, 0x11 + pin
*2, dest
);
497 reg
= io_apic_read(apic
, 0x10 + pin
*2);
498 reg
&= ~IO_APIC_REDIR_VECTOR_MASK
;
500 io_apic_modify(apic
, reg
);
507 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
509 struct irq_cfg
*cfg
= irq_cfg(irq
);
513 struct irq_desc
*desc
;
515 cpus_and(tmp
, mask
, cpu_online_map
);
519 if (assign_irq_vector(irq
, mask
))
522 cpus_and(tmp
, cfg
->domain
, mask
);
523 dest
= cpu_mask_to_apicid(tmp
);
526 * Only the high 8 bits are valid.
528 dest
= SET_APIC_LOGICAL_ID(dest
);
530 desc
= irq_to_desc(irq
);
531 spin_lock_irqsave(&ioapic_lock
, flags
);
532 __target_IO_APIC_irq(irq
, dest
, cfg
->vector
);
533 desc
->affinity
= mask
;
534 spin_unlock_irqrestore(&ioapic_lock
, flags
);
539 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
540 * shared ISA-space IRQs, so we have to support them. We are super
541 * fast in the common case, and fast for shared ISA-space IRQs.
543 int first_free_entry
;
544 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
547 struct irq_pin_list
*entry
;
549 BUG_ON(irq
>= nr_irqs
);
550 /* first time to refer irq_cfg, so with new */
551 cfg
= irq_cfg_alloc(irq
);
552 entry
= cfg
->irq_2_pin
;
554 entry
= get_one_free_irq_2_pin();
555 cfg
->irq_2_pin
= entry
;
558 printk(KERN_DEBUG
" 0 add_pin_to_irq: irq %d --> apic %d pin %d\n", irq
, apic
, pin
);
562 while (entry
->next
) {
563 /* not again, please */
564 if (entry
->apic
== apic
&& entry
->pin
== pin
)
570 entry
->next
= get_one_free_irq_2_pin();
574 printk(KERN_DEBUG
" x add_pin_to_irq: irq %d --> apic %d pin %d\n", irq
, apic
, pin
);
578 * Reroute an IRQ to a different pin.
580 static void __init
replace_pin_at_irq(unsigned int irq
,
581 int oldapic
, int oldpin
,
582 int newapic
, int newpin
)
584 struct irq_cfg
*cfg
= irq_cfg(irq
);
585 struct irq_pin_list
*entry
= cfg
->irq_2_pin
;
589 if (entry
->apic
== oldapic
&& entry
->pin
== oldpin
) {
590 entry
->apic
= newapic
;
593 /* every one is different, right? */
599 /* why? call replace before add? */
601 add_pin_to_irq(irq
, newapic
, newpin
);
605 #define DO_ACTION(name,R,ACTION, FINAL) \
607 static void name##_IO_APIC_irq (unsigned int irq) \
608 __DO_ACTION(R, ACTION, FINAL)
611 DO_ACTION(__mask
, 0, |= IO_APIC_REDIR_MASKED
, io_apic_sync(entry
->apic
))
614 DO_ACTION(__unmask
, 0, &= ~IO_APIC_REDIR_MASKED
, )
616 static void mask_IO_APIC_irq (unsigned int irq
)
620 spin_lock_irqsave(&ioapic_lock
, flags
);
621 __mask_IO_APIC_irq(irq
);
622 spin_unlock_irqrestore(&ioapic_lock
, flags
);
625 static void unmask_IO_APIC_irq (unsigned int irq
)
629 spin_lock_irqsave(&ioapic_lock
, flags
);
630 __unmask_IO_APIC_irq(irq
);
631 spin_unlock_irqrestore(&ioapic_lock
, flags
);
634 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
636 struct IO_APIC_route_entry entry
;
638 /* Check delivery_mode to be sure we're not clearing an SMI pin */
639 entry
= ioapic_read_entry(apic
, pin
);
640 if (entry
.delivery_mode
== dest_SMI
)
643 * Disable it in the IO-APIC irq-routing table:
645 ioapic_mask_entry(apic
, pin
);
648 static void clear_IO_APIC (void)
652 for (apic
= 0; apic
< nr_ioapics
; apic
++)
653 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
654 clear_IO_APIC_pin(apic
, pin
);
658 * Saves and masks all the unmasked IO-APIC RTE's
660 int save_mask_IO_APIC_setup(void)
662 union IO_APIC_reg_01 reg_01
;
667 * The number of IO-APIC IRQ registers (== #pins):
669 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
670 spin_lock_irqsave(&ioapic_lock
, flags
);
671 reg_01
.raw
= io_apic_read(apic
, 1);
672 spin_unlock_irqrestore(&ioapic_lock
, flags
);
673 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
676 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
677 early_ioapic_entries
[apic
] =
678 kzalloc(sizeof(struct IO_APIC_route_entry
) *
679 nr_ioapic_registers
[apic
], GFP_KERNEL
);
680 if (!early_ioapic_entries
[apic
])
684 for (apic
= 0; apic
< nr_ioapics
; apic
++)
685 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
686 struct IO_APIC_route_entry entry
;
688 entry
= early_ioapic_entries
[apic
][pin
] =
689 ioapic_read_entry(apic
, pin
);
692 ioapic_write_entry(apic
, pin
, entry
);
698 void restore_IO_APIC_setup(void)
702 for (apic
= 0; apic
< nr_ioapics
; apic
++)
703 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
704 ioapic_write_entry(apic
, pin
,
705 early_ioapic_entries
[apic
][pin
]);
708 void reinit_intr_remapped_IO_APIC(int intr_remapping
)
711 * for now plain restore of previous settings.
712 * TBD: In the case of OS enabling interrupt-remapping,
713 * IO-APIC RTE's need to be setup to point to interrupt-remapping
714 * table entries. for now, do a plain restore, and wait for
715 * the setup_IO_APIC_irqs() to do proper initialization.
717 restore_IO_APIC_setup();
720 int skip_ioapic_setup
;
723 static int __init
parse_noapic(char *str
)
725 disable_ioapic_setup();
728 early_param("noapic", parse_noapic
);
730 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
731 static int __init
disable_timer_pin_setup(char *arg
)
733 disable_timer_pin_1
= 1;
736 __setup("disable_timer_pin_1", disable_timer_pin_setup
);
740 * Find the IRQ entry number of a certain pin.
742 static int find_irq_entry(int apic
, int pin
, int type
)
746 for (i
= 0; i
< mp_irq_entries
; i
++)
747 if (mp_irqs
[i
].mp_irqtype
== type
&&
748 (mp_irqs
[i
].mp_dstapic
== mp_ioapics
[apic
].mp_apicid
||
749 mp_irqs
[i
].mp_dstapic
== MP_APIC_ALL
) &&
750 mp_irqs
[i
].mp_dstirq
== pin
)
757 * Find the pin to which IRQ[irq] (ISA) is connected
759 static int __init
find_isa_irq_pin(int irq
, int type
)
763 for (i
= 0; i
< mp_irq_entries
; i
++) {
764 int lbus
= mp_irqs
[i
].mp_srcbus
;
766 if (test_bit(lbus
, mp_bus_not_pci
) &&
767 (mp_irqs
[i
].mp_irqtype
== type
) &&
768 (mp_irqs
[i
].mp_srcbusirq
== irq
))
770 return mp_irqs
[i
].mp_dstirq
;
775 static int __init
find_isa_irq_apic(int irq
, int type
)
779 for (i
= 0; i
< mp_irq_entries
; i
++) {
780 int lbus
= mp_irqs
[i
].mp_srcbus
;
782 if (test_bit(lbus
, mp_bus_not_pci
) &&
783 (mp_irqs
[i
].mp_irqtype
== type
) &&
784 (mp_irqs
[i
].mp_srcbusirq
== irq
))
787 if (i
< mp_irq_entries
) {
789 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
790 if (mp_ioapics
[apic
].mp_apicid
== mp_irqs
[i
].mp_dstapic
)
799 * Find a specific PCI IRQ entry.
800 * Not an __init, possibly needed by modules
802 static int pin_2_irq(int idx
, int apic
, int pin
);
804 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
806 int apic
, i
, best_guess
= -1;
808 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
810 if (test_bit(bus
, mp_bus_not_pci
)) {
811 apic_printk(APIC_VERBOSE
, "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
814 for (i
= 0; i
< mp_irq_entries
; i
++) {
815 int lbus
= mp_irqs
[i
].mp_srcbus
;
817 for (apic
= 0; apic
< nr_ioapics
; apic
++)
818 if (mp_ioapics
[apic
].mp_apicid
== mp_irqs
[i
].mp_dstapic
||
819 mp_irqs
[i
].mp_dstapic
== MP_APIC_ALL
)
822 if (!test_bit(lbus
, mp_bus_not_pci
) &&
823 !mp_irqs
[i
].mp_irqtype
&&
825 (slot
== ((mp_irqs
[i
].mp_srcbusirq
>> 2) & 0x1f))) {
826 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mp_dstirq
);
828 if (!(apic
|| IO_APIC_IRQ(irq
)))
831 if (pin
== (mp_irqs
[i
].mp_srcbusirq
& 3))
834 * Use the first all-but-pin matching entry as a
835 * best-guess fuzzy result for broken mptables.
