1 // SPDX-License-Identifier: GPL-2.0-only
3 * Local APIC related interfaces to support IOAPIC, MSI, etc.
5 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 * Moved from arch/x86/kernel/apic/io_apic.c.
7 * Jiang Liu <jiang.liu@linux.intel.com>
8 * Enable support of hierarchical irqdomains
10 #include <linux/interrupt.h>
11 #include <linux/irq.h>
12 #include <linux/seq_file.h>
13 #include <linux/init.h>
14 #include <linux/compiler.h>
15 #include <linux/slab.h>
16 #include <asm/irqdomain.h>
17 #include <asm/hw_irq.h>
18 #include <asm/traps.h>
20 #include <asm/i8259.h>
22 #include <asm/irq_remapping.h>
24 #include <asm/trace/irq_vectors.h>
26 struct apic_chip_data {
27 struct irq_cfg hw_irq_cfg;
29 unsigned int prev_vector;
31 unsigned int prev_cpu;
33 struct hlist_node clist;
34 unsigned int move_in_progress : 1,
40 struct irq_domain *x86_vector_domain;
41 EXPORT_SYMBOL_GPL(x86_vector_domain);
42 static DEFINE_RAW_SPINLOCK(vector_lock);
43 static cpumask_var_t vector_searchmask;
44 static struct irq_chip lapic_controller;
45 static struct irq_matrix *vector_matrix;
47 static DEFINE_PER_CPU(struct hlist_head, cleanup_list);
50 void lock_vector_lock(void)
52 /* Used to the online set of cpus does not change
53 * during assign_irq_vector.
55 raw_spin_lock(&vector_lock);
58 void unlock_vector_lock(void)
60 raw_spin_unlock(&vector_lock);
63 void init_irq_alloc_info(struct irq_alloc_info *info,
64 const struct cpumask *mask)
66 memset(info, 0, sizeof(*info));
70 void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
75 memset(dst, 0, sizeof(*dst));
78 static struct apic_chip_data *apic_chip_data(struct irq_data *irqd)
83 while (irqd->parent_data)
84 irqd = irqd->parent_data;
86 return irqd->chip_data;
89 struct irq_cfg *irqd_cfg(struct irq_data *irqd)
91 struct apic_chip_data *apicd = apic_chip_data(irqd);
93 return apicd ? &apicd->hw_irq_cfg : NULL;
95 EXPORT_SYMBOL_GPL(irqd_cfg);
97 struct irq_cfg *irq_cfg(unsigned int irq)
99 return irqd_cfg(irq_get_irq_data(irq));
102 static struct apic_chip_data *alloc_apic_chip_data(int node)
104 struct apic_chip_data *apicd;
106 apicd = kzalloc_node(sizeof(*apicd), GFP_KERNEL, node);
108 INIT_HLIST_NODE(&apicd->clist);
112 static void free_apic_chip_data(struct apic_chip_data *apicd)
117 static void apic_update_irq_cfg(struct irq_data *irqd, unsigned int vector,
120 struct apic_chip_data *apicd = apic_chip_data(irqd);
122 lockdep_assert_held(&vector_lock);
124 apicd->hw_irq_cfg.vector = vector;
125 apicd->hw_irq_cfg.dest_apicid = apic->calc_dest_apicid(cpu);
126 irq_data_update_effective_affinity(irqd, cpumask_of(cpu));
127 trace_vector_config(irqd->irq, vector, cpu,
128 apicd->hw_irq_cfg.dest_apicid);
131 static void apic_update_vector(struct irq_data *irqd, unsigned int newvec,
134 struct apic_chip_data *apicd = apic_chip_data(irqd);
135 struct irq_desc *desc = irq_data_to_desc(irqd);
136 bool managed = irqd_affinity_is_managed(irqd);
138 lockdep_assert_held(&vector_lock);
140 trace_vector_update(irqd->irq, newvec, newcpu, apicd->vector,
144 * If there is no vector associated or if the associated vector is
145 * the shutdown vector, which is associated to make PCI/MSI
146 * shutdown mode work, then there is nothing to release. Clear out
147 * prev_vector for this and the offlined target case.
149 apicd->prev_vector = 0;
150 if (!apicd->vector || apicd->vector == MANAGED_IRQ_SHUTDOWN_VECTOR)
153 * If the target CPU of the previous vector is online, then mark
154 * the vector as move in progress and store it for cleanup when the
155 * first interrupt on the new vector arrives. If the target CPU is
156 * offline then the regular release mechanism via the cleanup
157 * vector is not possible and the vector can be immediately freed
158 * in the underlying matrix allocator.
