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/mc146818rtc.h>
29 #include <linux/compiler.h>
30 #include <linux/acpi.h>
31 #include <linux/module.h>
32 #include <linux/sysdev.h>
33 #include <linux/pci.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h> /* time_after() */
43 #include <asm/timer.h>
44 #include <asm/i8259.h>
46 #include <asm/msidef.h>
47 #include <asm/hypertransport.h>
49 #include <mach_apic.h>
50 #include <mach_apicdef.h>
52 int (*ioapic_renumber_irq
)(int ioapic
, int irq
);
53 atomic_t irq_mis_count
;
55 /* Where if anywhere is the i8259 connect in external int mode */
56 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
58 static DEFINE_SPINLOCK(ioapic_lock
);
59 static DEFINE_SPINLOCK(vector_lock
);
61 int timer_through_8259 __initdata
;
64 * Is the SiS APIC rmw bug present ?
65 * -1 = don't know, 0 = no, 1 = yes
67 int sis_apic_bug
= -1;
70 * # of IRQ routing registers
72 int nr_ioapic_registers
[MAX_IO_APICS
];
74 /* I/O APIC entries */
75 struct mpc_config_ioapic mp_ioapics
[MAX_IO_APICS
];
78 /* MP IRQ source entries */
79 struct mpc_config_intsrc mp_irqs
[MAX_IRQ_SOURCES
];
81 /* # of MP IRQ source entries */
84 static int disable_timer_pin_1 __initdata
;
87 * Rough estimation of how many shared IRQs there are, can
90 #define MAX_PLUS_SHARED_IRQS NR_IRQS
91 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
94 * This is performance-critical, we want to do it O(1)
96 * the indexing order of this array favors 1:1 mappings
97 * between pins and IRQs.
100 static struct irq_pin_list
{
102 } irq_2_pin
[PIN_MAP_SIZE
];
106 unsigned int unused
[3];
110 static __attribute_const__
struct io_apic __iomem
*io_apic_base(int idx
)
112 return (void __iomem
*) __fix_to_virt(FIX_IO_APIC_BASE_0
+ idx
)
113 + (mp_ioapics
[idx
].mpc_apicaddr
& ~PAGE_MASK
);
116 static inline unsigned int io_apic_read(unsigned int apic
, unsigned int reg
)
118 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
119 writel(reg
, &io_apic
->index
);
120 return readl(&io_apic
->data
);
123 static inline void io_apic_write(unsigned int apic
, unsigned int reg
, unsigned int value
)
125 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
126 writel(reg
, &io_apic
->index
);
127 writel(value
, &io_apic
->data
);
131 * Re-write a value: to be used for read-modify-write
132 * cycles where the read already set up the index register.
134 * Older SiS APIC requires we rewrite the index register
136 static inline void io_apic_modify(unsigned int apic
, unsigned int reg
, unsigned int value
)
138 volatile struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
140 writel(reg
, &io_apic
->index
);
141 writel(value
, &io_apic
->data
);
145 struct { u32 w1
, w2
; };
146 struct IO_APIC_route_entry entry
;
149 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
151 union entry_union eu
;
153 spin_lock_irqsave(&ioapic_lock
, flags
);
154 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
155 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
156 spin_unlock_irqrestore(&ioapic_lock
, flags
);
161 * When we write a new IO APIC routing entry, we need to write the high
162 * word first! If the mask bit in the low word is clear, we will enable
163 * the interrupt, and we need to make sure the entry is fully populated
164 * before that happens.
167 __ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
169 union entry_union eu
;
171 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
172 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
175 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
178 spin_lock_irqsave(&ioapic_lock
, flags
);
179 __ioapic_write_entry(apic
, pin
, e
);
180 spin_unlock_irqrestore(&ioapic_lock
, flags
);
184 * When we mask an IO APIC routing entry, we need to write the low
185 * word first, in order to set the mask bit before we change the
188 static void ioapic_mask_entry(int apic
, int pin
)
191 union entry_union eu
= { .entry
.mask
= 1 };
193 spin_lock_irqsave(&ioapic_lock
, flags
);
194 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
195 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
196 spin_unlock_irqrestore(&ioapic_lock
, flags
);
200 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
201 * shared ISA-space IRQs, so we have to support them. We are super
202 * fast in the common case, and fast for shared ISA-space IRQs.
204 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
206 static int first_free_entry
= NR_IRQS
;
207 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
210 entry
= irq_2_pin
+ entry
->next
;
212 if (entry
->pin
!= -1) {
213 entry
->next
= first_free_entry
;
214 entry
= irq_2_pin
+ entry
->next
;
215 if (++first_free_entry
>= PIN_MAP_SIZE
)
216 panic("io_apic.c: whoops");
223 * Reroute an IRQ to a different pin.
225 static void __init
replace_pin_at_irq(unsigned int irq
,
226 int oldapic
, int oldpin
,
227 int newapic
, int newpin
)
229 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
232 if (entry
->apic
== oldapic
&& entry
->pin
== oldpin
) {
233 entry
->apic
= newapic
;
238 entry
= irq_2_pin
+ entry
->next
;
242 static void __modify_IO_APIC_irq (unsigned int irq
, unsigned long enable
, unsigned long disable
)
244 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
245 unsigned int pin
, reg
;
251 reg
= io_apic_read(entry
->apic
, 0x10 + pin
*2);
254 io_apic_modify(entry
->apic
, 0x10 + pin
*2, reg
);
257 entry
= irq_2_pin
+ entry
->next
;
262 static void __mask_IO_APIC_irq (unsigned int irq
)
264 __modify_IO_APIC_irq(irq
, 0x00010000, 0);
268 static void __unmask_IO_APIC_irq (unsigned int irq
)
270 __modify_IO_APIC_irq(irq
, 0, 0x00010000);
273 /* mask = 1, trigger = 0 */
274 static void __mask_and_edge_IO_APIC_irq (unsigned int irq
)
276 __modify_IO_APIC_irq(irq
, 0x00010000, 0x00008000);
279 /* mask = 0, trigger = 1 */
280 static void __unmask_and_level_IO_APIC_irq (unsigned int irq
)
282 __modify_IO_APIC_irq(irq
, 0x00008000, 0x00010000);
285 static void mask_IO_APIC_irq (unsigned int irq
)
289 spin_lock_irqsave(&ioapic_lock
, flags
);
290 __mask_IO_APIC_irq(irq
);
291 spin_unlock_irqrestore(&ioapic_lock
, flags
);
294 static void unmask_IO_APIC_irq (unsigned int irq
)
298 spin_lock_irqsave(&ioapic_lock
, flags
);
299 __unmask_IO_APIC_irq(irq
);
300 spin_unlock_irqrestore(&ioapic_lock
, flags
);
303 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
305 struct IO_APIC_route_entry entry
;
307 /* Check delivery_mode to be sure we're not clearing an SMI pin */
308 entry
= ioapic_read_entry(apic
, pin
);
309 if (entry
.delivery_mode
== dest_SMI
)
313 * Disable it in the IO-APIC irq-routing table:
315 ioapic_mask_entry(apic
, pin
);
318 static void clear_IO_APIC (void)
322 for (apic
= 0; apic
< nr_ioapics
; apic
++)
323 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
324 clear_IO_APIC_pin(apic
, pin
);
328 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t cpumask
)
332 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
333 unsigned int apicid_value
;
336 cpus_and(tmp
, cpumask
, cpu_online_map
);
340 cpus_and(cpumask
, tmp
, CPU_MASK_ALL
);
342 apicid_value
= cpu_mask_to_apicid(cpumask
);
343 /* Prepare to do the io_apic_write */
344 apicid_value
= apicid_value
<< 24;
345 spin_lock_irqsave(&ioapic_lock
, flags
);
350 io_apic_write(entry
->apic
, 0x10 + 1 + pin
*2, apicid_value
);
353 entry
= irq_2_pin
+ entry
->next
;
355 irq_desc
[irq
].affinity
= cpumask
;
356 spin_unlock_irqrestore(&ioapic_lock
, flags
);
359 #if defined(CONFIG_IRQBALANCE)
360 # include <asm/processor.h> /* kernel_thread() */
361 # include <linux/kernel_stat.h> /* kstat */
362 # include <linux/slab.h> /* kmalloc() */
363 # include <linux/timer.h>
365 #define IRQBALANCE_CHECK_ARCH -999
366 #define MAX_BALANCED_IRQ_INTERVAL (5*HZ)
367 #define MIN_BALANCED_IRQ_INTERVAL (HZ/2)
368 #define BALANCED_IRQ_MORE_DELTA (HZ/10)
369 #define BALANCED_IRQ_LESS_DELTA (HZ)
371 static int irqbalance_disabled __read_mostly
= IRQBALANCE_CHECK_ARCH
;
372 static int physical_balance __read_mostly
;
373 static long balanced_irq_interval __read_mostly
= MAX_BALANCED_IRQ_INTERVAL
;
375 static struct irq_cpu_info
{
376 unsigned long * last_irq
;
377 unsigned long * irq_delta
;
379 } irq_cpu_data
[NR_CPUS
];
381 #define CPU_IRQ(cpu) (irq_cpu_data[cpu].irq)
382 #define LAST_CPU_IRQ(cpu,irq) (irq_cpu_data[cpu].last_irq[irq])
383 #define IRQ_DELTA(cpu,irq) (irq_cpu_data[cpu].irq_delta[irq])
385 #define IDLE_ENOUGH(cpu,now) \
386 (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
388 #define IRQ_ALLOWED(cpu, allowed_mask) cpu_isset(cpu, allowed_mask)