841 BUG_ON(best_guess
>= nr_irqs
);
845 /* ISA interrupts are always polarity zero edge triggered,
846 * when listed as conforming in the MP table. */
848 #define default_ISA_trigger(idx) (0)
849 #define default_ISA_polarity(idx) (0)
851 /* PCI interrupts are always polarity one level triggered,
852 * when listed as conforming in the MP table. */
854 #define default_PCI_trigger(idx) (1)
855 #define default_PCI_polarity(idx) (1)
857 static int MPBIOS_polarity(int idx
)
859 int bus
= mp_irqs
[idx
].mp_srcbus
;
863 * Determine IRQ line polarity (high active or low active):
865 switch (mp_irqs
[idx
].mp_irqflag
& 3)
867 case 0: /* conforms, ie. bus-type dependent polarity */
868 if (test_bit(bus
, mp_bus_not_pci
))
869 polarity
= default_ISA_polarity(idx
);
871 polarity
= default_PCI_polarity(idx
);
873 case 1: /* high active */
878 case 2: /* reserved */
880 printk(KERN_WARNING
"broken BIOS!!\n");
884 case 3: /* low active */
889 default: /* invalid */
891 printk(KERN_WARNING
"broken BIOS!!\n");
899 static int MPBIOS_trigger(int idx
)
901 int bus
= mp_irqs
[idx
].mp_srcbus
;
905 * Determine IRQ trigger mode (edge or level sensitive):
907 switch ((mp_irqs
[idx
].mp_irqflag
>>2) & 3)
909 case 0: /* conforms, ie. bus-type dependent */
910 if (test_bit(bus
, mp_bus_not_pci
))
911 trigger
= default_ISA_trigger(idx
);
913 trigger
= default_PCI_trigger(idx
);
920 case 2: /* reserved */
922 printk(KERN_WARNING
"broken BIOS!!\n");
931 default: /* invalid */
933 printk(KERN_WARNING
"broken BIOS!!\n");
941 static inline int irq_polarity(int idx
)
943 return MPBIOS_polarity(idx
);
946 static inline int irq_trigger(int idx
)
948 return MPBIOS_trigger(idx
);
951 static int pin_2_irq(int idx
, int apic
, int pin
)
954 int bus
= mp_irqs
[idx
].mp_srcbus
;
957 * Debugging check, we are in big trouble if this message pops up!
959 if (mp_irqs
[idx
].mp_dstirq
!= pin
)
960 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
962 if (test_bit(bus
, mp_bus_not_pci
)) {
963 irq
= mp_irqs
[idx
].mp_srcbusirq
;
966 * PCI IRQs are mapped in order
970 irq
+= nr_ioapic_registers
[i
++];
973 BUG_ON(irq
>= nr_irqs
);
977 void lock_vector_lock(void)
979 /* Used to the online set of cpus does not change
980 * during assign_irq_vector.
982 spin_lock(&vector_lock
);
985 void unlock_vector_lock(void)
987 spin_unlock(&vector_lock
);
990 static int __assign_irq_vector(int irq
, cpumask_t mask
)
993 * NOTE! The local APIC isn't very good at handling
994 * multiple interrupts at the same interrupt level.
995 * As the interrupt level is determined by taking the
996 * vector number and shifting that right by 4, we
997 * want to spread these out a bit so that they don't
998 * all fall in the same interrupt level.
1000 * Also, we've got to be careful not to trash gate
1001 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1003 static int current_vector
= FIRST_DEVICE_VECTOR
, current_offset
= 0;
1004 unsigned int old_vector
;
1006 struct irq_cfg
*cfg
;
1008 BUG_ON((unsigned)irq
>= nr_irqs
);
1011 /* Only try and allocate irqs on cpus that are present */
1012 cpus_and(mask
, mask
, cpu_online_map
);
1014 if ((cfg
->move_in_progress
) || cfg
->move_cleanup_count
)
1017 old_vector
= cfg
->vector
;
1020 cpus_and(tmp
, cfg
->domain
, mask
);
1021 if (!cpus_empty(tmp
))
1025 for_each_cpu_mask_nr(cpu
, mask
) {
1026 cpumask_t domain
, new_mask
;
1030 domain
= vector_allocation_domain(cpu
);
1031 cpus_and(new_mask
, domain
, cpu_online_map
);
1033 vector
= current_vector
;
1034 offset
= current_offset
;
1037 if (vector
>= first_system_vector
) {
1038 /* If we run out of vectors on large boxen, must share them. */
1039 offset
= (offset
+ 1) % 8;
1040 vector
= FIRST_DEVICE_VECTOR
+ offset
;
1042 if (unlikely(current_vector
== vector
))
1044 if (vector
== IA32_SYSCALL_VECTOR
)
1046 for_each_cpu_mask_nr(new_cpu
, new_mask
)
1047 if (per_cpu(vector_irq
, new_cpu
)[vector
] != -1)
1050 current_vector
= vector
;
1051 current_offset
= offset
;
1053 cfg
->move_in_progress
= 1;
1054 cfg
->old_domain
= cfg
->domain
;
1056 for_each_cpu_mask_nr(new_cpu
, new_mask
)
1057 per_cpu(vector_irq
, new_cpu
)[vector
] = irq
;
1058 cfg
->vector
= vector
;
1059 cfg
->domain
= domain
;
1065 static int assign_irq_vector(int irq
, cpumask_t mask
)
1068 unsigned long flags
;
1070 spin_lock_irqsave(&vector_lock
, flags
);
1071 err
= __assign_irq_vector(irq
, mask
);
1072 spin_unlock_irqrestore(&vector_lock
, flags
);
1076 static void __clear_irq_vector(int irq
)
1078 struct irq_cfg
*cfg
;
1082 BUG_ON((unsigned)irq
>= nr_irqs
);
1084 BUG_ON(!cfg
->vector
);
1086 vector
= cfg
->vector
;
1087 cpus_and(mask
, cfg
->domain
, cpu_online_map
);
1088 for_each_cpu_mask_nr(cpu
, mask
)
1089 per_cpu(vector_irq
, cpu
)[vector
] = -1;
1092 cpus_clear(cfg
->domain
);
1095 void __setup_vector_irq(int cpu
)
1097 /* Initialize vector_irq on a new cpu */
1098 /* This function must be called with vector_lock held */
1101 /* Mark the inuse vectors */
1102 for (irq
= 0; irq
< nr_irqs
; ++irq
) {
1103 struct irq_cfg
*cfg
= irq_cfg(irq
);
1105 if (!cpu_isset(cpu
, cfg
->domain
))
1107 vector
= cfg
->vector
;
1108 per_cpu(vector_irq
, cpu
)[vector
] = irq
;
1110 /* Mark the free vectors */
1111 for (vector
= 0; vector
< NR_VECTORS
; ++vector
) {
1112 struct irq_cfg
*cfg
;
1114 irq
= per_cpu(vector_irq
, cpu
)[vector
];
1119 if (!cpu_isset(cpu
, cfg
->domain
))
1120 per_cpu(vector_irq
, cpu
)[vector
] = -1;
1124 static struct irq_chip ioapic_chip
;
1125 #ifdef CONFIG_INTR_REMAP
1126 static struct irq_chip ir_ioapic_chip
;
1129 static void ioapic_register_intr(int irq
, unsigned long trigger
)
1131 struct irq_desc
*desc
;
1133 desc
= irq_to_desc(irq
);
1135 desc
->status
|= IRQ_LEVEL
;
1137 desc
->status
&= ~IRQ_LEVEL
;
1139 #ifdef CONFIG_INTR_REMAP
1140 if (irq_remapped(irq
)) {
1141 desc
->status
|= IRQ_MOVE_PCNTXT
;
1143 set_irq_chip_and_handler_name(irq
, &ir_ioapic_chip
,
1147 set_irq_chip_and_handler_name(irq
, &ir_ioapic_chip
,
1148 handle_edge_irq
, "edge");
1153 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
1157 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
1158 handle_edge_irq
, "edge");
1161 static int setup_ioapic_entry(int apic
, int irq
,
1162 struct IO_APIC_route_entry
*entry
,
1163 unsigned int destination
, int trigger
,
1164 int polarity
, int vector
)
1167 * add it to the IO-APIC irq-routing table:
1169 memset(entry
,0,sizeof(*entry
));
1171 #ifdef CONFIG_INTR_REMAP
1172 if (intr_remapping_enabled
) {
1173 struct intel_iommu
*iommu
= map_ioapic_to_ir(apic
);
1175 struct IR_IO_APIC_route_entry
*ir_entry
=
1176 (struct IR_IO_APIC_route_entry
*) entry
;
1180 panic("No mapping iommu for ioapic %d\n", apic
);
1182 index
= alloc_irte(iommu
, irq
, 1);
1184 panic("Failed to allocate IRTE for ioapic %d\n", apic
);
1186 memset(&irte
, 0, sizeof(irte
));
1189 irte
.dst_mode
= INT_DEST_MODE
;
1190 irte
.trigger_mode
= trigger
;
1191 irte
.dlvry_mode
= INT_DELIVERY_MODE
;
1192 irte
.vector
= vector
;
1193 irte
.dest_id
= IRTE_DEST(destination
);
1195 modify_irte(irq
, &irte
);
1197 ir_entry
->index2
= (index
>> 15) & 0x1;
1199 ir_entry
->format
= 1;
1200 ir_entry
->index
= (index
& 0x7fff);
1204 entry
->delivery_mode
= INT_DELIVERY_MODE
;
1205 entry
->dest_mode
= INT_DEST_MODE
;
1206 entry
->dest
= destination
;
1209 entry
->mask
= 0; /* enable IRQ */
1210 entry
->trigger
= trigger
;
1211 entry
->polarity
= polarity
;
1212 entry
->vector
= vector
;
1214 /* Mask level triggered irqs.
