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
67 unsigned move_cleanup_count;
69 u8 move_in_progress : 1;
72 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
73 static struct irq_cfg irq_cfg_legacy[] __initdata = {
74 [0] = { .irq = 0, .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
75 [1] = { .irq = 1, .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
76 [2] = { .irq = 2, .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
77 [3] = { .irq = 3, .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR, },
78 [4] = { .irq = 4, .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR, },
79 [5] = { .irq = 5, .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR, },
80 [6] = { .irq = 6, .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR, },
81 [7] = { .irq = 7, .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR, },
82 [8] = { .irq = 8, .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR, },
83 [9] = { .irq = 9, .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR, },
84 [10] = { .irq = 10, .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
85 [11] = { .irq = 11, .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
86 [12] = { .irq = 12, .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
87 [13] = { .irq = 13, .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
88 [14] = { .irq = 14, .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
89 [15] = { .irq = 15, .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
92 static struct irq_cfg irq_cfg_init = { .irq = -1U, };
93 /* need to be biger than size of irq_cfg_legacy */
94 static int nr_irq_cfg = 32;
96 static int __init parse_nr_irq_cfg(char *arg)
99 nr_irq_cfg = simple_strtoul(arg, NULL, 0);
106 early_param("nr_irq_cfg", parse_nr_irq_cfg);
108 static void init_one_irq_cfg(struct irq_cfg *cfg)
110 memcpy(cfg, &irq_cfg_init, sizeof(struct irq_cfg));
113 static void __init init_work(void *data)
115 struct dyn_array *da = data;
121 memcpy(cfg, irq_cfg_legacy, sizeof(irq_cfg_legacy));
123 i = sizeof(irq_cfg_legacy)/sizeof(irq_cfg_legacy[0]);
124 for (; i < *da->nr; i++)
125 init_one_irq_cfg(&cfg[i]);
127 for (i = 1; i < *da->nr; i++)
128 cfg[i-1].next = &cfg[i];
131 static struct irq_cfg *irq_cfgx;
132 DEFINE_DYN_ARRAY(irq_cfgx, sizeof(struct irq_cfg), nr_irq_cfg, PAGE_SIZE, init_work);
134 static struct irq_cfg *irq_cfg(unsigned int irq)
154 static struct irq_cfg *irq_cfg_alloc(unsigned int irq)
156 struct irq_cfg *cfg, *cfg_pri;
162 cfg_pri = cfg = &irq_cfgx[0];
167 if (cfg->irq == -1U) {
177 * we run out of pre-allocate ones, allocate more
179 printk(KERN_DEBUG "try to get more irq_cfg %d\n", nr_irq_cfg);
182 cfg = kzalloc(sizeof(struct irq_cfg)*nr_irq_cfg, GFP_ATOMIC);
184 cfg = __alloc_bootmem_nopanic(sizeof(struct irq_cfg)*nr_irq_cfg, PAGE_SIZE, 0);
187 panic("please boot with nr_irq_cfg= %d\n", count * 2);
189 for (i = 0; i < nr_irq_cfg; i++)
190 init_one_irq_cfg(&cfg[i]);
192 for (i = 1; i < nr_irq_cfg; i++)
193 cfg[i-1].next = &cfg[i];
201 static int assign_irq_vector(int irq, cpumask_t mask);
203 int first_system_vector = 0xfe;
205 char system_vectors[NR_VECTORS] = { [0 ... NR_VECTORS-1] = SYS_VECTOR_FREE};
207 int sis_apic_bug; /* not actually supported, dummy for compile */
209 static int no_timer_check;
211 static int disable_timer_pin_1 __initdata;
213 int timer_through_8259 __initdata;
215 /* Where if anywhere is the i8259 connect in external int mode */
216 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
218 static DEFINE_SPINLOCK(ioapic_lock);
219 static DEFINE_SPINLOCK(vector_lock);
222 * # of IRQ routing registers
224 int nr_ioapic_registers[MAX_IO_APICS];
226 /* I/O APIC RTE contents at the OS boot up */
227 struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
229 /* I/O APIC entries */
230 struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
233 /* MP IRQ source entries */
234 struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
236 /* # of MP IRQ source entries */
239 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
242 * Rough estimation of how many shared IRQs there are, can
243 * be changed anytime.
249 * This is performance-critical, we want to do it O(1)
251 * the indexing order of this array favors 1:1 mappings
252 * between pins and IRQs.
255 static struct irq_pin_list {
260 DEFINE_DYN_ARRAY(irq_2_pin, sizeof(struct irq_pin_list), pin_map_size, sizeof(struct irq_pin_list), NULL);
265 unsigned int unused[3];
269 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
271 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
272 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
275 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
277 struct io_apic __iomem *io_apic = io_apic_base(apic);
278 writel(reg, &io_apic->index);
279 return readl(&io_apic->data);
282 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
284 struct io_apic __iomem *io_apic = io_apic_base(apic);
285 writel(reg, &io_apic->index);
286 writel(value, &io_apic->data);
290 * Re-write a value: to be used for read-modify-write
291 * cycles where the read already set up the index register.
293 static inline void io_apic_modify(unsigned int apic, unsigned int value)
295 struct io_apic __iomem *io_apic = io_apic_base(apic);
296 writel(value, &io_apic->data);
299 static bool io_apic_level_ack_pending(unsigned int irq)
301 struct irq_pin_list *entry;
304 spin_lock_irqsave(&ioapic_lock, flags);
305 entry = irq_2_pin + irq;
313 reg = io_apic_read(entry->apic, 0x10 + pin*2);
314 /* Is the remote IRR bit set? */
315 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
316 spin_unlock_irqrestore(&ioapic_lock, flags);
321 entry = irq_2_pin + entry->next;
323 spin_unlock_irqrestore(&ioapic_lock, flags);
329 * Synchronize the IO-APIC and the CPU by doing
330 * a dummy read from the IO-APIC
332 static inline void io_apic_sync(unsigned int apic)
334 struct io_apic __iomem *io_apic = io_apic_base(apic);
335 readl(&io_apic->data);
338 #define __DO_ACTION(R, ACTION, FINAL) \
342 struct irq_pin_list *entry = irq_2_pin + irq; \
344 BUG_ON(irq >= nr_irqs); \
350 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
352 io_apic_modify(entry->apic, reg); \
356 entry = irq_2_pin + entry->next; \
361 struct { u32 w1, w2; };
362 struct IO_APIC_route_entry entry;
365 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
367 union entry_union eu;
369 spin_lock_irqsave(&ioapic_lock, flags);
370 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
371 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
372 spin_unlock_irqrestore(&ioapic_lock, flags);
377 * When we write a new IO APIC routing entry, we need to write the high
378 * word first! If the mask bit in the low word is clear, we will enable
379 * the interrupt, and we need to make sure the entry is fully populated
380 * before that happens.
383 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
385 union entry_union eu;
387 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
388 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
391 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
394 spin_lock_irqsave(&ioapic_lock, flags);
395 __ioapic_write_entry(apic, pin, e);
396 spin_unlock_irqrestore(&ioapic_lock, flags);
400 * When we mask an IO APIC routing entry, we need to write the low
401 * word first, in order to set the mask bit before we change the
404 static void ioapic_mask_entry(int apic, int pin)
407 union entry_union eu = { .entry.mask = 1 };
409 spin_lock_irqsave(&ioapic_lock, flags);
410 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
411 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
412 spin_unlock_irqrestore(&ioapic_lock, flags);
416 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
419 struct irq_pin_list *entry = irq_2_pin + irq;
421 BUG_ON(irq >= nr_irqs);
429 * With interrupt-remapping, destination information comes
430 * from interrupt-remapping table entry.
432 if (!irq_remapped(irq))
433 io_apic_write(apic, 0x11 + pin*2, dest);
434 reg = io_apic_read(apic, 0x10 + pin*2);
435 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
437 io_apic_modify(apic, reg);
440 entry = irq_2_pin + entry->next;
444 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
446 struct irq_cfg *cfg = irq_cfg(irq);
450 struct irq_desc *desc;
452 cpus_and(tmp, mask, cpu_online_map);
456 if (assign_irq_vector(irq, mask))
459 cpus_and(tmp, cfg->domain, mask);
460 dest = cpu_mask_to_apicid(tmp);
463 * Only the high 8 bits are valid.
465 dest = SET_APIC_LOGICAL_ID(dest);
467 desc = irq_to_desc(irq);
468 spin_lock_irqsave(&ioapic_lock, flags);
469 __target_IO_APIC_irq(irq, dest, cfg->vector);
470 desc->affinity = mask;
471 spin_unlock_irqrestore(&ioapic_lock, flags);
476 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
477 * shared ISA-space IRQs, so we have to support them. We are super
478 * fast in the common case, and fast for shared ISA-space IRQs.
