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
63 unsigned move_cleanup_count
;
65 u8 move_in_progress
: 1;
68 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
69 static struct irq_cfg irq_cfg_legacy
[] __initdata
= {
70 [0] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ0_VECTOR
, },
71 [1] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ1_VECTOR
, },
72 [2] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ2_VECTOR
, },
73 [3] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ3_VECTOR
, },
74 [4] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ4_VECTOR
, },
75 [5] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ5_VECTOR
, },
76 [6] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ6_VECTOR
, },
77 [7] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ7_VECTOR
, },
78 [8] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ8_VECTOR
, },
79 [9] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ9_VECTOR
, },
80 [10] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ10_VECTOR
, },
81 [11] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ11_VECTOR
, },
82 [12] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ12_VECTOR
, },
83 [13] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ13_VECTOR
, },
84 [14] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ14_VECTOR
, },
85 [15] = { .domain
= CPU_MASK_ALL
, .vector
= IRQ15_VECTOR
, },
88 static struct irq_cfg
*irq_cfg
;
90 static void __init
init_work(void *data
)
92 struct dyn_array
*da
= data
;
94 memcpy(*da
->name
, irq_cfg_legacy
, sizeof(irq_cfg_legacy
));
97 DEFINE_DYN_ARRAY(irq_cfg
, sizeof(struct irq_cfg
), nr_irqs
, PAGE_SIZE
, init_work
);
99 static int assign_irq_vector(int irq
, cpumask_t mask
);
101 int first_system_vector
= 0xfe;
103 char system_vectors
[NR_VECTORS
] = { [0 ... NR_VECTORS
-1] = SYS_VECTOR_FREE
};
105 int sis_apic_bug
; /* not actually supported, dummy for compile */
107 static int no_timer_check
;
109 static int disable_timer_pin_1 __initdata
;
111 int timer_through_8259 __initdata
;
113 /* Where if anywhere is the i8259 connect in external int mode */
114 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
116 static DEFINE_SPINLOCK(ioapic_lock
);
117 static DEFINE_SPINLOCK(vector_lock
);
120 * # of IRQ routing registers
122 int nr_ioapic_registers
[MAX_IO_APICS
];
124 /* I/O APIC RTE contents at the OS boot up */
125 struct IO_APIC_route_entry
*early_ioapic_entries
[MAX_IO_APICS
];
127 /* I/O APIC entries */
128 struct mp_config_ioapic mp_ioapics
[MAX_IO_APICS
];
131 /* MP IRQ source entries */
132 struct mp_config_intsrc mp_irqs
[MAX_IRQ_SOURCES
];
134 /* # of MP IRQ source entries */
137 DECLARE_BITMAP(mp_bus_not_pci
, MAX_MP_BUSSES
);
140 * Rough estimation of how many shared IRQs there are, can
141 * be changed anytime.
147 * This is performance-critical, we want to do it O(1)
149 * the indexing order of this array favors 1:1 mappings
150 * between pins and IRQs.
153 static struct irq_pin_list
{
158 DEFINE_DYN_ARRAY(irq_2_pin
, sizeof(struct irq_pin_list
), pin_map_size
, sizeof(struct irq_pin_list
), NULL
);
163 unsigned int unused
[3];
167 static __attribute_const__
struct io_apic __iomem
*io_apic_base(int idx
)
169 return (void __iomem
*) __fix_to_virt(FIX_IO_APIC_BASE_0
+ idx
)
170 + (mp_ioapics
[idx
].mp_apicaddr
& ~PAGE_MASK
);
173 static inline unsigned int io_apic_read(unsigned int apic
, unsigned int reg
)
175 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
176 writel(reg
, &io_apic
->index
);
177 return readl(&io_apic
->data
);
180 static inline void io_apic_write(unsigned int apic
, unsigned int reg
, unsigned int value
)
182 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
183 writel(reg
, &io_apic
->index
);
184 writel(value
, &io_apic
->data
);
188 * Re-write a value: to be used for read-modify-write
189 * cycles where the read already set up the index register.
191 static inline void io_apic_modify(unsigned int apic
, unsigned int value
)
193 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
194 writel(value
, &io_apic
->data
);
197 static bool io_apic_level_ack_pending(unsigned int irq
)
199 struct irq_pin_list
*entry
;
202 spin_lock_irqsave(&ioapic_lock
, flags
);
203 entry
= irq_2_pin
+ irq
;
211 reg
= io_apic_read(entry
->apic
, 0x10 + pin
*2);
212 /* Is the remote IRR bit set? */
213 if (reg
& IO_APIC_REDIR_REMOTE_IRR
) {
214 spin_unlock_irqrestore(&ioapic_lock
, flags
);
219 entry
= irq_2_pin
+ entry
->next
;
221 spin_unlock_irqrestore(&ioapic_lock
, flags
);
227 * Synchronize the IO-APIC and the CPU by doing
228 * a dummy read from the IO-APIC
230 static inline void io_apic_sync(unsigned int apic
)
232 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
233 readl(&io_apic
->data
);
236 #define __DO_ACTION(R, ACTION, FINAL) \
240 struct irq_pin_list *entry = irq_2_pin + irq; \
242 BUG_ON(irq >= nr_irqs); \
248 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
250 io_apic_modify(entry->apic, reg); \
254 entry = irq_2_pin + entry->next; \
259 struct { u32 w1
, w2
; };
260 struct IO_APIC_route_entry entry
;
263 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
265 union entry_union eu
;
267 spin_lock_irqsave(&ioapic_lock
, flags
);
268 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
269 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
270 spin_unlock_irqrestore(&ioapic_lock
, flags
);
275 * When we write a new IO APIC routing entry, we need to write the high
276 * word first! If the mask bit in the low word is clear, we will enable
277 * the interrupt, and we need to make sure the entry is fully populated
278 * before that happens.
281 __ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
283 union entry_union eu
;
285 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
286 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
289 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
292 spin_lock_irqsave(&ioapic_lock
, flags
);
293 __ioapic_write_entry(apic
, pin
, e
);
294 spin_unlock_irqrestore(&ioapic_lock
, flags
);
298 * When we mask an IO APIC routing entry, we need to write the low
299 * word first, in order to set the mask bit before we change the
302 static void ioapic_mask_entry(int apic
, int pin
)
305 union entry_union eu
= { .entry
.mask
= 1 };
307 spin_lock_irqsave(&ioapic_lock
, flags
);
308 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
309 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
310 spin_unlock_irqrestore(&ioapic_lock
, flags
);
314 static void __target_IO_APIC_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
317 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
319 BUG_ON(irq
>= nr_irqs
);
327 * With interrupt-remapping, destination information comes
328 * from interrupt-remapping table entry.
330 if (!irq_remapped(irq
))
331 io_apic_write(apic
, 0x11 + pin
*2, dest
);
332 reg
= io_apic_read(apic
, 0x10 + pin
*2);
333 reg
&= ~IO_APIC_REDIR_VECTOR_MASK
;
335 io_apic_modify(apic
, reg
);
338 entry
= irq_2_pin
+ entry
->next
;
342 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
344 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
349 cpus_and(tmp
, mask
, cpu_online_map
);
353 if (assign_irq_vector(irq
, mask
))
356 cpus_and(tmp
, cfg
->domain
, mask
);
357 dest
= cpu_mask_to_apicid(tmp
);
360 * Only the high 8 bits are valid.
362 dest
= SET_APIC_LOGICAL_ID(dest
);
364 spin_lock_irqsave(&ioapic_lock
, flags
);
365 __target_IO_APIC_irq(irq
, dest
, cfg
->vector
);
366 irq_desc
[irq
].affinity
= mask
;
367 spin_unlock_irqrestore(&ioapic_lock
, flags
);
372 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
373 * shared ISA-space IRQs, so we have to support them. We are super
374 * fast in the common case, and fast for shared ISA-space IRQs.
376 int first_free_entry
;
377 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
379 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
381 BUG_ON(irq
>= nr_irqs
);
383 entry
= irq_2_pin
+ entry
->next
;
385 if (entry
->pin
!= -1) {
386 entry
->next
= first_free_entry
;
387 entry
= irq_2_pin
+ entry
->next
;
388 if (++first_free_entry
>= pin_map_size
)
389 panic("io_apic.c: ran out of irq_2_pin entries!");
396 * Reroute an IRQ to a different pin.