390 #define CPU_TO_PACKAGEINDEX(i) (first_cpu(per_cpu(cpu_sibling_map, i)))
392 static cpumask_t balance_irq_affinity
[NR_IRQS
] = {
393 [0 ... NR_IRQS
-1] = CPU_MASK_ALL
396 void set_balance_irq_affinity(unsigned int irq
, cpumask_t mask
)
398 balance_irq_affinity
[irq
] = mask
;
401 static unsigned long move(int curr_cpu
, cpumask_t allowed_mask
,
402 unsigned long now
, int direction
)
410 if (unlikely(cpu
== curr_cpu
))
413 if (direction
== 1) {
422 } while (!cpu_online(cpu
) || !IRQ_ALLOWED(cpu
,allowed_mask
) ||
423 (search_idle
&& !IDLE_ENOUGH(cpu
,now
)));
428 static inline void balance_irq(int cpu
, int irq
)
430 unsigned long now
= jiffies
;
431 cpumask_t allowed_mask
;
432 unsigned int new_cpu
;
434 if (irqbalance_disabled
)
437 cpus_and(allowed_mask
, cpu_online_map
, balance_irq_affinity
[irq
]);
438 new_cpu
= move(cpu
, allowed_mask
, now
, 1);
439 if (cpu
!= new_cpu
) {
440 set_pending_irq(irq
, cpumask_of_cpu(new_cpu
));
444 static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold
)
448 for_each_online_cpu(i
) {
449 for (j
= 0; j
< NR_IRQS
; j
++) {
450 if (!irq_desc
[j
].action
)
452 /* Is it a significant load ? */
453 if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i
),j
) <
454 useful_load_threshold
)
459 balanced_irq_interval
= max((long)MIN_BALANCED_IRQ_INTERVAL
,
460 balanced_irq_interval
- BALANCED_IRQ_LESS_DELTA
);
464 static void do_irq_balance(void)
467 unsigned long max_cpu_irq
= 0, min_cpu_irq
= (~0);
468 unsigned long move_this_load
= 0;
469 int max_loaded
= 0, min_loaded
= 0;
471 unsigned long useful_load_threshold
= balanced_irq_interval
+ 10;
473 int tmp_loaded
, first_attempt
= 1;
474 unsigned long tmp_cpu_irq
;
475 unsigned long imbalance
= 0;
476 cpumask_t allowed_mask
, target_cpu_mask
, tmp
;
478 for_each_possible_cpu(i
) {
483 package_index
= CPU_TO_PACKAGEINDEX(i
);
484 for (j
= 0; j
< NR_IRQS
; j
++) {
485 unsigned long value_now
, delta
;
486 /* Is this an active IRQ or balancing disabled ? */
487 if (!irq_desc
[j
].action
|| irq_balancing_disabled(j
))
489 if ( package_index
== i
)
490 IRQ_DELTA(package_index
,j
) = 0;
491 /* Determine the total count per processor per IRQ */
492 value_now
= (unsigned long) kstat_cpu(i
).irqs
[j
];
494 /* Determine the activity per processor per IRQ */
495 delta
= value_now
- LAST_CPU_IRQ(i
,j
);
497 /* Update last_cpu_irq[][] for the next time */
498 LAST_CPU_IRQ(i
,j
) = value_now
;
500 /* Ignore IRQs whose rate is less than the clock */
501 if (delta
< useful_load_threshold
)
503 /* update the load for the processor or package total */
504 IRQ_DELTA(package_index
,j
) += delta
;
506 /* Keep track of the higher numbered sibling as well */
507 if (i
!= package_index
)
510 * We have sibling A and sibling B in the package
512 * cpu_irq[A] = load for cpu A + load for cpu B
513 * cpu_irq[B] = load for cpu B
515 CPU_IRQ(package_index
) += delta
;
518 /* Find the least loaded processor package */
519 for_each_online_cpu(i
) {
520 if (i
!= CPU_TO_PACKAGEINDEX(i
))
522 if (min_cpu_irq
> CPU_IRQ(i
)) {
523 min_cpu_irq
= CPU_IRQ(i
);
527 max_cpu_irq
= ULONG_MAX
;
530 /* Look for heaviest loaded processor.
531 * We may come back to get the next heaviest loaded processor.
532 * Skip processors with trivial loads.
536 for_each_online_cpu(i
) {
537 if (i
!= CPU_TO_PACKAGEINDEX(i
))
539 if (max_cpu_irq
<= CPU_IRQ(i
))
541 if (tmp_cpu_irq
< CPU_IRQ(i
)) {
542 tmp_cpu_irq
= CPU_IRQ(i
);
547 if (tmp_loaded
== -1) {
548 /* In the case of small number of heavy interrupt sources,
549 * loading some of the cpus too much. We use Ingo's original
550 * approach to rotate them around.
552 if (!first_attempt
&& imbalance
>= useful_load_threshold
) {
553 rotate_irqs_among_cpus(useful_load_threshold
);
556 goto not_worth_the_effort
;
559 first_attempt
= 0; /* heaviest search */
560 max_cpu_irq
= tmp_cpu_irq
; /* load */
561 max_loaded
= tmp_loaded
; /* processor */
562 imbalance
= (max_cpu_irq
- min_cpu_irq
) / 2;
564 /* if imbalance is less than approx 10% of max load, then
565 * observe diminishing returns action. - quit
567 if (imbalance
< (max_cpu_irq
>> 3))
568 goto not_worth_the_effort
;
571 /* if we select an IRQ to move that can't go where we want, then
572 * see if there is another one to try.
576 for (j
= 0; j
< NR_IRQS
; j
++) {
577 /* Is this an active IRQ? */
578 if (!irq_desc
[j
].action
)
580 if (imbalance
<= IRQ_DELTA(max_loaded
,j
))
582 /* Try to find the IRQ that is closest to the imbalance
583 * without going over.
585 if (move_this_load
< IRQ_DELTA(max_loaded
,j
)) {
586 move_this_load
= IRQ_DELTA(max_loaded
,j
);
590 if (selected_irq
== -1) {
594 imbalance
= move_this_load
;
596 /* For physical_balance case, we accumulated both load
597 * values in the one of the siblings cpu_irq[],
598 * to use the same code for physical and logical processors
599 * as much as possible.
601 * NOTE: the cpu_irq[] array holds the sum of the load for
602 * sibling A and sibling B in the slot for the lowest numbered
603 * sibling (A), _AND_ the load for sibling B in the slot for
604 * the higher numbered sibling.
606 * We seek the least loaded sibling by making the comparison
609 load
= CPU_IRQ(min_loaded
) >> 1;
610 for_each_cpu_mask(j
, per_cpu(cpu_sibling_map
, min_loaded
)) {
611 if (load
> CPU_IRQ(j
)) {
612 /* This won't change cpu_sibling_map[min_loaded] */
618 cpus_and(allowed_mask
,
620 balance_irq_affinity
[selected_irq
]);
621 target_cpu_mask
= cpumask_of_cpu(min_loaded
);
622 cpus_and(tmp
, target_cpu_mask
, allowed_mask
);
624 if (!cpus_empty(tmp
)) {
625 /* mark for change destination */
626 set_pending_irq(selected_irq
, cpumask_of_cpu(min_loaded
));
628 /* Since we made a change, come back sooner to
629 * check for more variation.
631 balanced_irq_interval
= max((long)MIN_BALANCED_IRQ_INTERVAL
,
632 balanced_irq_interval
- BALANCED_IRQ_LESS_DELTA
);
637 not_worth_the_effort
:
639 * if we did not find an IRQ to move, then adjust the time interval
642 balanced_irq_interval
= min((long)MAX_BALANCED_IRQ_INTERVAL
,
643 balanced_irq_interval
+ BALANCED_IRQ_MORE_DELTA
);
647 static int balanced_irq(void *unused
)
650 unsigned long prev_balance_time
= jiffies
;
651 long time_remaining
= balanced_irq_interval
;
653 /* push everything to CPU 0 to give us a starting point. */
654 for (i
= 0 ; i
< NR_IRQS
; i
++) {
655 irq_desc
[i
].pending_mask
= cpumask_of_cpu(0);
656 set_pending_irq(i
, cpumask_of_cpu(0));
661 time_remaining
= schedule_timeout_interruptible(time_remaining
);
663 if (time_after(jiffies
,
664 prev_balance_time
+balanced_irq_interval
)) {
667 prev_balance_time
= jiffies
;
668 time_remaining
= balanced_irq_interval
;
675 static int __init
balanced_irq_init(void)
678 struct cpuinfo_x86
*c
;
681 cpus_shift_right(tmp
, cpu_online_map
, 2);
683 /* When not overwritten by the command line ask subarchitecture. */
684 if (irqbalance_disabled
== IRQBALANCE_CHECK_ARCH
)
685 irqbalance_disabled
= NO_BALANCE_IRQ
;
686 if (irqbalance_disabled
)
689 /* disable irqbalance completely if there is only one processor online */
690 if (num_online_cpus() < 2) {
691 irqbalance_disabled
= 1;
695 * Enable physical balance only if more than 1 physical processor
698 if (smp_num_siblings
> 1 && !cpus_empty(tmp
))
699 physical_balance
= 1;
701 for_each_online_cpu(i
) {
702 irq_cpu_data
[i
].irq_delta
= kmalloc(sizeof(unsigned long) * NR_IRQS
, GFP_KERNEL
);
703 irq_cpu_data
[i
].last_irq
= kmalloc(sizeof(unsigned long) * NR_IRQS
, GFP_KERNEL
);
704 if (irq_cpu_data
[i
].irq_delta
== NULL
|| irq_cpu_data
[i
].last_irq
== NULL
) {
705 printk(KERN_ERR
"balanced_irq_init: out of memory");
708 memset(irq_cpu_data
[i
].irq_delta
,0,sizeof(unsigned long) * NR_IRQS
);
709 memset(irq_cpu_data
[i
].last_irq
,0,sizeof(unsigned long) * NR_IRQS
);
712 printk(KERN_INFO
"Starting balanced_irq\n");
713 if (!IS_ERR(kthread_run(balanced_irq
, NULL
, "kirqd")))
715 printk(KERN_ERR
"balanced_irq_init: failed to spawn balanced_irq");
717 for_each_possible_cpu(i
) {
718 kfree(irq_cpu_data
[i
].irq_delta
);
719 irq_cpu_data
[i
].irq_delta
= NULL
;
720 kfree(irq_cpu_data
[i
].last_irq
);
721 irq_cpu_data
[i
].last_irq
= NULL
;
726 int __devinit
irqbalance_disable(char *str
)
728 irqbalance_disabled
= 1;
732 __setup("noirqbalance", irqbalance_disable
);
734 late_initcall(balanced_irq_init
);
735 #endif /* CONFIG_IRQBALANCE */
736 #endif /* CONFIG_SMP */
739 void send_IPI_self(int vector
)
746 apic_wait_icr_idle();
747 cfg
= APIC_DM_FIXED
| APIC_DEST_SELF
| vector
| APIC_DEST_LOGICAL
;
749 * Send the IPI. The write to APIC_ICR fires this off.