480 int first_free_entry;
481 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
483 struct irq_pin_list *entry = irq_2_pin + irq;
485 BUG_ON(irq >= nr_irqs);
489 entry = irq_2_pin + entry->next;
491 if (entry->pin != -1) {
492 entry->next = first_free_entry;
493 entry = irq_2_pin + entry->next;
494 if (++first_free_entry >= pin_map_size)
495 panic("io_apic.c: ran out of irq_2_pin entries!");
502 * Reroute an IRQ to a different pin.
504 static void __init replace_pin_at_irq(unsigned int irq,
505 int oldapic, int oldpin,
506 int newapic, int newpin)
508 struct irq_pin_list *entry = irq_2_pin + irq;
511 if (entry->apic == oldapic && entry->pin == oldpin) {
512 entry->apic = newapic;
517 entry = irq_2_pin + entry->next;
522 #define DO_ACTION(name,R,ACTION, FINAL) \
524 static void name##_IO_APIC_irq (unsigned int irq) \
525 __DO_ACTION(R, ACTION, FINAL)
528 DO_ACTION(__mask, 0, |= IO_APIC_REDIR_MASKED, io_apic_sync(entry->apic))
531 DO_ACTION(__unmask, 0, &= ~IO_APIC_REDIR_MASKED, )
533 static void mask_IO_APIC_irq (unsigned int irq)
537 spin_lock_irqsave(&ioapic_lock, flags);
538 __mask_IO_APIC_irq(irq);
539 spin_unlock_irqrestore(&ioapic_lock, flags);
542 static void unmask_IO_APIC_irq (unsigned int irq)
546 spin_lock_irqsave(&ioapic_lock, flags);
547 __unmask_IO_APIC_irq(irq);
548 spin_unlock_irqrestore(&ioapic_lock, flags);
551 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
553 struct IO_APIC_route_entry entry;
555 /* Check delivery_mode to be sure we're not clearing an SMI pin */
556 entry = ioapic_read_entry(apic, pin);
557 if (entry.delivery_mode == dest_SMI)
560 * Disable it in the IO-APIC irq-routing table:
562 ioapic_mask_entry(apic, pin);
565 static void clear_IO_APIC (void)
569 for (apic = 0; apic < nr_ioapics; apic++)
570 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
571 clear_IO_APIC_pin(apic, pin);
575 * Saves and masks all the unmasked IO-APIC RTE's
577 int save_mask_IO_APIC_setup(void)
579 union IO_APIC_reg_01 reg_01;
584 * The number of IO-APIC IRQ registers (== #pins):
586 for (apic = 0; apic < nr_ioapics; apic++) {
587 spin_lock_irqsave(&ioapic_lock, flags);
588 reg_01.raw = io_apic_read(apic, 1);
589 spin_unlock_irqrestore(&ioapic_lock, flags);
590 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
593 for (apic = 0; apic < nr_ioapics; apic++) {
594 early_ioapic_entries[apic] =
595 kzalloc(sizeof(struct IO_APIC_route_entry) *
596 nr_ioapic_registers[apic], GFP_KERNEL);
597 if (!early_ioapic_entries[apic])
601 for (apic = 0; apic < nr_ioapics; apic++)
602 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
603 struct IO_APIC_route_entry entry;
605 entry = early_ioapic_entries[apic][pin] =
606 ioapic_read_entry(apic, pin);
609 ioapic_write_entry(apic, pin, entry);
615 void restore_IO_APIC_setup(void)
619 for (apic = 0; apic < nr_ioapics; apic++)
620 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
621 ioapic_write_entry(apic, pin,
622 early_ioapic_entries[apic][pin]);
625 void reinit_intr_remapped_IO_APIC(int intr_remapping)
628 * for now plain restore of previous settings.
629 * TBD: In the case of OS enabling interrupt-remapping,
630 * IO-APIC RTE's need to be setup to point to interrupt-remapping
631 * table entries. for now, do a plain restore, and wait for
632 * the setup_IO_APIC_irqs() to do proper initialization.
634 restore_IO_APIC_setup();
637 int skip_ioapic_setup;
640 static int __init parse_noapic(char *str)
642 disable_ioapic_setup();
645 early_param("noapic", parse_noapic);
647 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
648 static int __init disable_timer_pin_setup(char *arg)
650 disable_timer_pin_1 = 1;
653 __setup("disable_timer_pin_1", disable_timer_pin_setup);
657 * Find the IRQ entry number of a certain pin.
659 static int find_irq_entry(int apic, int pin, int type)
663 for (i = 0; i < mp_irq_entries; i++)
664 if (mp_irqs[i].mp_irqtype == type &&
665 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
666 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
667 mp_irqs[i].mp_dstirq == pin)
674 * Find the pin to which IRQ[irq] (ISA) is connected
676 static int __init find_isa_irq_pin(int irq, int type)
680 for (i = 0; i < mp_irq_entries; i++) {
681 int lbus = mp_irqs[i].mp_srcbus;
683 if (test_bit(lbus, mp_bus_not_pci) &&
684 (mp_irqs[i].mp_irqtype == type) &&
685 (mp_irqs[i].mp_srcbusirq == irq))
687 return mp_irqs[i].mp_dstirq;
692 static int __init find_isa_irq_apic(int irq, int type)
696 for (i = 0; i < mp_irq_entries; i++) {
697 int lbus = mp_irqs[i].mp_srcbus;
699 if (test_bit(lbus, mp_bus_not_pci) &&
700 (mp_irqs[i].mp_irqtype == type) &&
701 (mp_irqs[i].mp_srcbusirq == irq))
704 if (i < mp_irq_entries) {
706 for(apic = 0; apic < nr_ioapics; apic++) {
707 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
716 * Find a specific PCI IRQ entry.
717 * Not an __init, possibly needed by modules
719 static int pin_2_irq(int idx, int apic, int pin);
721 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
723 int apic, i, best_guess = -1;
725 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
727 if (test_bit(bus, mp_bus_not_pci)) {
728 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
731 for (i = 0; i < mp_irq_entries; i++) {
732 int lbus = mp_irqs[i].mp_srcbus;
734 for (apic = 0; apic < nr_ioapics; apic++)
735 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
736 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
739 if (!test_bit(lbus, mp_bus_not_pci) &&
740 !mp_irqs[i].mp_irqtype &&
742 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
743 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
745 if (!(apic || IO_APIC_IRQ(irq)))
748 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
751 * Use the first all-but-pin matching entry as a
752 * best-guess fuzzy result for broken mptables.
758 BUG_ON(best_guess >= nr_irqs);
762 /* ISA interrupts are always polarity zero edge triggered,
763 * when listed as conforming in the MP table. */
765 #define default_ISA_trigger(idx) (0)
766 #define default_ISA_polarity(idx) (0)
768 /* PCI interrupts are always polarity one level triggered,
769 * when listed as conforming in the MP table. */
771 #define default_PCI_trigger(idx) (1)
772 #define default_PCI_polarity(idx) (1)
774 static int MPBIOS_polarity(int idx)
776 int bus = mp_irqs[idx].mp_srcbus;
780 * Determine IRQ line polarity (high active or low active):
782 switch (mp_irqs[idx].mp_irqflag & 3)
784 case 0: /* conforms, ie. bus-type dependent polarity */
785 if (test_bit(bus, mp_bus_not_pci))
786 polarity = default_ISA_polarity(idx);
788 polarity = default_PCI_polarity(idx);
790 case 1: /* high active */
795 case 2: /* reserved */
797 printk(KERN_WARNING "broken BIOS!!\n");
801 case 3: /* low active */
806 default: /* invalid */
808 printk(KERN_WARNING "broken BIOS!!\n");
816 static int MPBIOS_trigger(int idx)
818 int bus = mp_irqs[idx].mp_srcbus;
822 * Determine IRQ trigger mode (edge or level sensitive):
824 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
826 case 0: /* conforms, ie. bus-type dependent */
827 if (test_bit(bus, mp_bus_not_pci))
828 trigger = default_ISA_trigger(idx);
830 trigger = default_PCI_trigger(idx);
837 case 2: /* reserved */
839 printk(KERN_WARNING "broken BIOS!!\n");
848 default: /* invalid */
850 printk(KERN_WARNING "broken BIOS!!\n");
858 static inline int irq_polarity(int idx)
860 return MPBIOS_polarity(idx);
863 static inline int irq_trigger(int idx)
865 return MPBIOS_trigger(idx);
868 static int pin_2_irq(int idx, int apic, int pin)
871 int bus = mp_irqs[idx].mp_srcbus;
874 * Debugging check, we are in big trouble if this message pops up!