398 static void __init
replace_pin_at_irq(unsigned int irq
,
399 int oldapic
, int oldpin
,
400 int newapic
, int newpin
)
402 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
405 if (entry
->apic
== oldapic
&& entry
->pin
== oldpin
) {
406 entry
->apic
= newapic
;
411 entry
= irq_2_pin
+ entry
->next
;
416 #define DO_ACTION(name,R,ACTION, FINAL) \
418 static void name##_IO_APIC_irq (unsigned int irq) \
419 __DO_ACTION(R, ACTION, FINAL)
422 DO_ACTION(__mask
, 0, |= IO_APIC_REDIR_MASKED
, io_apic_sync(entry
->apic
))
425 DO_ACTION(__unmask
, 0, &= ~IO_APIC_REDIR_MASKED
, )
427 static void mask_IO_APIC_irq (unsigned int irq
)
431 spin_lock_irqsave(&ioapic_lock
, flags
);
432 __mask_IO_APIC_irq(irq
);
433 spin_unlock_irqrestore(&ioapic_lock
, flags
);
436 static void unmask_IO_APIC_irq (unsigned int irq
)
440 spin_lock_irqsave(&ioapic_lock
, flags
);
441 __unmask_IO_APIC_irq(irq
);
442 spin_unlock_irqrestore(&ioapic_lock
, flags
);
445 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
447 struct IO_APIC_route_entry entry
;
449 /* Check delivery_mode to be sure we're not clearing an SMI pin */
450 entry
= ioapic_read_entry(apic
, pin
);
451 if (entry
.delivery_mode
== dest_SMI
)
454 * Disable it in the IO-APIC irq-routing table:
456 ioapic_mask_entry(apic
, pin
);
459 static void clear_IO_APIC (void)
463 for (apic
= 0; apic
< nr_ioapics
; apic
++)
464 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
465 clear_IO_APIC_pin(apic
, pin
);
469 * Saves and masks all the unmasked IO-APIC RTE's
471 int save_mask_IO_APIC_setup(void)
473 union IO_APIC_reg_01 reg_01
;
478 * The number of IO-APIC IRQ registers (== #pins):
480 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
481 spin_lock_irqsave(&ioapic_lock
, flags
);
482 reg_01
.raw
= io_apic_read(apic
, 1);
483 spin_unlock_irqrestore(&ioapic_lock
, flags
);
484 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
487 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
488 early_ioapic_entries
[apic
] =
489 kzalloc(sizeof(struct IO_APIC_route_entry
) *
490 nr_ioapic_registers
[apic
], GFP_KERNEL
);
491 if (!early_ioapic_entries
[apic
])
495 for (apic
= 0; apic
< nr_ioapics
; apic
++)
496 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
497 struct IO_APIC_route_entry entry
;
499 entry
= early_ioapic_entries
[apic
][pin
] =
500 ioapic_read_entry(apic
, pin
);
503 ioapic_write_entry(apic
, pin
, entry
);
509 void restore_IO_APIC_setup(void)
513 for (apic
= 0; apic
< nr_ioapics
; apic
++)
514 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
515 ioapic_write_entry(apic
, pin
,
516 early_ioapic_entries
[apic
][pin
]);
519 void reinit_intr_remapped_IO_APIC(int intr_remapping
)
522 * for now plain restore of previous settings.
523 * TBD: In the case of OS enabling interrupt-remapping,
524 * IO-APIC RTE's need to be setup to point to interrupt-remapping
525 * table entries. for now, do a plain restore, and wait for
526 * the setup_IO_APIC_irqs() to do proper initialization.
528 restore_IO_APIC_setup();
531 int skip_ioapic_setup
;
534 static int __init
parse_noapic(char *str
)
536 disable_ioapic_setup();
539 early_param("noapic", parse_noapic
);
541 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
542 static int __init
disable_timer_pin_setup(char *arg
)
544 disable_timer_pin_1
= 1;
547 __setup("disable_timer_pin_1", disable_timer_pin_setup
);
551 * Find the IRQ entry number of a certain pin.
553 static int find_irq_entry(int apic
, int pin
, int type
)
557 for (i
= 0; i
< mp_irq_entries
; i
++)
558 if (mp_irqs
[i
].mp_irqtype
== type
&&
559 (mp_irqs
[i
].mp_dstapic
== mp_ioapics
[apic
].mp_apicid
||
560 mp_irqs
[i
].mp_dstapic
== MP_APIC_ALL
) &&
561 mp_irqs
[i
].mp_dstirq
== pin
)
568 * Find the pin to which IRQ[irq] (ISA) is connected
570 static int __init
find_isa_irq_pin(int irq
, int type
)
574 for (i
= 0; i
< mp_irq_entries
; i
++) {
575 int lbus
= mp_irqs
[i
].mp_srcbus
;
577 if (test_bit(lbus
, mp_bus_not_pci
) &&
578 (mp_irqs
[i
].mp_irqtype
== type
) &&
579 (mp_irqs
[i
].mp_srcbusirq
== irq
))
581 return mp_irqs
[i
].mp_dstirq
;
586 static int __init
find_isa_irq_apic(int irq
, int type
)
590 for (i
= 0; i
< mp_irq_entries
; i
++) {
591 int lbus
= mp_irqs
[i
].mp_srcbus
;
593 if (test_bit(lbus
, mp_bus_not_pci
) &&
594 (mp_irqs
[i
].mp_irqtype
== type
) &&
595 (mp_irqs
[i
].mp_srcbusirq
== irq
))
598 if (i
< mp_irq_entries
) {
600 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
601 if (mp_ioapics
[apic
].mp_apicid
== mp_irqs
[i
].mp_dstapic
)
610 * Find a specific PCI IRQ entry.
611 * Not an __init, possibly needed by modules
613 static int pin_2_irq(int idx
, int apic
, int pin
);
615 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
617 int apic
, i
, best_guess
= -1;
619 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
621 if (test_bit(bus
, mp_bus_not_pci
)) {
622 apic_printk(APIC_VERBOSE
, "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
625 for (i
= 0; i
< mp_irq_entries
; i
++) {
626 int lbus
= mp_irqs
[i
].mp_srcbus
;
628 for (apic
= 0; apic
< nr_ioapics
; apic
++)
629 if (mp_ioapics
[apic
].mp_apicid
== mp_irqs
[i
].mp_dstapic
||
630 mp_irqs
[i
].mp_dstapic
== MP_APIC_ALL
)
633 if (!test_bit(lbus
, mp_bus_not_pci
) &&
634 !mp_irqs
[i
].mp_irqtype
&&
636 (slot
== ((mp_irqs
[i
].mp_srcbusirq
>> 2) & 0x1f))) {
637 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mp_dstirq
);
639 if (!(apic
|| IO_APIC_IRQ(irq
)))
642 if (pin
== (mp_irqs
[i
].mp_srcbusirq
& 3))
645 * Use the first all-but-pin matching entry as a
646 * best-guess fuzzy result for broken mptables.
652 BUG_ON(best_guess
>= nr_irqs
);
656 /* ISA interrupts are always polarity zero edge triggered,
657 * when listed as conforming in the MP table. */
659 #define default_ISA_trigger(idx) (0)
660 #define default_ISA_polarity(idx) (0)
662 /* PCI interrupts are always polarity one level triggered,
663 * when listed as conforming in the MP table. */
665 #define default_PCI_trigger(idx) (1)
666 #define default_PCI_polarity(idx) (1)
668 static int MPBIOS_polarity(int idx
)
670 int bus
= mp_irqs
[idx
].mp_srcbus
;
674 * Determine IRQ line polarity (high active or low active):
676 switch (mp_irqs
[idx
].mp_irqflag
& 3)
678 case 0: /* conforms, ie. bus-type dependent polarity */
679 if (test_bit(bus
, mp_bus_not_pci
))
680 polarity
= default_ISA_polarity(idx
);
682 polarity
= default_PCI_polarity(idx
);
684 case 1: /* high active */
689 case 2: /* reserved */
691 printk(KERN_WARNING
"broken BIOS!!\n");
695 case 3: /* low active */
700 default: /* invalid */
702 printk(KERN_WARNING
"broken BIOS!!\n");
710 static int MPBIOS_trigger(int idx
)
712 int bus
= mp_irqs
[idx
].mp_srcbus
;
716 * Determine IRQ trigger mode (edge or level sensitive):
718 switch ((mp_irqs
[idx
].mp_irqflag
>>2) & 3)
720 case 0: /* conforms, ie. bus-type dependent */
721 if (test_bit(bus
, mp_bus_not_pci
))
722 trigger
= default_ISA_trigger(idx
);
724 trigger
= default_PCI_trigger(idx
);
731 case 2: /* reserved */
733 printk(KERN_WARNING
"broken BIOS!!\n");
742 default: /* invalid */
744 printk(KERN_WARNING
"broken BIOS!!\n");
752 static inline int irq_polarity(int idx
)
754 return MPBIOS_polarity(idx
);
757 static inline int irq_trigger(int idx
)
759 return MPBIOS_trigger(idx
);
762 static int pin_2_irq(int idx
, int apic
, int pin
)
765 int bus
= mp_irqs
[idx
].mp_srcbus
;
768 * Debugging check, we are in big trouble if this message pops up!
770 if (mp_irqs
[idx
].mp_dstirq
!= pin
)
771 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
773 if (test_bit(bus
, mp_bus_not_pci
)) {
774 irq
= mp_irqs
[idx
].mp_srcbusirq
;
777 * PCI IRQs are mapped in order
781 irq
+= nr_ioapic_registers
[i
++];
784 BUG_ON(irq
>= nr_irqs
);
788 void lock_vector_lock(void)
790 /* Used to the online set of cpus does not change
791 * during assign_irq_vector.
793 spin_lock(&vector_lock
);
796 void unlock_vector_lock(void)
798 spin_unlock(&vector_lock
);
801 static int __assign_irq_vector(int irq
, cpumask_t mask
)
804 * NOTE! The local APIC isn't very good at handling
805 * multiple interrupts at the same interrupt level.
806 * As the interrupt level is determined by taking the
807 * vector number and shifting that right by 4, we
808 * want to spread these out a bit so that they don't
809 * all fall in the same interrupt level.
