x86: Sanitize smp_record and move it to x86_init_ops
[safe/jmp/linux-2.6] / arch / x86 / kernel / apic / io_apic.c
index 82376e0..d2ed6c5 100644 (file)
@@ -59,6 +59,7 @@
 #include <asm/setup.h>
 #include <asm/irq_remapping.h>
 #include <asm/hpet.h>
+#include <asm/hw_irq.h>
 #include <asm/uv/uv_hub.h>
 #include <asm/uv/uv_irq.h>
 
@@ -176,16 +177,18 @@ int __init arch_early_irq_init(void)
        struct irq_cfg *cfg;
        struct irq_desc *desc;
        int count;
+       int node;
        int i;
 
        cfg = irq_cfgx;
        count = ARRAY_SIZE(irq_cfgx);
+       node= cpu_to_node(boot_cpu_id);
 
        for (i = 0; i < count; i++) {
                desc = irq_to_desc(i);
                desc->chip_data = &cfg[i];
-               alloc_bootmem_cpumask_var(&cfg[i].domain);
-               alloc_bootmem_cpumask_var(&cfg[i].old_domain);
+               zalloc_cpumask_var_node(&cfg[i].domain, GFP_NOWAIT, node);
+               zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_NOWAIT, node);
                if (i < NR_IRQS_LEGACY)
                        cpumask_setall(cfg[i].domain);
        }
@@ -459,7 +462,8 @@ static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
 static void
 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 {
-       union entry_union eu;
+       union entry_union eu = {{0, 0}};
+
        eu.entry = e;
        io_apic_write(apic, 0x11 + 2*pin, eu.w2);
        io_apic_write(apic, 0x10 + 2*pin, eu.w1);
@@ -489,121 +493,6 @@ static void ioapic_mask_entry(int apic, int pin)
        spin_unlock_irqrestore(&ioapic_lock, flags);
 }
 
-#ifdef CONFIG_SMP
-static void send_cleanup_vector(struct irq_cfg *cfg)
-{
-       cpumask_var_t cleanup_mask;
-
-       if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
-               unsigned int i;
-               cfg->move_cleanup_count = 0;
-               for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
-                       cfg->move_cleanup_count++;
-               for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
-                       apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
-       } else {
-               cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
-               cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
-               apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
-               free_cpumask_var(cleanup_mask);
-       }
-       cfg->move_in_progress = 0;
-}
-
-static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
-{
-       int apic, pin;
-       struct irq_pin_list *entry;
-       u8 vector = cfg->vector;
-
-       entry = cfg->irq_2_pin;
-       for (;;) {
-               unsigned int reg;
-
-               if (!entry)
-                       break;
-
-               apic = entry->apic;
-               pin = entry->pin;
-               /*
-                * With interrupt-remapping, destination information comes
-                * from interrupt-remapping table entry.
-                */
-               if (!irq_remapped(irq))
-                       io_apic_write(apic, 0x11 + pin*2, dest);
-               reg = io_apic_read(apic, 0x10 + pin*2);
-               reg &= ~IO_APIC_REDIR_VECTOR_MASK;
-               reg |= vector;
-               io_apic_modify(apic, 0x10 + pin*2, reg);
-               if (!entry->next)
-                       break;
-               entry = entry->next;
-       }
-}
-
-static int
-assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
-
-/*
- * Either sets desc->affinity to a valid value, and returns
- * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
- * leaves desc->affinity untouched.
- */
-static unsigned int
-set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
-{
-       struct irq_cfg *cfg;
-       unsigned int irq;
-
-       if (!cpumask_intersects(mask, cpu_online_mask))
-               return BAD_APICID;
-
-       irq = desc->irq;
-       cfg = desc->chip_data;
-       if (assign_irq_vector(irq, cfg, mask))
-               return BAD_APICID;
-
-       cpumask_copy(desc->affinity, mask);
-
-       return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
-}
-
-static int
-set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
-{
-       struct irq_cfg *cfg;
-       unsigned long flags;
-       unsigned int dest;
-       unsigned int irq;
-       int ret = -1;
-
-       irq = desc->irq;
-       cfg = desc->chip_data;
-
-       spin_lock_irqsave(&ioapic_lock, flags);
-       dest = set_desc_affinity(desc, mask);
-       if (dest != BAD_APICID) {
-               /* Only the high 8 bits are valid. */
-               dest = SET_APIC_LOGICAL_ID(dest);
-               __target_IO_APIC_irq(irq, dest, cfg);
-               ret = 0;
-       }
-       spin_unlock_irqrestore(&ioapic_lock, flags);
-
-       return ret;
-}
-
-static int
-set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
-{
-       struct irq_desc *desc;
-
-       desc = irq_to_desc(irq);
-
-       return set_ioapic_affinity_irq_desc(desc, mask);
-}
-#endif /* CONFIG_SMP */
-
 /*
  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
  * shared ISA-space IRQs, so we have to support them. We are super
@@ -822,7 +711,6 @@ static int __init ioapic_pirq_setup(char *str)
 __setup("pirq=", ioapic_pirq_setup);
 #endif /* CONFIG_X86_32 */
 