876 if (mp_irqs[idx].mp_dstirq != pin)
877 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
879 if (test_bit(bus, mp_bus_not_pci)) {
880 irq = mp_irqs[idx].mp_srcbusirq;
883 * PCI IRQs are mapped in order
887 irq += nr_ioapic_registers[i++];
890 BUG_ON(irq >= nr_irqs);
894 void lock_vector_lock(void)
896 /* Used to the online set of cpus does not change
897 * during assign_irq_vector.
899 spin_lock(&vector_lock);
902 void unlock_vector_lock(void)
904 spin_unlock(&vector_lock);
907 static int __assign_irq_vector(int irq, cpumask_t mask)
910 * NOTE! The local APIC isn't very good at handling
911 * multiple interrupts at the same interrupt level.
912 * As the interrupt level is determined by taking the
913 * vector number and shifting that right by 4, we
914 * want to spread these out a bit so that they don't
915 * all fall in the same interrupt level.
917 * Also, we've got to be careful not to trash gate
918 * 0x80, because int 0x80 is hm, kind of importantish. ;)
920 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
921 unsigned int old_vector;
925 BUG_ON((unsigned)irq >= nr_irqs);
928 /* Only try and allocate irqs on cpus that are present */
929 cpus_and(mask, mask, cpu_online_map);
931 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
934 old_vector = cfg->vector;
937 cpus_and(tmp, cfg->domain, mask);
938 if (!cpus_empty(tmp))
942 for_each_cpu_mask_nr(cpu, mask) {
943 cpumask_t domain, new_mask;
947 domain = vector_allocation_domain(cpu);
948 cpus_and(new_mask, domain, cpu_online_map);
950 vector = current_vector;
951 offset = current_offset;
954 if (vector >= first_system_vector) {
955 /* If we run out of vectors on large boxen, must share them. */
956 offset = (offset + 1) % 8;
957 vector = FIRST_DEVICE_VECTOR + offset;
959 if (unlikely(current_vector == vector))
961 if (vector == IA32_SYSCALL_VECTOR)
963 for_each_cpu_mask_nr(new_cpu, new_mask)
964 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
967 current_vector = vector;
968 current_offset = offset;
970 cfg->move_in_progress = 1;
971 cfg->old_domain = cfg->domain;
973 for_each_cpu_mask_nr(new_cpu, new_mask)
974 per_cpu(vector_irq, new_cpu)[vector] = irq;
975 cfg->vector = vector;
976 cfg->domain = domain;
982 static int assign_irq_vector(int irq, cpumask_t mask)
987 spin_lock_irqsave(&vector_lock, flags);
988 err = __assign_irq_vector(irq, mask);
989 spin_unlock_irqrestore(&vector_lock, flags);
993 static void __clear_irq_vector(int irq)
999 BUG_ON((unsigned)irq >= nr_irqs);
1001 BUG_ON(!cfg->vector);
1003 vector = cfg->vector;
1004 cpus_and(mask, cfg->domain, cpu_online_map);
1005 for_each_cpu_mask_nr(cpu, mask)
1006 per_cpu(vector_irq, cpu)[vector] = -1;
1009 cpus_clear(cfg->domain);
1012 void __setup_vector_irq(int cpu)
1014 /* Initialize vector_irq on a new cpu */
1015 /* This function must be called with vector_lock held */
1018 /* Mark the inuse vectors */
1019 for (irq = 0; irq < nr_irqs; ++irq) {
1020 struct irq_cfg *cfg = irq_cfg(irq);
1022 if (!cpu_isset(cpu, cfg->domain))
1024 vector = cfg->vector;
1025 per_cpu(vector_irq, cpu)[vector] = irq;
1027 /* Mark the free vectors */
1028 for (vector = 0; vector < NR_VECTORS; ++vector) {
1029 struct irq_cfg *cfg;
1031 irq = per_cpu(vector_irq, cpu)[vector];
1036 if (!cpu_isset(cpu, cfg->domain))
1037 per_cpu(vector_irq, cpu)[vector] = -1;
1041 static struct irq_chip ioapic_chip;
1042 #ifdef CONFIG_INTR_REMAP
1043 static struct irq_chip ir_ioapic_chip;
1046 static void ioapic_register_intr(int irq, unsigned long trigger)
1048 struct irq_desc *desc;
1050 desc = irq_to_desc(irq);
1052 desc->status |= IRQ_LEVEL;
1054 desc->status &= ~IRQ_LEVEL;
1056 #ifdef CONFIG_INTR_REMAP
1057 if (irq_remapped(irq)) {
1058 desc->status |= IRQ_MOVE_PCNTXT;
1060 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1064 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1065 handle_edge_irq, "edge");
1070 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1074 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1075 handle_edge_irq, "edge");
1078 static int setup_ioapic_entry(int apic, int irq,
1079 struct IO_APIC_route_entry *entry,
1080 unsigned int destination, int trigger,
1081 int polarity, int vector)
1084 * add it to the IO-APIC irq-routing table:
1086 memset(entry,0,sizeof(*entry));
1088 #ifdef CONFIG_INTR_REMAP
1089 if (intr_remapping_enabled) {
1090 struct intel_iommu *iommu = map_ioapic_to_ir(apic);
1092 struct IR_IO_APIC_route_entry *ir_entry =
1093 (struct IR_IO_APIC_route_entry *) entry;
1097 panic("No mapping iommu for ioapic %d\n", apic);
1099 index = alloc_irte(iommu, irq, 1);
1101 panic("Failed to allocate IRTE for ioapic %d\n", apic);
1103 memset(&irte, 0, sizeof(irte));
1106 irte.dst_mode = INT_DEST_MODE;
1107 irte.trigger_mode = trigger;
1108 irte.dlvry_mode = INT_DELIVERY_MODE;
1109 irte.vector = vector;
1110 irte.dest_id = IRTE_DEST(destination);
1112 modify_irte(irq, &irte);
1114 ir_entry->index2 = (index >> 15) & 0x1;
1116 ir_entry->format = 1;
1117 ir_entry->index = (index & 0x7fff);
1121 entry->delivery_mode = INT_DELIVERY_MODE;
1122 entry->dest_mode = INT_DEST_MODE;
1123 entry->dest = destination;
1126 entry->mask = 0; /* enable IRQ */
1127 entry->trigger = trigger;
1128 entry->polarity = polarity;
1129 entry->vector = vector;
1131 /* Mask level triggered irqs.
1132 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1139 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
1140 int trigger, int polarity)
1142 struct irq_cfg *cfg;
1143 struct IO_APIC_route_entry entry;
1146 if (!IO_APIC_IRQ(irq))
1152 if (assign_irq_vector(irq, mask))
1155 cpus_and(mask, cfg->domain, mask);
1157 apic_printk(APIC_VERBOSE,KERN_DEBUG
1158 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1159 "IRQ %d Mode:%i Active:%i)\n",
1160 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
1161 irq, trigger, polarity);
1164 if (setup_ioapic_entry(mp_ioapics[apic].mp_apicid, irq, &entry,
1165 cpu_mask_to_apicid(mask), trigger, polarity,
1167 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1168 mp_ioapics[apic].mp_apicid, pin);
1169 __clear_irq_vector(irq);
1173 ioapic_register_intr(irq, trigger);
1175 disable_8259A_irq(irq);
1177 ioapic_write_entry(apic, pin, entry);
1180 static void __init setup_IO_APIC_irqs(void)
1182 int apic, pin, idx, irq, first_notcon = 1;
1184 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1186 for (apic = 0; apic < nr_ioapics; apic++) {
1187 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1189 idx = find_irq_entry(apic,pin,mp_INT);
1192 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mp_apicid, pin);
1195 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mp_apicid, pin);
1198 if (!first_notcon) {
1199 apic_printk(APIC_VERBOSE, " not connected.\n");
1203 irq = pin_2_irq(idx, apic, pin);
1204 add_pin_to_irq(irq, apic, pin);
1206 setup_IO_APIC_irq(apic, pin, irq,
1207 irq_trigger(idx), irq_polarity(idx));
1212 apic_printk(APIC_VERBOSE, " not connected.\n");
1216 * Set up the timer pin, possibly with the 8259A-master behind.
1218 static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1221 struct IO_APIC_route_entry entry;
1223 if (intr_remapping_enabled)
1226 memset(&entry, 0, sizeof(entry));
1229 * We use logical delivery to get the timer IRQ
1232 entry.dest_mode = INT_DEST_MODE;
1233 entry.mask = 1; /* mask IRQ now */
1234 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
1235 entry.delivery_mode = INT_DELIVERY_MODE;
1238 entry.vector = vector;
1241 * The timer IRQ doesn't have to know that behind the
1242 * scene we may have a 8259A-master in AEOI mode ...