751 apic_write_around(APIC_ICR
, cfg
);
753 #endif /* !CONFIG_SMP */
757 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
758 * specific CPU-side IRQs.
762 static int pirq_entries
[MAX_PIRQS
];
763 static int pirqs_enabled
;
764 int skip_ioapic_setup
;
766 static int __init
ioapic_pirq_setup(char *str
)
769 int ints
[MAX_PIRQS
+1];
771 get_options(str
, ARRAY_SIZE(ints
), ints
);
773 for (i
= 0; i
< MAX_PIRQS
; i
++)
774 pirq_entries
[i
] = -1;
777 apic_printk(APIC_VERBOSE
, KERN_INFO
778 "PIRQ redirection, working around broken MP-BIOS.\n");
780 if (ints
[0] < MAX_PIRQS
)
783 for (i
= 0; i
< max
; i
++) {
784 apic_printk(APIC_VERBOSE
, KERN_DEBUG
785 "... PIRQ%d -> IRQ %d\n", i
, ints
[i
+1]);
787 * PIRQs are mapped upside down, usually.
789 pirq_entries
[MAX_PIRQS
-i
-1] = ints
[i
+1];
794 __setup("pirq=", ioapic_pirq_setup
);
797 * Find the IRQ entry number of a certain pin.
799 static int find_irq_entry(int apic
, int pin
, int type
)
803 for (i
= 0; i
< mp_irq_entries
; i
++)
804 if (mp_irqs
[i
].mpc_irqtype
== type
&&
805 (mp_irqs
[i
].mpc_dstapic
== mp_ioapics
[apic
].mpc_apicid
||
806 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
) &&
807 mp_irqs
[i
].mpc_dstirq
== pin
)
814 * Find the pin to which IRQ[irq] (ISA) is connected
816 static int __init
find_isa_irq_pin(int irq
, int type
)
820 for (i
= 0; i
< mp_irq_entries
; i
++) {
821 int lbus
= mp_irqs
[i
].mpc_srcbus
;
823 if (test_bit(lbus
, mp_bus_not_pci
) &&
824 (mp_irqs
[i
].mpc_irqtype
== type
) &&
825 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
827 return mp_irqs
[i
].mpc_dstirq
;
832 static int __init
find_isa_irq_apic(int irq
, int type
)
836 for (i
= 0; i
< mp_irq_entries
; i
++) {
837 int lbus
= mp_irqs
[i
].mpc_srcbus
;
839 if (test_bit(lbus
, mp_bus_not_pci
) &&
840 (mp_irqs
[i
].mpc_irqtype
== type
) &&
841 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
844 if (i
< mp_irq_entries
) {
846 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
847 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
)
856 * Find a specific PCI IRQ entry.
857 * Not an __init, possibly needed by modules
859 static int pin_2_irq(int idx
, int apic
, int pin
);
861 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
863 int apic
, i
, best_guess
= -1;
865 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, "
866 "slot:%d, pin:%d.\n", bus
, slot
, pin
);
867 if (mp_bus_id_to_pci_bus
[bus
] == -1) {
868 printk(KERN_WARNING
"PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
871 for (i
= 0; i
< mp_irq_entries
; i
++) {
872 int lbus
= mp_irqs
[i
].mpc_srcbus
;
874 for (apic
= 0; apic
< nr_ioapics
; apic
++)
875 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
||
876 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
)
879 if (!test_bit(lbus
, mp_bus_not_pci
) &&
880 !mp_irqs
[i
].mpc_irqtype
&&
882 (slot
== ((mp_irqs
[i
].mpc_srcbusirq
>> 2) & 0x1f))) {
883 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mpc_dstirq
);
885 if (!(apic
|| IO_APIC_IRQ(irq
)))
888 if (pin
== (mp_irqs
[i
].mpc_srcbusirq
& 3))
891 * Use the first all-but-pin matching entry as a
892 * best-guess fuzzy result for broken mptables.
900 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector
);
903 * This function currently is only a helper for the i386 smp boot process where
904 * we need to reprogram the ioredtbls to cater for the cpus which have come online
905 * so mask in all cases should simply be TARGET_CPUS
908 void __init
setup_ioapic_dest(void)
910 int pin
, ioapic
, irq
, irq_entry
;
912 if (skip_ioapic_setup
== 1)
915 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++) {
916 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
917 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
920 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
921 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);
928 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
930 * EISA Edge/Level control register, ELCR
932 static int EISA_ELCR(unsigned int irq
)
935 unsigned int port
= 0x4d0 + (irq
>> 3);
936 return (inb(port
) >> (irq
& 7)) & 1;
938 apic_printk(APIC_VERBOSE
, KERN_INFO
939 "Broken MPtable reports ISA irq %d\n", irq
);
944 /* ISA interrupts are always polarity zero edge triggered,
945 * when listed as conforming in the MP table. */
947 #define default_ISA_trigger(idx) (0)
948 #define default_ISA_polarity(idx) (0)
950 /* EISA interrupts are always polarity zero and can be edge or level
951 * trigger depending on the ELCR value. If an interrupt is listed as
952 * EISA conforming in the MP table, that means its trigger type must
953 * be read in from the ELCR */
955 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
956 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
958 /* PCI interrupts are always polarity one level triggered,
959 * when listed as conforming in the MP table. */
961 #define default_PCI_trigger(idx) (1)
962 #define default_PCI_polarity(idx) (1)
964 /* MCA interrupts are always polarity zero level triggered,
965 * when listed as conforming in the MP table. */
967 #define default_MCA_trigger(idx) (1)
968 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
970 static int MPBIOS_polarity(int idx
)
972 int bus
= mp_irqs
[idx
].mpc_srcbus
;
976 * Determine IRQ line polarity (high active or low active):
978 switch (mp_irqs
[idx
].mpc_irqflag
& 3)
980 case 0: /* conforms, ie. bus-type dependent polarity */
982 polarity
= test_bit(bus
, mp_bus_not_pci
)?
983 default_ISA_polarity(idx
):
984 default_PCI_polarity(idx
);
987 case 1: /* high active */
992 case 2: /* reserved */
994 printk(KERN_WARNING
"broken BIOS!!\n");
998 case 3: /* low active */
1003 default: /* invalid */
1005 printk(KERN_WARNING
"broken BIOS!!\n");
1013 static int MPBIOS_trigger(int idx
)
1015 int bus
= mp_irqs
[idx
].mpc_srcbus
;
1019 * Determine IRQ trigger mode (edge or level sensitive):
1021 switch ((mp_irqs
[idx
].mpc_irqflag
>>2) & 3)
1023 case 0: /* conforms, ie. bus-type dependent */
1025 trigger
= test_bit(bus
, mp_bus_not_pci
)?
1026 default_ISA_trigger(idx
):
1027 default_PCI_trigger(idx
);
1028 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1029 switch (mp_bus_id_to_type
[bus
])
1031 case MP_BUS_ISA
: /* ISA pin */
1033 /* set before the switch */
1036 case MP_BUS_EISA
: /* EISA pin */
1038 trigger
= default_EISA_trigger(idx
);
1041 case MP_BUS_PCI
: /* PCI pin */
1043 /* set before the switch */
1046 case MP_BUS_MCA
: /* MCA pin */
1048 trigger
= default_MCA_trigger(idx
);
1053 printk(KERN_WARNING
"broken BIOS!!\n");
1066 case 2: /* reserved */
1068 printk(KERN_WARNING
"broken BIOS!!\n");
1077 default: /* invalid */
1079 printk(KERN_WARNING
"broken BIOS!!\n");
1087 static inline int irq_polarity(int idx
)
1089 return MPBIOS_polarity(idx
);
1092 static inline int irq_trigger(int idx
)
1094 return MPBIOS_trigger(idx
);
1097 static int pin_2_irq(int idx
, int apic
, int pin
)
1100 int bus
= mp_irqs
[idx
].mpc_srcbus
;
1103 * Debugging check, we are in big trouble if this message pops up!