160 if (cpu_online(apicd->cpu)) {
161 apicd->move_in_progress = true;
162 apicd->prev_vector = apicd->vector;
163 apicd->prev_cpu = apicd->cpu;
164 WARN_ON_ONCE(apicd->cpu == newcpu);
166 irq_matrix_free(vector_matrix, apicd->cpu, apicd->vector,
171 apicd->vector = newvec;
173 BUG_ON(!IS_ERR_OR_NULL(per_cpu(vector_irq, newcpu)[newvec]));
174 per_cpu(vector_irq, newcpu)[newvec] = desc;
177 static void vector_assign_managed_shutdown(struct irq_data *irqd)
179 unsigned int cpu = cpumask_first(cpu_online_mask);
181 apic_update_irq_cfg(irqd, MANAGED_IRQ_SHUTDOWN_VECTOR, cpu);
184 static int reserve_managed_vector(struct irq_data *irqd)
186 const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
187 struct apic_chip_data *apicd = apic_chip_data(irqd);
191 raw_spin_lock_irqsave(&vector_lock, flags);
192 apicd->is_managed = true;
193 ret = irq_matrix_reserve_managed(vector_matrix, affmsk);
194 raw_spin_unlock_irqrestore(&vector_lock, flags);
195 trace_vector_reserve_managed(irqd->irq, ret);
199 static void reserve_irq_vector_locked(struct irq_data *irqd)
201 struct apic_chip_data *apicd = apic_chip_data(irqd);
203 irq_matrix_reserve(vector_matrix);
204 apicd->can_reserve = true;
205 apicd->has_reserved = true;
206 irqd_set_can_reserve(irqd);
207 trace_vector_reserve(irqd->irq, 0);
208 vector_assign_managed_shutdown(irqd);
211 static int reserve_irq_vector(struct irq_data *irqd)
215 raw_spin_lock_irqsave(&vector_lock, flags);
216 reserve_irq_vector_locked(irqd);
217 raw_spin_unlock_irqrestore(&vector_lock, flags);
222 assign_vector_locked(struct irq_data *irqd, const struct cpumask *dest)
224 struct apic_chip_data *apicd = apic_chip_data(irqd);
225 bool resvd = apicd->has_reserved;
226 unsigned int cpu = apicd->cpu;
227 int vector = apicd->vector;
229 lockdep_assert_held(&vector_lock);
232 * If the current target CPU is online and in the new requested
233 * affinity mask, there is no point in moving the interrupt from
234 * one CPU to another.
236 if (vector && cpu_online(cpu) && cpumask_test_cpu(cpu, dest))
240 * Careful here. @apicd might either have move_in_progress set or
241 * be enqueued for cleanup. Assigning a new vector would either
242 * leave a stale vector on some CPU around or in case of a pending
243 * cleanup corrupt the hlist.
245 if (apicd->move_in_progress || !hlist_unhashed(&apicd->clist))
248 vector = irq_matrix_alloc(vector_matrix, dest, resvd, &cpu);
249 trace_vector_alloc(irqd->irq, vector, resvd, vector);
252 apic_update_vector(irqd, vector, cpu);
253 apic_update_irq_cfg(irqd, vector, cpu);
258 static int assign_irq_vector(struct irq_data *irqd, const struct cpumask *dest)
263 raw_spin_lock_irqsave(&vector_lock, flags);
264 cpumask_and(vector_searchmask, dest, cpu_online_mask);
265 ret = assign_vector_locked(irqd, vector_searchmask);
266 raw_spin_unlock_irqrestore(&vector_lock, flags);
270 static int assign_irq_vector_any_locked(struct irq_data *irqd)
272 /* Get the affinity mask - either irq_default_affinity or (user) set */
273 const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
274 int node = irq_data_get_node(irqd);
276 if (node == NUMA_NO_NODE)
278 /* Try the intersection of @affmsk and node mask */
279 cpumask_and(vector_searchmask, cpumask_of_node(node), affmsk);
280 if (!assign_vector_locked(irqd, vector_searchmask))
282 /* Try the node mask */
283 if (!assign_vector_locked(irqd, cpumask_of_node(node)))
286 /* Try the full affinity mask */
287 cpumask_and(vector_searchmask, affmsk, cpu_online_mask);
288 if (!assign_vector_locked(irqd, vector_searchmask))
290 /* Try the full online mask */
291 return assign_vector_locked(irqd, cpu_online_mask);
295 assign_irq_vector_policy(struct irq_data *irqd, struct irq_alloc_info *info)
297 if (irqd_affinity_is_managed(irqd))
298 return reserve_managed_vector(irqd);
300 return assign_irq_vector(irqd, info->mask);
302 * Make only a global reservation with no guarantee. A real vector
303 * is associated at activation time.