1244 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1247 * Add it to the IO-APIC irq-routing table:
1249 ioapic_write_entry(apic, pin, entry);
1253 __apicdebuginit(void) print_IO_APIC(void)
1256 union IO_APIC_reg_00 reg_00;
1257 union IO_APIC_reg_01 reg_01;
1258 union IO_APIC_reg_02 reg_02;
1259 unsigned long flags;
1261 if (apic_verbosity == APIC_QUIET)
1264 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1265 for (i = 0; i < nr_ioapics; i++)
1266 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1267 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1270 * We are a bit conservative about what we expect. We have to
1271 * know about every hardware change ASAP.
1273 printk(KERN_INFO "testing the IO APIC.......................\n");
1275 for (apic = 0; apic < nr_ioapics; apic++) {
1277 spin_lock_irqsave(&ioapic_lock, flags);
1278 reg_00.raw = io_apic_read(apic, 0);
1279 reg_01.raw = io_apic_read(apic, 1);
1280 if (reg_01.bits.version >= 0x10)
1281 reg_02.raw = io_apic_read(apic, 2);
1282 spin_unlock_irqrestore(&ioapic_lock, flags);
1285 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1286 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1287 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1289 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1290 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1292 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1293 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1295 if (reg_01.bits.version >= 0x10) {
1296 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1297 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1300 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1302 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1303 " Stat Dmod Deli Vect: \n");
1305 for (i = 0; i <= reg_01.bits.entries; i++) {
1306 struct IO_APIC_route_entry entry;
1308 entry = ioapic_read_entry(apic, i);
1310 printk(KERN_DEBUG " %02x %03X ",
1315 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1320 entry.delivery_status,
1322 entry.delivery_mode,
1327 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1328 for (i = 0; i < nr_irqs; i++) {
1329 struct irq_pin_list *entry = irq_2_pin + i;
1332 printk(KERN_DEBUG "IRQ%d ", i);
1334 printk("-> %d:%d", entry->apic, entry->pin);
1337 entry = irq_2_pin + entry->next;
1342 printk(KERN_INFO ".................................... done.\n");
1347 __apicdebuginit(void) print_APIC_bitfield(int base)
1352 if (apic_verbosity == APIC_QUIET)
1355 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1356 for (i = 0; i < 8; i++) {
1357 v = apic_read(base + i*0x10);
1358 for (j = 0; j < 32; j++) {
1368 __apicdebuginit(void) print_local_APIC(void *dummy)
1370 unsigned int v, ver, maxlvt;
1373 if (apic_verbosity == APIC_QUIET)
1376 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1377 smp_processor_id(), hard_smp_processor_id());
1378 v = apic_read(APIC_ID);
1379 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1380 v = apic_read(APIC_LVR);
1381 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1382 ver = GET_APIC_VERSION(v);
1383 maxlvt = lapic_get_maxlvt();
1385 v = apic_read(APIC_TASKPRI);
1386 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1388 v = apic_read(APIC_ARBPRI);
1389 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1390 v & APIC_ARBPRI_MASK);
1391 v = apic_read(APIC_PROCPRI);
1392 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1394 v = apic_read(APIC_EOI);
1395 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1396 v = apic_read(APIC_RRR);
1397 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1398 v = apic_read(APIC_LDR);
1399 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1400 v = apic_read(APIC_DFR);
1401 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1402 v = apic_read(APIC_SPIV);
1403 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1405 printk(KERN_DEBUG "... APIC ISR field:\n");
1406 print_APIC_bitfield(APIC_ISR);
1407 printk(KERN_DEBUG "... APIC TMR field:\n");
1408 print_APIC_bitfield(APIC_TMR);
1409 printk(KERN_DEBUG "... APIC IRR field:\n");
1410 print_APIC_bitfield(APIC_IRR);
1412 v = apic_read(APIC_ESR);
1413 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1415 icr = apic_icr_read();
1416 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1417 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1419 v = apic_read(APIC_LVTT);
1420 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1422 if (maxlvt > 3) { /* PC is LVT#4. */
1423 v = apic_read(APIC_LVTPC);
1424 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1426 v = apic_read(APIC_LVT0);
1427 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1428 v = apic_read(APIC_LVT1);
1429 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1431 if (maxlvt > 2) { /* ERR is LVT#3. */
1432 v = apic_read(APIC_LVTERR);
1433 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1436 v = apic_read(APIC_TMICT);
1437 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1438 v = apic_read(APIC_TMCCT);
1439 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1440 v = apic_read(APIC_TDCR);
1441 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1445 __apicdebuginit(void) print_all_local_APICs(void)
1447 on_each_cpu(print_local_APIC, NULL, 1);
1450 __apicdebuginit(void) print_PIC(void)
1453 unsigned long flags;
1455 if (apic_verbosity == APIC_QUIET)
1458 printk(KERN_DEBUG "\nprinting PIC contents\n");
1460 spin_lock_irqsave(&i8259A_lock, flags);
1462 v = inb(0xa1) << 8 | inb(0x21);
1463 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1465 v = inb(0xa0) << 8 | inb(0x20);
1466 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1470 v = inb(0xa0) << 8 | inb(0x20);
1474 spin_unlock_irqrestore(&i8259A_lock, flags);
1476 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1478 v = inb(0x4d1) << 8 | inb(0x4d0);
1479 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1482 __apicdebuginit(int) print_all_ICs(void)
1485 print_all_local_APICs();
1491 fs_initcall(print_all_ICs);
1494 void __init enable_IO_APIC(void)
1496 union IO_APIC_reg_01 reg_01;
1497 int i8259_apic, i8259_pin;
1499 unsigned long flags;
1501 for (i = 0; i < pin_map_size; i++) {
1502 irq_2_pin[i].pin = -1;
1503 irq_2_pin[i].next = 0;
1507 * The number of IO-APIC IRQ registers (== #pins):
1509 for (apic = 0; apic < nr_ioapics; apic++) {
1510 spin_lock_irqsave(&ioapic_lock, flags);
1511 reg_01.raw = io_apic_read(apic, 1);
1512 spin_unlock_irqrestore(&ioapic_lock, flags);
1513 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1515 for(apic = 0; apic < nr_ioapics; apic++) {
1517 /* See if any of the pins is in ExtINT mode */
1518 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1519 struct IO_APIC_route_entry entry;
1520 entry = ioapic_read_entry(apic, pin);
1522 /* If the interrupt line is enabled and in ExtInt mode
1523 * I have found the pin where the i8259 is connected.
1525 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1526 ioapic_i8259.apic = apic;
1527 ioapic_i8259.pin = pin;
1533 /* Look to see what if the MP table has reported the ExtINT */
1534 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1535 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1536 /* Trust the MP table if nothing is setup in the hardware */
1537 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1538 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1539 ioapic_i8259.pin = i8259_pin;
1540 ioapic_i8259.apic = i8259_apic;
1542 /* Complain if the MP table and the hardware disagree */
1543 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1544 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1546 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1550 * Do not trust the IO-APIC being empty at bootup
1556 * Not an __init, needed by the reboot code
1558 void disable_IO_APIC(void)
1561 * Clear the IO-APIC before rebooting:
1566 * If the i8259 is routed through an IOAPIC
1567 * Put that IOAPIC in virtual wire mode
1568 * so legacy interrupts can be delivered.
1570 if (ioapic_i8259.pin != -1) {
1571 struct IO_APIC_route_entry entry;
1573 memset(&entry, 0, sizeof(entry));
1574 entry.mask = 0; /* Enabled */
1575 entry.trigger = 0; /* Edge */
1577 entry.polarity = 0; /* High */
1578 entry.delivery_status = 0;
1579 entry.dest_mode = 0; /* Physical */
1580 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1582 entry.dest = read_apic_id();
1585 * Add it to the IO-APIC irq-routing table:
1587 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1590 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1594 * There is a nasty bug in some older SMP boards, their mptable lies
1595 * about the timer IRQ. We do the following to work around the situation:
1597 * - timer IRQ defaults to IO-APIC IRQ
1598 * - if this function detects that timer IRQs are defunct, then we fall
1599 * back to ISA timer IRQs
1601 static int __init timer_irq_works(void)
1603 unsigned long t1 = jiffies;
1604 unsigned long flags;
1606 local_save_flags(flags);
1608 /* Let ten ticks pass... */
1609 mdelay((10 * 1000) / HZ);
1610 local_irq_restore(flags);
1613 * Expect a few ticks at least, to be sure some possible
1614 * glue logic does not lock up after one or two first
1615 * ticks in a non-ExtINT mode. Also the local APIC
1616 * might have cached one ExtINT interrupt. Finally, at
1617 * least one tick may be lost due to delays.
1621 if (time_after(jiffies, t1 + 4))
1627 * In the SMP+IOAPIC case it might happen that there are an unspecified
1628 * number of pending IRQ events unhandled. These cases are very rare,
1629 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1630 * better to do it this way as thus we do not have to be aware of
1631 * 'pending' interrupts in the IRQ path, except at this point.