811 * Also, we've got to be careful not to trash gate
812 * 0x80, because int 0x80 is hm, kind of importantish. ;)
814 static int current_vector
= FIRST_DEVICE_VECTOR
, current_offset
= 0;
815 unsigned int old_vector
;
819 BUG_ON((unsigned)irq
>= nr_irqs
);
822 /* Only try and allocate irqs on cpus that are present */
823 cpus_and(mask
, mask
, cpu_online_map
);
825 if ((cfg
->move_in_progress
) || cfg
->move_cleanup_count
)
828 old_vector
= cfg
->vector
;
831 cpus_and(tmp
, cfg
->domain
, mask
);
832 if (!cpus_empty(tmp
))
836 for_each_cpu_mask_nr(cpu
, mask
) {
837 cpumask_t domain
, new_mask
;
841 domain
= vector_allocation_domain(cpu
);
842 cpus_and(new_mask
, domain
, cpu_online_map
);
844 vector
= current_vector
;
845 offset
= current_offset
;
848 if (vector
>= first_system_vector
) {
849 /* If we run out of vectors on large boxen, must share them. */
850 offset
= (offset
+ 1) % 8;
851 vector
= FIRST_DEVICE_VECTOR
+ offset
;
853 if (unlikely(current_vector
== vector
))
855 if (vector
== IA32_SYSCALL_VECTOR
)
857 for_each_cpu_mask_nr(new_cpu
, new_mask
)
858 if (per_cpu(vector_irq
, new_cpu
)[vector
] != -1)
861 current_vector
= vector
;
862 current_offset
= offset
;
864 cfg
->move_in_progress
= 1;
865 cfg
->old_domain
= cfg
->domain
;
867 for_each_cpu_mask_nr(new_cpu
, new_mask
)
868 per_cpu(vector_irq
, new_cpu
)[vector
] = irq
;
869 cfg
->vector
= vector
;
870 cfg
->domain
= domain
;
876 static int assign_irq_vector(int irq
, cpumask_t mask
)
881 spin_lock_irqsave(&vector_lock
, flags
);
882 err
= __assign_irq_vector(irq
, mask
);
883 spin_unlock_irqrestore(&vector_lock
, flags
);
887 static void __clear_irq_vector(int irq
)
893 BUG_ON((unsigned)irq
>= nr_irqs
);
895 BUG_ON(!cfg
->vector
);
897 vector
= cfg
->vector
;
898 cpus_and(mask
, cfg
->domain
, cpu_online_map
);
899 for_each_cpu_mask_nr(cpu
, mask
)
900 per_cpu(vector_irq
, cpu
)[vector
] = -1;
903 cpus_clear(cfg
->domain
);
906 void __setup_vector_irq(int cpu
)
908 /* Initialize vector_irq on a new cpu */
909 /* This function must be called with vector_lock held */
912 /* Mark the inuse vectors */
913 for (irq
= 0; irq
< nr_irqs
; ++irq
) {
914 if (!cpu_isset(cpu
, irq_cfg
[irq
].domain
))
916 vector
= irq_cfg
[irq
].vector
;
917 per_cpu(vector_irq
, cpu
)[vector
] = irq
;
919 /* Mark the free vectors */
920 for (vector
= 0; vector
< NR_VECTORS
; ++vector
) {
921 irq
= per_cpu(vector_irq
, cpu
)[vector
];
924 if (!cpu_isset(cpu
, irq_cfg
[irq
].domain
))
925 per_cpu(vector_irq
, cpu
)[vector
] = -1;
929 static struct irq_chip ioapic_chip
;
930 #ifdef CONFIG_INTR_REMAP
931 static struct irq_chip ir_ioapic_chip
;
934 static void ioapic_register_intr(int irq
, unsigned long trigger
)
937 irq_desc
[irq
].status
|= IRQ_LEVEL
;
939 irq_desc
[irq
].status
&= ~IRQ_LEVEL
;
941 #ifdef CONFIG_INTR_REMAP
942 if (irq_remapped(irq
)) {
943 irq_desc
[irq
].status
|= IRQ_MOVE_PCNTXT
;
945 set_irq_chip_and_handler_name(irq
, &ir_ioapic_chip
,
949 set_irq_chip_and_handler_name(irq
, &ir_ioapic_chip
,
950 handle_edge_irq
, "edge");
955 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
959 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
960 handle_edge_irq
, "edge");
963 static int setup_ioapic_entry(int apic
, int irq
,
964 struct IO_APIC_route_entry
*entry
,
965 unsigned int destination
, int trigger
,
966 int polarity
, int vector
)
969 * add it to the IO-APIC irq-routing table:
971 memset(entry
,0,sizeof(*entry
));
973 #ifdef CONFIG_INTR_REMAP
974 if (intr_remapping_enabled
) {
975 struct intel_iommu
*iommu
= map_ioapic_to_ir(apic
);
977 struct IR_IO_APIC_route_entry
*ir_entry
=
978 (struct IR_IO_APIC_route_entry
*) entry
;
982 panic("No mapping iommu for ioapic %d\n", apic
);
984 index
= alloc_irte(iommu
, irq
, 1);
986 panic("Failed to allocate IRTE for ioapic %d\n", apic
);
988 memset(&irte
, 0, sizeof(irte
));
991 irte
.dst_mode
= INT_DEST_MODE
;
992 irte
.trigger_mode
= trigger
;
993 irte
.dlvry_mode
= INT_DELIVERY_MODE
;
994 irte
.vector
= vector
;
995 irte
.dest_id
= IRTE_DEST(destination
);
997 modify_irte(irq
, &irte
);
999 ir_entry
->index2
= (index
>> 15) & 0x1;
1001 ir_entry
->format
= 1;
1002 ir_entry
->index
= (index
& 0x7fff);
1006 entry
->delivery_mode
= INT_DELIVERY_MODE
;
1007 entry
->dest_mode
= INT_DEST_MODE
;
1008 entry
->dest
= destination
;
1011 entry
->mask
= 0; /* enable IRQ */
1012 entry
->trigger
= trigger
;
1013 entry
->polarity
= polarity
;
1014 entry
->vector
= vector
;
1016 /* Mask level triggered irqs.
1017 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1024 static void setup_IO_APIC_irq(int apic
, int pin
, unsigned int irq
,
1025 int trigger
, int polarity
)
1027 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
1028 struct IO_APIC_route_entry entry
;
1031 if (!IO_APIC_IRQ(irq
))
1035 if (assign_irq_vector(irq
, mask
))
1038 cpus_and(mask
, cfg
->domain
, mask
);
1040 apic_printk(APIC_VERBOSE
,KERN_DEBUG
1041 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1042 "IRQ %d Mode:%i Active:%i)\n",
1043 apic
, mp_ioapics
[apic
].mp_apicid
, pin
, cfg
->vector
,
1044 irq
, trigger
, polarity
);
1047 if (setup_ioapic_entry(mp_ioapics
[apic
].mp_apicid
, irq
, &entry
,
1048 cpu_mask_to_apicid(mask
), trigger
, polarity
,
1050 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1051 mp_ioapics
[apic
].mp_apicid
, pin
);
1052 __clear_irq_vector(irq
);
1056 ioapic_register_intr(irq
, trigger
);
1058 disable_8259A_irq(irq
);
1060 ioapic_write_entry(apic
, pin
, entry
);
1063 static void __init
setup_IO_APIC_irqs(void)
1065 int apic
, pin
, idx
, irq
, first_notcon
= 1;
1067 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
1069 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1070 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1072 idx
= find_irq_entry(apic
,pin
,mp_INT
);
1075 apic_printk(APIC_VERBOSE
, KERN_DEBUG
" IO-APIC (apicid-pin) %d-%d", mp_ioapics
[apic
].mp_apicid
, pin
);
1078 apic_printk(APIC_VERBOSE
, ", %d-%d", mp_ioapics
[apic
].mp_apicid
, pin
);
1081 if (!first_notcon
) {
1082 apic_printk(APIC_VERBOSE
, " not connected.\n");
1086 irq
= pin_2_irq(idx
, apic
, pin
);
1087 add_pin_to_irq(irq
, apic
, pin
);
1089 setup_IO_APIC_irq(apic
, pin
, irq
,
1090 irq_trigger(idx
), irq_polarity(idx
));
1095 apic_printk(APIC_VERBOSE
, " not connected.\n");
1099 * Set up the timer pin, possibly with the 8259A-master behind.
1101 static void __init
setup_timer_IRQ0_pin(unsigned int apic
, unsigned int pin
,
1104 struct IO_APIC_route_entry entry
;
1106 if (intr_remapping_enabled
)
1109 memset(&entry
, 0, sizeof(entry
));
1112 * We use logical delivery to get the timer IRQ
1115 entry
.dest_mode
= INT_DEST_MODE
;
1116 entry
.mask
= 1; /* mask IRQ now */
1117 entry
.dest
= cpu_mask_to_apicid(TARGET_CPUS
);
1118 entry
.delivery_mode
= INT_DELIVERY_MODE
;
1121 entry
.vector
= vector
;
1124 * The timer IRQ doesn't have to know that behind the
1125 * scene we may have a 8259A-master in AEOI mode ...
1127 set_irq_chip_and_handler_name(0, &ioapic_chip
, handle_edge_irq
, "edge");
1130 * Add it to the IO-APIC irq-routing table:
1132 ioapic_write_entry(apic
, pin
, entry
);
1136 __apicdebuginit(void) print_IO_APIC(void)
1139 union IO_APIC_reg_00 reg_00
;
1140 union IO_APIC_reg_01 reg_01
;
1141 union IO_APIC_reg_02 reg_02
;
1142 unsigned long flags
;
1144 if (apic_verbosity
== APIC_QUIET
)
1147 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
1148 for (i
= 0; i
< nr_ioapics
; i
++)
1149 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
1150 mp_ioapics
[i
].mp_apicid
, nr_ioapic_registers
[i
]);
1153 * We are a bit conservative about what we expect. We have to
1154 * know about every hardware change ASAP.