1215 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1222 static void setup_IO_APIC_irq(int apic
, int pin
, unsigned int irq
,
1223 int trigger
, int polarity
)
1225 struct irq_cfg
*cfg
;
1226 struct IO_APIC_route_entry entry
;
1229 if (!IO_APIC_IRQ(irq
))
1235 if (assign_irq_vector(irq
, mask
))
1238 cpus_and(mask
, cfg
->domain
, mask
);
1240 apic_printk(APIC_VERBOSE
,KERN_DEBUG
1241 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1242 "IRQ %d Mode:%i Active:%i)\n",
1243 apic
, mp_ioapics
[apic
].mp_apicid
, pin
, cfg
->vector
,
1244 irq
, trigger
, polarity
);
1247 if (setup_ioapic_entry(mp_ioapics
[apic
].mp_apicid
, irq
, &entry
,
1248 cpu_mask_to_apicid(mask
), trigger
, polarity
,
1250 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1251 mp_ioapics
[apic
].mp_apicid
, pin
);
1252 __clear_irq_vector(irq
);
1256 ioapic_register_intr(irq
, trigger
);
1258 disable_8259A_irq(irq
);
1260 ioapic_write_entry(apic
, pin
, entry
);
1263 static void __init
setup_IO_APIC_irqs(void)
1265 int apic
, pin
, idx
, irq
, first_notcon
= 1;
1267 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
1269 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1270 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1272 idx
= find_irq_entry(apic
,pin
,mp_INT
);
1275 apic_printk(APIC_VERBOSE
, KERN_DEBUG
" IO-APIC (apicid-pin) %d-%d", mp_ioapics
[apic
].mp_apicid
, pin
);
1278 apic_printk(APIC_VERBOSE
, ", %d-%d", mp_ioapics
[apic
].mp_apicid
, pin
);
1281 if (!first_notcon
) {
1282 apic_printk(APIC_VERBOSE
, " not connected.\n");
1286 irq
= pin_2_irq(idx
, apic
, pin
);
1287 add_pin_to_irq(irq
, apic
, pin
);
1289 setup_IO_APIC_irq(apic
, pin
, irq
,
1290 irq_trigger(idx
), irq_polarity(idx
));
1295 apic_printk(APIC_VERBOSE
, " not connected.\n");
1299 * Set up the timer pin, possibly with the 8259A-master behind.
1301 static void __init
setup_timer_IRQ0_pin(unsigned int apic
, unsigned int pin
,
1304 struct IO_APIC_route_entry entry
;
1306 if (intr_remapping_enabled
)
1309 memset(&entry
, 0, sizeof(entry
));
1312 * We use logical delivery to get the timer IRQ
1315 entry
.dest_mode
= INT_DEST_MODE
;
1316 entry
.mask
= 1; /* mask IRQ now */
1317 entry
.dest
= cpu_mask_to_apicid(TARGET_CPUS
);
1318 entry
.delivery_mode
= INT_DELIVERY_MODE
;
1321 entry
.vector
= vector
;
1324 * The timer IRQ doesn't have to know that behind the
1325 * scene we may have a 8259A-master in AEOI mode ...
1327 set_irq_chip_and_handler_name(0, &ioapic_chip
, handle_edge_irq
, "edge");
1330 * Add it to the IO-APIC irq-routing table:
1332 ioapic_write_entry(apic
, pin
, entry
);
1336 __apicdebuginit(void) print_IO_APIC(void)
1339 union IO_APIC_reg_00 reg_00
;
1340 union IO_APIC_reg_01 reg_01
;
1341 union IO_APIC_reg_02 reg_02
;
1342 unsigned long flags
;
1344 if (apic_verbosity
== APIC_QUIET
)
1347 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
1348 for (i
= 0; i
< nr_ioapics
; i
++)
1349 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
1350 mp_ioapics
[i
].mp_apicid
, nr_ioapic_registers
[i
]);
1353 * We are a bit conservative about what we expect. We have to
1354 * know about every hardware change ASAP.
1356 printk(KERN_INFO
"testing the IO APIC.......................\n");
1358 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1360 spin_lock_irqsave(&ioapic_lock
, flags
);
1361 reg_00
.raw
= io_apic_read(apic
, 0);
1362 reg_01
.raw
= io_apic_read(apic
, 1);
1363 if (reg_01
.bits
.version
>= 0x10)
1364 reg_02
.raw
= io_apic_read(apic
, 2);
1365 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1368 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].mp_apicid
);
1369 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
1370 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
1372 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_01
);
1373 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
1375 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
1376 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
1378 if (reg_01
.bits
.version
>= 0x10) {
1379 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
1380 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
1383 printk(KERN_DEBUG
".... IRQ redirection table:\n");
1385 printk(KERN_DEBUG
" NR Dst Mask Trig IRR Pol"
1386 " Stat Dmod Deli Vect: \n");
1388 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
1389 struct IO_APIC_route_entry entry
;
1391 entry
= ioapic_read_entry(apic
, i
);
1393 printk(KERN_DEBUG
" %02x %03X ",
1398 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1403 entry
.delivery_status
,
1405 entry
.delivery_mode
,
1410 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
1411 for (i
= 0; i
< nr_irqs
; i
++) {
1412 struct irq_cfg
*cfg
= irq_cfg(i
);
1413 struct irq_pin_list
*entry
= cfg
->irq_2_pin
;
1416 printk(KERN_DEBUG
"IRQ%d ", i
);
1418 printk("-> %d:%d", entry
->apic
, entry
->pin
);
1421 entry
= entry
->next
;
1426 printk(KERN_INFO
".................................... done.\n");
1431 __apicdebuginit(void) print_APIC_bitfield(int base
)
1436 if (apic_verbosity
== APIC_QUIET
)
1439 printk(KERN_DEBUG
"0123456789abcdef0123456789abcdef\n" KERN_DEBUG
);
1440 for (i
= 0; i
< 8; i
++) {
1441 v
= apic_read(base
+ i
*0x10);
1442 for (j
= 0; j
< 32; j
++) {
1452 __apicdebuginit(void) print_local_APIC(void *dummy
)
1454 unsigned int v
, ver
, maxlvt
;
1457 if (apic_verbosity
== APIC_QUIET
)
1460 printk("\n" KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
1461 smp_processor_id(), hard_smp_processor_id());
1462 v
= apic_read(APIC_ID
);
1463 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
, read_apic_id());
1464 v
= apic_read(APIC_LVR
);
1465 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
1466 ver
= GET_APIC_VERSION(v
);
1467 maxlvt
= lapic_get_maxlvt();
1469 v
= apic_read(APIC_TASKPRI
);
1470 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
1472 v
= apic_read(APIC_ARBPRI
);
1473 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
1474 v
& APIC_ARBPRI_MASK
);
1475 v
= apic_read(APIC_PROCPRI
);
1476 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
1478 v
= apic_read(APIC_EOI
);
1479 printk(KERN_DEBUG
"... APIC EOI: %08x\n", v
);
1480 v
= apic_read(APIC_RRR
);
1481 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
1482 v
= apic_read(APIC_LDR
);
1483 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
1484 v
= apic_read(APIC_DFR
);
1485 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
1486 v
= apic_read(APIC_SPIV
);
1487 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
1489 printk(KERN_DEBUG
"... APIC ISR field:\n");
1490 print_APIC_bitfield(APIC_ISR
);
1491 printk(KERN_DEBUG
"... APIC TMR field:\n");
1492 print_APIC_bitfield(APIC_TMR
);
1493 printk(KERN_DEBUG
"... APIC IRR field:\n");
1494 print_APIC_bitfield(APIC_IRR
);
1496 v
= apic_read(APIC_ESR
);
1497 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
1499 icr
= apic_icr_read();
1500 printk(KERN_DEBUG
"... APIC ICR: %08x\n", (u32
)icr
);
1501 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", (u32
)(icr
>> 32));
1503 v
= apic_read(APIC_LVTT
);
1504 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1506 if (maxlvt
> 3) { /* PC is LVT#4. */
1507 v
= apic_read(APIC_LVTPC
);
1508 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
1510 v
= apic_read(APIC_LVT0
);
1511 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
1512 v
= apic_read(APIC_LVT1
);
1513 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
1515 if (maxlvt
> 2) { /* ERR is LVT#3. */
1516 v
= apic_read(APIC_LVTERR
);
1517 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
1520 v
= apic_read(APIC_TMICT
);
1521 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
1522 v
= apic_read(APIC_TMCCT
);
1523 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
1524 v
= apic_read(APIC_TDCR
);
1525 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1529 __apicdebuginit(void) print_all_local_APICs(void)
1531 on_each_cpu(print_local_APIC
, NULL
, 1);
1534 __apicdebuginit(void) print_PIC(void)
1537 unsigned long flags
;
1539 if (apic_verbosity
== APIC_QUIET
)
1542 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1544 spin_lock_irqsave(&i8259A_lock
, flags
);
1546 v
= inb(0xa1) << 8 | inb(0x21);
1547 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1549 v
= inb(0xa0) << 8 | inb(0x20);
1550 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1554 v
= inb(0xa0) << 8 | inb(0x20);
1558 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1560 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1562 v
= inb(0x4d1) << 8 | inb(0x4d0);
1563 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1566 __apicdebuginit(int) print_all_ICs(void)
1569 print_all_local_APICs();
1575 fs_initcall(print_all_ICs
);
1578 void __init
enable_IO_APIC(void)
1580 union IO_APIC_reg_01 reg_01
;
1581 int i8259_apic
, i8259_pin
;
1583 unsigned long flags
;
1586 * The number of IO-APIC IRQ registers (== #pins):
1588 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1589 spin_lock_irqsave(&ioapic_lock
, flags
);
1590 reg_01
.raw
= io_apic_read(apic
, 1);
1591 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1592 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
1594 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1596 /* See if any of the pins is in ExtINT mode */
1597 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1598 struct IO_APIC_route_entry entry
;
1599 entry
= ioapic_read_entry(apic
, pin
);
1601 /* If the interrupt line is enabled and in ExtInt mode
1602 * I have found the pin where the i8259 is connected.