-#ifdef CONFIG_INTR_REMAP
 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
 {
        int apic;
@@ -920,20 +808,6 @@ int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
        return 0;
 }
 
-void reinit_intr_remapped_IO_APIC(int intr_remapping,
-       struct IO_APIC_route_entry **ioapic_entries)
-
-{
-       /*
-        * for now plain restore of previous settings.
-        * TBD: In the case of OS enabling interrupt-remapping,
-        * IO-APIC RTE's need to be setup to point to interrupt-remapping
-        * table entries. for now, do a plain restore, and wait for
-        * the setup_IO_APIC_irqs() to do proper initialization.
-        */
-       restore_IO_APIC_setup(ioapic_entries);
-}
-
 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
 {
        int apic;
@@ -943,7 +817,6 @@ void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
 
        kfree(ioapic_entries);
 }
-#endif
 
 /*
  * Find the IRQ entry number of a certain pin.
@@ -1004,54 +877,6 @@ static int __init find_isa_irq_apic(int irq, int type)
        return -1;
 }
 
-/*
- * Find a specific PCI IRQ entry.
- * Not an __init, possibly needed by modules
- */
-static int pin_2_irq(int idx, int apic, int pin);
-
-int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
-{
-       int apic, i, best_guess = -1;
-
-       apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
-               bus, slot, pin);
-       if (test_bit(bus, mp_bus_not_pci)) {
-               apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
-               return -1;
-       }
-       for (i = 0; i < mp_irq_entries; i++) {
-               int lbus = mp_irqs[i].srcbus;
-
-               for (apic = 0; apic < nr_ioapics; apic++)
-                       if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
-                           mp_irqs[i].dstapic == MP_APIC_ALL)
-                               break;
-
-               if (!test_bit(lbus, mp_bus_not_pci) &&
-                   !mp_irqs[i].irqtype &&
-                   (bus == lbus) &&
-                   (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
-                       int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
-
-                       if (!(apic || IO_APIC_IRQ(irq)))
-                               continue;
-
-                       if (pin == (mp_irqs[i].srcbusirq & 3))
-                               return irq;
-                       /*
-                        * Use the first all-but-pin matching entry as a
-                        * best-guess fuzzy result for broken mptables.
-                        */
-                       if (best_guess < 0)
-                               best_guess = irq;
-               }
-       }
-       return best_guess;
-}
-
-EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
-
 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
 /*
  * EISA Edge/Level control register, ELCR
@@ -1270,6 +1095,64 @@ static int pin_2_irq(int idx, int apic, int pin)
        return irq;
 }
 
+/*
+ * Find a specific PCI IRQ entry.
+ * Not an __init, possibly needed by modules
+ */
+int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
+                               struct io_apic_irq_attr *irq_attr)
+{
+       int apic, i, best_guess = -1;
+
+       apic_printk(APIC_DEBUG,
+                   "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
+                   bus, slot, pin);
+       if (test_bit(bus, mp_bus_not_pci)) {
+               apic_printk(APIC_VERBOSE,
+                           "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
+               return -1;
+       }
+       for (i = 0; i < mp_irq_entries; i++) {
+               int lbus = mp_irqs[i].srcbus;
+
+               for (apic = 0; apic < nr_ioapics; apic++)
+                       if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
+                           mp_irqs[i].dstapic == MP_APIC_ALL)
+                               break;
+
+               if (!test_bit(lbus, mp_bus_not_pci) &&
+                   !mp_irqs[i].irqtype &&
+                   (bus == lbus) &&
+                   (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
+                       int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
+
+                       if (!(apic || IO_APIC_IRQ(irq)))
+                               continue;
+
+                       if (pin == (mp_irqs[i].srcbusirq & 3)) {
+                               set_io_apic_irq_attr(irq_attr, apic,
+                                                    mp_irqs[i].dstirq,
+                                                    irq_trigger(i),
+                                                    irq_polarity(i));
+                               return irq;
+                       }
+                       /*
+                        * Use the first all-but-pin matching entry as a
+                        * best-guess fuzzy result for broken mptables.
+                        */
+                       if (best_guess < 0) {
+                               set_io_apic_irq_attr(irq_attr, apic,
+                                                    mp_irqs[i].dstirq,
+                                                    irq_trigger(i),
+                                                    irq_polarity(i));
+                               best_guess = irq;
+                       }
+               }
+       }
+       return best_guess;
+}
+EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
+
 void lock_vector_lock(void)
 {
        /* Used to the online set of cpus does not change
@@ -1531,6 +1414,9 @@ int setup_ioapic_entry(int apic_id, int irq,
                irte.vector = vector;
                irte.dest_id = IRTE_DEST(destination);
 