1156 printk(KERN_INFO
"testing the IO APIC.......................\n");
1158 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1160 spin_lock_irqsave(&ioapic_lock
, flags
);
1161 reg_00
.raw
= io_apic_read(apic
, 0);
1162 reg_01
.raw
= io_apic_read(apic
, 1);
1163 if (reg_01
.bits
.version
>= 0x10)
1164 reg_02
.raw
= io_apic_read(apic
, 2);
1165 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1168 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].mp_apicid
);
1169 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
1170 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
1172 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_01
);
1173 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
1175 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
1176 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
1178 if (reg_01
.bits
.version
>= 0x10) {
1179 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
1180 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
1183 printk(KERN_DEBUG
".... IRQ redirection table:\n");
1185 printk(KERN_DEBUG
" NR Dst Mask Trig IRR Pol"
1186 " Stat Dmod Deli Vect: \n");
1188 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
1189 struct IO_APIC_route_entry entry
;
1191 entry
= ioapic_read_entry(apic
, i
);
1193 printk(KERN_DEBUG
" %02x %03X ",
1198 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1203 entry
.delivery_status
,
1205 entry
.delivery_mode
,
1210 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
1211 for (i
= 0; i
< nr_irqs
; i
++) {
1212 struct irq_pin_list
*entry
= irq_2_pin
+ i
;
1215 printk(KERN_DEBUG
"IRQ%d ", i
);
1217 printk("-> %d:%d", entry
->apic
, entry
->pin
);
1220 entry
= irq_2_pin
+ entry
->next
;
1225 printk(KERN_INFO
".................................... done.\n");
1230 __apicdebuginit(void) print_APIC_bitfield(int base
)
1235 if (apic_verbosity
== APIC_QUIET
)
1238 printk(KERN_DEBUG
"0123456789abcdef0123456789abcdef\n" KERN_DEBUG
);
1239 for (i
= 0; i
< 8; i
++) {
1240 v
= apic_read(base
+ i
*0x10);
1241 for (j
= 0; j
< 32; j
++) {
1251 __apicdebuginit(void) print_local_APIC(void *dummy
)
1253 unsigned int v
, ver
, maxlvt
;
1256 if (apic_verbosity
== APIC_QUIET
)
1259 printk("\n" KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
1260 smp_processor_id(), hard_smp_processor_id());
1261 v
= apic_read(APIC_ID
);
1262 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
, read_apic_id());
1263 v
= apic_read(APIC_LVR
);
1264 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
1265 ver
= GET_APIC_VERSION(v
);
1266 maxlvt
= lapic_get_maxlvt();
1268 v
= apic_read(APIC_TASKPRI
);
1269 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
1271 v
= apic_read(APIC_ARBPRI
);
1272 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
1273 v
& APIC_ARBPRI_MASK
);
1274 v
= apic_read(APIC_PROCPRI
);
1275 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
1277 v
= apic_read(APIC_EOI
);
1278 printk(KERN_DEBUG
"... APIC EOI: %08x\n", v
);
1279 v
= apic_read(APIC_RRR
);
1280 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
1281 v
= apic_read(APIC_LDR
);
1282 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
1283 v
= apic_read(APIC_DFR
);
1284 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
1285 v
= apic_read(APIC_SPIV
);
1286 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
1288 printk(KERN_DEBUG
"... APIC ISR field:\n");
1289 print_APIC_bitfield(APIC_ISR
);
1290 printk(KERN_DEBUG
"... APIC TMR field:\n");
1291 print_APIC_bitfield(APIC_TMR
);
1292 printk(KERN_DEBUG
"... APIC IRR field:\n");
1293 print_APIC_bitfield(APIC_IRR
);
1295 v
= apic_read(APIC_ESR
);
1296 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
1298 icr
= apic_icr_read();
1299 printk(KERN_DEBUG
"... APIC ICR: %08x\n", (u32
)icr
);
1300 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", (u32
)(icr
>> 32));
1302 v
= apic_read(APIC_LVTT
);
1303 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1305 if (maxlvt
> 3) { /* PC is LVT#4. */
1306 v
= apic_read(APIC_LVTPC
);
1307 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
1309 v
= apic_read(APIC_LVT0
);
1310 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
1311 v
= apic_read(APIC_LVT1
);
1312 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
1314 if (maxlvt
> 2) { /* ERR is LVT#3. */
1315 v
= apic_read(APIC_LVTERR
);
1316 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
1319 v
= apic_read(APIC_TMICT
);
1320 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
1321 v
= apic_read(APIC_TMCCT
);
1322 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
1323 v
= apic_read(APIC_TDCR
);
1324 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1328 __apicdebuginit(void) print_all_local_APICs(void)
1330 on_each_cpu(print_local_APIC
, NULL
, 1);
1333 __apicdebuginit(void) print_PIC(void)
1336 unsigned long flags
;
1338 if (apic_verbosity
== APIC_QUIET
)
1341 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1343 spin_lock_irqsave(&i8259A_lock
, flags
);
1345 v
= inb(0xa1) << 8 | inb(0x21);
1346 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1348 v
= inb(0xa0) << 8 | inb(0x20);
1349 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1353 v
= inb(0xa0) << 8 | inb(0x20);
1357 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1359 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1361 v
= inb(0x4d1) << 8 | inb(0x4d0);
1362 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1365 __apicdebuginit(int) print_all_ICs(void)
1368 print_all_local_APICs();
1374 fs_initcall(print_all_ICs
);
1377 void __init
enable_IO_APIC(void)
1379 union IO_APIC_reg_01 reg_01
;
1380 int i8259_apic
, i8259_pin
;
1382 unsigned long flags
;
1384 for (i
= 0; i
< pin_map_size
; i
++) {
1385 irq_2_pin
[i
].pin
= -1;
1386 irq_2_pin
[i
].next
= 0;
1390 * The number of IO-APIC IRQ registers (== #pins):
1392 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1393 spin_lock_irqsave(&ioapic_lock
, flags
);
1394 reg_01
.raw
= io_apic_read(apic
, 1);
1395 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1396 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
1398 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1400 /* See if any of the pins is in ExtINT mode */
1401 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1402 struct IO_APIC_route_entry entry
;
1403 entry
= ioapic_read_entry(apic
, pin
);
1405 /* If the interrupt line is enabled and in ExtInt mode
1406 * I have found the pin where the i8259 is connected.
1408 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1409 ioapic_i8259
.apic
= apic
;
1410 ioapic_i8259
.pin
= pin
;
1416 /* Look to see what if the MP table has reported the ExtINT */
1417 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1418 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1419 /* Trust the MP table if nothing is setup in the hardware */
1420 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1421 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1422 ioapic_i8259
.pin
= i8259_pin
;
1423 ioapic_i8259
.apic
= i8259_apic
;
1425 /* Complain if the MP table and the hardware disagree */
1426 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1427 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1429 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1433 * Do not trust the IO-APIC being empty at bootup
1439 * Not an __init, needed by the reboot code
1441 void disable_IO_APIC(void)
1444 * Clear the IO-APIC before rebooting:
1449 * If the i8259 is routed through an IOAPIC
1450 * Put that IOAPIC in virtual wire mode
1451 * so legacy interrupts can be delivered.
1453 if (ioapic_i8259
.pin
!= -1) {
1454 struct IO_APIC_route_entry entry
;
1456 memset(&entry
, 0, sizeof(entry
));
1457 entry
.mask
= 0; /* Enabled */
1458 entry
.trigger
= 0; /* Edge */
1460 entry
.polarity
= 0; /* High */
1461 entry
.delivery_status
= 0;
1462 entry
.dest_mode
= 0; /* Physical */
1463 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1465 entry
.dest
= read_apic_id();
1468 * Add it to the IO-APIC irq-routing table:
1470 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1473 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
1477 * There is a nasty bug in some older SMP boards, their mptable lies
1478 * about the timer IRQ. We do the following to work around the situation:
1480 * - timer IRQ defaults to IO-APIC IRQ
1481 * - if this function detects that timer IRQs are defunct, then we fall
1482 * back to ISA timer IRQs
1484 static int __init
timer_irq_works(void)
1486 unsigned long t1
= jiffies
;
1487 unsigned long flags
;
1489 local_save_flags(flags
);
1491 /* Let ten ticks pass... */
1492 mdelay((10 * 1000) / HZ
);
1493 local_irq_restore(flags
);
1496 * Expect a few ticks at least, to be sure some possible
1497 * glue logic does not lock up after one or two first
1498 * ticks in a non-ExtINT mode. Also the local APIC
1499 * might have cached one ExtINT interrupt. Finally, at
1500 * least one tick may be lost due to delays.
1504 if (time_after(jiffies
, t1
+ 4))
1510 * In the SMP+IOAPIC case it might happen that there are an unspecified
1511 * number of pending IRQ events unhandled. These cases are very rare,
1512 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1513 * better to do it this way as thus we do not have to be aware of
1514 * 'pending' interrupts in the IRQ path, except at this point.
1517 * Edge triggered needs to resend any interrupt
1518 * that was delayed but this is now handled in the device
1523 * Starting up a edge-triggered IO-APIC interrupt is
1524 * nasty - we need to make sure that we get the edge.
1525 * If it is already asserted for some reason, we need
1526 * return 1 to indicate that is was pending.
1528 * This is not complete - we should be able to fake
1529 * an edge even if it isn't on the 8259A...
1532 static unsigned int startup_ioapic_irq(unsigned int irq
)
1534 int was_pending
= 0;
1535 unsigned long flags
;
1537 spin_lock_irqsave(&ioapic_lock
, flags
);
1539 disable_8259A_irq(irq
);
1540 if (i8259A_irq_pending(irq
))
1543 __unmask_IO_APIC_irq(irq
);
1544 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1549 static int ioapic_retrigger_irq(unsigned int irq
)
1551 struct irq_cfg
*cfg
= &irq_cfg
[irq
];
1552 unsigned long flags
;
1554 spin_lock_irqsave(&vector_lock
, flags
);
1555 send_IPI_mask(cpumask_of_cpu(first_cpu(cfg
->domain
)), cfg
->vector
);
1556 spin_unlock_irqrestore(&vector_lock
, flags
);
1562 * Level and edge triggered IO-APIC interrupts need different handling,
1563 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1564 * handled with the level-triggered descriptor, but that one has slightly
1565 * more overhead. Level-triggered interrupts cannot be handled with the
1566 * edge-triggered handler, without risking IRQ storms and other ugly
1572 #ifdef CONFIG_INTR_REMAP
1573 static void ir_irq_migration(struct work_struct
*work
);
1575 static DECLARE_DELAYED_WORK(ir_migration_work
, ir_irq_migration
);
1578 * Migrate the IO-APIC irq in the presence of intr-remapping.