1604 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1605 ioapic_i8259
.apic
= apic
;
1606 ioapic_i8259
.pin
= pin
;
1612 /* Look to see what if the MP table has reported the ExtINT */
1613 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1614 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1615 /* Trust the MP table if nothing is setup in the hardware */
1616 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1617 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1618 ioapic_i8259
.pin
= i8259_pin
;
1619 ioapic_i8259
.apic
= i8259_apic
;
1621 /* Complain if the MP table and the hardware disagree */
1622 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1623 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1625 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1629 * Do not trust the IO-APIC being empty at bootup
1635 * Not an __init, needed by the reboot code
1637 void disable_IO_APIC(void)
1640 * Clear the IO-APIC before rebooting:
1645 * If the i8259 is routed through an IOAPIC
1646 * Put that IOAPIC in virtual wire mode
1647 * so legacy interrupts can be delivered.
1649 if (ioapic_i8259
.pin
!= -1) {
1650 struct IO_APIC_route_entry entry
;
1652 memset(&entry
, 0, sizeof(entry
));
1653 entry
.mask
= 0; /* Enabled */
1654 entry
.trigger
= 0; /* Edge */
1656 entry
.polarity
= 0; /* High */
1657 entry
.delivery_status
= 0;
1658 entry
.dest_mode
= 0; /* Physical */
1659 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1661 entry
.dest
= read_apic_id();
1664 * Add it to the IO-APIC irq-routing table:
1666 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1669 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
1673 * There is a nasty bug in some older SMP boards, their mptable lies
1674 * about the timer IRQ. We do the following to work around the situation:
1676 * - timer IRQ defaults to IO-APIC IRQ
1677 * - if this function detects that timer IRQs are defunct, then we fall
1678 * back to ISA timer IRQs
1680 static int __init
timer_irq_works(void)
1682 unsigned long t1
= jiffies
;
1683 unsigned long flags
;
1685 local_save_flags(flags
);
1687 /* Let ten ticks pass... */
1688 mdelay((10 * 1000) / HZ
);
1689 local_irq_restore(flags
);
1692 * Expect a few ticks at least, to be sure some possible
1693 * glue logic does not lock up after one or two first
1694 * ticks in a non-ExtINT mode. Also the local APIC
1695 * might have cached one ExtINT interrupt. Finally, at
1696 * least one tick may be lost due to delays.
1700 if (time_after(jiffies
, t1
+ 4))
1706 * In the SMP+IOAPIC case it might happen that there are an unspecified
1707 * number of pending IRQ events unhandled. These cases are very rare,
1708 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1709 * better to do it this way as thus we do not have to be aware of
1710 * 'pending' interrupts in the IRQ path, except at this point.
1713 * Edge triggered needs to resend any interrupt
1714 * that was delayed but this is now handled in the device
1719 * Starting up a edge-triggered IO-APIC interrupt is
1720 * nasty - we need to make sure that we get the edge.
1721 * If it is already asserted for some reason, we need
1722 * return 1 to indicate that is was pending.
1724 * This is not complete - we should be able to fake
1725 * an edge even if it isn't on the 8259A...
1728 static unsigned int startup_ioapic_irq(unsigned int irq
)
1730 int was_pending
= 0;
1731 unsigned long flags
;
1733 spin_lock_irqsave(&ioapic_lock
, flags
);
1735 disable_8259A_irq(irq
);
1736 if (i8259A_irq_pending(irq
))
1739 __unmask_IO_APIC_irq(irq
);
1740 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1745 static int ioapic_retrigger_irq(unsigned int irq
)
1747 struct irq_cfg
*cfg
= irq_cfg(irq
);
1748 unsigned long flags
;
1750 spin_lock_irqsave(&vector_lock
, flags
);
1751 send_IPI_mask(cpumask_of_cpu(first_cpu(cfg
->domain
)), cfg
->vector
);
1752 spin_unlock_irqrestore(&vector_lock
, flags
);
1758 * Level and edge triggered IO-APIC interrupts need different handling,
1759 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1760 * handled with the level-triggered descriptor, but that one has slightly
1761 * more overhead. Level-triggered interrupts cannot be handled with the
1762 * edge-triggered handler, without risking IRQ storms and other ugly
1768 #ifdef CONFIG_INTR_REMAP
1769 static void ir_irq_migration(struct work_struct
*work
);
1771 static DECLARE_DELAYED_WORK(ir_migration_work
, ir_irq_migration
);
1774 * Migrate the IO-APIC irq in the presence of intr-remapping.
1776 * For edge triggered, irq migration is a simple atomic update(of vector
1777 * and cpu destination) of IRTE and flush the hardware cache.
1779 * For level triggered, we need to modify the io-apic RTE aswell with the update
1780 * vector information, along with modifying IRTE with vector and destination.
1781 * So irq migration for level triggered is little bit more complex compared to
1782 * edge triggered migration. But the good news is, we use the same algorithm
1783 * for level triggered migration as we have today, only difference being,
1784 * we now initiate the irq migration from process context instead of the
1785 * interrupt context.
1787 * In future, when we do a directed EOI (combined with cpu EOI broadcast
1788 * suppression) to the IO-APIC, level triggered irq migration will also be
1789 * as simple as edge triggered migration and we can do the irq migration
1790 * with a simple atomic update to IO-APIC RTE.
1792 static void migrate_ioapic_irq(int irq
, cpumask_t mask
)
1794 struct irq_cfg
*cfg
;
1795 struct irq_desc
*desc
;
1796 cpumask_t tmp
, cleanup_mask
;
1798 int modify_ioapic_rte
;
1800 unsigned long flags
;
1802 cpus_and(tmp
, mask
, cpu_online_map
);
1803 if (cpus_empty(tmp
))
1806 if (get_irte(irq
, &irte
))
1809 if (assign_irq_vector(irq
, mask
))
1813 cpus_and(tmp
, cfg
->domain
, mask
);
1814 dest
= cpu_mask_to_apicid(tmp
);
1816 desc
= irq_to_desc(irq
);
1817 modify_ioapic_rte
= desc
->status
& IRQ_LEVEL
;
1818 if (modify_ioapic_rte
) {
1819 spin_lock_irqsave(&ioapic_lock
, flags
);
1820 __target_IO_APIC_irq(irq
, dest
, cfg
->vector
);
1821 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1824 irte
.vector
= cfg
->vector
;
1825 irte
.dest_id
= IRTE_DEST(dest
);
1828 * Modified the IRTE and flushes the Interrupt entry cache.
1830 modify_irte(irq
, &irte
);
1832 if (cfg
->move_in_progress
) {
1833 cpus_and(cleanup_mask
, cfg
->old_domain
, cpu_online_map
);
1834 cfg
->move_cleanup_count
= cpus_weight(cleanup_mask
);
1835 send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
1836 cfg
->move_in_progress
= 0;
1839 desc
->affinity
= mask
;
1842 static int migrate_irq_remapped_level(int irq
)
1845 struct irq_desc
*desc
= irq_to_desc(irq
);
1847 mask_IO_APIC_irq(irq
);
1849 if (io_apic_level_ack_pending(irq
)) {
1851 * Interrupt in progress. Migrating irq now will change the
1852 * vector information in the IO-APIC RTE and that will confuse
1853 * the EOI broadcast performed by cpu.
1854 * So, delay the irq migration to the next instance.