+               /* Set source-id of interrupt request */
+               set_ioapic_sid(&irte, apic_id);
+
                modify_irte(irq, &irte);
 
                ir_entry->index2 = (index >> 15) & 0x1;
@@ -1600,9 +1486,13 @@ static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq
        ioapic_write_entry(apic_id, pin, entry);
 }
 
+static struct {
+       DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
+} mp_ioapic_routing[MAX_IO_APICS];
+
 static void __init setup_IO_APIC_irqs(void)
 {
-       int apic_id, pin, idx, irq;
+       int apic_id = 0, pin, idx, irq;
        int notcon = 0;
        struct irq_desc *desc;
        struct irq_cfg *cfg;
@@ -1610,48 +1500,56 @@ static void __init setup_IO_APIC_irqs(void)
 
        apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
 
-       for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
-               for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
-
-                       idx = find_irq_entry(apic_id, pin, mp_INT);
-                       if (idx == -1) {
-                               if (!notcon) {
-                                       notcon = 1;
-                                       apic_printk(APIC_VERBOSE,
-                                               KERN_DEBUG " %d-%d",
-                                               mp_ioapics[apic_id].apicid, pin);
-                               } else
-                                       apic_printk(APIC_VERBOSE, " %d-%d",
-                                               mp_ioapics[apic_id].apicid, pin);
-                               continue;
-                       }
-                       if (notcon) {
-                               apic_printk(APIC_VERBOSE,
-                                       " (apicid-pin) not connected\n");
-                               notcon = 0;
-                       }
+#ifdef CONFIG_ACPI
+       if (!acpi_disabled && acpi_ioapic) {
+               apic_id = mp_find_ioapic(0);
+               if (apic_id < 0)
+                       apic_id = 0;
+       }
+#endif
 
-                       irq = pin_2_irq(idx, apic_id, pin);
+       for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
+               idx = find_irq_entry(apic_id, pin, mp_INT);
+               if (idx == -1) {
+                       if (!notcon) {
+                               notcon = 1;
+                               apic_printk(APIC_VERBOSE,
+                                       KERN_DEBUG " %d-%d",
+                                       mp_ioapics[apic_id].apicid, pin);
+                       } else
+                               apic_printk(APIC_VERBOSE, " %d-%d",
+                                       mp_ioapics[apic_id].apicid, pin);
+                       continue;
+               }
+               if (notcon) {
+                       apic_printk(APIC_VERBOSE,
+                               " (apicid-pin) not connected\n");
+                       notcon = 0;
+               }
 
-                       /*
-                        * Skip the timer IRQ if there's a quirk handler
-                        * installed and if it returns 1:
-                        */
-                       if (apic->multi_timer_check &&
-                                       apic->multi_timer_check(apic_id, irq))
-                               continue;
+               irq = pin_2_irq(idx, apic_id, pin);
 
-                       desc = irq_to_desc_alloc_node(irq, node);
-                       if (!desc) {
-                               printk(KERN_INFO "can not get irq_desc for %d\n", irq);
-                               continue;
-                       }
-                       cfg = desc->chip_data;
-                       add_pin_to_irq_node(cfg, node, apic_id, pin);
+               /*
+                * Skip the timer IRQ if there's a quirk handler
+                * installed and if it returns 1:
+                */
+               if (apic->multi_timer_check &&
+                               apic->multi_timer_check(apic_id, irq))
+                       continue;
 