1580 * For edge triggered, irq migration is a simple atomic update(of vector
1581 * and cpu destination) of IRTE and flush the hardware cache.
1583 * For level triggered, we need to modify the io-apic RTE aswell with the update
1584 * vector information, along with modifying IRTE with vector and destination.
1585 * So irq migration for level triggered is little bit more complex compared to
1586 * edge triggered migration. But the good news is, we use the same algorithm
1587 * for level triggered migration as we have today, only difference being,
1588 * we now initiate the irq migration from process context instead of the
1589 * interrupt context.
1591 * In future, when we do a directed EOI (combined with cpu EOI broadcast
1592 * suppression) to the IO-APIC, level triggered irq migration will also be
1593 * as simple as edge triggered migration and we can do the irq migration
1594 * with a simple atomic update to IO-APIC RTE.
1596 static void migrate_ioapic_irq(int irq
, cpumask_t mask
)
1598 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
1599 struct irq_desc
*desc
= irq_desc
+ irq
;
1600 cpumask_t tmp
, cleanup_mask
;
1602 int modify_ioapic_rte
= desc
->status
& IRQ_LEVEL
;
1604 unsigned long flags
;
1606 cpus_and(tmp
, mask
, cpu_online_map
);
1607 if (cpus_empty(tmp
))
1610 if (get_irte(irq
, &irte
))
1613 if (assign_irq_vector(irq
, mask
))
1616 cpus_and(tmp
, cfg
->domain
, mask
);
1617 dest
= cpu_mask_to_apicid(tmp
);
1619 if (modify_ioapic_rte
) {
1620 spin_lock_irqsave(&ioapic_lock
, flags
);
1621 __target_IO_APIC_irq(irq
, dest
, cfg
->vector
);
1622 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1625 irte
.vector
= cfg
->vector
;
1626 irte
.dest_id
= IRTE_DEST(dest
);
1629 * Modified the IRTE and flushes the Interrupt entry cache.
1631 modify_irte(irq
, &irte
);
1633 if (cfg
->move_in_progress
) {
1634 cpus_and(cleanup_mask
, cfg
->old_domain
, cpu_online_map
);
1635 cfg
->move_cleanup_count
= cpus_weight(cleanup_mask
);
1636 send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
1637 cfg
->move_in_progress
= 0;
1640 irq_desc
[irq
].affinity
= mask
;
1643 static int migrate_irq_remapped_level(int irq
)
1647 mask_IO_APIC_irq(irq
);
1649 if (io_apic_level_ack_pending(irq
)) {
1651 * Interrupt in progress. Migrating irq now will change the
1652 * vector information in the IO-APIC RTE and that will confuse
1653 * the EOI broadcast performed by cpu.
1654 * So, delay the irq migration to the next instance.
1656 schedule_delayed_work(&ir_migration_work
, 1);
1660 /* everthing is clear. we have right of way */
1661 migrate_ioapic_irq(irq
, irq_desc
[irq
].pending_mask
);
1664 irq_desc
[irq
].status
&= ~IRQ_MOVE_PENDING
;
1665 cpus_clear(irq_desc
[irq
].pending_mask
);
1668 unmask_IO_APIC_irq(irq
);
1672 static void ir_irq_migration(struct work_struct
*work
)
1676 for (irq
= 0; irq
< nr_irqs
; irq
++) {
1677 struct irq_desc
*desc
= irq_desc
+ irq
;
1678 if (desc
->status
& IRQ_MOVE_PENDING
) {
1679 unsigned long flags
;
1681 spin_lock_irqsave(&desc
->lock
, flags
);
1682 if (!desc
->chip
->set_affinity
||
1683 !(desc
->status
& IRQ_MOVE_PENDING
)) {
1684 desc
->status
&= ~IRQ_MOVE_PENDING
;
1685 spin_unlock_irqrestore(&desc
->lock
, flags
);
1689 desc
->chip
->set_affinity(irq
,
1690 irq_desc
[irq
].pending_mask
);
1691 spin_unlock_irqrestore(&desc
->lock
, flags
);
1697 * Migrates the IRQ destination in the process context.
1699 static void set_ir_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
1701 if (irq_desc
[irq
].status
& IRQ_LEVEL
) {
1702 irq_desc
[irq
].status
|= IRQ_MOVE_PENDING
;
1703 irq_desc
[irq
].pending_mask
= mask
;
1704 migrate_irq_remapped_level(irq
);
1708 migrate_ioapic_irq(irq
, mask
);
1712 asmlinkage
void smp_irq_move_cleanup_interrupt(void)
1714 unsigned vector
, me
;
1719 me
= smp_processor_id();
1720 for (vector
= FIRST_EXTERNAL_VECTOR
; vector
< NR_VECTORS
; vector
++) {
1722 struct irq_desc
*desc
;
1723 struct irq_cfg
*cfg
;
1724 irq
= __get_cpu_var(vector_irq
)[vector
];
1728 desc
= irq_desc
+ irq
;
1729 cfg
= irq_cfg
+ irq
;
1730 spin_lock(&desc
->lock
);
1731 if (!cfg
->move_cleanup_count
)
1734 if ((vector
== cfg
->vector
) && cpu_isset(me
, cfg
->domain
))
1737 __get_cpu_var(vector_irq
)[vector
] = -1;
1738 cfg
->move_cleanup_count
--;
1740 spin_unlock(&desc
->lock
);
1746 static void irq_complete_move(unsigned int irq
)
1748 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
1749 unsigned vector
, me
;
1751 if (likely(!cfg
->move_in_progress
))
1754 vector
= ~get_irq_regs()->orig_ax
;
1755 me
= smp_processor_id();
1756 if ((vector
== cfg
->vector
) && cpu_isset(me
, cfg
->domain
)) {
1757 cpumask_t cleanup_mask
;
1759 cpus_and(cleanup_mask
, cfg
->old_domain
, cpu_online_map
);
1760 cfg
->move_cleanup_count
= cpus_weight(cleanup_mask
);
1761 send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
1762 cfg
->move_in_progress
= 0;
1766 static inline void irq_complete_move(unsigned int irq
) {}
1768 #ifdef CONFIG_INTR_REMAP
1769 static void ack_x2apic_level(unsigned int irq
)
1774 static void ack_x2apic_edge(unsigned int irq
)
1780 static void ack_apic_edge(unsigned int irq
)
1782 irq_complete_move(irq
);
1783 move_native_irq(irq
);
1787 static void ack_apic_level(unsigned int irq
)
1789 int do_unmask_irq
= 0;
1791 irq_complete_move(irq
);
1792 #ifdef CONFIG_GENERIC_PENDING_IRQ
1793 /* If we are moving the irq we need to mask it */
1794 if (unlikely(irq_desc
[irq
].status
& IRQ_MOVE_PENDING
)) {
1796 mask_IO_APIC_irq(irq
);
1801 * We must acknowledge the irq before we move it or the acknowledge will
1802 * not propagate properly.
1806 /* Now we can move and renable the irq */
1807 if (unlikely(do_unmask_irq
)) {
1808 /* Only migrate the irq if the ack has been received.
1810 * On rare occasions the broadcast level triggered ack gets
1811 * delayed going to ioapics, and if we reprogram the
1812 * vector while Remote IRR is still set the irq will never
1815 * To prevent this scenario we read the Remote IRR bit
1816 * of the ioapic. This has two effects.
1817 * - On any sane system the read of the ioapic will
1818 * flush writes (and acks) going to the ioapic from
1820 * - We get to see if the ACK has actually been delivered.
1822 * Based on failed experiments of reprogramming the
1823 * ioapic entry from outside of irq context starting
1824 * with masking the ioapic entry and then polling until
1825 * Remote IRR was clear before reprogramming the
1826 * ioapic I don't trust the Remote IRR bit to be
1827 * completey accurate.
1829 * However there appears to be no other way to plug
1830 * this race, so if the Remote IRR bit is not
1831 * accurate and is causing problems then it is a hardware bug
1832 * and you can go talk to the chipset vendor about it.
1834 if (!io_apic_level_ack_pending(irq
))
1835 move_masked_irq(irq
);
1836 unmask_IO_APIC_irq(irq
);
1840 static struct irq_chip ioapic_chip __read_mostly
= {
1842 .startup
= startup_ioapic_irq
,
1843 .mask
= mask_IO_APIC_irq
,
1844 .unmask
= unmask_IO_APIC_irq
,
1845 .ack
= ack_apic_edge
,
1846 .eoi
= ack_apic_level
,
1848 .set_affinity
= set_ioapic_affinity_irq
,
1850 .retrigger
= ioapic_retrigger_irq
,
1853 #ifdef CONFIG_INTR_REMAP
1854 static struct irq_chip ir_ioapic_chip __read_mostly
= {
1855 .name
= "IR-IO-APIC",
1856 .startup
= startup_ioapic_irq
,
1857 .mask
= mask_IO_APIC_irq
,
1858 .unmask
= unmask_IO_APIC_irq
,
1859 .ack
= ack_x2apic_edge
,
1860 .eoi
= ack_x2apic_level
,
1862 .set_affinity
= set_ir_ioapic_affinity_irq
,
1864 .retrigger
= ioapic_retrigger_irq
,
1868 static inline void init_IO_APIC_traps(void)
1873 * NOTE! The local APIC isn't very good at handling
1874 * multiple interrupts at the same interrupt level.
1875 * As the interrupt level is determined by taking the
1876 * vector number and shifting that right by 4, we
1877 * want to spread these out a bit so that they don't
1878 * all fall in the same interrupt level.