305 return reserve_irq_vector(irqd);
309 assign_managed_vector(struct irq_data *irqd, const struct cpumask *dest)
311 const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
312 struct apic_chip_data *apicd = apic_chip_data(irqd);
315 cpumask_and(vector_searchmask, dest, affmsk);
317 /* set_affinity might call here for nothing */
318 if (apicd->vector && cpumask_test_cpu(apicd->cpu, vector_searchmask))
320 vector = irq_matrix_alloc_managed(vector_matrix, vector_searchmask,
322 trace_vector_alloc_managed(irqd->irq, vector, vector);
325 apic_update_vector(irqd, vector, cpu);
326 apic_update_irq_cfg(irqd, vector, cpu);
330 static void clear_irq_vector(struct irq_data *irqd)
332 struct apic_chip_data *apicd = apic_chip_data(irqd);
333 bool managed = irqd_affinity_is_managed(irqd);
334 unsigned int vector = apicd->vector;
336 lockdep_assert_held(&vector_lock);
341 trace_vector_clear(irqd->irq, vector, apicd->cpu, apicd->prev_vector,
344 per_cpu(vector_irq, apicd->cpu)[vector] = VECTOR_SHUTDOWN;
345 irq_matrix_free(vector_matrix, apicd->cpu, vector, managed);
348 /* Clean up move in progress */
349 vector = apicd->prev_vector;
353 per_cpu(vector_irq, apicd->prev_cpu)[vector] = VECTOR_SHUTDOWN;
354 irq_matrix_free(vector_matrix, apicd->prev_cpu, vector, managed);
355 apicd->prev_vector = 0;
356 apicd->move_in_progress = 0;
357 hlist_del_init(&apicd->clist);
360 static void x86_vector_deactivate(struct irq_domain *dom, struct irq_data *irqd)
362 struct apic_chip_data *apicd = apic_chip_data(irqd);
365 trace_vector_deactivate(irqd->irq, apicd->is_managed,
366 apicd->can_reserve, false);
368 /* Regular fixed assigned interrupt */
369 if (!apicd->is_managed && !apicd->can_reserve)
371 /* If the interrupt has a global reservation, nothing to do */
372 if (apicd->has_reserved)
375 raw_spin_lock_irqsave(&vector_lock, flags);
376 clear_irq_vector(irqd);
377 if (apicd->can_reserve)
378 reserve_irq_vector_locked(irqd);
380 vector_assign_managed_shutdown(irqd);
381 raw_spin_unlock_irqrestore(&vector_lock, flags);
384 static int activate_reserved(struct irq_data *irqd)
386 struct apic_chip_data *apicd = apic_chip_data(irqd);
389 ret = assign_irq_vector_any_locked(irqd);
391 apicd->has_reserved = false;
393 * Core might have disabled reservation mode after
394 * allocating the irq descriptor. Ideally this should
395 * happen before allocation time, but that would require
396 * completely convoluted ways of transporting that
399 if (!irqd_can_reserve(irqd))
400 apicd->can_reserve = false;
404 * Check to ensure that the effective affinity mask is a subset
405 * the user supplied affinity mask, and warn the user if it is not
407 if (!cpumask_subset(irq_data_get_effective_affinity_mask(irqd),
408 irq_data_get_affinity_mask(irqd))) {
409 pr_warn("irq %u: Affinity broken due to vector space exhaustion.\n",
416 static int activate_managed(struct irq_data *irqd)
418 const struct cpumask *dest = irq_data_get_affinity_mask(irqd);
421 cpumask_and(vector_searchmask, dest, cpu_online_mask);
422 if (WARN_ON_ONCE(cpumask_empty(vector_searchmask))) {
423 /* Something in the core code broke! Survive gracefully */
424 pr_err("Managed startup for irq %u, but no CPU\n", irqd->irq);
428 ret = assign_managed_vector(irqd, vector_searchmask);
430 * This should not happen. The vector reservation got buggered. Handle
433 if (WARN_ON_ONCE(ret < 0)) {
434 pr_err("Managed startup irq %u, no vector available\n",
440 static int x86_vector_activate(struct irq_domain *dom, struct irq_data *irqd,
443 struct apic_chip_data *apicd = apic_chip_data(irqd);
447 trace_vector_activate(irqd->irq, apicd->is_managed,
448 apicd->can_reserve, reserve);
450 raw_spin_lock_irqsave(&vector_lock, flags);
451 if (!apicd->can_reserve && !apicd->is_managed)
452 assign_irq_vector_any_locked(irqd);
453 else if (reserve || irqd_is_managed_and_shutdown(irqd))
454 vector_assign_managed_shutdown(irqd);
455 else if (apicd->is_managed)
456 ret = activate_managed(irqd);
457 else if (apicd->has_reserved)
458 ret = activate_reserved(irqd);
459 raw_spin_unlock_irqrestore(&vector_lock, flags);
463 static void vector_free_reserved_and_managed(struct irq_data *irqd)
465 const struct cpumask *dest = irq_data_get_affinity_mask(irqd);
466 struct apic_chip_data *apicd = apic_chip_data(irqd);
468 trace_vector_teardown(irqd->irq, apicd->is_managed,
469 apicd->has_reserved);
471 if (apicd->has_reserved)
472 irq_matrix_remove_reserved(vector_matrix);
473 if (apicd->is_managed)
474 irq_matrix_remove_managed(vector_matrix, dest);
477 static void x86_vector_free_irqs(struct irq_domain *domain,
478 unsigned int virq, unsigned int nr_irqs)
480 struct apic_chip_data *apicd;
481 struct irq_data *irqd;
485 for (i = 0; i < nr_irqs; i++) {
486 irqd = irq_domain_get_irq_data(x86_vector_domain, virq + i);
487 if (irqd && irqd->chip_data) {
488 raw_spin_lock_irqsave(&vector_lock, flags);
489 clear_irq_vector(irqd);
490 vector_free_reserved_and_managed(irqd);
491 apicd = irqd->chip_data;
492 irq_domain_reset_irq_data(irqd);
493 raw_spin_unlock_irqrestore(&vector_lock, flags);
494 free_apic_chip_data(apicd);
499 static bool vector_configure_legacy(unsigned int virq, struct irq_data *irqd,
500 struct apic_chip_data *apicd)
503 bool realloc = false;
505 apicd->vector = ISA_IRQ_VECTOR(virq);
508 raw_spin_lock_irqsave(&vector_lock, flags);
510 * If the interrupt is activated, then it must stay at this vector
511 * position. That's usually the timer interrupt (0).