-                       setup_IO_APIC_irq(apic_id, pin, irq, desc,
-                                       irq_trigger(idx), irq_polarity(idx));
+               desc = irq_to_desc_alloc_node(irq, node);
+               if (!desc) {
+                       printk(KERN_INFO "can not get irq_desc for %d\n", irq);
+                       continue;
                }
+               cfg = desc->chip_data;
+               add_pin_to_irq_node(cfg, node, apic_id, pin);
+               /*
+                * don't mark it in pin_programmed, so later acpi could
+                * set it correctly when irq < 16
+                */
+               setup_IO_APIC_irq(apic_id, pin, irq, desc,
+                               irq_trigger(idx), irq_polarity(idx));
        }
 
        if (notcon)
@@ -1818,36 +1716,30 @@ __apicdebuginit(void) print_IO_APIC(void)
        return;
 }
 
-__apicdebuginit(void) print_APIC_bitfield(int base)
+__apicdebuginit(void) print_APIC_field(int base)
 {
-       unsigned int v;
-       int i, j;
+       int i;
 
        if (apic_verbosity == APIC_QUIET)
                return;
 
-       printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
-       for (i = 0; i < 8; i++) {
-               v = apic_read(base + i*0x10);
-               for (j = 0; j < 32; j++) {
-                       if (v & (1<<j))
-                               printk("1");
-                       else
-                               printk("0");
-               }
-               printk("\n");
-       }
+       printk(KERN_DEBUG);
+
+       for (i = 0; i < 8; i++)
+               printk(KERN_CONT "%08x", apic_read(base + i*0x10));
+
+       printk(KERN_CONT "\n");
 }
 
 __apicdebuginit(void) print_local_APIC(void *dummy)
 {
-       unsigned int v, ver, maxlvt;
+       unsigned int i, v, ver, maxlvt;
        u64 icr;
 
        if (apic_verbosity == APIC_QUIET)
                return;
 
-       printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
+       printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
                smp_processor_id(), hard_smp_processor_id());
        v = apic_read(APIC_ID);
        printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
@@ -1888,11 +1780,11 @@ __apicdebuginit(void) print_local_APIC(void *dummy)
        printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
 
        printk(KERN_DEBUG "... APIC ISR field:\n");
-       print_APIC_bitfield(APIC_ISR);
+       print_APIC_field(APIC_ISR);
        printk(KERN_DEBUG "... APIC TMR field:\n");
-       print_APIC_bitfield(APIC_TMR);
+       print_APIC_field(APIC_TMR);
        printk(KERN_DEBUG "... APIC IRR field:\n");
-       print_APIC_bitfield(APIC_IRR);
+       print_APIC_field(APIC_IRR);
 
        if (APIC_INTEGRATED(ver)) {             /* !82489DX */
                if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
@@ -1929,6 +1821,18 @@ __apicdebuginit(void) print_local_APIC(void *dummy)
        printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
        v = apic_read(APIC_TDCR);
        printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
+
+       if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
+               v = apic_read(APIC_EFEAT);
+               maxlvt = (v >> 16) & 0xff;
+               printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
+               v = apic_read(APIC_ECTRL);
+               printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
+               for (i = 0; i < maxlvt; i++) {
+                       v = apic_read(APIC_EILVTn(i));
+                       printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
+               }
+       }
        printk("\n");
 }
 
@@ -1977,6 +1881,11 @@ __apicdebuginit(void) print_PIC(void)
 __apicdebuginit(int) print_all_ICs(void)
 {
        print_PIC();
+
+       /* don't print out if apic is not there */
+       if (!cpu_has_apic || disable_apic)
+               return 0;
+
        print_all_local_APICs();
        print_IO_APIC();
 
@@ -2092,7 +2001,9 @@ void disable_IO_APIC(void)
        /*
         * Use virtual wire A mode when interrupt remapping is enabled.
         */
-       disconnect_bsp_APIC(!intr_remapping_enabled && ioapic_i8259.pin != -1);
+       if (cpu_has_apic)
+               disconnect_bsp_APIC(!intr_remapping_enabled &&
+                               ioapic_i8259.pin != -1);
 }
 
 #ifdef CONFIG_X86_32
@@ -2332,11 +2243,123 @@ static int ioapic_retrigger_irq(unsigned int irq)
  */
 
 #ifdef CONFIG_SMP
+static void send_cleanup_vector(struct irq_cfg *cfg)
+{
+       cpumask_var_t cleanup_mask;
 