1856 schedule_delayed_work(&ir_migration_work
, 1);
1860 /* everthing is clear. we have right of way */
1861 migrate_ioapic_irq(irq
, desc
->pending_mask
);
1864 desc
->status
&= ~IRQ_MOVE_PENDING
;
1865 cpus_clear(desc
->pending_mask
);
1868 unmask_IO_APIC_irq(irq
);
1872 static void ir_irq_migration(struct work_struct
*work
)
1875 struct irq_desc
*desc
;
1877 for_each_irq_desc(irq
, desc
) {
1878 if (desc
->status
& IRQ_MOVE_PENDING
) {
1879 unsigned long flags
;
1881 spin_lock_irqsave(&desc
->lock
, flags
);
1882 if (!desc
->chip
->set_affinity
||
1883 !(desc
->status
& IRQ_MOVE_PENDING
)) {
1884 desc
->status
&= ~IRQ_MOVE_PENDING
;
1885 spin_unlock_irqrestore(&desc
->lock
, flags
);
1889 desc
->chip
->set_affinity(irq
, desc
->pending_mask
);
1890 spin_unlock_irqrestore(&desc
->lock
, flags
);
1896 * Migrates the IRQ destination in the process context.
1898 static void set_ir_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
1900 struct irq_desc
*desc
= irq_to_desc(irq
);
1902 if (desc
->status
& IRQ_LEVEL
) {
1903 desc
->status
|= IRQ_MOVE_PENDING
;
1904 desc
->pending_mask
= mask
;
1905 migrate_irq_remapped_level(irq
);
1909 migrate_ioapic_irq(irq
, mask
);
1913 asmlinkage
void smp_irq_move_cleanup_interrupt(void)
1915 unsigned vector
, me
;
1920 me
= smp_processor_id();
1921 for (vector
= FIRST_EXTERNAL_VECTOR
; vector
< NR_VECTORS
; vector
++) {
1923 struct irq_desc
*desc
;
1924 struct irq_cfg
*cfg
;
1925 irq
= __get_cpu_var(vector_irq
)[vector
];
1929 desc
= irq_to_desc(irq
);
1931 spin_lock(&desc
->lock
);
1932 if (!cfg
->move_cleanup_count
)
1935 if ((vector
== cfg
->vector
) && cpu_isset(me
, cfg
->domain
))
1938 __get_cpu_var(vector_irq
)[vector
] = -1;
1939 cfg
->move_cleanup_count
--;
1941 spin_unlock(&desc
->lock
);
1947 static void irq_complete_move(unsigned int irq
)
1949 struct irq_cfg
*cfg
= irq_cfg(irq
);
1950 unsigned vector
, me
;
1952 if (likely(!cfg
->move_in_progress
))
1955 vector
= ~get_irq_regs()->orig_ax
;
1956 me
= smp_processor_id();
1957 if ((vector
== cfg
->vector
) && cpu_isset(me
, cfg
->domain
)) {
1958 cpumask_t cleanup_mask
;
1960 cpus_and(cleanup_mask
, cfg
->old_domain
, cpu_online_map
);
1961 cfg
->move_cleanup_count
= cpus_weight(cleanup_mask
);
1962 send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
1963 cfg
->move_in_progress
= 0;
1967 static inline void irq_complete_move(unsigned int irq
) {}
1969 #ifdef CONFIG_INTR_REMAP
1970 static void ack_x2apic_level(unsigned int irq
)
1975 static void ack_x2apic_edge(unsigned int irq
)
1981 static void ack_apic_edge(unsigned int irq
)
1983 irq_complete_move(irq
);
1984 move_native_irq(irq
);
1988 static void ack_apic_level(unsigned int irq
)
1990 int do_unmask_irq
= 0;
1992 irq_complete_move(irq
);
1993 #ifdef CONFIG_GENERIC_PENDING_IRQ
1994 /* If we are moving the irq we need to mask it */
1995 if (unlikely(irq_to_desc(irq
)->status
& IRQ_MOVE_PENDING
)) {
1997 mask_IO_APIC_irq(irq
);
2002 * We must acknowledge the irq before we move it or the acknowledge will
2003 * not propagate properly.
2007 /* Now we can move and renable the irq */
2008 if (unlikely(do_unmask_irq
)) {
2009 /* Only migrate the irq if the ack has been received.
2011 * On rare occasions the broadcast level triggered ack gets
2012 * delayed going to ioapics, and if we reprogram the
2013 * vector while Remote IRR is still set the irq will never
2016 * To prevent this scenario we read the Remote IRR bit
2017 * of the ioapic. This has two effects.
2018 * - On any sane system the read of the ioapic will
2019 * flush writes (and acks) going to the ioapic from
2021 * - We get to see if the ACK has actually been delivered.
2023 * Based on failed experiments of reprogramming the
2024 * ioapic entry from outside of irq context starting
2025 * with masking the ioapic entry and then polling until
2026 * Remote IRR was clear before reprogramming the
2027 * ioapic I don't trust the Remote IRR bit to be
2028 * completey accurate.
2030 * However there appears to be no other way to plug
2031 * this race, so if the Remote IRR bit is not
2032 * accurate and is causing problems then it is a hardware bug
2033 * and you can go talk to the chipset vendor about it.
2035 if (!io_apic_level_ack_pending(irq
))
2036 move_masked_irq(irq
);
2037 unmask_IO_APIC_irq(irq
);
2041 static struct irq_chip ioapic_chip __read_mostly
= {
2043 .startup
= startup_ioapic_irq
,
2044 .mask
= mask_IO_APIC_irq
,
2045 .unmask
= unmask_IO_APIC_irq
,
2046 .ack
= ack_apic_edge
,
2047 .eoi
= ack_apic_level
,
2049 .set_affinity
= set_ioapic_affinity_irq
,
2051 .retrigger
= ioapic_retrigger_irq
,
2054 #ifdef CONFIG_INTR_REMAP
2055 static struct irq_chip ir_ioapic_chip __read_mostly
= {
2056 .name
= "IR-IO-APIC",
2057 .startup
= startup_ioapic_irq
,
2058 .mask
= mask_IO_APIC_irq
,
2059 .unmask
= unmask_IO_APIC_irq
,
2060 .ack
= ack_x2apic_edge
,
2061 .eoi
= ack_x2apic_level
,
2063 .set_affinity
= set_ir_ioapic_affinity_irq
,
2065 .retrigger
= ioapic_retrigger_irq
,
2069 static inline void init_IO_APIC_traps(void)
2072 struct irq_desc
*desc
;
2075 * NOTE! The local APIC isn't very good at handling
2076 * multiple interrupts at the same interrupt level.
2077 * As the interrupt level is determined by taking the
2078 * vector number and shifting that right by 4, we
2079 * want to spread these out a bit so that they don't
2080 * all fall in the same interrupt level.
2082 * Also, we've got to be careful not to trash gate
2083 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2085 for (irq
= 0; irq
< nr_irqs
; irq
++) {
2086 struct irq_cfg
*cfg
;
2089 if (IO_APIC_IRQ(irq
) && !cfg
->vector
) {
2091 * Hmm.. We don't have an entry for this,
2092 * so default to an old-fashioned 8259
2093 * interrupt if we can..
2096 make_8259A_irq(irq
);
2098 desc
= irq_to_desc(irq
);
2099 /* Strange. Oh, well.. */
2100 desc
->chip
= &no_irq_chip
;
2106 static void unmask_lapic_irq(unsigned int irq
)
2110 v
= apic_read(APIC_LVT0
);
2111 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
2114 static void mask_lapic_irq(unsigned int irq
)
2118 v
= apic_read(APIC_LVT0
);
2119 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
2122 static void ack_lapic_irq (unsigned int irq
)
2127 static struct irq_chip lapic_chip __read_mostly
= {
2128 .name
= "local-APIC",
2129 .mask
= mask_lapic_irq
,
2130 .unmask
= unmask_lapic_irq
,
2131 .ack
= ack_lapic_irq
,
2134 static void lapic_register_intr(int irq
)
2136 struct irq_desc
*desc
;
2138 desc
= irq_to_desc(irq
);
2139 desc
->status
&= ~IRQ_LEVEL
;
2140 set_irq_chip_and_handler_name(irq
, &lapic_chip
, handle_edge_irq
,
2144 static void __init
setup_nmi(void)
2147 * Dirty trick to enable the NMI watchdog ...
2148 * We put the 8259A master into AEOI mode and
2149 * unmask on all local APICs LVT0 as NMI.
2151 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2152 * is from Maciej W. Rozycki - so we do not have to EOI from
2153 * the NMI handler or the timer interrupt.
2155 printk(KERN_INFO
"activating NMI Watchdog ...");
2157 enable_NMI_through_LVT0();
2163 * This looks a bit hackish but it's about the only one way of sending
2164 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2165 * not support the ExtINT mode, unfortunately. We need to send these
2166 * cycles as some i82489DX-based boards have glue logic that keeps the
2167 * 8259A interrupt line asserted until INTA. --macro
2169 static inline void __init
unlock_ExtINT_logic(void)
2172 struct IO_APIC_route_entry entry0
, entry1
;
2173 unsigned char save_control
, save_freq_select
;
2175 pin
= find_isa_irq_pin(8, mp_INT
);
2176 apic
= find_isa_irq_apic(8, mp_INT
);
2180 entry0
= ioapic_read_entry(apic
, pin
);
2182 clear_IO_APIC_pin(apic
, pin
);
2184 memset(&entry1
, 0, sizeof(entry1
));
2186 entry1
.dest_mode
= 0; /* physical delivery */
2187 entry1
.mask
= 0; /* unmask IRQ now */
2188 entry1
.dest
= hard_smp_processor_id();
2189 entry1
.delivery_mode
= dest_ExtINT
;
2190 entry1
.polarity
= entry0
.polarity
;
2194 ioapic_write_entry(apic
, pin
, entry1
);
2196 save_control
= CMOS_READ(RTC_CONTROL
);
2197 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
2198 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
2200 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
2205 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
2209 CMOS_WRITE(save_control
, RTC_CONTROL
);
2210 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
2211 clear_IO_APIC_pin(apic
, pin
);
2213 ioapic_write_entry(apic
, pin
, entry0
);
2217 * This code may look a bit paranoid, but it's supposed to cooperate with
2218 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2219 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2220 * fanatically on his truly buggy board.