1634 * Edge triggered needs to resend any interrupt
1635 * that was delayed but this is now handled in the device
1640 * Starting up a edge-triggered IO-APIC interrupt is
1641 * nasty - we need to make sure that we get the edge.
1642 * If it is already asserted for some reason, we need
1643 * return 1 to indicate that is was pending.
1645 * This is not complete - we should be able to fake
1646 * an edge even if it isn't on the 8259A...
1649 static unsigned int startup_ioapic_irq(unsigned int irq)
1651 int was_pending = 0;
1652 unsigned long flags;
1654 spin_lock_irqsave(&ioapic_lock, flags);
1656 disable_8259A_irq(irq);
1657 if (i8259A_irq_pending(irq))
1660 __unmask_IO_APIC_irq(irq);
1661 spin_unlock_irqrestore(&ioapic_lock, flags);
1666 static int ioapic_retrigger_irq(unsigned int irq)
1668 struct irq_cfg *cfg = irq_cfg(irq);
1669 unsigned long flags;
1671 spin_lock_irqsave(&vector_lock, flags);
1672 send_IPI_mask(cpumask_of_cpu(first_cpu(cfg->domain)), cfg->vector);
1673 spin_unlock_irqrestore(&vector_lock, flags);
1679 * Level and edge triggered IO-APIC interrupts need different handling,
1680 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1681 * handled with the level-triggered descriptor, but that one has slightly
1682 * more overhead. Level-triggered interrupts cannot be handled with the
1683 * edge-triggered handler, without risking IRQ storms and other ugly
1689 #ifdef CONFIG_INTR_REMAP
1690 static void ir_irq_migration(struct work_struct *work);
1692 static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
1695 * Migrate the IO-APIC irq in the presence of intr-remapping.
1697 * For edge triggered, irq migration is a simple atomic update(of vector
1698 * and cpu destination) of IRTE and flush the hardware cache.
1700 * For level triggered, we need to modify the io-apic RTE aswell with the update
1701 * vector information, along with modifying IRTE with vector and destination.
1702 * So irq migration for level triggered is little bit more complex compared to
1703 * edge triggered migration. But the good news is, we use the same algorithm
1704 * for level triggered migration as we have today, only difference being,
1705 * we now initiate the irq migration from process context instead of the
1706 * interrupt context.
1708 * In future, when we do a directed EOI (combined with cpu EOI broadcast
1709 * suppression) to the IO-APIC, level triggered irq migration will also be
1710 * as simple as edge triggered migration and we can do the irq migration
1711 * with a simple atomic update to IO-APIC RTE.
1713 static void migrate_ioapic_irq(int irq, cpumask_t mask)
1715 struct irq_cfg *cfg;
1716 struct irq_desc *desc;
1717 cpumask_t tmp, cleanup_mask;
1719 int modify_ioapic_rte;
1721 unsigned long flags;
1723 cpus_and(tmp, mask, cpu_online_map);
1724 if (cpus_empty(tmp))
1727 if (get_irte(irq, &irte))
1730 if (assign_irq_vector(irq, mask))
1734 cpus_and(tmp, cfg->domain, mask);
1735 dest = cpu_mask_to_apicid(tmp);
1737 desc = irq_to_desc(irq);
1738 modify_ioapic_rte = desc->status & IRQ_LEVEL;
1739 if (modify_ioapic_rte) {
1740 spin_lock_irqsave(&ioapic_lock, flags);
1741 __target_IO_APIC_irq(irq, dest, cfg->vector);
1742 spin_unlock_irqrestore(&ioapic_lock, flags);
1745 irte.vector = cfg->vector;
1746 irte.dest_id = IRTE_DEST(dest);
1749 * Modified the IRTE and flushes the Interrupt entry cache.
1751 modify_irte(irq, &irte);
1753 if (cfg->move_in_progress) {
1754 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1755 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1756 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1757 cfg->move_in_progress = 0;
1760 desc->affinity = mask;
1763 static int migrate_irq_remapped_level(int irq)
1766 struct irq_desc *desc = irq_to_desc(irq);
1768 mask_IO_APIC_irq(irq);
1770 if (io_apic_level_ack_pending(irq)) {
1772 * Interrupt in progress. Migrating irq now will change the
1773 * vector information in the IO-APIC RTE and that will confuse
1774 * the EOI broadcast performed by cpu.
1775 * So, delay the irq migration to the next instance.
1777 schedule_delayed_work(&ir_migration_work, 1);
1781 /* everthing is clear. we have right of way */
1782 migrate_ioapic_irq(irq, desc->pending_mask);
1785 desc->status &= ~IRQ_MOVE_PENDING;
1786 cpus_clear(desc->pending_mask);
1789 unmask_IO_APIC_irq(irq);
1793 static void ir_irq_migration(struct work_struct *work)
1797 for (irq = 0; irq < nr_irqs; irq++) {
1798 struct irq_desc *desc = irq_to_desc(irq);
1799 if (desc->status & IRQ_MOVE_PENDING) {
1800 unsigned long flags;
1802 spin_lock_irqsave(&desc->lock, flags);
1803 if (!desc->chip->set_affinity ||
1804 !(desc->status & IRQ_MOVE_PENDING)) {
1805 desc->status &= ~IRQ_MOVE_PENDING;
1806 spin_unlock_irqrestore(&desc->lock, flags);
1810 desc->chip->set_affinity(irq, desc->pending_mask);
1811 spin_unlock_irqrestore(&desc->lock, flags);
1817 * Migrates the IRQ destination in the process context.
1819 static void set_ir_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
1821 struct irq_desc *desc = irq_to_desc(irq);
1823 if (desc->status & IRQ_LEVEL) {
1824 desc->status |= IRQ_MOVE_PENDING;
1825 desc->pending_mask = mask;
1826 migrate_irq_remapped_level(irq);
1830 migrate_ioapic_irq(irq, mask);
1834 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1836 unsigned vector, me;
1841 me = smp_processor_id();
1842 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1844 struct irq_desc *desc;
1845 struct irq_cfg *cfg;
1846 irq = __get_cpu_var(vector_irq)[vector];
1850 desc = irq_to_desc(irq);
1852 spin_lock(&desc->lock);
1853 if (!cfg->move_cleanup_count)
1856 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1859 __get_cpu_var(vector_irq)[vector] = -1;
1860 cfg->move_cleanup_count--;
1862 spin_unlock(&desc->lock);
1868 static void irq_complete_move(unsigned int irq)
1870 struct irq_cfg *cfg = irq_cfg(irq);
1871 unsigned vector, me;
1873 if (likely(!cfg->move_in_progress))
1876 vector = ~get_irq_regs()->orig_ax;
1877 me = smp_processor_id();
1878 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1879 cpumask_t cleanup_mask;
1881 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1882 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1883 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1884 cfg->move_in_progress = 0;
1888 static inline void irq_complete_move(unsigned int irq) {}
1890 #ifdef CONFIG_INTR_REMAP
1891 static void ack_x2apic_level(unsigned int irq)
1896 static void ack_x2apic_edge(unsigned int irq)
1902 static void ack_apic_edge(unsigned int irq)
1904 irq_complete_move(irq);
1905 move_native_irq(irq);
1909 static void ack_apic_level(unsigned int irq)
1911 int do_unmask_irq = 0;
1913 irq_complete_move(irq);
1914 #ifdef CONFIG_GENERIC_PENDING_IRQ
1915 /* If we are moving the irq we need to mask it */
1916 if (unlikely(irq_to_desc(irq)->status & IRQ_MOVE_PENDING)) {
1918 mask_IO_APIC_irq(irq);
1923 * We must acknowledge the irq before we move it or the acknowledge will
1924 * not propagate properly.
1928 /* Now we can move and renable the irq */
1929 if (unlikely(do_unmask_irq)) {
1930 /* Only migrate the irq if the ack has been received.
1932 * On rare occasions the broadcast level triggered ack gets
1933 * delayed going to ioapics, and if we reprogram the
1934 * vector while Remote IRR is still set the irq will never
1937 * To prevent this scenario we read the Remote IRR bit
1938 * of the ioapic. This has two effects.
1939 * - On any sane system the read of the ioapic will
1940 * flush writes (and acks) going to the ioapic from
1942 * - We get to see if the ACK has actually been delivered.
1944 * Based on failed experiments of reprogramming the
1945 * ioapic entry from outside of irq context starting
1946 * with masking the ioapic entry and then polling until
1947 * Remote IRR was clear before reprogramming the
1948 * ioapic I don't trust the Remote IRR bit to be
1949 * completey accurate.
1951 * However there appears to be no other way to plug
1952 * this race, so if the Remote IRR bit is not
1953 * accurate and is causing problems then it is a hardware bug
1954 * and you can go talk to the chipset vendor about it.