-#ifdef CONFIG_INTR_REMAP
-
-/*
- * Migrate the IO-APIC irq in the presence of intr-remapping.
+       if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
+               unsigned int i;
+               cfg->move_cleanup_count = 0;
+               for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
+                       cfg->move_cleanup_count++;
+               for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
+                       apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
+       } else {
+               cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
+               cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
+               apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
+               free_cpumask_var(cleanup_mask);
+       }
+       cfg->move_in_progress = 0;
+}
+
+static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
+{
+       int apic, pin;
+       struct irq_pin_list *entry;
+       u8 vector = cfg->vector;
+
+       entry = cfg->irq_2_pin;
+       for (;;) {
+               unsigned int reg;
+
+               if (!entry)
+                       break;
+
+               apic = entry->apic;
+               pin = entry->pin;
+               /*
+                * With interrupt-remapping, destination information comes
+                * from interrupt-remapping table entry.
+                */
+               if (!irq_remapped(irq))
+                       io_apic_write(apic, 0x11 + pin*2, dest);
+               reg = io_apic_read(apic, 0x10 + pin*2);
+               reg &= ~IO_APIC_REDIR_VECTOR_MASK;
+               reg |= vector;
+               io_apic_modify(apic, 0x10 + pin*2, reg);
+               if (!entry->next)
+                       break;
+               entry = entry->next;
+       }
+}
+
+static int
+assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
+
+/*
+ * Either sets desc->affinity to a valid value, and returns
+ * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
+ * leaves desc->affinity untouched.
+ */
+static unsigned int
+set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
+{
+       struct irq_cfg *cfg;
+       unsigned int irq;
+
+       if (!cpumask_intersects(mask, cpu_online_mask))
+               return BAD_APICID;
+
+       irq = desc->irq;
+       cfg = desc->chip_data;
+       if (assign_irq_vector(irq, cfg, mask))
+               return BAD_APICID;
+
+       cpumask_copy(desc->affinity, mask);
+
+       return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
+}
+
+static int
+set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
+{
+       struct irq_cfg *cfg;
+       unsigned long flags;
+       unsigned int dest;
+       unsigned int irq;
+       int ret = -1;
+
+       irq = desc->irq;
+       cfg = desc->chip_data;
+
+       spin_lock_irqsave(&ioapic_lock, flags);
+       dest = set_desc_affinity(desc, mask);
+       if (dest != BAD_APICID) {
+               /* Only the high 8 bits are valid. */
+               dest = SET_APIC_LOGICAL_ID(dest);
+               __target_IO_APIC_irq(irq, dest, cfg);
+               ret = 0;
+       }
+       spin_unlock_irqrestore(&ioapic_lock, flags);
+
+       return ret;
+}
+
+static int
+set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
+{
+       struct irq_desc *desc;
+
+       desc = irq_to_desc(irq);
+
+       return set_ioapic_affinity_irq_desc(desc, mask);
+}
+
+#ifdef CONFIG_INTR_REMAP
+
+/*
+ * Migrate the IO-APIC irq in the presence of intr-remapping.
  *
  * For both level and edge triggered, irq migration is a simple atomic
  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
@@ -2478,53 +2501,6 @@ static void irq_complete_move(struct irq_desc **descp)
 static inline void irq_complete_move(struct irq_desc **descp) {}
 #endif
 
-static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
-{
-       int apic, pin;
-       struct irq_pin_list *entry;
-
-       entry = cfg->irq_2_pin;
-       for (;;) {
-
-               if (!entry)
-                       break;
-
-               apic = entry->apic;
-               pin = entry->pin;
-               io_apic_eoi(apic, pin);
-               entry = entry->next;
-       }
-}
-
-static void
-eoi_ioapic_irq(struct irq_desc *desc)
-{
-       struct irq_cfg *cfg;
-       unsigned long flags;
-       unsigned int irq;
-
-       irq = desc->irq;
-       cfg = desc->chip_data;
-
-       spin_lock_irqsave(&ioapic_lock, flags);
-       __eoi_ioapic_irq(irq, cfg);
-       spin_unlock_irqrestore(&ioapic_lock, flags);
-}
-
-#ifdef CONFIG_X86_X2APIC
-static void ack_x2apic_level(unsigned int irq)
-{
-       struct irq_desc *desc = irq_to_desc(irq);
-       ack_x2APIC_irq();
-       eoi_ioapic_irq(desc);
-}
-
-static void ack_x2apic_edge(unsigned int irq)
-{
-       ack_x2APIC_irq();
-}
-#endif
-
 static void ack_apic_edge(unsigned int irq)
 {
        struct irq_desc *desc = irq_to_desc(irq);
@@ -2588,9 +2564,6 @@ static void ack_apic_level(unsigned int irq)
         */
        ack_APIC_irq();
 