1105 if (mp_irqs
[idx
].mpc_dstirq
!= pin
)
1106 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
1108 if (test_bit(bus
, mp_bus_not_pci
))
1109 irq
= mp_irqs
[idx
].mpc_srcbusirq
;
1112 * PCI IRQs are mapped in order
1116 irq
+= nr_ioapic_registers
[i
++];
1120 * For MPS mode, so far only needed by ES7000 platform
1122 if (ioapic_renumber_irq
)
1123 irq
= ioapic_renumber_irq(apic
, irq
);
1127 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1129 if ((pin
>= 16) && (pin
<= 23)) {
1130 if (pirq_entries
[pin
-16] != -1) {
1131 if (!pirq_entries
[pin
-16]) {
1132 apic_printk(APIC_VERBOSE
, KERN_DEBUG
1133 "disabling PIRQ%d\n", pin
-16);
1135 irq
= pirq_entries
[pin
-16];
1136 apic_printk(APIC_VERBOSE
, KERN_DEBUG
1137 "using PIRQ%d -> IRQ %d\n",
1145 static inline int IO_APIC_irq_trigger(int irq
)
1149 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1150 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1151 idx
= find_irq_entry(apic
,pin
,mp_INT
);
1152 if ((idx
!= -1) && (irq
== pin_2_irq(idx
,apic
,pin
)))
1153 return irq_trigger(idx
);
1157 * nonexistent IRQs are edge default
1162 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1163 static u8 irq_vector
[NR_IRQ_VECTORS
] __read_mostly
= { FIRST_DEVICE_VECTOR
, 0 };
1165 static int __assign_irq_vector(int irq
)
1167 static int current_vector
= FIRST_DEVICE_VECTOR
, current_offset
= 0;
1170 BUG_ON((unsigned)irq
>= NR_IRQ_VECTORS
);
1172 if (irq_vector
[irq
] > 0)
1173 return irq_vector
[irq
];
1175 vector
= current_vector
;
1176 offset
= current_offset
;
1179 if (vector
>= FIRST_SYSTEM_VECTOR
) {
1180 offset
= (offset
+ 1) % 8;
1181 vector
= FIRST_DEVICE_VECTOR
+ offset
;
1183 if (vector
== current_vector
)
1185 if (test_and_set_bit(vector
, used_vectors
))
1188 current_vector
= vector
;
1189 current_offset
= offset
;
1190 irq_vector
[irq
] = vector
;
1195 static int assign_irq_vector(int irq
)
1197 unsigned long flags
;
1200 spin_lock_irqsave(&vector_lock
, flags
);
1201 vector
= __assign_irq_vector(irq
);
1202 spin_unlock_irqrestore(&vector_lock
, flags
);
1206 static struct irq_chip ioapic_chip
;
1208 #define IOAPIC_AUTO -1
1209 #define IOAPIC_EDGE 0
1210 #define IOAPIC_LEVEL 1
1212 static void ioapic_register_intr(int irq
, int vector
, unsigned long trigger
)
1214 if ((trigger
== IOAPIC_AUTO
&& IO_APIC_irq_trigger(irq
)) ||
1215 trigger
== IOAPIC_LEVEL
) {
1216 irq_desc
[irq
].status
|= IRQ_LEVEL
;
1217 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
1218 handle_fasteoi_irq
, "fasteoi");
1220 irq_desc
[irq
].status
&= ~IRQ_LEVEL
;
1221 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
1222 handle_edge_irq
, "edge");
1224 set_intr_gate(vector
, interrupt
[irq
]);
1227 static void __init
setup_IO_APIC_irqs(void)
1229 struct IO_APIC_route_entry entry
;
1230 int apic
, pin
, idx
, irq
, first_notcon
= 1, vector
;
1232 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
1234 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1235 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1238 * add it to the IO-APIC irq-routing table:
1240 memset(&entry
,0,sizeof(entry
));
1242 entry
.delivery_mode
= INT_DELIVERY_MODE
;
1243 entry
.dest_mode
= INT_DEST_MODE
;
1244 entry
.mask
= 0; /* enable IRQ */
1245 entry
.dest
.logical
.logical_dest
=
1246 cpu_mask_to_apicid(TARGET_CPUS
);
1248 idx
= find_irq_entry(apic
,pin
,mp_INT
);
1251 apic_printk(APIC_VERBOSE
, KERN_DEBUG
1252 " IO-APIC (apicid-pin) %d-%d",
1253 mp_ioapics
[apic
].mpc_apicid
,
1257 apic_printk(APIC_VERBOSE
, ", %d-%d",
1258 mp_ioapics
[apic
].mpc_apicid
, pin
);
1262 if (!first_notcon
) {
1263 apic_printk(APIC_VERBOSE
, " not connected.\n");
1267 entry
.trigger
= irq_trigger(idx
);
1268 entry
.polarity
= irq_polarity(idx
);
1270 if (irq_trigger(idx
)) {
1275 irq
= pin_2_irq(idx
, apic
, pin
);
1277 * skip adding the timer int on secondary nodes, which causes
1278 * a small but painful rift in the time-space continuum
1280 if (multi_timer_check(apic
, irq
))
1283 add_pin_to_irq(irq
, apic
, pin
);
1285 if (!apic
&& !IO_APIC_IRQ(irq
))
1288 if (IO_APIC_IRQ(irq
)) {
1289 vector
= assign_irq_vector(irq
);
1290 entry
.vector
= vector
;
1291 ioapic_register_intr(irq
, vector
, IOAPIC_AUTO
);
1293 if (!apic
&& (irq
< 16))
1294 disable_8259A_irq(irq
);
1296 ioapic_write_entry(apic
, pin
, entry
);
1301 apic_printk(APIC_VERBOSE
, " not connected.\n");
1305 * Set up the timer pin, possibly with the 8259A-master behind.
1307 static void __init
setup_timer_IRQ0_pin(unsigned int apic
, unsigned int pin
,
1310 struct IO_APIC_route_entry entry
;
1312 memset(&entry
,0,sizeof(entry
));
1315 * We use logical delivery to get the timer IRQ
1318 entry
.dest_mode
= INT_DEST_MODE
;
1319 entry
.mask
= 1; /* mask IRQ now */
1320 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
1321 entry
.delivery_mode
= INT_DELIVERY_MODE
;
1324 entry
.vector
= vector
;
1327 * The timer IRQ doesn't have to know that behind the
1328 * scene we may have a 8259A-master in AEOI mode ...
1330 ioapic_register_intr(0, vector
, IOAPIC_EDGE
);
1333 * Add it to the IO-APIC irq-routing table:
1335 ioapic_write_entry(apic
, pin
, entry
);
1338 void __init
print_IO_APIC(void)
1341 union IO_APIC_reg_00 reg_00
;
1342 union IO_APIC_reg_01 reg_01
;
1343 union IO_APIC_reg_02 reg_02
;
1344 union IO_APIC_reg_03 reg_03
;
1345 unsigned long flags
;
1347 if (apic_verbosity
== APIC_QUIET
)
1350 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
1351 for (i
= 0; i
< nr_ioapics
; i
++)
1352 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
1353 mp_ioapics
[i
].mpc_apicid
, nr_ioapic_registers
[i
]);
1356 * We are a bit conservative about what we expect. We have to
1357 * know about every hardware change ASAP.
1359 printk(KERN_INFO
"testing the IO APIC.......................\n");
1361 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1363 spin_lock_irqsave(&ioapic_lock
, flags
);
1364 reg_00
.raw
= io_apic_read(apic
, 0);
1365 reg_01
.raw
= io_apic_read(apic
, 1);
1366 if (reg_01
.bits
.version
>= 0x10)
1367 reg_02
.raw
= io_apic_read(apic
, 2);
1368 if (reg_01
.bits
.version
>= 0x20)
1369 reg_03
.raw
= io_apic_read(apic
, 3);
1370 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1372 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].mpc_apicid
);
1373 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
1374 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
1375 printk(KERN_DEBUG
"....... : Delivery Type: %X\n", reg_00
.bits
.delivery_type
);
1376 printk(KERN_DEBUG
"....... : LTS : %X\n", reg_00
.bits
.LTS
);
1378 printk(KERN_DEBUG
".... register #01: %08X\n", reg_01
.raw
);
1379 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
1381 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
1382 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
1385 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1386 * but the value of reg_02 is read as the previous read register
1387 * value, so ignore it if reg_02 == reg_01.
1389 if (reg_01
.bits
.version
>= 0x10 && reg_02
.raw
!= reg_01
.raw
) {
1390 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
1391 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
1395 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1396 * or reg_03, but the value of reg_0[23] is read as the previous read
1397 * register value, so ignore it if reg_03 == reg_0[12].