1880 * Also, we've got to be careful not to trash gate
1881 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1883 for (irq
= 0; irq
< nr_irqs
; irq
++) {
1884 if (IO_APIC_IRQ(irq
) && !irq_cfg
[irq
].vector
) {
1886 * Hmm.. We don't have an entry for this,
1887 * so default to an old-fashioned 8259
1888 * interrupt if we can..
1891 make_8259A_irq(irq
);
1893 /* Strange. Oh, well.. */
1894 irq_desc
[irq
].chip
= &no_irq_chip
;
1899 static void unmask_lapic_irq(unsigned int irq
)
1903 v
= apic_read(APIC_LVT0
);
1904 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
1907 static void mask_lapic_irq(unsigned int irq
)
1911 v
= apic_read(APIC_LVT0
);
1912 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
1915 static void ack_lapic_irq (unsigned int irq
)
1920 static struct irq_chip lapic_chip __read_mostly
= {
1921 .name
= "local-APIC",
1922 .mask
= mask_lapic_irq
,
1923 .unmask
= unmask_lapic_irq
,
1924 .ack
= ack_lapic_irq
,
1927 static void lapic_register_intr(int irq
)
1929 irq_desc
[irq
].status
&= ~IRQ_LEVEL
;
1930 set_irq_chip_and_handler_name(irq
, &lapic_chip
, handle_edge_irq
,
1934 static void __init
setup_nmi(void)
1937 * Dirty trick to enable the NMI watchdog ...
1938 * We put the 8259A master into AEOI mode and
1939 * unmask on all local APICs LVT0 as NMI.
1941 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1942 * is from Maciej W. Rozycki - so we do not have to EOI from
1943 * the NMI handler or the timer interrupt.
1945 printk(KERN_INFO
"activating NMI Watchdog ...");
1947 enable_NMI_through_LVT0();
1953 * This looks a bit hackish but it's about the only one way of sending
1954 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1955 * not support the ExtINT mode, unfortunately. We need to send these
1956 * cycles as some i82489DX-based boards have glue logic that keeps the
1957 * 8259A interrupt line asserted until INTA. --macro
1959 static inline void __init
unlock_ExtINT_logic(void)
1962 struct IO_APIC_route_entry entry0
, entry1
;
1963 unsigned char save_control
, save_freq_select
;
1965 pin
= find_isa_irq_pin(8, mp_INT
);
1966 apic
= find_isa_irq_apic(8, mp_INT
);
1970 entry0
= ioapic_read_entry(apic
, pin
);
1972 clear_IO_APIC_pin(apic
, pin
);
1974 memset(&entry1
, 0, sizeof(entry1
));
1976 entry1
.dest_mode
= 0; /* physical delivery */
1977 entry1
.mask
= 0; /* unmask IRQ now */
1978 entry1
.dest
= hard_smp_processor_id();
1979 entry1
.delivery_mode
= dest_ExtINT
;
1980 entry1
.polarity
= entry0
.polarity
;
1984 ioapic_write_entry(apic
, pin
, entry1
);
1986 save_control
= CMOS_READ(RTC_CONTROL
);
1987 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
1988 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
1990 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
1995 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
1999 CMOS_WRITE(save_control
, RTC_CONTROL
);
2000 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
2001 clear_IO_APIC_pin(apic
, pin
);
2003 ioapic_write_entry(apic
, pin
, entry0
);
2007 * This code may look a bit paranoid, but it's supposed to cooperate with
2008 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2009 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2010 * fanatically on his truly buggy board.
2012 * FIXME: really need to revamp this for modern platforms only.
2014 static inline void __init
check_timer(void)
2016 struct irq_cfg
*cfg
= irq_cfg
+ 0;
2017 int apic1
, pin1
, apic2
, pin2
;
2018 unsigned long flags
;
2021 local_irq_save(flags
);
2024 * get/set the timer IRQ vector:
2026 disable_8259A_irq(0);
2027 assign_irq_vector(0, TARGET_CPUS
);
2030 * As IRQ0 is to be enabled in the 8259A, the virtual
2031 * wire has to be disabled in the local APIC.
2033 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
2036 pin1
= find_isa_irq_pin(0, mp_INT
);
2037 apic1
= find_isa_irq_apic(0, mp_INT
);
2038 pin2
= ioapic_i8259
.pin
;
2039 apic2
= ioapic_i8259
.apic
;
2041 apic_printk(APIC_QUIET
, KERN_INFO
"..TIMER: vector=0x%02X "
2042 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2043 cfg
->vector
, apic1
, pin1
, apic2
, pin2
);
2046 * Some BIOS writers are clueless and report the ExtINTA
2047 * I/O APIC input from the cascaded 8259A as the timer
2048 * interrupt input. So just in case, if only one pin
2049 * was found above, try it both directly and through the
2053 if (intr_remapping_enabled
)
2054 panic("BIOS bug: timer not connected to IO-APIC");
2058 } else if (pin2
== -1) {
2065 * Ok, does IRQ0 through the IOAPIC work?
2068 add_pin_to_irq(0, apic1
, pin1
);
2069 setup_timer_IRQ0_pin(apic1
, pin1
, cfg
->vector
);
2071 unmask_IO_APIC_irq(0);
2072 if (!no_timer_check
&& timer_irq_works()) {
2073 if (nmi_watchdog
== NMI_IO_APIC
) {
2075 enable_8259A_irq(0);
2077 if (disable_timer_pin_1
> 0)
2078 clear_IO_APIC_pin(0, pin1
);
2081 if (intr_remapping_enabled
)
2082 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2083 clear_IO_APIC_pin(apic1
, pin1
);
2085 apic_printk(APIC_QUIET
, KERN_ERR
"..MP-BIOS bug: "
2086 "8254 timer not connected to IO-APIC\n");
2088 apic_printk(APIC_QUIET
, KERN_INFO
"...trying to set up timer "
2089 "(IRQ0) through the 8259A ...\n");
2090 apic_printk(APIC_QUIET
, KERN_INFO
2091 "..... (found apic %d pin %d) ...\n", apic2
, pin2
);
2093 * legacy devices should be connected to IO APIC #0
2095 replace_pin_at_irq(0, apic1
, pin1
, apic2
, pin2
);
2096 setup_timer_IRQ0_pin(apic2
, pin2
, cfg
->vector
);
2097 unmask_IO_APIC_irq(0);
2098 enable_8259A_irq(0);
2099 if (timer_irq_works()) {
2100 apic_printk(APIC_QUIET
, KERN_INFO
"....... works.\n");
2101 timer_through_8259
= 1;
2102 if (nmi_watchdog
== NMI_IO_APIC
) {
2103 disable_8259A_irq(0);
2105 enable_8259A_irq(0);
2110 * Cleanup, just in case ...
2112 disable_8259A_irq(0);
2113 clear_IO_APIC_pin(apic2
, pin2
);
2114 apic_printk(APIC_QUIET
, KERN_INFO
"....... failed.\n");
2117 if (nmi_watchdog
== NMI_IO_APIC
) {
2118 apic_printk(APIC_QUIET
, KERN_WARNING
"timer doesn't work "
2119 "through the IO-APIC - disabling NMI Watchdog!\n");
2120 nmi_watchdog
= NMI_NONE
;
2123 apic_printk(APIC_QUIET
, KERN_INFO
2124 "...trying to set up timer as Virtual Wire IRQ...\n");
2126 lapic_register_intr(0);
2127 apic_write(APIC_LVT0
, APIC_DM_FIXED
| cfg
->vector
); /* Fixed mode */
2128 enable_8259A_irq(0);
2130 if (timer_irq_works()) {
2131 apic_printk(APIC_QUIET
, KERN_INFO
"..... works.\n");
2134 disable_8259A_irq(0);
2135 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| cfg
->vector
);
2136 apic_printk(APIC_QUIET
, KERN_INFO
"..... failed.\n");
2138 apic_printk(APIC_QUIET
, KERN_INFO
2139 "...trying to set up timer as ExtINT IRQ...\n");
2143 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
2145 unlock_ExtINT_logic();
2147 if (timer_irq_works()) {
2148 apic_printk(APIC_QUIET
, KERN_INFO
"..... works.\n");
2151 apic_printk(APIC_QUIET
, KERN_INFO
"..... failed :(.\n");
2152 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2153 "report. Then try booting with the 'noapic' option.\n");
2155 local_irq_restore(flags
);
2158 static int __init
notimercheck(char *s
)
2163 __setup("no_timer_check", notimercheck
);
2166 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2167 * to devices. However there may be an I/O APIC pin available for
2168 * this interrupt regardless. The pin may be left unconnected, but
2169 * typically it will be reused as an ExtINT cascade interrupt for
2170 * the master 8259A. In the MPS case such a pin will normally be
2171 * reported as an ExtINT interrupt in the MP table. With ACPI
2172 * there is no provision for ExtINT interrupts, and in the absence
2173 * of an override it would be treated as an ordinary ISA I/O APIC
2174 * interrupt, that is edge-triggered and unmasked by default. We
2175 * used to do this, but it caused problems on some systems because
2176 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2177 * the same ExtINT cascade interrupt to drive the local APIC of the
2178 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2179 * the I/O APIC in all cases now. No actual device should request
2180 * it anyway. --macro
2182 #define PIC_IRQS (1<<2)
2184 void __init
setup_IO_APIC(void)
2188 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2191 io_apic_irqs
= ~PIC_IRQS
;
2193 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
2196 setup_IO_APIC_irqs();
2197 init_IO_APIC_traps();
2201 struct sysfs_ioapic_data
{
2202 struct sys_device dev
;
2203 struct IO_APIC_route_entry entry
[0];