1956 if (!io_apic_level_ack_pending(irq))
1957 move_masked_irq(irq);
1958 unmask_IO_APIC_irq(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_apic_edge,
1968 .eoi = ack_apic_level,
1970 .set_affinity = set_ioapic_affinity_irq,
1972 .retrigger = ioapic_retrigger_irq,
1975 #ifdef CONFIG_INTR_REMAP
1976 static struct irq_chip ir_ioapic_chip __read_mostly = {
1977 .name = "IR-IO-APIC",
1978 .startup = startup_ioapic_irq,
1979 .mask = mask_IO_APIC_irq,
1980 .unmask = unmask_IO_APIC_irq,
1981 .ack = ack_x2apic_edge,
1982 .eoi = ack_x2apic_level,
1984 .set_affinity = set_ir_ioapic_affinity_irq,
1986 .retrigger = ioapic_retrigger_irq,
1990 static inline void init_IO_APIC_traps(void)
1993 struct irq_desc *desc;
1996 * NOTE! The local APIC isn't very good at handling
1997 * multiple interrupts at the same interrupt level.
1998 * As the interrupt level is determined by taking the
1999 * vector number and shifting that right by 4, we
2000 * want to spread these out a bit so that they don't
2001 * all fall in the same interrupt level.
2003 * Also, we've got to be careful not to trash gate
2004 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2006 for (irq = 0; irq < nr_irqs ; irq++) {
2007 struct irq_cfg *cfg;
2010 if (IO_APIC_IRQ(irq) && !cfg->vector) {
2012 * Hmm.. We don't have an entry for this,
2013 * so default to an old-fashioned 8259
2014 * interrupt if we can..
2017 make_8259A_irq(irq);
2019 desc = irq_to_desc(irq);
2020 /* Strange. Oh, well.. */
2021 desc->chip = &no_irq_chip;
2027 static void unmask_lapic_irq(unsigned int irq)
2031 v = apic_read(APIC_LVT0);
2032 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2035 static void mask_lapic_irq(unsigned int irq)
2039 v = apic_read(APIC_LVT0);
2040 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2043 static void ack_lapic_irq (unsigned int irq)
2048 static struct irq_chip lapic_chip __read_mostly = {
2049 .name = "local-APIC",
2050 .mask = mask_lapic_irq,
2051 .unmask = unmask_lapic_irq,
2052 .ack = ack_lapic_irq,
2055 static void lapic_register_intr(int irq)
2057 struct irq_desc *desc;
2059 desc = irq_to_desc(irq);
2060 desc->status &= ~IRQ_LEVEL;
2061 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2065 static void __init setup_nmi(void)
2068 * Dirty trick to enable the NMI watchdog ...
2069 * We put the 8259A master into AEOI mode and
2070 * unmask on all local APICs LVT0 as NMI.
2072 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2073 * is from Maciej W. Rozycki - so we do not have to EOI from
2074 * the NMI handler or the timer interrupt.
2076 printk(KERN_INFO "activating NMI Watchdog ...");
2078 enable_NMI_through_LVT0();
2084 * This looks a bit hackish but it's about the only one way of sending
2085 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2086 * not support the ExtINT mode, unfortunately. We need to send these
2087 * cycles as some i82489DX-based boards have glue logic that keeps the
2088 * 8259A interrupt line asserted until INTA. --macro
2090 static inline void __init unlock_ExtINT_logic(void)
2093 struct IO_APIC_route_entry entry0, entry1;
2094 unsigned char save_control, save_freq_select;
2096 pin = find_isa_irq_pin(8, mp_INT);
2097 apic = find_isa_irq_apic(8, mp_INT);
2101 entry0 = ioapic_read_entry(apic, pin);
2103 clear_IO_APIC_pin(apic, pin);
2105 memset(&entry1, 0, sizeof(entry1));
2107 entry1.dest_mode = 0; /* physical delivery */
2108 entry1.mask = 0; /* unmask IRQ now */
2109 entry1.dest = hard_smp_processor_id();
2110 entry1.delivery_mode = dest_ExtINT;
2111 entry1.polarity = entry0.polarity;
2115 ioapic_write_entry(apic, pin, entry1);
2117 save_control = CMOS_READ(RTC_CONTROL);
2118 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2119 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2121 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2126 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2130 CMOS_WRITE(save_control, RTC_CONTROL);
2131 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2132 clear_IO_APIC_pin(apic, pin);
2134 ioapic_write_entry(apic, pin, entry0);
2138 * This code may look a bit paranoid, but it's supposed to cooperate with
2139 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2140 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2141 * fanatically on his truly buggy board.
2143 * FIXME: really need to revamp this for modern platforms only.
2145 static inline void __init check_timer(void)
2147 struct irq_cfg *cfg = irq_cfg(0);
2148 int apic1, pin1, apic2, pin2;
2149 unsigned long flags;
2152 local_irq_save(flags);
2155 * get/set the timer IRQ vector:
2157 disable_8259A_irq(0);
2158 assign_irq_vector(0, TARGET_CPUS);
2161 * As IRQ0 is to be enabled in the 8259A, the virtual
2162 * wire has to be disabled in the local APIC.
2164 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2167 pin1 = find_isa_irq_pin(0, mp_INT);
2168 apic1 = find_isa_irq_apic(0, mp_INT);
2169 pin2 = ioapic_i8259.pin;
2170 apic2 = ioapic_i8259.apic;
2172 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2173 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2174 cfg->vector, apic1, pin1, apic2, pin2);
2177 * Some BIOS writers are clueless and report the ExtINTA
2178 * I/O APIC input from the cascaded 8259A as the timer
2179 * interrupt input. So just in case, if only one pin
2180 * was found above, try it both directly and through the
2184 if (intr_remapping_enabled)
2185 panic("BIOS bug: timer not connected to IO-APIC");
2189 } else if (pin2 == -1) {
2196 * Ok, does IRQ0 through the IOAPIC work?
2199 add_pin_to_irq(0, apic1, pin1);
2200 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2202 unmask_IO_APIC_irq(0);
2203 if (!no_timer_check && timer_irq_works()) {
2204 if (nmi_watchdog == NMI_IO_APIC) {
2206 enable_8259A_irq(0);
2208 if (disable_timer_pin_1 > 0)
2209 clear_IO_APIC_pin(0, pin1);
2212 if (intr_remapping_enabled)
2213 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2214 clear_IO_APIC_pin(apic1, pin1);
2216 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2217 "8254 timer not connected to IO-APIC\n");
2219 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2220 "(IRQ0) through the 8259A ...\n");
2221 apic_printk(APIC_QUIET, KERN_INFO
2222 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2224 * legacy devices should be connected to IO APIC #0
2226 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2227 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2228 unmask_IO_APIC_irq(0);
2229 enable_8259A_irq(0);
2230 if (timer_irq_works()) {
2231 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2232 timer_through_8259 = 1;
2233 if (nmi_watchdog == NMI_IO_APIC) {
2234 disable_8259A_irq(0);
2236 enable_8259A_irq(0);
2241 * Cleanup, just in case ...