-       if (irq_remapped(irq))
-               eoi_ioapic_irq(desc);
-
        /* Now we can move and renable the irq */
        if (unlikely(do_unmask_irq)) {
                /* Only migrate the irq if the ack has been received.
@@ -2637,22 +2610,50 @@ static void ack_apic_level(unsigned int irq)
 }
 
 #ifdef CONFIG_INTR_REMAP
+static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
+{
+       int apic, pin;
+       struct irq_pin_list *entry;
+
+       entry = cfg->irq_2_pin;
+       for (;;) {
+
+               if (!entry)
+                       break;
+
+               apic = entry->apic;
+               pin = entry->pin;
+               io_apic_eoi(apic, pin);
+               entry = entry->next;
+       }
+}
+
+static void
+eoi_ioapic_irq(struct irq_desc *desc)
+{
+       struct irq_cfg *cfg;
+       unsigned long flags;
+       unsigned int irq;
+
+       irq = desc->irq;
+       cfg = desc->chip_data;
+
+       spin_lock_irqsave(&ioapic_lock, flags);
+       __eoi_ioapic_irq(irq, cfg);
+       spin_unlock_irqrestore(&ioapic_lock, flags);
+}
+
 static void ir_ack_apic_edge(unsigned int irq)
 {
-#ifdef CONFIG_X86_X2APIC
-       if (x2apic_enabled())
-               return ack_x2apic_edge(irq);
-#endif
-       return ack_apic_edge(irq);
+       ack_APIC_irq();
 }
 
 static void ir_ack_apic_level(unsigned int irq)
 {
-#ifdef CONFIG_X86_X2APIC
-       if (x2apic_enabled())
-               return ack_x2apic_level(irq);
-#endif
-       return ack_apic_level(irq);
+       struct irq_desc *desc = irq_to_desc(irq);
+
+       ack_APIC_irq();
+       eoi_ioapic_irq(desc);
 }
 #endif /* CONFIG_INTR_REMAP */
 
@@ -3172,14 +3173,13 @@ static int nr_irqs_gsi = NR_IRQS_LEGACY;
 /*
  * Dynamic irq allocate and deallocation
  */
-unsigned int create_irq_nr(unsigned int irq_want)
+unsigned int create_irq_nr(unsigned int irq_want, int node)
 {
        /* Allocate an unused irq */
        unsigned int irq;
        unsigned int new;
        unsigned long flags;
        struct irq_cfg *cfg_new = NULL;
-       int node = cpu_to_node(boot_cpu_id);
        struct irq_desc *desc_new = NULL;
 
        irq = 0;
@@ -3197,6 +3197,9 @@ unsigned int create_irq_nr(unsigned int irq_want)
 
                if (cfg_new->vector != 0)
                        continue;
+
+               desc_new = move_irq_desc(desc_new, node);
+
                if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
                        irq = new;
                break;
@@ -3214,11 +3217,12 @@ unsigned int create_irq_nr(unsigned int irq_want)
 
 int create_irq(void)
 {
+       int node = cpu_to_node(boot_cpu_id);
        unsigned int irq_want;
        int irq;
 
        irq_want = nr_irqs_gsi;
-       irq = create_irq_nr(irq_want);
+       irq = create_irq_nr(irq_want, node);
 
        if (irq == 0)
                irq = -1;
@@ -3283,6 +3287,9 @@ static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_ms
                irte.vector = cfg->vector;
                irte.dest_id = IRTE_DEST(dest);
 
+               /* Set source-id of interrupt request */
+               set_msi_sid(&irte, pdev);
+
                modify_irte(irq, &irte);
 
                msg->address_hi = MSI_ADDR_BASE_HI;
@@ -3476,15 +3483,17 @@ int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
        unsigned int irq_want;
        struct intel_iommu *iommu = NULL;
        int index = 0;
+       int node;
 
        /* x86 doesn't support multiple MSI yet */
        if (type == PCI_CAP_ID_MSI && nvec > 1)
                return 1;
 
+       node = dev_to_node(&dev->dev);
        irq_want = nr_irqs_gsi;
        sub_handle = 0;
        list_for_each_entry(msidesc, &dev->msi_list, list) {
-               irq = create_irq_nr(irq_want);
+               irq = create_irq_nr(irq_want, node);
                if (irq == 0)
                        return -1;
                irq_want = irq + 1;
@@ -3561,7 +3570,7 @@ static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
 