1399 if (reg_01
.bits
.version
>= 0x20 && reg_03
.raw
!= reg_02
.raw
&&
1400 reg_03
.raw
!= reg_01
.raw
) {
1401 printk(KERN_DEBUG
".... register #03: %08X\n", reg_03
.raw
);
1402 printk(KERN_DEBUG
"....... : Boot DT : %X\n", reg_03
.bits
.boot_DT
);
1405 printk(KERN_DEBUG
".... IRQ redirection table:\n");
1407 printk(KERN_DEBUG
" NR Log Phy Mask Trig IRR Pol"
1408 " Stat Dest Deli Vect: \n");
1410 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
1411 struct IO_APIC_route_entry entry
;
1413 entry
= ioapic_read_entry(apic
, i
);
1415 printk(KERN_DEBUG
" %02x %03X %02X ",
1417 entry
.dest
.logical
.logical_dest
,
1418 entry
.dest
.physical
.physical_dest
1421 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1426 entry
.delivery_status
,
1428 entry
.delivery_mode
,
1433 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
1434 for (i
= 0; i
< NR_IRQS
; i
++) {
1435 struct irq_pin_list
*entry
= irq_2_pin
+ i
;
1438 printk(KERN_DEBUG
"IRQ%d ", i
);
1440 printk("-> %d:%d", entry
->apic
, entry
->pin
);
1443 entry
= irq_2_pin
+ entry
->next
;
1448 printk(KERN_INFO
".................................... done.\n");
1455 static void print_APIC_bitfield (int base
)
1460 if (apic_verbosity
== APIC_QUIET
)
1463 printk(KERN_DEBUG
"0123456789abcdef0123456789abcdef\n" KERN_DEBUG
);
1464 for (i
= 0; i
< 8; i
++) {
1465 v
= apic_read(base
+ i
*0x10);
1466 for (j
= 0; j
< 32; j
++) {
1476 void /*__init*/ print_local_APIC(void * dummy
)
1478 unsigned int v
, ver
, maxlvt
;
1480 if (apic_verbosity
== APIC_QUIET
)
1483 printk("\n" KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
1484 smp_processor_id(), hard_smp_processor_id());
1485 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
,
1486 GET_APIC_ID(read_apic_id()));
1487 v
= apic_read(APIC_LVR
);
1488 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
1489 ver
= GET_APIC_VERSION(v
);
1490 maxlvt
= lapic_get_maxlvt();
1492 v
= apic_read(APIC_TASKPRI
);
1493 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
1495 if (APIC_INTEGRATED(ver
)) { /* !82489DX */
1496 v
= apic_read(APIC_ARBPRI
);
1497 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
1498 v
& APIC_ARBPRI_MASK
);
1499 v
= apic_read(APIC_PROCPRI
);
1500 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
1503 v
= apic_read(APIC_EOI
);
1504 printk(KERN_DEBUG
"... APIC EOI: %08x\n", v
);
1505 v
= apic_read(APIC_RRR
);
1506 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
1507 v
= apic_read(APIC_LDR
);
1508 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
1509 v
= apic_read(APIC_DFR
);
1510 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
1511 v
= apic_read(APIC_SPIV
);
1512 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
1514 printk(KERN_DEBUG
"... APIC ISR field:\n");
1515 print_APIC_bitfield(APIC_ISR
);
1516 printk(KERN_DEBUG
"... APIC TMR field:\n");
1517 print_APIC_bitfield(APIC_TMR
);
1518 printk(KERN_DEBUG
"... APIC IRR field:\n");
1519 print_APIC_bitfield(APIC_IRR
);
1521 if (APIC_INTEGRATED(ver
)) { /* !82489DX */
1522 if (maxlvt
> 3) /* Due to the Pentium erratum 3AP. */
1523 apic_write(APIC_ESR
, 0);
1524 v
= apic_read(APIC_ESR
);
1525 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
1528 v
= apic_read(APIC_ICR
);
1529 printk(KERN_DEBUG
"... APIC ICR: %08x\n", v
);
1530 v
= apic_read(APIC_ICR2
);
1531 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", v
);
1533 v
= apic_read(APIC_LVTT
);
1534 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1536 if (maxlvt
> 3) { /* PC is LVT#4. */
1537 v
= apic_read(APIC_LVTPC
);
1538 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
1540 v
= apic_read(APIC_LVT0
);
1541 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
1542 v
= apic_read(APIC_LVT1
);
1543 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
1545 if (maxlvt
> 2) { /* ERR is LVT#3. */
1546 v
= apic_read(APIC_LVTERR
);
1547 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
1550 v
= apic_read(APIC_TMICT
);
1551 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
1552 v
= apic_read(APIC_TMCCT
);
1553 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
1554 v
= apic_read(APIC_TDCR
);
1555 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1559 void print_all_local_APICs (void)
1561 on_each_cpu(print_local_APIC
, NULL
, 1, 1);
1564 void /*__init*/ print_PIC(void)
1567 unsigned long flags
;
1569 if (apic_verbosity
== APIC_QUIET
)
1572 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1574 spin_lock_irqsave(&i8259A_lock
, flags
);
1576 v
= inb(0xa1) << 8 | inb(0x21);
1577 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1579 v
= inb(0xa0) << 8 | inb(0x20);
1580 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1584 v
= inb(0xa0) << 8 | inb(0x20);
1588 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1590 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1592 v
= inb(0x4d1) << 8 | inb(0x4d0);
1593 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1598 static void __init
enable_IO_APIC(void)
1600 union IO_APIC_reg_01 reg_01
;
1601 int i8259_apic
, i8259_pin
;
1603 unsigned long flags
;
1605 for (i
= 0; i
< PIN_MAP_SIZE
; i
++) {
1606 irq_2_pin
[i
].pin
= -1;
1607 irq_2_pin
[i
].next
= 0;
1610 for (i
= 0; i
< MAX_PIRQS
; i
++)
1611 pirq_entries
[i
] = -1;
1614 * The number of IO-APIC IRQ registers (== #pins):
1616 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1617 spin_lock_irqsave(&ioapic_lock
, flags
);
1618 reg_01
.raw
= io_apic_read(apic
, 1);
1619 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1620 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
1622 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1624 /* See if any of the pins is in ExtINT mode */
1625 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1626 struct IO_APIC_route_entry entry
;
1627 entry
= ioapic_read_entry(apic
, pin
);
1630 /* If the interrupt line is enabled and in ExtInt mode
1631 * I have found the pin where the i8259 is connected.
1633 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1634 ioapic_i8259
.apic
= apic
;
1635 ioapic_i8259
.pin
= pin
;
1641 /* Look to see what if the MP table has reported the ExtINT */
1642 /* If we could not find the appropriate pin by looking at the ioapic
1643 * the i8259 probably is not connected the ioapic but give the
1644 * mptable a chance anyway.
1646 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1647 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1648 /* Trust the MP table if nothing is setup in the hardware */
1649 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1650 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1651 ioapic_i8259
.pin
= i8259_pin
;
1652 ioapic_i8259
.apic
= i8259_apic
;
1654 /* Complain if the MP table and the hardware disagree */
1655 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1656 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1658 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1662 * Do not trust the IO-APIC being empty at bootup
1668 * Not an __init, needed by the reboot code
1670 void disable_IO_APIC(void)
1673 * Clear the IO-APIC before rebooting:
1678 * If the i8259 is routed through an IOAPIC
1679 * Put that IOAPIC in virtual wire mode
1680 * so legacy interrupts can be delivered.
1682 if (ioapic_i8259
.pin
!= -1) {
1683 struct IO_APIC_route_entry entry
;
1685 memset(&entry
, 0, sizeof(entry
));
1686 entry
.mask
= 0; /* Enabled */
1687 entry
.trigger
= 0; /* Edge */
1689 entry
.polarity
= 0; /* High */
1690 entry
.delivery_status
= 0;
1691 entry
.dest_mode
= 0; /* Physical */
1692 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1694 entry
.dest
.physical
.physical_dest
=
1695 GET_APIC_ID(read_apic_id());
1698 * Add it to the IO-APIC irq-routing table:
1700 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1702 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
1706 * function to set the IO-APIC physical IDs based on the
1707 * values stored in the MPC table.
1709 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1712 #ifndef CONFIG_X86_NUMAQ
1713 static void __init
setup_ioapic_ids_from_mpc(void)
1715 union IO_APIC_reg_00 reg_00
;
1716 physid_mask_t phys_id_present_map
;
1719 unsigned char old_id
;
1720 unsigned long flags
;
1723 * Don't check I/O APIC IDs for xAPIC systems. They have
1724 * no meaning without the serial APIC bus.
1726 if (!(boot_cpu_data
.x86_vendor
== X86_VENDOR_INTEL
)
1727 || APIC_XAPIC(apic_version
[boot_cpu_physical_apicid
]))
1730 * This is broken; anything with a real cpu count has to
1731 * circumvent this idiocy regardless.
1733 phys_id_present_map
= ioapic_phys_id_map(phys_cpu_present_map
);
1736 * Set the IOAPIC ID to the value stored in the MPC table.
1738 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1740 /* Read the register 0 value */
1741 spin_lock_irqsave(&ioapic_lock
, flags
);
1742 reg_00
.raw
= io_apic_read(apic
, 0);
1743 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1745 old_id
= mp_ioapics
[apic
].mpc_apicid
;
1747 if (mp_ioapics
[apic
].mpc_apicid
>= get_physical_broadcast()) {
1748 printk(KERN_ERR
"BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1749 apic
, mp_ioapics
[apic
].mpc_apicid
);
1750 printk(KERN_ERR
"... fixing up to %d. (tell your hw vendor)\n",
1752 mp_ioapics
[apic
].mpc_apicid
= reg_00
.bits
.ID
;
1756 * Sanity check, is the ID really free? Every APIC in a
1757 * system must have a unique ID or we get lots of nice
1758 * 'stuck on smp_invalidate_needed IPI wait' messages.
1760 if (check_apicid_used(phys_id_present_map
,
1761 mp_ioapics
[apic
].mpc_apicid
)) {
1762 printk(KERN_ERR
"BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1763 apic
, mp_ioapics
[apic
].mpc_apicid
);
1764 for (i
= 0; i
< get_physical_broadcast(); i
++)
1765 if (!physid_isset(i
, phys_id_present_map
))
1767 if (i
>= get_physical_broadcast())
1768 panic("Max APIC ID exceeded!\n");
1769 printk(KERN_ERR
"... fixing up to %d. (tell your hw vendor)\n",
1771 physid_set(i
, phys_id_present_map
);
1772 mp_ioapics
[apic
].mpc_apicid
= i
;
1775 tmp
= apicid_to_cpu_present(mp_ioapics
[apic
].mpc_apicid
);
1776 apic_printk(APIC_VERBOSE
, "Setting %d in the "
1777 "phys_id_present_map\n",
1778 mp_ioapics
[apic
].mpc_apicid
);
1779 physids_or(phys_id_present_map
, phys_id_present_map
, tmp
);
1784 * We need to adjust the IRQ routing table
1785 * if the ID changed.