2222 * FIXME: really need to revamp this for modern platforms only.
2224 static inline void __init
check_timer(void)
2226 struct irq_cfg
*cfg
= irq_cfg(0);
2227 int apic1
, pin1
, apic2
, pin2
;
2228 unsigned long flags
;
2231 local_irq_save(flags
);
2234 * get/set the timer IRQ vector:
2236 disable_8259A_irq(0);
2237 assign_irq_vector(0, TARGET_CPUS
);
2240 * As IRQ0 is to be enabled in the 8259A, the virtual
2241 * wire has to be disabled in the local APIC.
2243 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
2246 pin1
= find_isa_irq_pin(0, mp_INT
);
2247 apic1
= find_isa_irq_apic(0, mp_INT
);
2248 pin2
= ioapic_i8259
.pin
;
2249 apic2
= ioapic_i8259
.apic
;
2251 apic_printk(APIC_QUIET
, KERN_INFO
"..TIMER: vector=0x%02X "
2252 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2253 cfg
->vector
, apic1
, pin1
, apic2
, pin2
);
2256 * Some BIOS writers are clueless and report the ExtINTA
2257 * I/O APIC input from the cascaded 8259A as the timer
2258 * interrupt input. So just in case, if only one pin
2259 * was found above, try it both directly and through the
2263 if (intr_remapping_enabled
)
2264 panic("BIOS bug: timer not connected to IO-APIC");
2268 } else if (pin2
== -1) {
2275 * Ok, does IRQ0 through the IOAPIC work?
2278 add_pin_to_irq(0, apic1
, pin1
);
2279 setup_timer_IRQ0_pin(apic1
, pin1
, cfg
->vector
);
2281 unmask_IO_APIC_irq(0);
2282 if (!no_timer_check
&& timer_irq_works()) {
2283 if (nmi_watchdog
== NMI_IO_APIC
) {
2285 enable_8259A_irq(0);
2287 if (disable_timer_pin_1
> 0)
2288 clear_IO_APIC_pin(0, pin1
);
2291 if (intr_remapping_enabled
)
2292 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2293 clear_IO_APIC_pin(apic1
, pin1
);
2295 apic_printk(APIC_QUIET
, KERN_ERR
"..MP-BIOS bug: "
2296 "8254 timer not connected to IO-APIC\n");
2298 apic_printk(APIC_QUIET
, KERN_INFO
"...trying to set up timer "
2299 "(IRQ0) through the 8259A ...\n");
2300 apic_printk(APIC_QUIET
, KERN_INFO
2301 "..... (found apic %d pin %d) ...\n", apic2
, pin2
);
2303 * legacy devices should be connected to IO APIC #0
2305 replace_pin_at_irq(0, apic1
, pin1
, apic2
, pin2
);
2306 setup_timer_IRQ0_pin(apic2
, pin2
, cfg
->vector
);
2307 unmask_IO_APIC_irq(0);
2308 enable_8259A_irq(0);
2309 if (timer_irq_works()) {
2310 apic_printk(APIC_QUIET
, KERN_INFO
"....... works.\n");
2311 timer_through_8259
= 1;
2312 if (nmi_watchdog
== NMI_IO_APIC
) {
2313 disable_8259A_irq(0);
2315 enable_8259A_irq(0);
2320 * Cleanup, just in case ...
2322 disable_8259A_irq(0);
2323 clear_IO_APIC_pin(apic2
, pin2
);
2324 apic_printk(APIC_QUIET
, KERN_INFO
"....... failed.\n");
2327 if (nmi_watchdog
== NMI_IO_APIC
) {
2328 apic_printk(APIC_QUIET
, KERN_WARNING
"timer doesn't work "
2329 "through the IO-APIC - disabling NMI Watchdog!\n");
2330 nmi_watchdog
= NMI_NONE
;
2333 apic_printk(APIC_QUIET
, KERN_INFO
2334 "...trying to set up timer as Virtual Wire IRQ...\n");
2336 lapic_register_intr(0);
2337 apic_write(APIC_LVT0
, APIC_DM_FIXED
| cfg
->vector
); /* Fixed mode */
2338 enable_8259A_irq(0);
2340 if (timer_irq_works()) {
2341 apic_printk(APIC_QUIET
, KERN_INFO
"..... works.\n");
2344 disable_8259A_irq(0);
2345 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| cfg
->vector
);
2346 apic_printk(APIC_QUIET
, KERN_INFO
"..... failed.\n");
2348 apic_printk(APIC_QUIET
, KERN_INFO
2349 "...trying to set up timer as ExtINT IRQ...\n");
2353 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
2355 unlock_ExtINT_logic();
2357 if (timer_irq_works()) {
2358 apic_printk(APIC_QUIET
, KERN_INFO
"..... works.\n");
2361 apic_printk(APIC_QUIET
, KERN_INFO
"..... failed :(.\n");
2362 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2363 "report. Then try booting with the 'noapic' option.\n");
2365 local_irq_restore(flags
);
2368 static int __init
notimercheck(char *s
)
2373 __setup("no_timer_check", notimercheck
);
2376 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2377 * to devices. However there may be an I/O APIC pin available for
2378 * this interrupt regardless. The pin may be left unconnected, but
2379 * typically it will be reused as an ExtINT cascade interrupt for
2380 * the master 8259A. In the MPS case such a pin will normally be
2381 * reported as an ExtINT interrupt in the MP table. With ACPI
2382 * there is no provision for ExtINT interrupts, and in the absence
2383 * of an override it would be treated as an ordinary ISA I/O APIC
2384 * interrupt, that is edge-triggered and unmasked by default. We
2385 * used to do this, but it caused problems on some systems because
2386 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2387 * the same ExtINT cascade interrupt to drive the local APIC of the
2388 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2389 * the I/O APIC in all cases now. No actual device should request
2390 * it anyway. --macro
2392 #define PIC_IRQS (1<<2)
2394 void __init
setup_IO_APIC(void)
2398 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2401 io_apic_irqs
= ~PIC_IRQS
;
2403 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
2406 setup_IO_APIC_irqs();
2407 init_IO_APIC_traps();
2411 struct sysfs_ioapic_data
{
2412 struct sys_device dev
;
2413 struct IO_APIC_route_entry entry
[0];
2415 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
2417 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
2419 struct IO_APIC_route_entry
*entry
;
2420 struct sysfs_ioapic_data
*data
;
2423 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
2424 entry
= data
->entry
;
2425 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++, entry
++ )
2426 *entry
= ioapic_read_entry(dev
->id
, i
);
2431 static int ioapic_resume(struct sys_device
*dev
)
2433 struct IO_APIC_route_entry
*entry
;
2434 struct sysfs_ioapic_data
*data
;
2435 unsigned long flags
;
2436 union IO_APIC_reg_00 reg_00
;
2439 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
2440 entry
= data
->entry
;
2442 spin_lock_irqsave(&ioapic_lock
, flags
);
2443 reg_00
.raw
= io_apic_read(dev
->id
, 0);
2444 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].mp_apicid
) {
2445 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].mp_apicid
;
2446 io_apic_write(dev
->id
, 0, reg_00
.raw
);
2448 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2449 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
2450 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
2455 static struct sysdev_class ioapic_sysdev_class
= {
2457 .suspend
= ioapic_suspend
,
2458 .resume
= ioapic_resume
,
2461 static int __init
ioapic_init_sysfs(void)
2463 struct sys_device
* dev
;
2466 error
= sysdev_class_register(&ioapic_sysdev_class
);
2470 for (i
= 0; i
< nr_ioapics
; i
++ ) {
2471 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
2472 * sizeof(struct IO_APIC_route_entry
);
2473 mp_ioapic_data
[i
] = kzalloc(size
, GFP_KERNEL
);
2474 if (!mp_ioapic_data
[i
]) {
2475 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
2478 dev
= &mp_ioapic_data
[i
]->dev
;
2480 dev
->cls
= &ioapic_sysdev_class
;
2481 error
= sysdev_register(dev
);
2483 kfree(mp_ioapic_data
[i
]);
2484 mp_ioapic_data
[i
] = NULL
;
2485 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
2493 device_initcall(ioapic_init_sysfs
);
2496 * Dynamic irq allocate and deallocation
2498 int create_irq(void)
2500 /* Allocate an unused irq */
2503 unsigned long flags
;
2504 struct irq_cfg
*cfg_new
;
2507 spin_lock_irqsave(&vector_lock
, flags
);
2508 for (new = (nr_irqs
- 1); new >= 0; new--) {
2509 if (platform_legacy_irq(new))
2511 cfg_new
= irq_cfg(new);
2512 if (cfg_new
&& cfg_new
->vector
!= 0)
2514 /* check if need to create one */
2516 cfg_new
= irq_cfg_alloc(new);
2517 if (__assign_irq_vector(new, TARGET_CPUS
) == 0)
2521 spin_unlock_irqrestore(&vector_lock
, flags
);
2524 dynamic_irq_init(irq
);
2529 void destroy_irq(unsigned int irq
)
2531 unsigned long flags
;
2533 dynamic_irq_cleanup(irq
);
2535 #ifdef CONFIG_INTR_REMAP
2538 spin_lock_irqsave(&vector_lock
, flags
);
2539 __clear_irq_vector(irq
);
2540 spin_unlock_irqrestore(&vector_lock
, flags
);
2544 * MSI message composition
2546 #ifdef CONFIG_PCI_MSI
2547 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
2549 struct irq_cfg
*cfg
;
2555 err
= assign_irq_vector(irq
, tmp
);
2560 cpus_and(tmp
, cfg
->domain
, tmp
);
2561 dest
= cpu_mask_to_apicid(tmp
);
2563 #ifdef CONFIG_INTR_REMAP
2564 if (irq_remapped(irq
)) {
2569 ir_index
= map_irq_to_irte_handle(irq
, &sub_handle
);
2570 BUG_ON(ir_index
== -1);
2572 memset (&irte
, 0, sizeof(irte
));
2575 irte
.dst_mode
= INT_DEST_MODE
;
2576 irte
.trigger_mode
= 0; /* edge */
2577 irte
.dlvry_mode
= INT_DELIVERY_MODE
;
2578 irte
.vector
= cfg
->vector
;
2579 irte
.dest_id
= IRTE_DEST(dest
);
2581 modify_irte(irq
, &irte
);
2583 msg
->address_hi
= MSI_ADDR_BASE_HI
;
2584 msg
->data
= sub_handle
;
2585 msg
->address_lo
= MSI_ADDR_BASE_LO
| MSI_ADDR_IR_EXT_INT
|
2587 MSI_ADDR_IR_INDEX1(ir_index
) |
2588 MSI_ADDR_IR_INDEX2(ir_index
);
2592 msg
->address_hi
= MSI_ADDR_BASE_HI
;
2595 ((INT_DEST_MODE
== 0) ?