 #endif /* CONFIG_SMP */
 
-struct irq_chip dmar_msi_type = {
+static struct irq_chip dmar_msi_type = {
        .name = "DMAR_MSI",
        .unmask = dmar_msi_unmask,
        .mask = dmar_msi_mask,
@@ -3758,6 +3767,8 @@ int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
        unsigned long flags;
        int err;
 
+       BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
+
        cfg = irq_cfg(irq);
 
        err = assign_irq_vector(irq, cfg, eligible_cpu);
@@ -3771,19 +3782,20 @@ int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
 
        mmr_value = 0;
        entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
-       BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
-
-       entry->vector = cfg->vector;
-       entry->delivery_mode = apic->irq_delivery_mode;
-       entry->dest_mode = apic->irq_dest_mode;
-       entry->polarity = 0;
-       entry->trigger = 0;
-       entry->mask = 0;
-       entry->dest = apic->cpu_mask_to_apicid(eligible_cpu);
+       entry->vector           = cfg->vector;
+       entry->delivery_mode    = apic->irq_delivery_mode;
+       entry->dest_mode        = apic->irq_dest_mode;
+       entry->polarity         = 0;
+       entry->trigger          = 0;
+       entry->mask             = 0;
+       entry->dest             = apic->cpu_mask_to_apicid(eligible_cpu);
 
        mmr_pnode = uv_blade_to_pnode(mmr_blade);
        uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
 
+       if (cfg->move_in_progress)
+               send_cleanup_vector(cfg);
+
        return irq;
 }
 
@@ -3797,10 +3809,10 @@ void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
        struct uv_IO_APIC_route_entry *entry;
        int mmr_pnode;
 
+       BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
+
        mmr_value = 0;
        entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
-       BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
-
        entry->mask = 1;
 
        mmr_pnode = uv_blade_to_pnode(mmr_blade);
@@ -3864,6 +3876,71 @@ int __init arch_probe_nr_irqs(void)
 }
 #endif
 
+static int __io_apic_set_pci_routing(struct device *dev, int irq,
+                               struct io_apic_irq_attr *irq_attr)
+{
+       struct irq_desc *desc;
+       struct irq_cfg *cfg;
+       int node;
+       int ioapic, pin;
+       int trigger, polarity;
+
+       ioapic = irq_attr->ioapic;
+       if (!IO_APIC_IRQ(irq)) {
+               apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
+                       ioapic);
+               return -EINVAL;
+       }
+
+       if (dev)
+               node = dev_to_node(dev);
+       else
+               node = cpu_to_node(boot_cpu_id);
+
+       desc = irq_to_desc_alloc_node(irq, node);
+       if (!desc) {
+               printk(KERN_INFO "can not get irq_desc %d\n", irq);
+               return 0;
+       }
+
+       pin = irq_attr->ioapic_pin;
+       trigger = irq_attr->trigger;
+       polarity = irq_attr->polarity;
+
+       /*
+        * IRQs < 16 are already in the irq_2_pin[] map
+        */
+       if (irq >= NR_IRQS_LEGACY) {
+               cfg = desc->chip_data;
+               add_pin_to_irq_node(cfg, node, ioapic, pin);
+       }
+
+       setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
+
+       return 0;
+}
+
+int io_apic_set_pci_routing(struct device *dev, int irq,
+                               struct io_apic_irq_attr *irq_attr)
+{
+       int ioapic, pin;
+       /*
+        * Avoid pin reprogramming.  PRTs typically include entries
+        * with redundant pin->gsi mappings (but unique PCI devices);
+        * we only program the IOAPIC on the first.
+        */
+       ioapic = irq_attr->ioapic;
+       pin = irq_attr->ioapic_pin;
+       if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
+               pr_debug("Pin %d-%d already programmed\n",
+                        mp_ioapics[ioapic].apicid, pin);
+               return 0;
+       }
+       set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
+
+       return __io_apic_set_pci_routing(dev, irq, irq_attr);
+}
+
 /* --------------------------------------------------------------------------
                           ACPI-based IOAPIC Configuration
    -------------------------------------------------------------------------- */
@@ -3944,6 +4021,7 @@ int __init io_apic_get_unique_id(int ioapic, int apic_id)
 
        return apic_id;
 }
+#endif
 
 int __init io_apic_get_version(int ioapic)
 {
@@ -3956,45 +4034,6 @@ int __init io_apic_get_version(int ioapic)
 