1787 if (old_id
!= mp_ioapics
[apic
].mpc_apicid
)
1788 for (i
= 0; i
< mp_irq_entries
; i
++)
1789 if (mp_irqs
[i
].mpc_dstapic
== old_id
)
1790 mp_irqs
[i
].mpc_dstapic
1791 = mp_ioapics
[apic
].mpc_apicid
;
1794 * Read the right value from the MPC table and
1795 * write it into the ID register.
1797 apic_printk(APIC_VERBOSE
, KERN_INFO
1798 "...changing IO-APIC physical APIC ID to %d ...",
1799 mp_ioapics
[apic
].mpc_apicid
);
1801 reg_00
.bits
.ID
= mp_ioapics
[apic
].mpc_apicid
;
1802 spin_lock_irqsave(&ioapic_lock
, flags
);
1803 io_apic_write(apic
, 0, reg_00
.raw
);
1804 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1809 spin_lock_irqsave(&ioapic_lock
, flags
);
1810 reg_00
.raw
= io_apic_read(apic
, 0);
1811 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1812 if (reg_00
.bits
.ID
!= mp_ioapics
[apic
].mpc_apicid
)
1813 printk("could not set ID!\n");
1815 apic_printk(APIC_VERBOSE
, " ok.\n");
1819 static void __init
setup_ioapic_ids_from_mpc(void) { }
1822 int no_timer_check __initdata
;
1824 static int __init
notimercheck(char *s
)
1829 __setup("no_timer_check", notimercheck
);
1832 * There is a nasty bug in some older SMP boards, their mptable lies
1833 * about the timer IRQ. We do the following to work around the situation:
1835 * - timer IRQ defaults to IO-APIC IRQ
1836 * - if this function detects that timer IRQs are defunct, then we fall
1837 * back to ISA timer IRQs
1839 static int __init
timer_irq_works(void)
1841 unsigned long t1
= jiffies
;
1842 unsigned long flags
;
1847 local_save_flags(flags
);
1849 /* Let ten ticks pass... */
1850 mdelay((10 * 1000) / HZ
);
1851 local_irq_restore(flags
);
1854 * Expect a few ticks at least, to be sure some possible
1855 * glue logic does not lock up after one or two first
1856 * ticks in a non-ExtINT mode. Also the local APIC
1857 * might have cached one ExtINT interrupt. Finally, at
1858 * least one tick may be lost due to delays.
1860 if (time_after(jiffies
, t1
+ 4))
1867 * In the SMP+IOAPIC case it might happen that there are an unspecified
1868 * number of pending IRQ events unhandled. These cases are very rare,
1869 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1870 * better to do it this way as thus we do not have to be aware of
1871 * 'pending' interrupts in the IRQ path, except at this point.
1874 * Edge triggered needs to resend any interrupt
1875 * that was delayed but this is now handled in the device
1882 * Starting up a edge-triggered IO-APIC interrupt is
1883 * nasty - we need to make sure that we get the edge.
1884 * If it is already asserted for some reason, we need
1885 * return 1 to indicate that is was pending.
1887 * This is not complete - we should be able to fake
1888 * an edge even if it isn't on the 8259A...
1890 * (We do this for level-triggered IRQs too - it cannot hurt.)
1892 static unsigned int startup_ioapic_irq(unsigned int irq
)
1894 int was_pending
= 0;
1895 unsigned long flags
;
1897 spin_lock_irqsave(&ioapic_lock
, flags
);
1899 disable_8259A_irq(irq
);
1900 if (i8259A_irq_pending(irq
))
1903 __unmask_IO_APIC_irq(irq
);
1904 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1909 static void ack_ioapic_irq(unsigned int irq
)
1911 move_native_irq(irq
);
1915 static void ack_ioapic_quirk_irq(unsigned int irq
)
1920 move_native_irq(irq
);
1922 * It appears there is an erratum which affects at least version 0x11
1923 * of I/O APIC (that's the 82093AA and cores integrated into various
1924 * chipsets). Under certain conditions a level-triggered interrupt is
1925 * erroneously delivered as edge-triggered one but the respective IRR
1926 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1927 * message but it will never arrive and further interrupts are blocked
1928 * from the source. The exact reason is so far unknown, but the
1929 * phenomenon was observed when two consecutive interrupt requests
1930 * from a given source get delivered to the same CPU and the source is
1931 * temporarily disabled in between.
1933 * A workaround is to simulate an EOI message manually. We achieve it
1934 * by setting the trigger mode to edge and then to level when the edge
1935 * trigger mode gets detected in the TMR of a local APIC for a
1936 * level-triggered interrupt. We mask the source for the time of the
1937 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1938 * The idea is from Manfred Spraul. --macro
1940 i
= irq_vector
[irq
];
1942 v
= apic_read(APIC_TMR
+ ((i
& ~0x1f) >> 1));
1946 if (!(v
& (1 << (i
& 0x1f)))) {
1947 atomic_inc(&irq_mis_count
);
1948 spin_lock(&ioapic_lock
);
1949 __mask_and_edge_IO_APIC_irq(irq
);
1950 __unmask_and_level_IO_APIC_irq(irq
);
1951 spin_unlock(&ioapic_lock
);
1955 static int ioapic_retrigger_irq(unsigned int irq
)
1957 send_IPI_self(irq_vector
[irq
]);
1962 static struct irq_chip ioapic_chip __read_mostly
= {
1964 .startup
= startup_ioapic_irq
,
1965 .mask
= mask_IO_APIC_irq
,
1966 .unmask
= unmask_IO_APIC_irq
,
1967 .ack
= ack_ioapic_irq
,
1968 .eoi
= ack_ioapic_quirk_irq
,
1970 .set_affinity
= set_ioapic_affinity_irq
,
1972 .retrigger
= ioapic_retrigger_irq
,
1976 static inline void init_IO_APIC_traps(void)
1981 * NOTE! The local APIC isn't very good at handling
1982 * multiple interrupts at the same interrupt level.
1983 * As the interrupt level is determined by taking the
1984 * vector number and shifting that right by 4, we
1985 * want to spread these out a bit so that they don't
1986 * all fall in the same interrupt level.
1988 * Also, we've got to be careful not to trash gate
1989 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1991 for (irq
= 0; irq
< NR_IRQS
; irq
++) {
1992 if (IO_APIC_IRQ(irq
) && !irq_vector
[irq
]) {
1994 * Hmm.. We don't have an entry for this,
1995 * so default to an old-fashioned 8259
1996 * interrupt if we can..
1999 make_8259A_irq(irq
);
2001 /* Strange. Oh, well.. */
2002 irq_desc
[irq
].chip
= &no_irq_chip
;
2008 * The local APIC irq-chip implementation:
2011 static void ack_apic(unsigned int irq
)
2016 static void mask_lapic_irq (unsigned int irq
)
2020 v
= apic_read(APIC_LVT0
);
2021 apic_write_around(APIC_LVT0
, v
| APIC_LVT_MASKED
);
2024 static void unmask_lapic_irq (unsigned int irq
)
2028 v
= apic_read(APIC_LVT0
);
2029 apic_write_around(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
2032 static struct irq_chip lapic_chip __read_mostly
= {
2033 .name
= "local-APIC",
2034 .mask
= mask_lapic_irq
,
2035 .unmask
= unmask_lapic_irq
,
2039 static void __init
setup_nmi(void)
2042 * Dirty trick to enable the NMI watchdog ...
2043 * We put the 8259A master into AEOI mode and
2044 * unmask on all local APICs LVT0 as NMI.
2046 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2047 * is from Maciej W. Rozycki - so we do not have to EOI from
2048 * the NMI handler or the timer interrupt.
2050 apic_printk(APIC_VERBOSE
, KERN_INFO
"activating NMI Watchdog ...");
2052 enable_NMI_through_LVT0();
2054 apic_printk(APIC_VERBOSE
, " done.\n");
2058 * This looks a bit hackish but it's about the only one way of sending
2059 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2060 * not support the ExtINT mode, unfortunately. We need to send these
2061 * cycles as some i82489DX-based boards have glue logic that keeps the
2062 * 8259A interrupt line asserted until INTA. --macro
2064 static inline void __init
unlock_ExtINT_logic(void)
2067 struct IO_APIC_route_entry entry0
, entry1
;
2068 unsigned char save_control
, save_freq_select
;
2070 pin
= find_isa_irq_pin(8, mp_INT
);
2075 apic
= find_isa_irq_apic(8, mp_INT
);
2081 entry0
= ioapic_read_entry(apic
, pin
);
2082 clear_IO_APIC_pin(apic
, pin
);
2084 memset(&entry1
, 0, sizeof(entry1
));
2086 entry1
.dest_mode
= 0; /* physical delivery */
2087 entry1
.mask
= 0; /* unmask IRQ now */
2088 entry1
.dest
.physical
.physical_dest
= hard_smp_processor_id();
2089 entry1
.delivery_mode
= dest_ExtINT
;
2090 entry1
.polarity
= entry0
.polarity
;
2094 ioapic_write_entry(apic
, pin
, entry1
);
2096 save_control
= CMOS_READ(RTC_CONTROL
);
2097 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
2098 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
2100 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
2105 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
2109 CMOS_WRITE(save_control
, RTC_CONTROL
);
2110 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
2111 clear_IO_APIC_pin(apic
, pin
);
2113 ioapic_write_entry(apic
, pin
, entry0
);
2117 * This code may look a bit paranoid, but it's supposed to cooperate with
2118 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2119 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2120 * fanatically on his truly buggy board.