2205 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
2207 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
2209 struct IO_APIC_route_entry
*entry
;
2210 struct sysfs_ioapic_data
*data
;
2213 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
2214 entry
= data
->entry
;
2215 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++, entry
++ )
2216 *entry
= ioapic_read_entry(dev
->id
, i
);
2221 static int ioapic_resume(struct sys_device
*dev
)
2223 struct IO_APIC_route_entry
*entry
;
2224 struct sysfs_ioapic_data
*data
;
2225 unsigned long flags
;
2226 union IO_APIC_reg_00 reg_00
;
2229 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
2230 entry
= data
->entry
;
2232 spin_lock_irqsave(&ioapic_lock
, flags
);
2233 reg_00
.raw
= io_apic_read(dev
->id
, 0);
2234 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].mp_apicid
) {
2235 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].mp_apicid
;
2236 io_apic_write(dev
->id
, 0, reg_00
.raw
);
2238 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2239 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
2240 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
2245 static struct sysdev_class ioapic_sysdev_class
= {
2247 .suspend
= ioapic_suspend
,
2248 .resume
= ioapic_resume
,
2251 static int __init
ioapic_init_sysfs(void)
2253 struct sys_device
* dev
;
2256 error
= sysdev_class_register(&ioapic_sysdev_class
);
2260 for (i
= 0; i
< nr_ioapics
; i
++ ) {
2261 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
2262 * sizeof(struct IO_APIC_route_entry
);
2263 mp_ioapic_data
[i
] = kzalloc(size
, GFP_KERNEL
);
2264 if (!mp_ioapic_data
[i
]) {
2265 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
2268 dev
= &mp_ioapic_data
[i
]->dev
;
2270 dev
->cls
= &ioapic_sysdev_class
;
2271 error
= sysdev_register(dev
);
2273 kfree(mp_ioapic_data
[i
]);
2274 mp_ioapic_data
[i
] = NULL
;
2275 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
2283 device_initcall(ioapic_init_sysfs
);
2286 * Dynamic irq allocate and deallocation
2288 int create_irq(void)
2290 /* Allocate an unused irq */
2293 unsigned long flags
;
2296 spin_lock_irqsave(&vector_lock
, flags
);
2297 for (new = (nr_irqs
- 1); new >= 0; new--) {
2298 if (platform_legacy_irq(new))
2300 if (irq_cfg
[new].vector
!= 0)
2302 if (__assign_irq_vector(new, TARGET_CPUS
) == 0)
2306 spin_unlock_irqrestore(&vector_lock
, flags
);
2309 dynamic_irq_init(irq
);
2314 void destroy_irq(unsigned int irq
)
2316 unsigned long flags
;
2318 dynamic_irq_cleanup(irq
);
2320 #ifdef CONFIG_INTR_REMAP
2323 spin_lock_irqsave(&vector_lock
, flags
);
2324 __clear_irq_vector(irq
);
2325 spin_unlock_irqrestore(&vector_lock
, flags
);
2329 * MSI message composition
2331 #ifdef CONFIG_PCI_MSI
2332 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
2334 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
2340 err
= assign_irq_vector(irq
, tmp
);
2344 cpus_and(tmp
, cfg
->domain
, tmp
);
2345 dest
= cpu_mask_to_apicid(tmp
);
2347 #ifdef CONFIG_INTR_REMAP
2348 if (irq_remapped(irq
)) {
2353 ir_index
= map_irq_to_irte_handle(irq
, &sub_handle
);
2354 BUG_ON(ir_index
== -1);
2356 memset (&irte
, 0, sizeof(irte
));
2359 irte
.dst_mode
= INT_DEST_MODE
;
2360 irte
.trigger_mode
= 0; /* edge */
2361 irte
.dlvry_mode
= INT_DELIVERY_MODE
;
2362 irte
.vector
= cfg
->vector
;
2363 irte
.dest_id
= IRTE_DEST(dest
);
2365 modify_irte(irq
, &irte
);
2367 msg
->address_hi
= MSI_ADDR_BASE_HI
;
2368 msg
->data
= sub_handle
;
2369 msg
->address_lo
= MSI_ADDR_BASE_LO
| MSI_ADDR_IR_EXT_INT
|
2371 MSI_ADDR_IR_INDEX1(ir_index
) |
2372 MSI_ADDR_IR_INDEX2(ir_index
);
2376 msg
->address_hi
= MSI_ADDR_BASE_HI
;
2379 ((INT_DEST_MODE
== 0) ?
2380 MSI_ADDR_DEST_MODE_PHYSICAL
:
2381 MSI_ADDR_DEST_MODE_LOGICAL
) |
2382 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2383 MSI_ADDR_REDIRECTION_CPU
:
2384 MSI_ADDR_REDIRECTION_LOWPRI
) |
2385 MSI_ADDR_DEST_ID(dest
);
2388 MSI_DATA_TRIGGER_EDGE
|
2389 MSI_DATA_LEVEL_ASSERT
|
2390 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2391 MSI_DATA_DELIVERY_FIXED
:
2392 MSI_DATA_DELIVERY_LOWPRI
) |
2393 MSI_DATA_VECTOR(cfg
->vector
);
2399 static void set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
2401 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
2406 cpus_and(tmp
, mask
, cpu_online_map
);
2407 if (cpus_empty(tmp
))
2410 if (assign_irq_vector(irq
, mask
))
2413 cpus_and(tmp
, cfg
->domain
, mask
);
2414 dest
= cpu_mask_to_apicid(tmp
);
2416 read_msi_msg(irq
, &msg
);
2418 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
2419 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
2420 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
2421 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
2423 write_msi_msg(irq
, &msg
);
2424 irq_desc
[irq
].affinity
= mask
;
2427 #ifdef CONFIG_INTR_REMAP
2429 * Migrate the MSI irq to another cpumask. This migration is
2430 * done in the process context using interrupt-remapping hardware.
2432 static void ir_set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
2434 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
2436 cpumask_t tmp
, cleanup_mask
;
2439 cpus_and(tmp
, mask
, cpu_online_map
);
2440 if (cpus_empty(tmp
))
2443 if (get_irte(irq
, &irte
))
2446 if (assign_irq_vector(irq
, mask
))
2449 cpus_and(tmp
, cfg
->domain
, mask
);
2450 dest
= cpu_mask_to_apicid(tmp
);
2452 irte
.vector
= cfg
->vector
;
2453 irte
.dest_id
= IRTE_DEST(dest
);
2456 * atomically update the IRTE with the new destination and vector.
2458 modify_irte(irq
, &irte
);
2461 * After this point, all the interrupts will start arriving
2462 * at the new destination. So, time to cleanup the previous
2463 * vector allocation.
2465 if (cfg
->move_in_progress
) {
2466 cpus_and(cleanup_mask
, cfg
->old_domain
, cpu_online_map
);
2467 cfg
->move_cleanup_count
= cpus_weight(cleanup_mask
);
2468 send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
2469 cfg
->move_in_progress
= 0;
2472 irq_desc
[irq
].affinity
= mask
;
2475 #endif /* CONFIG_SMP */
2478 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2479 * which implement the MSI or MSI-X Capability Structure.
2481 static struct irq_chip msi_chip
= {
2483 .unmask
= unmask_msi_irq
,
2484 .mask
= mask_msi_irq
,
2485 .ack
= ack_apic_edge
,
2487 .set_affinity
= set_msi_irq_affinity
,
2489 .retrigger
= ioapic_retrigger_irq
,
2492 #ifdef CONFIG_INTR_REMAP
2493 static struct irq_chip msi_ir_chip
= {
2494 .name
= "IR-PCI-MSI",
2495 .unmask
= unmask_msi_irq
,
2496 .mask
= mask_msi_irq
,
2497 .ack
= ack_x2apic_edge
,
2499 .set_affinity
= ir_set_msi_irq_affinity
,
2501 .retrigger
= ioapic_retrigger_irq
,
2505 * Map the PCI dev to the corresponding remapping hardware unit
2506 * and allocate 'nvec' consecutive interrupt-remapping table entries
2509 static int msi_alloc_irte(struct pci_dev
*dev
, int irq
, int nvec
)
2511 struct intel_iommu
*iommu
;
2514 iommu
= map_dev_to_ir(dev
);
2517 "Unable to map PCI %s to iommu\n", pci_name(dev
));
2521 index
= alloc_irte(iommu
, irq
, nvec
);
2524 "Unable to allocate %d IRTE for PCI %s\n", nvec
,
2532 static int setup_msi_irq(struct pci_dev
*dev
, struct msi_desc
*desc
, int irq
)
2537 ret
= msi_compose_msg(dev
, irq
, &msg
);
2541 set_irq_msi(irq
, desc
);
2542 write_msi_msg(irq
, &msg
);
2544 #ifdef CONFIG_INTR_REMAP
2545 if (irq_remapped(irq
)) {
2546 struct irq_desc
*desc
= irq_desc
+ irq
;
2548 * irq migration in process context
2550 desc
->status
|= IRQ_MOVE_PCNTXT
;
2551 set_irq_chip_and_handler_name(irq
, &msi_ir_chip
, handle_edge_irq
, "edge");
2554 set_irq_chip_and_handler_name(irq
, &msi_chip
, handle_edge_irq
, "edge");
2559 int arch_setup_msi_irq(struct pci_dev
*dev
, struct msi_desc
*desc
)
2567 #ifdef CONFIG_INTR_REMAP
2568 if (!intr_remapping_enabled
)
2571 ret
= msi_alloc_irte(dev
, irq
, 1);
2576 ret
= setup_msi_irq(dev
, desc
, irq
);
2583 #ifdef CONFIG_INTR_REMAP
2590 int arch_setup_msi_irqs(struct pci_dev
*dev
, int nvec
, int type
)
2592 int irq
, ret
, sub_handle
;
2593 struct msi_desc
*desc
;
2594 #ifdef CONFIG_INTR_REMAP
2595 struct intel_iommu
*iommu
= 0;
2600 list_for_each_entry(desc
, &dev
->msi_list
, list
) {
2604 #ifdef CONFIG_INTR_REMAP
2605 if (!intr_remapping_enabled
)
2610 * allocate the consecutive block of IRTE's
2613 index
= msi_alloc_irte(dev
, irq
, nvec
);
2619 iommu
= map_dev_to_ir(dev
);
2625 * setup the mapping between the irq and the IRTE
2626 * base index, the sub_handle pointing to the
2627 * appropriate interrupt remap table entry.