513 if (irqd_is_activated(irqd)) {
514 trace_vector_setup(virq, true, 0);
515 apic_update_irq_cfg(irqd, apicd->vector, apicd->cpu);
517 /* Release the vector */
518 apicd->can_reserve = true;
519 irqd_set_can_reserve(irqd);
520 clear_irq_vector(irqd);
523 raw_spin_unlock_irqrestore(&vector_lock, flags);
527 static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
528 unsigned int nr_irqs, void *arg)
530 struct irq_alloc_info *info = arg;
531 struct apic_chip_data *apicd;
532 struct irq_data *irqd;
538 /* Currently vector allocator can't guarantee contiguous allocations */
539 if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
542 for (i = 0; i < nr_irqs; i++) {
543 irqd = irq_domain_get_irq_data(domain, virq + i);
545 node = irq_data_get_node(irqd);
546 WARN_ON_ONCE(irqd->chip_data);
547 apicd = alloc_apic_chip_data(node);
553 apicd->irq = virq + i;
554 irqd->chip = &lapic_controller;
555 irqd->chip_data = apicd;
556 irqd->hwirq = virq + i;
557 irqd_set_single_target(irqd);
559 * Prevent that any of these interrupts is invoked in
560 * non interrupt context via e.g. generic_handle_irq()
561 * as that can corrupt the affinity move state.
563 irqd_set_handle_enforce_irqctx(irqd);
565 /* Don't invoke affinity setter on deactivated interrupts */
566 irqd_set_affinity_on_activate(irqd);
569 * Legacy vectors are already assigned when the IOAPIC
570 * takes them over. They stay on the same vector. This is
571 * required for check_timer() to work correctly as it might
572 * switch back to legacy mode. Only update the hardware
575 if (info->flags & X86_IRQ_ALLOC_LEGACY) {
576 if (!vector_configure_legacy(virq + i, irqd, apicd))
580 err = assign_irq_vector_policy(irqd, info);
581 trace_vector_setup(virq + i, false, err);
583 irqd->chip_data = NULL;
584 free_apic_chip_data(apicd);
592 x86_vector_free_irqs(domain, virq, i);
596 #ifdef CONFIG_GENERIC_IRQ_DEBUGFS
597 static void x86_vector_debug_show(struct seq_file *m, struct irq_domain *d,
598 struct irq_data *irqd, int ind)
600 struct apic_chip_data apicd;
605 irq_matrix_debug_show(m, vector_matrix, ind);
610 if (irq < nr_legacy_irqs() && !test_bit(irq, &io_apic_irqs)) {
611 seq_printf(m, "%*sVector: %5d\n", ind, "", ISA_IRQ_VECTOR(irq));
612 seq_printf(m, "%*sTarget: Legacy PIC all CPUs\n", ind, "");
616 if (!irqd->chip_data) {
617 seq_printf(m, "%*sVector: Not assigned\n", ind, "");
621 raw_spin_lock_irqsave(&vector_lock, flags);
622 memcpy(&apicd, irqd->chip_data, sizeof(apicd));
623 raw_spin_unlock_irqrestore(&vector_lock, flags);
625 seq_printf(m, "%*sVector: %5u\n", ind, "", apicd.vector);
626 seq_printf(m, "%*sTarget: %5u\n", ind, "", apicd.cpu);
627 if (apicd.prev_vector) {
628 seq_printf(m, "%*sPrevious vector: %5u\n", ind, "", apicd.prev_vector);
629 seq_printf(m, "%*sPrevious target: %5u\n", ind, "", apicd.prev_cpu);
631 seq_printf(m, "%*smove_in_progress: %u\n", ind, "", apicd.move_in_progress ? 1 : 0);
632 seq_printf(m, "%*sis_managed: %u\n", ind, "", apicd.is_managed ? 1 : 0);
633 seq_printf(m, "%*scan_reserve: %u\n", ind, "", apicd.can_reserve ? 1 : 0);
634 seq_printf(m, "%*shas_reserved: %u\n", ind, "", apicd.has_reserved ? 1 : 0);
635 seq_printf(m, "%*scleanup_pending: %u\n", ind, "", !hlist_unhashed(&apicd.clist));
639 int x86_fwspec_is_ioapic(struct irq_fwspec *fwspec)
641 if (fwspec->param_count != 1)
644 if (is_fwnode_irqchip(fwspec->fwnode)) {
645 const char *fwname = fwnode_get_name(fwspec->fwnode);
646 return fwname && !strncmp(fwname, "IO-APIC-", 8) &&
647 simple_strtol(fwname+8, NULL, 10) == fwspec->param[0];
649 return to_of_node(fwspec->fwnode) &&