2596 MSI_ADDR_DEST_MODE_PHYSICAL
:
2597 MSI_ADDR_DEST_MODE_LOGICAL
) |
2598 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2599 MSI_ADDR_REDIRECTION_CPU
:
2600 MSI_ADDR_REDIRECTION_LOWPRI
) |
2601 MSI_ADDR_DEST_ID(dest
);
2604 MSI_DATA_TRIGGER_EDGE
|
2605 MSI_DATA_LEVEL_ASSERT
|
2606 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2607 MSI_DATA_DELIVERY_FIXED
:
2608 MSI_DATA_DELIVERY_LOWPRI
) |
2609 MSI_DATA_VECTOR(cfg
->vector
);
2615 static void set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
2617 struct irq_cfg
*cfg
;
2621 struct irq_desc
*desc
;
2623 cpus_and(tmp
, mask
, cpu_online_map
);
2624 if (cpus_empty(tmp
))
2627 if (assign_irq_vector(irq
, mask
))
2631 cpus_and(tmp
, cfg
->domain
, mask
);
2632 dest
= cpu_mask_to_apicid(tmp
);
2634 read_msi_msg(irq
, &msg
);
2636 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
2637 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
2638 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
2639 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
2641 write_msi_msg(irq
, &msg
);
2642 desc
= irq_to_desc(irq
);
2643 desc
->affinity
= mask
;
2646 #ifdef CONFIG_INTR_REMAP
2648 * Migrate the MSI irq to another cpumask. This migration is
2649 * done in the process context using interrupt-remapping hardware.
2651 static void ir_set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
2653 struct irq_cfg
*cfg
;
2655 cpumask_t tmp
, cleanup_mask
;
2657 struct irq_desc
*desc
;
2659 cpus_and(tmp
, mask
, cpu_online_map
);
2660 if (cpus_empty(tmp
))
2663 if (get_irte(irq
, &irte
))
2666 if (assign_irq_vector(irq
, mask
))
2670 cpus_and(tmp
, cfg
->domain
, mask
);
2671 dest
= cpu_mask_to_apicid(tmp
);
2673 irte
.vector
= cfg
->vector
;
2674 irte
.dest_id
= IRTE_DEST(dest
);
2677 * atomically update the IRTE with the new destination and vector.
2679 modify_irte(irq
, &irte
);
2682 * After this point, all the interrupts will start arriving
2683 * at the new destination. So, time to cleanup the previous
2684 * vector allocation.
2686 if (cfg
->move_in_progress
) {
2687 cpus_and(cleanup_mask
, cfg
->old_domain
, cpu_online_map
);
2688 cfg
->move_cleanup_count
= cpus_weight(cleanup_mask
);
2689 send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
2690 cfg
->move_in_progress
= 0;
2693 desc
= irq_to_desc(irq
);
2694 desc
->affinity
= mask
;
2697 #endif /* CONFIG_SMP */
2700 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2701 * which implement the MSI or MSI-X Capability Structure.
2703 static struct irq_chip msi_chip
= {
2705 .unmask
= unmask_msi_irq
,
2706 .mask
= mask_msi_irq
,
2707 .ack
= ack_apic_edge
,
2709 .set_affinity
= set_msi_irq_affinity
,
2711 .retrigger
= ioapic_retrigger_irq
,
2714 #ifdef CONFIG_INTR_REMAP
2715 static struct irq_chip msi_ir_chip
= {
2716 .name
= "IR-PCI-MSI",
2717 .unmask
= unmask_msi_irq
,
2718 .mask
= mask_msi_irq
,
2719 .ack
= ack_x2apic_edge
,
2721 .set_affinity
= ir_set_msi_irq_affinity
,
2723 .retrigger
= ioapic_retrigger_irq
,
2727 * Map the PCI dev to the corresponding remapping hardware unit
2728 * and allocate 'nvec' consecutive interrupt-remapping table entries
2731 static int msi_alloc_irte(struct pci_dev
*dev
, int irq
, int nvec
)
2733 struct intel_iommu
*iommu
;
2736 iommu
= map_dev_to_ir(dev
);
2739 "Unable to map PCI %s to iommu\n", pci_name(dev
));
2743 index
= alloc_irte(iommu
, irq
, nvec
);
2746 "Unable to allocate %d IRTE for PCI %s\n", nvec
,
2754 static int setup_msi_irq(struct pci_dev
*dev
, struct msi_desc
*desc
, int irq
)
2759 ret
= msi_compose_msg(dev
, irq
, &msg
);
2763 set_irq_msi(irq
, desc
);
2764 write_msi_msg(irq
, &msg
);
2766 #ifdef CONFIG_INTR_REMAP
2767 if (irq_remapped(irq
)) {
2768 struct irq_desc
*desc
= irq_to_desc(irq
);
2770 * irq migration in process context
2772 desc
->status
|= IRQ_MOVE_PCNTXT
;
2773 set_irq_chip_and_handler_name(irq
, &msi_ir_chip
, handle_edge_irq
, "edge");
2776 set_irq_chip_and_handler_name(irq
, &msi_chip
, handle_edge_irq
, "edge");
2781 int arch_setup_msi_irq(struct pci_dev
*dev
, struct msi_desc
*desc
)
2789 #ifdef CONFIG_INTR_REMAP
2790 if (!intr_remapping_enabled
)
2793 ret
= msi_alloc_irte(dev
, irq
, 1);
2798 ret
= setup_msi_irq(dev
, desc
, irq
);
2805 #ifdef CONFIG_INTR_REMAP
2812 int arch_setup_msi_irqs(struct pci_dev
*dev
, int nvec
, int type
)
2814 int irq
, ret
, sub_handle
;
2815 struct msi_desc
*desc
;
2816 #ifdef CONFIG_INTR_REMAP
2817 struct intel_iommu
*iommu
= 0;
2822 list_for_each_entry(desc
, &dev
->msi_list
, list
) {
2826 #ifdef CONFIG_INTR_REMAP
2827 if (!intr_remapping_enabled
)
2832 * allocate the consecutive block of IRTE's
2835 index
= msi_alloc_irte(dev
, irq
, nvec
);
2841 iommu
= map_dev_to_ir(dev
);
2847 * setup the mapping between the irq and the IRTE
2848 * base index, the sub_handle pointing to the
2849 * appropriate interrupt remap table entry.