2243 disable_8259A_irq(0);
2244 clear_IO_APIC_pin(apic2, pin2);
2245 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2248 if (nmi_watchdog == NMI_IO_APIC) {
2249 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2250 "through the IO-APIC - disabling NMI Watchdog!\n");
2251 nmi_watchdog = NMI_NONE;
2254 apic_printk(APIC_QUIET, KERN_INFO
2255 "...trying to set up timer as Virtual Wire IRQ...\n");
2257 lapic_register_intr(0);
2258 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2259 enable_8259A_irq(0);
2261 if (timer_irq_works()) {
2262 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2265 disable_8259A_irq(0);
2266 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2267 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2269 apic_printk(APIC_QUIET, KERN_INFO
2270 "...trying to set up timer as ExtINT IRQ...\n");
2274 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2276 unlock_ExtINT_logic();
2278 if (timer_irq_works()) {
2279 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2282 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2283 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2284 "report. Then try booting with the 'noapic' option.\n");
2286 local_irq_restore(flags);
2289 static int __init notimercheck(char *s)
2294 __setup("no_timer_check", notimercheck);
2297 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2298 * to devices. However there may be an I/O APIC pin available for
2299 * this interrupt regardless. The pin may be left unconnected, but
2300 * typically it will be reused as an ExtINT cascade interrupt for
2301 * the master 8259A. In the MPS case such a pin will normally be
2302 * reported as an ExtINT interrupt in the MP table. With ACPI
2303 * there is no provision for ExtINT interrupts, and in the absence
2304 * of an override it would be treated as an ordinary ISA I/O APIC
2305 * interrupt, that is edge-triggered and unmasked by default. We
2306 * used to do this, but it caused problems on some systems because
2307 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2308 * the same ExtINT cascade interrupt to drive the local APIC of the
2309 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2310 * the I/O APIC in all cases now. No actual device should request
2311 * it anyway. --macro
2313 #define PIC_IRQS (1<<2)
2315 void __init setup_IO_APIC(void)
2319 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2322 io_apic_irqs = ~PIC_IRQS;
2324 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2327 setup_IO_APIC_irqs();
2328 init_IO_APIC_traps();
2332 struct sysfs_ioapic_data {
2333 struct sys_device dev;
2334 struct IO_APIC_route_entry entry[0];
2336 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2338 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2340 struct IO_APIC_route_entry *entry;
2341 struct sysfs_ioapic_data *data;
2344 data = container_of(dev, struct sysfs_ioapic_data, dev);
2345 entry = data->entry;
2346 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
2347 *entry = ioapic_read_entry(dev->id, i);
2352 static int ioapic_resume(struct sys_device *dev)
2354 struct IO_APIC_route_entry *entry;
2355 struct sysfs_ioapic_data *data;
2356 unsigned long flags;
2357 union IO_APIC_reg_00 reg_00;
2360 data = container_of(dev, struct sysfs_ioapic_data, dev);
2361 entry = data->entry;
2363 spin_lock_irqsave(&ioapic_lock, flags);
2364 reg_00.raw = io_apic_read(dev->id, 0);
2365 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
2366 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
2367 io_apic_write(dev->id, 0, reg_00.raw);
2369 spin_unlock_irqrestore(&ioapic_lock, flags);
2370 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
2371 ioapic_write_entry(dev->id, i, entry[i]);
2376 static struct sysdev_class ioapic_sysdev_class = {
2378 .suspend = ioapic_suspend,
2379 .resume = ioapic_resume,
2382 static int __init ioapic_init_sysfs(void)
2384 struct sys_device * dev;
2387 error = sysdev_class_register(&ioapic_sysdev_class);
2391 for (i = 0; i < nr_ioapics; i++ ) {
2392 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2393 * sizeof(struct IO_APIC_route_entry);
2394 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
2395 if (!mp_ioapic_data[i]) {
2396 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2399 dev = &mp_ioapic_data[i]->dev;
2401 dev->cls = &ioapic_sysdev_class;
2402 error = sysdev_register(dev);
2404 kfree(mp_ioapic_data[i]);
2405 mp_ioapic_data[i] = NULL;
2406 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2414 device_initcall(ioapic_init_sysfs);
2417 * Dynamic irq allocate and deallocation
2419 int create_irq(void)
2421 /* Allocate an unused irq */
2424 unsigned long flags;
2425 struct irq_cfg *cfg_new;
2428 spin_lock_irqsave(&vector_lock, flags);
2429 for (new = (nr_irqs - 1); new >= 0; new--) {
2430 if (platform_legacy_irq(new))
2432 cfg_new = irq_cfg(new);
2433 if (cfg_new && cfg_new->vector != 0)
2435 /* check if need to create one */
2437 cfg_new = irq_cfg_alloc(new);
2438 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
2442 spin_unlock_irqrestore(&vector_lock, flags);
2445 dynamic_irq_init(irq);
2450 void destroy_irq(unsigned int irq)
2452 unsigned long flags;
2454 dynamic_irq_cleanup(irq);
2456 #ifdef CONFIG_INTR_REMAP
2459 spin_lock_irqsave(&vector_lock, flags);
2460 __clear_irq_vector(irq);
2461 spin_unlock_irqrestore(&vector_lock, flags);
2465 * MSI message composition
2467 #ifdef CONFIG_PCI_MSI
2468 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2470 struct irq_cfg *cfg;
2476 err = assign_irq_vector(irq, tmp);
2481 cpus_and(tmp, cfg->domain, tmp);
2482 dest = cpu_mask_to_apicid(tmp);
2484 #ifdef CONFIG_INTR_REMAP
2485 if (irq_remapped(irq)) {
2490 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
2491 BUG_ON(ir_index == -1);
2493 memset (&irte, 0, sizeof(irte));
2496 irte.dst_mode = INT_DEST_MODE;
2497 irte.trigger_mode = 0; /* edge */
2498 irte.dlvry_mode = INT_DELIVERY_MODE;
2499 irte.vector = cfg->vector;
2500 irte.dest_id = IRTE_DEST(dest);
2502 modify_irte(irq, &irte);
2504 msg->address_hi = MSI_ADDR_BASE_HI;
2505 msg->data = sub_handle;
2506 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
2508 MSI_ADDR_IR_INDEX1(ir_index) |
2509 MSI_ADDR_IR_INDEX2(ir_index);
2513 msg->address_hi = MSI_ADDR_BASE_HI;
2516 ((INT_DEST_MODE == 0) ?
2517 MSI_ADDR_DEST_MODE_PHYSICAL:
2518 MSI_ADDR_DEST_MODE_LOGICAL) |
2519 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2520 MSI_ADDR_REDIRECTION_CPU:
2521 MSI_ADDR_REDIRECTION_LOWPRI) |
2522 MSI_ADDR_DEST_ID(dest);
2525 MSI_DATA_TRIGGER_EDGE |
2526 MSI_DATA_LEVEL_ASSERT |
2527 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2528 MSI_DATA_DELIVERY_FIXED:
2529 MSI_DATA_DELIVERY_LOWPRI) |
2530 MSI_DATA_VECTOR(cfg->vector);
2536 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2538 struct irq_cfg *cfg;
2542 struct irq_desc *desc;
2544 cpus_and(tmp, mask, cpu_online_map);
2545 if (cpus_empty(tmp))
2548 if (assign_irq_vector(irq, mask))
2552 cpus_and(tmp, cfg->domain, mask);
2553 dest = cpu_mask_to_apicid(tmp);
2555 read_msi_msg(irq, &msg);
2557 msg.data &= ~MSI_DATA_VECTOR_MASK;
2558 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2559 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2560 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2562 write_msi_msg(irq, &msg);
2563 desc = irq_to_desc(irq);
2564 desc->affinity = mask;
2567 #ifdef CONFIG_INTR_REMAP
2569 * Migrate the MSI irq to another cpumask. This migration is
2570 * done in the process context using interrupt-remapping hardware.
2572 static void ir_set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2574 struct irq_cfg *cfg;
2576 cpumask_t tmp, cleanup_mask;
2578 struct irq_desc *desc;
2580 cpus_and(tmp, mask, cpu_online_map);
2581 if (cpus_empty(tmp))
2584 if (get_irte(irq, &irte))
2587 if (assign_irq_vector(irq, mask))
2591 cpus_and(tmp, cfg->domain, mask);
2592 dest = cpu_mask_to_apicid(tmp);
2594 irte.vector = cfg->vector;
2595 irte.dest_id = IRTE_DEST(dest);
2598 * atomically update the IRTE with the new destination and vector.
2600 modify_irte(irq, &irte);
2603 * After this point, all the interrupts will start arriving
2604 * at the new destination. So, time to cleanup the previous
2605 * vector allocation.
2607 if (cfg->move_in_progress) {
2608 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
2609 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
2610 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2611 cfg->move_in_progress = 0;
2614 desc = irq_to_desc(irq);
2615 desc->affinity = mask;
2618 #endif /* CONFIG_SMP */
2621 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2622 * which implement the MSI or MSI-X Capability Structure.
2624 static struct irq_chip msi_chip = {
2626 .unmask = unmask_msi_irq,
2627 .mask = mask_msi_irq,
2628 .ack = ack_apic_edge,
2630 .set_affinity = set_msi_irq_affinity,
2632 .retrigger = ioapic_retrigger_irq,
2635 #ifdef CONFIG_INTR_REMAP
2636 static struct irq_chip msi_ir_chip = {
2637 .name = "IR-PCI-MSI",
2638 .unmask = unmask_msi_irq,
2639 .mask = mask_msi_irq,
2640 .ack = ack_x2apic_edge,
2642 .set_affinity = ir_set_msi_irq_affinity,
2644 .retrigger = ioapic_retrigger_irq,
2648 * Map the PCI dev to the corresponding remapping hardware unit
2649 * and allocate 'nvec' consecutive interrupt-remapping table entries
2652 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
2654 struct intel_iommu *iommu;
2657 iommu = map_dev_to_ir(dev);
2660 "Unable to map PCI %s to iommu\n", pci_name(dev));
2664 index = alloc_irte(iommu, irq, nvec);
2667 "Unable to allocate %d IRTE for PCI %s\n", nvec,
2675 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc, int irq)
2680 ret = msi_compose_msg(dev, irq, &msg);
2684 set_irq_msi(irq, desc);
2685 write_msi_msg(irq, &msg);
2687 #ifdef CONFIG_INTR_REMAP
2688 if (irq_remapped(irq)) {
2689 struct irq_desc *desc = irq_to_desc(irq);
2691 * irq migration in process context
2693 desc->status |= IRQ_MOVE_PCNTXT;
2694 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
2697 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2702 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2710 #ifdef CONFIG_INTR_REMAP
2711 if (!intr_remapping_enabled)
2714 ret = msi_alloc_irte(dev, irq, 1);
2719 ret = setup_msi_irq(dev, desc, irq);
2726 #ifdef CONFIG_INTR_REMAP
2733 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
2735 int irq, ret, sub_handle;
2736 struct msi_desc *desc;
2737 #ifdef CONFIG_INTR_REMAP
2738 struct intel_iommu *iommu = 0;
2743 list_for_each_entry(desc, &dev->msi_list, list) {
2747 #ifdef CONFIG_INTR_REMAP
2748 if (!intr_remapping_enabled)
2753 * allocate the consecutive block of IRTE's
2756 index = msi_alloc_irte(dev, irq, nvec);
2762 iommu = map_dev_to_ir(dev);
2768 * setup the mapping between the irq and the IRTE
2769 * base index, the sub_handle pointing to the
2770 * appropriate interrupt remap table entry.