650 of_device_is_compatible(to_of_node(fwspec->fwnode),
651 "intel,ce4100-ioapic");
654 int x86_fwspec_is_hpet(struct irq_fwspec *fwspec)
656 if (fwspec->param_count != 1)
659 if (is_fwnode_irqchip(fwspec->fwnode)) {
660 const char *fwname = fwnode_get_name(fwspec->fwnode);
661 return fwname && !strncmp(fwname, "HPET-MSI-", 9) &&
662 simple_strtol(fwname+9, NULL, 10) == fwspec->param[0];
667 static int x86_vector_select(struct irq_domain *d, struct irq_fwspec *fwspec,
668 enum irq_domain_bus_token bus_token)
671 * HPET and I/OAPIC cannot be parented in the vector domain
672 * if IRQ remapping is enabled. APIC IDs above 15 bits are
673 * only permitted if IRQ remapping is enabled, so check that.
675 if (apic->apic_id_valid(32768))
678 return x86_fwspec_is_ioapic(fwspec) || x86_fwspec_is_hpet(fwspec);
681 static const struct irq_domain_ops x86_vector_domain_ops = {
682 .select = x86_vector_select,
683 .alloc = x86_vector_alloc_irqs,
684 .free = x86_vector_free_irqs,
685 .activate = x86_vector_activate,
686 .deactivate = x86_vector_deactivate,
687 #ifdef CONFIG_GENERIC_IRQ_DEBUGFS
688 .debug_show = x86_vector_debug_show,
692 int __init arch_probe_nr_irqs(void)
696 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
697 nr_irqs = NR_VECTORS * nr_cpu_ids;
699 nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
700 #if defined(CONFIG_PCI_MSI)
702 * for MSI and HT dyn irq
704 if (gsi_top <= NR_IRQS_LEGACY)
705 nr += 8 * nr_cpu_ids;
713 * We don't know if PIC is present at this point so we need to do
714 * probe() to get the right number of legacy IRQs.
716 return legacy_pic->probe();
719 void lapic_assign_legacy_vector(unsigned int irq, bool replace)
722 * Use assign system here so it wont get accounted as allocated
723 * and moveable in the cpu hotplug check and it prevents managed
724 * irq reservation from touching it.
726 irq_matrix_assign_system(vector_matrix, ISA_IRQ_VECTOR(irq), replace);
729 void __init lapic_assign_system_vectors(void)
731 unsigned int i, vector = 0;
733 for_each_set_bit_from(vector, system_vectors, NR_VECTORS)
734 irq_matrix_assign_system(vector_matrix, vector, false);
736 if (nr_legacy_irqs() > 1)
737 lapic_assign_legacy_vector(PIC_CASCADE_IR, false);
739 /* System vectors are reserved, online it */
740 irq_matrix_online(vector_matrix);
742 /* Mark the preallocated legacy interrupts */
743 for (i = 0; i < nr_legacy_irqs(); i++) {
744 if (i != PIC_CASCADE_IR)
745 irq_matrix_assign(vector_matrix, ISA_IRQ_VECTOR(i));
749 int __init arch_early_irq_init(void)
751 struct fwnode_handle *fn;
753 fn = irq_domain_alloc_named_fwnode("VECTOR");
755 x86_vector_domain = irq_domain_create_tree(fn, &x86_vector_domain_ops,
757 BUG_ON(x86_vector_domain == NULL);
758 irq_set_default_host(x86_vector_domain);
760 BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
763 * Allocate the vector matrix allocator data structure and limit the
766 vector_matrix = irq_alloc_matrix(NR_VECTORS, FIRST_EXTERNAL_VECTOR,
767 FIRST_SYSTEM_VECTOR);
768 BUG_ON(!vector_matrix);
770 return arch_early_ioapic_init();
775 static struct irq_desc *__setup_vector_irq(int vector)
777 int isairq = vector - ISA_IRQ_VECTOR(0);
779 /* Check whether the irq is in the legacy space */
780 if (isairq < 0 || isairq >= nr_legacy_irqs())
781 return VECTOR_UNUSED;
782 /* Check whether the irq is handled by the IOAPIC */
783 if (test_bit(isairq, &io_apic_irqs))
784 return VECTOR_UNUSED;
785 return irq_to_desc(isairq);
788 /* Online the local APIC infrastructure and initialize the vectors */
789 void lapic_online(void)
793 lockdep_assert_held(&vector_lock);