2629 set_irte_irq(irq
, iommu
, index
, sub_handle
);
2633 ret
= setup_msi_irq(dev
, desc
, irq
);
2645 void arch_teardown_msi_irq(unsigned int irq
)
2652 static void dmar_msi_set_affinity(unsigned int irq
, cpumask_t mask
)
2654 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
2659 cpus_and(tmp
, mask
, cpu_online_map
);
2660 if (cpus_empty(tmp
))
2663 if (assign_irq_vector(irq
, mask
))
2666 cpus_and(tmp
, cfg
->domain
, mask
);
2667 dest
= cpu_mask_to_apicid(tmp
);
2669 dmar_msi_read(irq
, &msg
);
2671 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
2672 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
2673 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
2674 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
2676 dmar_msi_write(irq
, &msg
);
2677 irq_desc
[irq
].affinity
= mask
;
2679 #endif /* CONFIG_SMP */
2681 struct irq_chip dmar_msi_type
= {
2683 .unmask
= dmar_msi_unmask
,
2684 .mask
= dmar_msi_mask
,
2685 .ack
= ack_apic_edge
,
2687 .set_affinity
= dmar_msi_set_affinity
,
2689 .retrigger
= ioapic_retrigger_irq
,
2692 int arch_setup_dmar_msi(unsigned int irq
)
2697 ret
= msi_compose_msg(NULL
, irq
, &msg
);
2700 dmar_msi_write(irq
, &msg
);
2701 set_irq_chip_and_handler_name(irq
, &dmar_msi_type
, handle_edge_irq
,
2707 #endif /* CONFIG_PCI_MSI */
2709 * Hypertransport interrupt support
2711 #ifdef CONFIG_HT_IRQ
2715 static void target_ht_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
2717 struct ht_irq_msg msg
;
2718 fetch_ht_irq_msg(irq
, &msg
);
2720 msg
.address_lo
&= ~(HT_IRQ_LOW_VECTOR_MASK
| HT_IRQ_LOW_DEST_ID_MASK
);
2721 msg
.address_hi
&= ~(HT_IRQ_HIGH_DEST_ID_MASK
);
2723 msg
.address_lo
|= HT_IRQ_LOW_VECTOR(vector
) | HT_IRQ_LOW_DEST_ID(dest
);
2724 msg
.address_hi
|= HT_IRQ_HIGH_DEST_ID(dest
);
2726 write_ht_irq_msg(irq
, &msg
);
2729 static void set_ht_irq_affinity(unsigned int irq
, cpumask_t mask
)
2731 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
2735 cpus_and(tmp
, mask
, cpu_online_map
);
2736 if (cpus_empty(tmp
))
2739 if (assign_irq_vector(irq
, mask
))
2742 cpus_and(tmp
, cfg
->domain
, mask
);
2743 dest
= cpu_mask_to_apicid(tmp
);
2745 target_ht_irq(irq
, dest
, cfg
->vector
);
2746 irq_desc
[irq
].affinity
= mask
;
2750 static struct irq_chip ht_irq_chip
= {
2752 .mask
= mask_ht_irq
,
2753 .unmask
= unmask_ht_irq
,
2754 .ack
= ack_apic_edge
,
2756 .set_affinity
= set_ht_irq_affinity
,
2758 .retrigger
= ioapic_retrigger_irq
,
2761 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
2763 struct irq_cfg
*cfg
= irq_cfg
+ irq
;
2768 err
= assign_irq_vector(irq
, tmp
);
2770 struct ht_irq_msg msg
;
2773 cpus_and(tmp
, cfg
->domain
, tmp
);
2774 dest
= cpu_mask_to_apicid(tmp
);
2776 msg
.address_hi
= HT_IRQ_HIGH_DEST_ID(dest
);
2780 HT_IRQ_LOW_DEST_ID(dest
) |
2781 HT_IRQ_LOW_VECTOR(cfg
->vector
) |
2782 ((INT_DEST_MODE
== 0) ?
2783 HT_IRQ_LOW_DM_PHYSICAL
:
2784 HT_IRQ_LOW_DM_LOGICAL
) |
2785 HT_IRQ_LOW_RQEOI_EDGE
|
2786 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2787 HT_IRQ_LOW_MT_FIXED
:
2788 HT_IRQ_LOW_MT_ARBITRATED
) |
2789 HT_IRQ_LOW_IRQ_MASKED
;
2791 write_ht_irq_msg(irq
, &msg
);
2793 set_irq_chip_and_handler_name(irq
, &ht_irq_chip
,
2794 handle_edge_irq
, "edge");
2798 #endif /* CONFIG_HT_IRQ */
2800 /* --------------------------------------------------------------------------
2801 ACPI-based IOAPIC Configuration
2802 -------------------------------------------------------------------------- */
2806 #define IO_APIC_MAX_ID 0xFE
2808 int __init
io_apic_get_redir_entries (int ioapic
)
2810 union IO_APIC_reg_01 reg_01
;
2811 unsigned long flags
;
2813 spin_lock_irqsave(&ioapic_lock
, flags
);
2814 reg_01
.raw
= io_apic_read(ioapic
, 1);
2815 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2817 return reg_01
.bits
.entries
;
2821 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int triggering
, int polarity
)
2823 if (!IO_APIC_IRQ(irq
)) {
2824 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
2830 * IRQs < 16 are already in the irq_2_pin[] map
2833 add_pin_to_irq(irq
, ioapic
, pin
);
2835 setup_IO_APIC_irq(ioapic
, pin
, irq
, triggering
, polarity
);
2841 int acpi_get_override_irq(int bus_irq
, int *trigger
, int *polarity
)
2845 if (skip_ioapic_setup
)
2848 for (i
= 0; i
< mp_irq_entries
; i
++)
2849 if (mp_irqs
[i
].mp_irqtype
== mp_INT
&&
2850 mp_irqs
[i
].mp_srcbusirq
== bus_irq
)
2852 if (i
>= mp_irq_entries
)
2855 *trigger
= irq_trigger(i
);
2856 *polarity
= irq_polarity(i
);
2860 #endif /* CONFIG_ACPI */
2863 * This function currently is only a helper for the i386 smp boot process where
2864 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2865 * so mask in all cases should simply be TARGET_CPUS
2868 void __init
setup_ioapic_dest(void)
2870 int pin
, ioapic
, irq
, irq_entry
;
2872 if (skip_ioapic_setup
== 1)
2875 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++) {
2876 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
2877 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
2878 if (irq_entry
== -1)
2880 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
2882 /* setup_IO_APIC_irqs could fail to get vector for some device
2883 * when you have too many devices, because at that time only boot
2886 if (!irq_cfg
[irq
].vector
)
2887 setup_IO_APIC_irq(ioapic
, pin
, irq
,
2888 irq_trigger(irq_entry
),
2889 irq_polarity(irq_entry
));
2890 #ifdef CONFIG_INTR_REMAP
2891 else if (intr_remapping_enabled
)
2892 set_ir_ioapic_affinity_irq(irq
, TARGET_CPUS
);
2895 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);
2902 #define IOAPIC_RESOURCE_NAME_SIZE 11
2904 static struct resource
*ioapic_resources
;
2906 static struct resource
* __init
ioapic_setup_resources(void)
2909 struct resource
*res
;
2913 if (nr_ioapics
<= 0)
2916 n
= IOAPIC_RESOURCE_NAME_SIZE
+ sizeof(struct resource
);
2919 mem
= alloc_bootmem(n
);
2923 mem
+= sizeof(struct resource
) * nr_ioapics
;
2925 for (i
= 0; i
< nr_ioapics
; i
++) {
2927 res
[i
].flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
2928 sprintf(mem
, "IOAPIC %u", i
);
2929 mem
+= IOAPIC_RESOURCE_NAME_SIZE
;
2933 ioapic_resources
= res
;
2938 void __init
ioapic_init_mappings(void)
2940 unsigned long ioapic_phys
, idx
= FIX_IO_APIC_BASE_0
;
2941 struct resource
*ioapic_res
;
2944 ioapic_res
= ioapic_setup_resources();
2945 for (i
= 0; i
< nr_ioapics
; i
++) {
2946 if (smp_found_config
) {
2947 ioapic_phys
= mp_ioapics
[i
].mp_apicaddr
;
2949 ioapic_phys
= (unsigned long)
2950 alloc_bootmem_pages(PAGE_SIZE
);
2951 ioapic_phys
= __pa(ioapic_phys
);
2953 set_fixmap_nocache(idx
, ioapic_phys
);
2954 apic_printk(APIC_VERBOSE
,
2955 "mapped IOAPIC to %016lx (%016lx)\n",
2956 __fix_to_virt(idx
), ioapic_phys
);
2959 if (ioapic_res
!= NULL
) {
2960 ioapic_res
->start
= ioapic_phys
;
2961 ioapic_res
->end
= ioapic_phys
+ (4 * 1024) - 1;
2967 static int __init
ioapic_insert_resources(void)
2970 struct resource
*r
= ioapic_resources
;
2974 "IO APIC resources could be not be allocated.\n");
2978 for (i
= 0; i
< nr_ioapics
; i
++) {
2979 insert_resource(&iomem_resource
, r
);
2986 /* Insert the IO APIC resources after PCI initialization has occured to handle
2987 * IO APICS that are mapped in on a BAR in PCI space. */
2988 late_initcall(ioapic_insert_resources
);