2122 static inline void __init
check_timer(void)
2124 int apic1
, pin1
, apic2
, pin2
;
2127 unsigned long flags
;
2129 local_irq_save(flags
);
2131 ver
= apic_read(APIC_LVR
);
2132 ver
= GET_APIC_VERSION(ver
);
2135 * get/set the timer IRQ vector:
2137 disable_8259A_irq(0);
2138 vector
= assign_irq_vector(0);
2139 set_intr_gate(vector
, interrupt
[0]);
2142 * As IRQ0 is to be enabled in the 8259A, the virtual
2143 * wire has to be disabled in the local APIC. Also
2144 * timer interrupts need to be acknowledged manually in
2145 * the 8259A for the i82489DX when using the NMI
2146 * watchdog as that APIC treats NMIs as level-triggered.
2147 * The AEOI mode will finish them in the 8259A
2150 apic_write_around(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
2152 timer_ack
= (nmi_watchdog
== NMI_IO_APIC
&& !APIC_INTEGRATED(ver
));
2154 pin1
= find_isa_irq_pin(0, mp_INT
);
2155 apic1
= find_isa_irq_apic(0, mp_INT
);
2156 pin2
= ioapic_i8259
.pin
;
2157 apic2
= ioapic_i8259
.apic
;
2159 printk(KERN_INFO
"..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
2160 vector
, apic1
, pin1
, apic2
, pin2
);
2164 * Ok, does IRQ0 through the IOAPIC work?
2166 unmask_IO_APIC_irq(0);
2167 if (timer_irq_works()) {
2168 if (nmi_watchdog
== NMI_IO_APIC
) {
2170 enable_8259A_irq(0);
2172 if (disable_timer_pin_1
> 0)
2173 clear_IO_APIC_pin(0, pin1
);
2176 clear_IO_APIC_pin(apic1
, pin1
);
2177 printk(KERN_ERR
"..MP-BIOS bug: 8254 timer not connected to "
2181 printk(KERN_INFO
"...trying to set up timer (IRQ0) through the 8259A ... ");
2183 printk("\n..... (found pin %d) ...", pin2
);
2185 * legacy devices should be connected to IO APIC #0
2187 setup_timer_IRQ0_pin(apic2
, pin2
, vector
);
2188 unmask_IO_APIC_irq(0);
2189 enable_8259A_irq(0);
2190 if (timer_irq_works()) {
2192 timer_through_8259
= 1;
2194 replace_pin_at_irq(0, apic1
, pin1
, apic2
, pin2
);
2196 add_pin_to_irq(0, apic2
, pin2
);
2197 if (nmi_watchdog
== NMI_IO_APIC
) {
2198 disable_8259A_irq(0);
2200 enable_8259A_irq(0);
2205 * Cleanup, just in case ...
2207 disable_8259A_irq(0);
2208 clear_IO_APIC_pin(apic2
, pin2
);
2210 printk(" failed.\n");
2212 if (nmi_watchdog
== NMI_IO_APIC
) {
2213 printk(KERN_WARNING
"timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2218 printk(KERN_INFO
"...trying to set up timer as Virtual Wire IRQ...");
2220 set_irq_chip_and_handler_name(0, &lapic_chip
, handle_fasteoi_irq
,
2222 apic_write_around(APIC_LVT0
, APIC_DM_FIXED
| vector
); /* Fixed mode */
2223 enable_8259A_irq(0);
2225 if (timer_irq_works()) {
2226 printk(" works.\n");
2229 disable_8259A_irq(0);
2230 apic_write_around(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| vector
);
2231 printk(" failed.\n");
2233 printk(KERN_INFO
"...trying to set up timer as ExtINT IRQ...");
2237 apic_write_around(APIC_LVT0
, APIC_DM_EXTINT
);
2239 unlock_ExtINT_logic();
2241 if (timer_irq_works()) {
2242 printk(" works.\n");
2245 printk(" failed :(.\n");
2246 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2247 "report. Then try booting with the 'noapic' option");
2249 local_irq_restore(flags
);
2254 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2255 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2256 * Linux doesn't really care, as it's not actually used
2257 * for any interrupt handling anyway.
2259 #define PIC_IRQS (1 << PIC_CASCADE_IR)
2261 void __init
setup_IO_APIC(void)
2265 /* Reserve all the system vectors. */
2266 for (i
= FIRST_SYSTEM_VECTOR
; i
< NR_VECTORS
; i
++)
2267 set_bit(i
, used_vectors
);
2272 io_apic_irqs
= ~0; /* all IRQs go through IOAPIC */
2274 io_apic_irqs
= ~PIC_IRQS
;
2276 printk("ENABLING IO-APIC IRQs\n");
2279 * Set up IO-APIC IRQ routing.
2282 setup_ioapic_ids_from_mpc();
2284 setup_IO_APIC_irqs();
2285 init_IO_APIC_traps();
2292 * Called after all the initialization is done. If we didnt find any
2293 * APIC bugs then we can allow the modify fast path
2296 static int __init
io_apic_bug_finalize(void)
2298 if(sis_apic_bug
== -1)
2303 late_initcall(io_apic_bug_finalize
);
2305 struct sysfs_ioapic_data
{
2306 struct sys_device dev
;
2307 struct IO_APIC_route_entry entry
[0];
2309 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
2311 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
2313 struct IO_APIC_route_entry
*entry
;
2314 struct sysfs_ioapic_data
*data
;
2317 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
2318 entry
= data
->entry
;
2319 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
2320 entry
[i
] = ioapic_read_entry(dev
->id
, i
);
2325 static int ioapic_resume(struct sys_device
*dev
)
2327 struct IO_APIC_route_entry
*entry
;
2328 struct sysfs_ioapic_data
*data
;
2329 unsigned long flags
;
2330 union IO_APIC_reg_00 reg_00
;
2333 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
2334 entry
= data
->entry
;
2336 spin_lock_irqsave(&ioapic_lock
, flags
);
2337 reg_00
.raw
= io_apic_read(dev
->id
, 0);
2338 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].mpc_apicid
) {
2339 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].mpc_apicid
;
2340 io_apic_write(dev
->id
, 0, reg_00
.raw
);
2342 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2343 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
2344 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
2349 static struct sysdev_class ioapic_sysdev_class
= {
2351 .suspend
= ioapic_suspend
,
2352 .resume
= ioapic_resume
,
2355 static int __init
ioapic_init_sysfs(void)
2357 struct sys_device
* dev
;
2358 int i
, size
, error
= 0;
2360 error
= sysdev_class_register(&ioapic_sysdev_class
);
2364 for (i
= 0; i
< nr_ioapics
; i
++ ) {
2365 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
2366 * sizeof(struct IO_APIC_route_entry
);
2367 mp_ioapic_data
[i
] = kmalloc(size
, GFP_KERNEL
);
2368 if (!mp_ioapic_data
[i
]) {
2369 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
2372 memset(mp_ioapic_data
[i
], 0, size
);
2373 dev
= &mp_ioapic_data
[i
]->dev
;
2375 dev
->cls
= &ioapic_sysdev_class
;
2376 error
= sysdev_register(dev
);
2378 kfree(mp_ioapic_data
[i
]);
2379 mp_ioapic_data
[i
] = NULL
;
2380 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
2388 device_initcall(ioapic_init_sysfs
);
2391 * Dynamic irq allocate and deallocation
2393 int create_irq(void)
2395 /* Allocate an unused irq */
2396 int irq
, new, vector
= 0;
2397 unsigned long flags
;
2400 spin_lock_irqsave(&vector_lock
, flags
);
2401 for (new = (NR_IRQS
- 1); new >= 0; new--) {
2402 if (platform_legacy_irq(new))
2404 if (irq_vector
[new] != 0)
2406 vector
= __assign_irq_vector(new);
2407 if (likely(vector
> 0))
2411 spin_unlock_irqrestore(&vector_lock
, flags
);
2414 set_intr_gate(vector
, interrupt
[irq
]);
2415 dynamic_irq_init(irq
);
2420 void destroy_irq(unsigned int irq
)
2422 unsigned long flags
;
2424 dynamic_irq_cleanup(irq
);
2426 spin_lock_irqsave(&vector_lock
, flags
);
2427 clear_bit(irq_vector
[irq
], used_vectors
);
2428 irq_vector
[irq
] = 0;
2429 spin_unlock_irqrestore(&vector_lock
, flags
);
2433 * MSI message composition
2435 #ifdef CONFIG_PCI_MSI
2436 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
2441 vector
= assign_irq_vector(irq
);
2443 dest
= cpu_mask_to_apicid(TARGET_CPUS
);
2445 msg
->address_hi
= MSI_ADDR_BASE_HI
;
2448 ((INT_DEST_MODE
== 0) ?
2449 MSI_ADDR_DEST_MODE_PHYSICAL
:
2450 MSI_ADDR_DEST_MODE_LOGICAL
) |
2451 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2452 MSI_ADDR_REDIRECTION_CPU
:
2453 MSI_ADDR_REDIRECTION_LOWPRI
) |
2454 MSI_ADDR_DEST_ID(dest
);
2457 MSI_DATA_TRIGGER_EDGE
|
2458 MSI_DATA_LEVEL_ASSERT
|
2459 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2460 MSI_DATA_DELIVERY_FIXED
:
2461 MSI_DATA_DELIVERY_LOWPRI
) |
2462 MSI_DATA_VECTOR(vector
);
2468 static void set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
2475 cpus_and(tmp
, mask
, cpu_online_map
);
2476 if (cpus_empty(tmp
))
2479 vector
= assign_irq_vector(irq
);
2483 dest
= cpu_mask_to_apicid(mask
);
2485 read_msi_msg(irq
, &msg
);
2487 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
2488 msg
.data
|= MSI_DATA_VECTOR(vector
);
2489 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
2490 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
2492 write_msi_msg(irq
, &msg
);
2493 irq_desc
[irq
].affinity
= mask
;
2495 #endif /* CONFIG_SMP */
2498 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2499 * which implement the MSI or MSI-X Capability Structure.