795 /* Online the vector matrix array for this CPU */
796 irq_matrix_online(vector_matrix);
799 * The interrupt affinity logic never targets interrupts to offline
800 * CPUs. The exception are the legacy PIC interrupts. In general
801 * they are only targeted to CPU0, but depending on the platform
802 * they can be distributed to any online CPU in hardware. The
803 * kernel has no influence on that. So all active legacy vectors
804 * must be installed on all CPUs. All non legacy interrupts can be
807 for (vector = 0; vector < NR_VECTORS; vector++)
808 this_cpu_write(vector_irq[vector], __setup_vector_irq(vector));
811 void lapic_offline(void)
814 irq_matrix_offline(vector_matrix);
815 unlock_vector_lock();
818 static int apic_set_affinity(struct irq_data *irqd,
819 const struct cpumask *dest, bool force)
823 if (WARN_ON_ONCE(!irqd_is_activated(irqd)))
826 raw_spin_lock(&vector_lock);
827 cpumask_and(vector_searchmask, dest, cpu_online_mask);
828 if (irqd_affinity_is_managed(irqd))
829 err = assign_managed_vector(irqd, vector_searchmask);
831 err = assign_vector_locked(irqd, vector_searchmask);
832 raw_spin_unlock(&vector_lock);
833 return err ? err : IRQ_SET_MASK_OK;
837 # define apic_set_affinity NULL
840 static int apic_retrigger_irq(struct irq_data *irqd)
842 struct apic_chip_data *apicd = apic_chip_data(irqd);
845 raw_spin_lock_irqsave(&vector_lock, flags);
846 apic->send_IPI(apicd->cpu, apicd->vector);
847 raw_spin_unlock_irqrestore(&vector_lock, flags);
852 void apic_ack_irq(struct irq_data *irqd)
858 void apic_ack_edge(struct irq_data *irqd)
860 irq_complete_move(irqd_cfg(irqd));
864 static void x86_vector_msi_compose_msg(struct irq_data *data,
867 __irq_msi_compose_msg(irqd_cfg(data), msg, false);
870 static struct irq_chip lapic_controller = {
872 .irq_ack = apic_ack_edge,
873 .irq_set_affinity = apic_set_affinity,
874 .irq_compose_msi_msg = x86_vector_msi_compose_msg,
875 .irq_retrigger = apic_retrigger_irq,
880 static void free_moved_vector(struct apic_chip_data *apicd)
882 unsigned int vector = apicd->prev_vector;
883 unsigned int cpu = apicd->prev_cpu;
884 bool managed = apicd->is_managed;
887 * Managed interrupts are usually not migrated away
888 * from an online CPU, but CPU isolation 'managed_irq'
889 * can make that happen.
890 * 1) Activation does not take the isolation into account
891 * to keep the code simple
892 * 2) Migration away from an isolated CPU can happen when
893 * a non-isolated CPU which is in the calculated
894 * affinity mask comes online.
896 trace_vector_free_moved(apicd->irq, cpu, vector, managed);
897 irq_matrix_free(vector_matrix, cpu, vector, managed);
898 per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
899 hlist_del_init(&apicd->clist);
900 apicd->prev_vector = 0;
901 apicd->move_in_progress = 0;
904 DEFINE_IDTENTRY_SYSVEC(sysvec_irq_move_cleanup)
906 struct hlist_head *clhead = this_cpu_ptr(&cleanup_list);
907 struct apic_chip_data *apicd;
908 struct hlist_node *tmp;
911 /* Prevent vectors vanishing under us */
912 raw_spin_lock(&vector_lock);
914 hlist_for_each_entry_safe(apicd, tmp, clhead, clist) {
915 unsigned int irr, vector = apicd->prev_vector;
918 * Paranoia: Check if the vector that needs to be cleaned
919 * up is registered at the APICs IRR. If so, then this is
920 * not the best time to clean it up. Clean it up in the
921 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
922 * to this CPU. IRQ_MOVE_CLEANUP_VECTOR is the lowest
923 * priority external vector, so on return from this
924 * interrupt the device interrupt will happen first.