2851 set_irte_irq(irq
, iommu
, index
, sub_handle
);
2855 ret
= setup_msi_irq(dev
, desc
, irq
);
2867 void arch_teardown_msi_irq(unsigned int irq
)
2874 static void dmar_msi_set_affinity(unsigned int irq
, cpumask_t mask
)
2876 struct irq_cfg
*cfg
;
2880 struct irq_desc
*desc
;
2882 cpus_and(tmp
, mask
, cpu_online_map
);
2883 if (cpus_empty(tmp
))
2886 if (assign_irq_vector(irq
, mask
))
2890 cpus_and(tmp
, cfg
->domain
, mask
);
2891 dest
= cpu_mask_to_apicid(tmp
);
2893 dmar_msi_read(irq
, &msg
);
2895 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
2896 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
2897 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
2898 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
2900 dmar_msi_write(irq
, &msg
);
2901 desc
= irq_to_desc(irq
);
2902 desc
->affinity
= mask
;
2904 #endif /* CONFIG_SMP */
2906 struct irq_chip dmar_msi_type
= {
2908 .unmask
= dmar_msi_unmask
,
2909 .mask
= dmar_msi_mask
,
2910 .ack
= ack_apic_edge
,
2912 .set_affinity
= dmar_msi_set_affinity
,
2914 .retrigger
= ioapic_retrigger_irq
,
2917 int arch_setup_dmar_msi(unsigned int irq
)
2922 ret
= msi_compose_msg(NULL
, irq
, &msg
);
2925 dmar_msi_write(irq
, &msg
);
2926 set_irq_chip_and_handler_name(irq
, &dmar_msi_type
, handle_edge_irq
,
2932 #endif /* CONFIG_PCI_MSI */
2934 * Hypertransport interrupt support
2936 #ifdef CONFIG_HT_IRQ
2940 static void target_ht_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
2942 struct ht_irq_msg msg
;
2943 fetch_ht_irq_msg(irq
, &msg
);
2945 msg
.address_lo
&= ~(HT_IRQ_LOW_VECTOR_MASK
| HT_IRQ_LOW_DEST_ID_MASK
);
2946 msg
.address_hi
&= ~(HT_IRQ_HIGH_DEST_ID_MASK
);
2948 msg
.address_lo
|= HT_IRQ_LOW_VECTOR(vector
) | HT_IRQ_LOW_DEST_ID(dest
);
2949 msg
.address_hi
|= HT_IRQ_HIGH_DEST_ID(dest
);
2951 write_ht_irq_msg(irq
, &msg
);
2954 static void set_ht_irq_affinity(unsigned int irq
, cpumask_t mask
)
2956 struct irq_cfg
*cfg
;
2959 struct irq_desc
*desc
;
2961 cpus_and(tmp
, mask
, cpu_online_map
);
2962 if (cpus_empty(tmp
))
2965 if (assign_irq_vector(irq
, mask
))
2969 cpus_and(tmp
, cfg
->domain
, mask
);
2970 dest
= cpu_mask_to_apicid(tmp
);
2972 target_ht_irq(irq
, dest
, cfg
->vector
);
2973 desc
= irq_to_desc(irq
);
2974 desc
->affinity
= mask
;
2978 static struct irq_chip ht_irq_chip
= {
2980 .mask
= mask_ht_irq
,
2981 .unmask
= unmask_ht_irq
,
2982 .ack
= ack_apic_edge
,
2984 .set_affinity
= set_ht_irq_affinity
,
2986 .retrigger
= ioapic_retrigger_irq
,
2989 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
2991 struct irq_cfg
*cfg
;
2996 err
= assign_irq_vector(irq
, tmp
);
2998 struct ht_irq_msg msg
;
3002 cpus_and(tmp
, cfg
->domain
, tmp
);
3003 dest
= cpu_mask_to_apicid(tmp
);
3005 msg
.address_hi
= HT_IRQ_HIGH_DEST_ID(dest
);
3009 HT_IRQ_LOW_DEST_ID(dest
) |
3010 HT_IRQ_LOW_VECTOR(cfg
->vector
) |
3011 ((INT_DEST_MODE
== 0) ?
3012 HT_IRQ_LOW_DM_PHYSICAL
:
3013 HT_IRQ_LOW_DM_LOGICAL
) |
3014 HT_IRQ_LOW_RQEOI_EDGE
|
3015 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
3016 HT_IRQ_LOW_MT_FIXED
:
3017 HT_IRQ_LOW_MT_ARBITRATED
) |
3018 HT_IRQ_LOW_IRQ_MASKED
;
3020 write_ht_irq_msg(irq
, &msg
);
3022 set_irq_chip_and_handler_name(irq
, &ht_irq_chip
,
3023 handle_edge_irq
, "edge");
3027 #endif /* CONFIG_HT_IRQ */
3029 /* --------------------------------------------------------------------------
3030 ACPI-based IOAPIC Configuration
3031 -------------------------------------------------------------------------- */
3035 #define IO_APIC_MAX_ID 0xFE
3037 int __init
io_apic_get_redir_entries (int ioapic
)
3039 union IO_APIC_reg_01 reg_01
;
3040 unsigned long flags
;
3042 spin_lock_irqsave(&ioapic_lock
, flags
);
3043 reg_01
.raw
= io_apic_read(ioapic
, 1);
3044 spin_unlock_irqrestore(&ioapic_lock
, flags
);
3046 return reg_01
.bits
.entries
;
3050 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int triggering
, int polarity
)
3052 if (!IO_APIC_IRQ(irq
)) {
3053 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
3059 * IRQs < 16 are already in the irq_2_pin[] map
3062 add_pin_to_irq(irq
, ioapic
, pin
);
3064 setup_IO_APIC_irq(ioapic
, pin
, irq
, triggering
, polarity
);
3070 int acpi_get_override_irq(int bus_irq
, int *trigger
, int *polarity
)
3074 if (skip_ioapic_setup
)
3077 for (i
= 0; i
< mp_irq_entries
; i
++)
3078 if (mp_irqs
[i
].mp_irqtype
== mp_INT
&&
3079 mp_irqs
[i
].mp_srcbusirq
== bus_irq
)
3081 if (i
>= mp_irq_entries
)
3084 *trigger
= irq_trigger(i
);
3085 *polarity
= irq_polarity(i
);
3089 #endif /* CONFIG_ACPI */
3092 * This function currently is only a helper for the i386 smp boot process where
3093 * we need to reprogram the ioredtbls to cater for the cpus which have come online
3094 * so mask in all cases should simply be TARGET_CPUS
3097 void __init
setup_ioapic_dest(void)
3099 int pin
, ioapic
, irq
, irq_entry
;
3100 struct irq_cfg
*cfg
;
3102 if (skip_ioapic_setup
== 1)
3105 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++) {
3106 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
3107 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
3108 if (irq_entry
== -1)
3110 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
3112 /* setup_IO_APIC_irqs could fail to get vector for some device
3113 * when you have too many devices, because at that time only boot
3118 setup_IO_APIC_irq(ioapic
, pin
, irq
,
3119 irq_trigger(irq_entry
),
3120 irq_polarity(irq_entry
));
3121 #ifdef CONFIG_INTR_REMAP
3122 else if (intr_remapping_enabled
)
3123 set_ir_ioapic_affinity_irq(irq
, TARGET_CPUS
);
3126 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);
3133 #define IOAPIC_RESOURCE_NAME_SIZE 11
3135 static struct resource
*ioapic_resources
;
3137 static struct resource
* __init
ioapic_setup_resources(void)
3140 struct resource
*res
;
3144 if (nr_ioapics
<= 0)
3147 n
= IOAPIC_RESOURCE_NAME_SIZE
+ sizeof(struct resource
);
3150 mem
= alloc_bootmem(n
);
3154 mem
+= sizeof(struct resource
) * nr_ioapics
;
3156 for (i
= 0; i
< nr_ioapics
; i
++) {
3158 res
[i
].flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
3159 sprintf(mem
, "IOAPIC %u", i
);
3160 mem
+= IOAPIC_RESOURCE_NAME_SIZE
;
3164 ioapic_resources
= res
;
3169 void __init
ioapic_init_mappings(void)
3171 unsigned long ioapic_phys
, idx
= FIX_IO_APIC_BASE_0
;
3172 struct resource
*ioapic_res
;
3175 ioapic_res
= ioapic_setup_resources();
3176 for (i
= 0; i
< nr_ioapics
; i
++) {
3177 if (smp_found_config
) {
3178 ioapic_phys
= mp_ioapics
[i
].mp_apicaddr
;
3180 ioapic_phys
= (unsigned long)
3181 alloc_bootmem_pages(PAGE_SIZE
);
3182 ioapic_phys
= __pa(ioapic_phys
);
3184 set_fixmap_nocache(idx
, ioapic_phys
);
3185 apic_printk(APIC_VERBOSE
,
3186 "mapped IOAPIC to %016lx (%016lx)\n",
3187 __fix_to_virt(idx
), ioapic_phys
);
3190 if (ioapic_res
!= NULL
) {
3191 ioapic_res
->start
= ioapic_phys
;
3192 ioapic_res
->end
= ioapic_phys
+ (4 * 1024) - 1;
3198 static int __init
ioapic_insert_resources(void)
3201 struct resource
*r
= ioapic_resources
;
3205 "IO APIC resources could be not be allocated.\n");
3209 for (i
= 0; i
< nr_ioapics
; i
++) {
3210 insert_resource(&iomem_resource
, r
);
3217 /* Insert the IO APIC resources after PCI initialization has occured to handle
3218 * IO APICS that are mapped in on a BAR in PCI space. */
3219 late_initcall(ioapic_insert_resources
);