2501 static struct irq_chip msi_chip
= {
2503 .unmask
= unmask_msi_irq
,
2504 .mask
= mask_msi_irq
,
2505 .ack
= ack_ioapic_irq
,
2507 .set_affinity
= set_msi_irq_affinity
,
2509 .retrigger
= ioapic_retrigger_irq
,
2512 int arch_setup_msi_irq(struct pci_dev
*dev
, struct msi_desc
*desc
)
2520 ret
= msi_compose_msg(dev
, irq
, &msg
);
2526 set_irq_msi(irq
, desc
);
2527 write_msi_msg(irq
, &msg
);
2529 set_irq_chip_and_handler_name(irq
, &msi_chip
, handle_edge_irq
,
2535 void arch_teardown_msi_irq(unsigned int irq
)
2540 #endif /* CONFIG_PCI_MSI */
2543 * Hypertransport interrupt support
2545 #ifdef CONFIG_HT_IRQ
2549 static void target_ht_irq(unsigned int irq
, unsigned int dest
)
2551 struct ht_irq_msg msg
;
2552 fetch_ht_irq_msg(irq
, &msg
);
2554 msg
.address_lo
&= ~(HT_IRQ_LOW_DEST_ID_MASK
);
2555 msg
.address_hi
&= ~(HT_IRQ_HIGH_DEST_ID_MASK
);
2557 msg
.address_lo
|= HT_IRQ_LOW_DEST_ID(dest
);
2558 msg
.address_hi
|= HT_IRQ_HIGH_DEST_ID(dest
);
2560 write_ht_irq_msg(irq
, &msg
);
2563 static void set_ht_irq_affinity(unsigned int irq
, cpumask_t mask
)
2568 cpus_and(tmp
, mask
, cpu_online_map
);
2569 if (cpus_empty(tmp
))
2572 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
2574 dest
= cpu_mask_to_apicid(mask
);
2576 target_ht_irq(irq
, dest
);
2577 irq_desc
[irq
].affinity
= mask
;
2581 static struct irq_chip ht_irq_chip
= {
2583 .mask
= mask_ht_irq
,
2584 .unmask
= unmask_ht_irq
,
2585 .ack
= ack_ioapic_irq
,
2587 .set_affinity
= set_ht_irq_affinity
,
2589 .retrigger
= ioapic_retrigger_irq
,
2592 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
2596 vector
= assign_irq_vector(irq
);
2598 struct ht_irq_msg msg
;
2603 cpu_set(vector
>> 8, tmp
);
2604 dest
= cpu_mask_to_apicid(tmp
);
2606 msg
.address_hi
= HT_IRQ_HIGH_DEST_ID(dest
);
2610 HT_IRQ_LOW_DEST_ID(dest
) |
2611 HT_IRQ_LOW_VECTOR(vector
) |
2612 ((INT_DEST_MODE
== 0) ?
2613 HT_IRQ_LOW_DM_PHYSICAL
:
2614 HT_IRQ_LOW_DM_LOGICAL
) |
2615 HT_IRQ_LOW_RQEOI_EDGE
|
2616 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2617 HT_IRQ_LOW_MT_FIXED
:
2618 HT_IRQ_LOW_MT_ARBITRATED
) |
2619 HT_IRQ_LOW_IRQ_MASKED
;
2621 write_ht_irq_msg(irq
, &msg
);
2623 set_irq_chip_and_handler_name(irq
, &ht_irq_chip
,
2624 handle_edge_irq
, "edge");
2628 #endif /* CONFIG_HT_IRQ */
2630 /* --------------------------------------------------------------------------
2631 ACPI-based IOAPIC Configuration
2632 -------------------------------------------------------------------------- */
2636 int __init
io_apic_get_unique_id (int ioapic
, int apic_id
)
2638 union IO_APIC_reg_00 reg_00
;
2639 static physid_mask_t apic_id_map
= PHYSID_MASK_NONE
;
2641 unsigned long flags
;
2645 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2646 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2647 * supports up to 16 on one shared APIC bus.
2649 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2650 * advantage of new APIC bus architecture.
2653 if (physids_empty(apic_id_map
))
2654 apic_id_map
= ioapic_phys_id_map(phys_cpu_present_map
);
2656 spin_lock_irqsave(&ioapic_lock
, flags
);
2657 reg_00
.raw
= io_apic_read(ioapic
, 0);
2658 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2660 if (apic_id
>= get_physical_broadcast()) {
2661 printk(KERN_WARNING
"IOAPIC[%d]: Invalid apic_id %d, trying "
2662 "%d\n", ioapic
, apic_id
, reg_00
.bits
.ID
);
2663 apic_id
= reg_00
.bits
.ID
;
2667 * Every APIC in a system must have a unique ID or we get lots of nice
2668 * 'stuck on smp_invalidate_needed IPI wait' messages.
2670 if (check_apicid_used(apic_id_map
, apic_id
)) {
2672 for (i
= 0; i
< get_physical_broadcast(); i
++) {
2673 if (!check_apicid_used(apic_id_map
, i
))
2677 if (i
== get_physical_broadcast())
2678 panic("Max apic_id exceeded!\n");
2680 printk(KERN_WARNING
"IOAPIC[%d]: apic_id %d already used, "
2681 "trying %d\n", ioapic
, apic_id
, i
);
2686 tmp
= apicid_to_cpu_present(apic_id
);
2687 physids_or(apic_id_map
, apic_id_map
, tmp
);
2689 if (reg_00
.bits
.ID
!= apic_id
) {
2690 reg_00
.bits
.ID
= apic_id
;
2692 spin_lock_irqsave(&ioapic_lock
, flags
);
2693 io_apic_write(ioapic
, 0, reg_00
.raw
);
2694 reg_00
.raw
= io_apic_read(ioapic
, 0);
2695 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2698 if (reg_00
.bits
.ID
!= apic_id
) {
2699 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic
);
2704 apic_printk(APIC_VERBOSE
, KERN_INFO
2705 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic
, apic_id
);
2711 int __init
io_apic_get_version (int ioapic
)
2713 union IO_APIC_reg_01 reg_01
;
2714 unsigned long flags
;
2716 spin_lock_irqsave(&ioapic_lock
, flags
);
2717 reg_01
.raw
= io_apic_read(ioapic
, 1);
2718 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2720 return reg_01
.bits
.version
;
2724 int __init
io_apic_get_redir_entries (int ioapic
)
2726 union IO_APIC_reg_01 reg_01
;
2727 unsigned long flags
;
2729 spin_lock_irqsave(&ioapic_lock
, flags
);
2730 reg_01
.raw
= io_apic_read(ioapic
, 1);
2731 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2733 return reg_01
.bits
.entries
;
2737 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int edge_level
, int active_high_low
)
2739 struct IO_APIC_route_entry entry
;
2741 if (!IO_APIC_IRQ(irq
)) {
2742 printk(KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
2748 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2749 * Note that we mask (disable) IRQs now -- these get enabled when the
2750 * corresponding device driver registers for this IRQ.
2753 memset(&entry
,0,sizeof(entry
));
2755 entry
.delivery_mode
= INT_DELIVERY_MODE
;
2756 entry
.dest_mode
= INT_DEST_MODE
;
2757 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
2758 entry
.trigger
= edge_level
;
2759 entry
.polarity
= active_high_low
;
2763 * IRQs < 16 are already in the irq_2_pin[] map
2766 add_pin_to_irq(irq
, ioapic
, pin
);
2768 entry
.vector
= assign_irq_vector(irq
);
2770 apic_printk(APIC_DEBUG
, KERN_DEBUG
"IOAPIC[%d]: Set PCI routing entry "
2771 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic
,
2772 mp_ioapics
[ioapic
].mpc_apicid
, pin
, entry
.vector
, irq
,
2773 edge_level
, active_high_low
);
2775 ioapic_register_intr(irq
, entry
.vector
, edge_level
);
2777 if (!ioapic
&& (irq
< 16))
2778 disable_8259A_irq(irq
);
2780 ioapic_write_entry(ioapic
, pin
, entry
);
2785 int acpi_get_override_irq(int bus_irq
, int *trigger
, int *polarity
)
2789 if (skip_ioapic_setup
)
2792 for (i
= 0; i
< mp_irq_entries
; i
++)
2793 if (mp_irqs
[i
].mpc_irqtype
== mp_INT
&&
2794 mp_irqs
[i
].mpc_srcbusirq
== bus_irq
)
2796 if (i
>= mp_irq_entries
)
2799 *trigger
= irq_trigger(i
);
2800 *polarity
= irq_polarity(i
);
2804 #endif /* CONFIG_ACPI */
2806 static int __init
parse_disable_timer_pin_1(char *arg
)
2808 disable_timer_pin_1
= 1;
2811 early_param("disable_timer_pin_1", parse_disable_timer_pin_1
);
2813 static int __init
parse_enable_timer_pin_1(char *arg
)
2815 disable_timer_pin_1
= -1;
2818 early_param("enable_timer_pin_1", parse_enable_timer_pin_1
);
2820 static int __init
parse_noapic(char *arg
)
2822 /* disable IO-APIC */
2823 disable_ioapic_setup();
2826 early_param("noapic", parse_noapic
);