926 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
927 if (irr & (1U << (vector % 32))) {
928 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
931 free_moved_vector(apicd);
934 raw_spin_unlock(&vector_lock);
937 static void __send_cleanup_vector(struct apic_chip_data *apicd)
941 raw_spin_lock(&vector_lock);
942 apicd->move_in_progress = 0;
943 cpu = apicd->prev_cpu;
944 if (cpu_online(cpu)) {
945 hlist_add_head(&apicd->clist, per_cpu_ptr(&cleanup_list, cpu));
946 apic->send_IPI(cpu, IRQ_MOVE_CLEANUP_VECTOR);
948 apicd->prev_vector = 0;
950 raw_spin_unlock(&vector_lock);
953 void send_cleanup_vector(struct irq_cfg *cfg)
955 struct apic_chip_data *apicd;
957 apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg);
958 if (apicd->move_in_progress)
959 __send_cleanup_vector(apicd);
962 void irq_complete_move(struct irq_cfg *cfg)
964 struct apic_chip_data *apicd;
966 apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg);
967 if (likely(!apicd->move_in_progress))
971 * If the interrupt arrived on the new target CPU, cleanup the
972 * vector on the old target CPU. A vector check is not required
973 * because an interrupt can never move from one vector to another
976 if (apicd->cpu == smp_processor_id())
977 __send_cleanup_vector(apicd);
981 * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
983 void irq_force_complete_move(struct irq_desc *desc)
985 struct apic_chip_data *apicd;
986 struct irq_data *irqd;
990 * The function is called for all descriptors regardless of which
991 * irqdomain they belong to. For example if an IRQ is provided by
992 * an irq_chip as part of a GPIO driver, the chip data for that
993 * descriptor is specific to the irq_chip in question.
995 * Check first that the chip_data is what we expect
996 * (apic_chip_data) before touching it any further.
998 irqd = irq_domain_get_irq_data(x86_vector_domain,
999 irq_desc_get_irq(desc));
1003 raw_spin_lock(&vector_lock);
1004 apicd = apic_chip_data(irqd);
1009 * If prev_vector is empty, no action required.
1011 vector = apicd->prev_vector;
1016 * This is tricky. If the cleanup of the old vector has not been
1017 * done yet, then the following setaffinity call will fail with
1018 * -EBUSY. This can leave the interrupt in a stale state.
1020 * All CPUs are stuck in stop machine with interrupts disabled so
1021 * calling __irq_complete_move() would be completely pointless.
1023 * 1) The interrupt is in move_in_progress state. That means that we
1024 * have not seen an interrupt since the io_apic was reprogrammed to
1027 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
1028 * have not been processed yet.
1030 if (apicd->move_in_progress) {
1032 * In theory there is a race:
1034 * set_ioapic(new_vector) <-- Interrupt is raised before update
1035 * is effective, i.e. it's raised on
1038 * So if the target cpu cannot handle that interrupt before
1039 * the old vector is cleaned up, we get a spurious interrupt
1040 * and in the worst case the ioapic irq line becomes stale.
1042 * But in case of cpu hotplug this should be a non issue
1043 * because if the affinity update happens right before all
1044 * cpus rendevouz in stop machine, there is no way that the
1045 * interrupt can be blocked on the target cpu because all cpus
1046 * loops first with interrupts enabled in stop machine, so the
1047 * old vector is not yet cleaned up when the interrupt fires.
1049 * So the only way to run into this issue is if the delivery
1050 * of the interrupt on the apic/system bus would be delayed
1051 * beyond the point where the target cpu disables interrupts
1052 * in stop machine. I doubt that it can happen, but at least
1053 * there is a theroretical chance. Virtualization might be
1054 * able to expose this, but AFAICT the IOAPIC emulation is not
1055 * as stupid as the real hardware.
1057 * Anyway, there is nothing we can do about that at this point
1058 * w/o refactoring the whole fixup_irq() business completely.
1059 * We print at least the irq number and the old vector number,
1060 * so we have the necessary information when a problem in that
1063 pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
1066 free_moved_vector(apicd);
1068 raw_spin_unlock(&vector_lock);
1071 #ifdef CONFIG_HOTPLUG_CPU
1073 * Note, this is not accurate accounting, but at least good enough to
1074 * prevent that the actual interrupt move will run out of vectors.