        return reg_01.bits.version;
 }
-#endif
-
-int io_apic_set_pci_routing(struct device *dev, int ioapic, int pin, int irq,
-                                int triggering, int polarity)
-{
-       struct irq_desc *desc;
-       struct irq_cfg *cfg;
-       int node;
-
-       if (!IO_APIC_IRQ(irq)) {
-               apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
-                       ioapic);
-               return -EINVAL;
-       }
-
-       if (dev)
-               node = dev_to_node(dev);
-       else
-               node = cpu_to_node(boot_cpu_id);
-
-       desc = irq_to_desc_alloc_node(irq, node);
-       if (!desc) {
-               printk(KERN_INFO "can not get irq_desc %d\n", irq);
-               return 0;
-       }
-
-       /*
-        * IRQs < 16 are already in the irq_2_pin[] map
-        */
-       if (irq >= NR_IRQS_LEGACY) {
-               cfg = desc->chip_data;
-               add_pin_to_irq_node(cfg, node, ioapic, pin);
-       }
-
-       setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
-
-       return 0;
-}
-
 
 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
 {
@@ -4025,51 +4064,44 @@ int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
 #ifdef CONFIG_SMP
 void __init setup_ioapic_dest(void)
 {
-       int pin, ioapic, irq, irq_entry;
+       int pin, ioapic = 0, irq, irq_entry;
        struct irq_desc *desc;
-       struct irq_cfg *cfg;
        const struct cpumask *mask;
 
        if (skip_ioapic_setup == 1)
                return;
 
-       for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
-               for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
-                       irq_entry = find_irq_entry(ioapic, pin, mp_INT);
-                       if (irq_entry == -1)
-                               continue;
-                       irq = pin_2_irq(irq_entry, ioapic, pin);
-
-                       /* setup_IO_APIC_irqs could fail to get vector for some device
-                        * when you have too many devices, because at that time only boot
-                        * cpu is online.
-                        */
-                       desc = irq_to_desc(irq);
-                       cfg = desc->chip_data;
-                       if (!cfg->vector) {
-                               setup_IO_APIC_irq(ioapic, pin, irq, desc,
-                                                 irq_trigger(irq_entry),
-                                                 irq_polarity(irq_entry));
-                               continue;
+#ifdef CONFIG_ACPI
+       if (!acpi_disabled && acpi_ioapic) {
+               ioapic = mp_find_ioapic(0);
+               if (ioapic < 0)
+                       ioapic = 0;
+       }
+#endif
 
-                       }
+       for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
+               irq_entry = find_irq_entry(ioapic, pin, mp_INT);
+               if (irq_entry == -1)
+                       continue;
+               irq = pin_2_irq(irq_entry, ioapic, pin);
 
-                       /*
-                        * Honour affinities which have been set in early boot
-                        */
-                       if (desc->status &
-                           (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
-                               mask = desc->affinity;
-                       else
-                               mask = apic->target_cpus();
+               desc = irq_to_desc(irq);
 
-                       if (intr_remapping_enabled)
-                               set_ir_ioapic_affinity_irq_desc(desc, mask);
-                       else
-                               set_ioapic_affinity_irq_desc(desc, mask);
-               }
+               /*
+                * Honour affinities which have been set in early boot
+                */
+               if (desc->status &
+                   (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
+                       mask = desc->affinity;
+               else
+                       mask = apic->target_cpus();
 
+               if (intr_remapping_enabled)
+                       set_ir_ioapic_affinity_irq_desc(desc, mask);
+               else
+                       set_ioapic_affinity_irq_desc(desc, mask);
        }
+
 }
 #endif
 
@@ -4152,28 +4184,20 @@ fake_ioapic_page:
        }
 }
 
-static int __init ioapic_insert_resources(void)
+void __init ioapic_insert_resources(void)
 {
        int i;
        struct resource *r = ioapic_resources;
 
        if (!r) {
-               if (nr_ioapics > 0) {
+               if (nr_ioapics > 0)
                        printk(KERN_ERR
                                "IO APIC resources couldn't be allocated.\n");
-                       return -1;
-               }
-               return 0;
+               return;
        }
 
        for (i = 0; i < nr_ioapics; i++) {
                insert_resource(&iomem_resource, r);
                r++;
        }
-
-       return 0;
 }
-
-/* Insert the IO APIC resources after PCI initialization has occured to handle
- * IO APICS that are mapped in on a BAR in PCI space. */
-late_initcall(ioapic_insert_resources);