2772 set_irte_irq(irq, iommu, index, sub_handle);
2776 ret = setup_msi_irq(dev, desc, irq);
2788 void arch_teardown_msi_irq(unsigned int irq)
2795 static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
2797 struct irq_cfg *cfg;
2801 struct irq_desc *desc;
2803 cpus_and(tmp, mask, cpu_online_map);
2804 if (cpus_empty(tmp))
2807 if (assign_irq_vector(irq, mask))
2811 cpus_and(tmp, cfg->domain, mask);
2812 dest = cpu_mask_to_apicid(tmp);
2814 dmar_msi_read(irq, &msg);
2816 msg.data &= ~MSI_DATA_VECTOR_MASK;
2817 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2818 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2819 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2821 dmar_msi_write(irq, &msg);
2822 desc = irq_to_desc(irq);
2823 desc->affinity = mask;
2825 #endif /* CONFIG_SMP */
2827 struct irq_chip dmar_msi_type = {
2829 .unmask = dmar_msi_unmask,
2830 .mask = dmar_msi_mask,
2831 .ack = ack_apic_edge,
2833 .set_affinity = dmar_msi_set_affinity,
2835 .retrigger = ioapic_retrigger_irq,
2838 int arch_setup_dmar_msi(unsigned int irq)
2843 ret = msi_compose_msg(NULL, irq, &msg);
2846 dmar_msi_write(irq, &msg);
2847 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
2853 #endif /* CONFIG_PCI_MSI */
2855 * Hypertransport interrupt support
2857 #ifdef CONFIG_HT_IRQ
2861 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2863 struct ht_irq_msg msg;
2864 fetch_ht_irq_msg(irq, &msg);
2866 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2867 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2869 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2870 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2872 write_ht_irq_msg(irq, &msg);
2875 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2877 struct irq_cfg *cfg;
2880 struct irq_desc *desc;
2882 cpus_and(tmp, mask, cpu_online_map);
2883 if (cpus_empty(tmp))
2886 if (assign_irq_vector(irq, mask))
2890 cpus_and(tmp, cfg->domain, mask);
2891 dest = cpu_mask_to_apicid(tmp);
2893 target_ht_irq(irq, dest, cfg->vector);
2894 desc = irq_to_desc(irq);
2895 desc->affinity = mask;
2899 static struct irq_chip ht_irq_chip = {
2901 .mask = mask_ht_irq,
2902 .unmask = unmask_ht_irq,
2903 .ack = ack_apic_edge,
2905 .set_affinity = set_ht_irq_affinity,
2907 .retrigger = ioapic_retrigger_irq,
2910 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2912 struct irq_cfg *cfg;
2917 err = assign_irq_vector(irq, tmp);
2919 struct ht_irq_msg msg;
2923 cpus_and(tmp, cfg->domain, tmp);
2924 dest = cpu_mask_to_apicid(tmp);
2926 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2930 HT_IRQ_LOW_DEST_ID(dest) |
2931 HT_IRQ_LOW_VECTOR(cfg->vector) |
2932 ((INT_DEST_MODE == 0) ?
2933 HT_IRQ_LOW_DM_PHYSICAL :
2934 HT_IRQ_LOW_DM_LOGICAL) |
2935 HT_IRQ_LOW_RQEOI_EDGE |
2936 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2937 HT_IRQ_LOW_MT_FIXED :
2938 HT_IRQ_LOW_MT_ARBITRATED) |
2939 HT_IRQ_LOW_IRQ_MASKED;
2941 write_ht_irq_msg(irq, &msg);
2943 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2944 handle_edge_irq, "edge");
2948 #endif /* CONFIG_HT_IRQ */
2950 /* --------------------------------------------------------------------------
2951 ACPI-based IOAPIC Configuration
2952 -------------------------------------------------------------------------- */
2956 #define IO_APIC_MAX_ID 0xFE
2958 int __init io_apic_get_redir_entries (int ioapic)
2960 union IO_APIC_reg_01 reg_01;
2961 unsigned long flags;
2963 spin_lock_irqsave(&ioapic_lock, flags);
2964 reg_01.raw = io_apic_read(ioapic, 1);
2965 spin_unlock_irqrestore(&ioapic_lock, flags);
2967 return reg_01.bits.entries;
2971 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2973 if (!IO_APIC_IRQ(irq)) {
2974 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2980 * IRQs < 16 are already in the irq_2_pin[] map
2983 add_pin_to_irq(irq, ioapic, pin);
2985 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2991 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2995 if (skip_ioapic_setup)
2998 for (i = 0; i < mp_irq_entries; i++)
2999 if (mp_irqs[i].mp_irqtype == mp_INT &&
3000 mp_irqs[i].mp_srcbusirq == bus_irq)
3002 if (i >= mp_irq_entries)
3005 *trigger = irq_trigger(i);
3006 *polarity = irq_polarity(i);
3010 #endif /* CONFIG_ACPI */
3013 * This function currently is only a helper for the i386 smp boot process where
3014 * we need to reprogram the ioredtbls to cater for the cpus which have come online
3015 * so mask in all cases should simply be TARGET_CPUS
3018 void __init setup_ioapic_dest(void)
3020 int pin, ioapic, irq, irq_entry;
3021 struct irq_cfg *cfg;
3023 if (skip_ioapic_setup == 1)
3026 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
3027 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
3028 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
3029 if (irq_entry == -1)
3031 irq = pin_2_irq(irq_entry, ioapic, pin);
3033 /* setup_IO_APIC_irqs could fail to get vector for some device
3034 * when you have too many devices, because at that time only boot
3039 setup_IO_APIC_irq(ioapic, pin, irq,
3040 irq_trigger(irq_entry),
3041 irq_polarity(irq_entry));
3042 #ifdef CONFIG_INTR_REMAP
3043 else if (intr_remapping_enabled)
3044 set_ir_ioapic_affinity_irq(irq, TARGET_CPUS);
3047 set_ioapic_affinity_irq(irq, TARGET_CPUS);
3054 #define IOAPIC_RESOURCE_NAME_SIZE 11
3056 static struct resource *ioapic_resources;
3058 static struct resource * __init ioapic_setup_resources(void)
3061 struct resource *res;
3065 if (nr_ioapics <= 0)
3068 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
3071 mem = alloc_bootmem(n);
3075 mem += sizeof(struct resource) * nr_ioapics;
3077 for (i = 0; i < nr_ioapics; i++) {
3079 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
3080 sprintf(mem, "IOAPIC %u", i);
3081 mem += IOAPIC_RESOURCE_NAME_SIZE;
3085 ioapic_resources = res;
3090 void __init ioapic_init_mappings(void)
3092 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
3093 struct resource *ioapic_res;
3096 ioapic_res = ioapic_setup_resources();
3097 for (i = 0; i < nr_ioapics; i++) {
3098 if (smp_found_config) {
3099 ioapic_phys = mp_ioapics[i].mp_apicaddr;
3101 ioapic_phys = (unsigned long)
3102 alloc_bootmem_pages(PAGE_SIZE);
3103 ioapic_phys = __pa(ioapic_phys);
3105 set_fixmap_nocache(idx, ioapic_phys);
3106 apic_printk(APIC_VERBOSE,
3107 "mapped IOAPIC to %016lx (%016lx)\n",
3108 __fix_to_virt(idx), ioapic_phys);
3111 if (ioapic_res != NULL) {
3112 ioapic_res->start = ioapic_phys;
3113 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
3119 static int __init ioapic_insert_resources(void)
3122 struct resource *r = ioapic_resources;
3126 "IO APIC resources could be not be allocated.\n");
3130 for (i = 0; i < nr_ioapics; i++) {
3131 insert_resource(&iomem_resource, r);
3138 /* Insert the IO APIC resources after PCI initialization has occured to handle
3139 * IO APICS that are mapped in on a BAR in PCI space. */
3140 late_initcall(ioapic_insert_resources);