1076 int lapic_can_unplug_cpu(void)
1078 unsigned int rsvd, avl, tomove, cpu = smp_processor_id();
1081 raw_spin_lock(&vector_lock);
1082 tomove = irq_matrix_allocated(vector_matrix);
1083 avl = irq_matrix_available(vector_matrix, true);
1085 pr_warn("CPU %u has %u vectors, %u available. Cannot disable CPU\n",
1090 rsvd = irq_matrix_reserved(vector_matrix);
1092 pr_warn("Reserved vectors %u > available %u. IRQ request may fail\n",
1096 raw_spin_unlock(&vector_lock);
1099 #endif /* HOTPLUG_CPU */
1102 static void __init print_APIC_field(int base)
1108 for (i = 0; i < 8; i++)
1109 pr_cont("%08x", apic_read(base + i*0x10));
1114 static void __init print_local_APIC(void *dummy)
1116 unsigned int i, v, ver, maxlvt;
1119 pr_debug("printing local APIC contents on CPU#%d/%d:\n",
1120 smp_processor_id(), hard_smp_processor_id());
1121 v = apic_read(APIC_ID);
1122 pr_info("... APIC ID: %08x (%01x)\n", v, read_apic_id());
1123 v = apic_read(APIC_LVR);
1124 pr_info("... APIC VERSION: %08x\n", v);
1125 ver = GET_APIC_VERSION(v);
1126 maxlvt = lapic_get_maxlvt();
1128 v = apic_read(APIC_TASKPRI);
1129 pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1132 if (APIC_INTEGRATED(ver)) {
1133 if (!APIC_XAPIC(ver)) {
1134 v = apic_read(APIC_ARBPRI);
1135 pr_debug("... APIC ARBPRI: %08x (%02x)\n",
1136 v, v & APIC_ARBPRI_MASK);
1138 v = apic_read(APIC_PROCPRI);
1139 pr_debug("... APIC PROCPRI: %08x\n", v);
1143 * Remote read supported only in the 82489DX and local APIC for
1144 * Pentium processors.
1146 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1147 v = apic_read(APIC_RRR);
1148 pr_debug("... APIC RRR: %08x\n", v);
1151 v = apic_read(APIC_LDR);
1152 pr_debug("... APIC LDR: %08x\n", v);
1153 if (!x2apic_enabled()) {
1154 v = apic_read(APIC_DFR);
1155 pr_debug("... APIC DFR: %08x\n", v);
1157 v = apic_read(APIC_SPIV);
1158 pr_debug("... APIC SPIV: %08x\n", v);
1160 pr_debug("... APIC ISR field:\n");
1161 print_APIC_field(APIC_ISR);
1162 pr_debug("... APIC TMR field:\n");
1163 print_APIC_field(APIC_TMR);
1164 pr_debug("... APIC IRR field:\n");
1165 print_APIC_field(APIC_IRR);
1168 if (APIC_INTEGRATED(ver)) {
1169 /* Due to the Pentium erratum 3AP. */
1171 apic_write(APIC_ESR, 0);
1173 v = apic_read(APIC_ESR);
1174 pr_debug("... APIC ESR: %08x\n", v);
1177 icr = apic_icr_read();
1178 pr_debug("... APIC ICR: %08x\n", (u32)icr);
1179 pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
1181 v = apic_read(APIC_LVTT);
1182 pr_debug("... APIC LVTT: %08x\n", v);
1186 v = apic_read(APIC_LVTPC);
1187 pr_debug("... APIC LVTPC: %08x\n", v);
1189 v = apic_read(APIC_LVT0);
1190 pr_debug("... APIC LVT0: %08x\n", v);
1191 v = apic_read(APIC_LVT1);
1192 pr_debug("... APIC LVT1: %08x\n", v);
1196 v = apic_read(APIC_LVTERR);
1197 pr_debug("... APIC LVTERR: %08x\n", v);
1200 v = apic_read(APIC_TMICT);
1201 pr_debug("... APIC TMICT: %08x\n", v);
1202 v = apic_read(APIC_TMCCT);
1203 pr_debug("... APIC TMCCT: %08x\n", v);
1204 v = apic_read(APIC_TDCR);
1205 pr_debug("... APIC TDCR: %08x\n", v);
1207 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1208 v = apic_read(APIC_EFEAT);
1209 maxlvt = (v >> 16) & 0xff;
1210 pr_debug("... APIC EFEAT: %08x\n", v);
1211 v = apic_read(APIC_ECTRL);
1212 pr_debug("... APIC ECTRL: %08x\n", v);
1213 for (i = 0; i < maxlvt; i++) {
1214 v = apic_read(APIC_EILVTn(i));
1215 pr_debug("... APIC EILVT%d: %08x\n", i, v);
1221 static void __init print_local_APICs(int maxcpu)
1229 for_each_online_cpu(cpu) {
1232 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1237 static void __init print_PIC(void)
1240 unsigned long flags;
1242 if (!nr_legacy_irqs())
1245 pr_debug("\nprinting PIC contents\n");
1247 raw_spin_lock_irqsave(&i8259A_lock, flags);
1249 v = inb(0xa1) << 8 | inb(0x21);
1250 pr_debug("... PIC IMR: %04x\n", v);
1252 v = inb(0xa0) << 8 | inb(0x20);
1253 pr_debug("... PIC IRR: %04x\n", v);
1257 v = inb(0xa0) << 8 | inb(0x20);
1261 raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1263 pr_debug("... PIC ISR: %04x\n", v);
1265 v = inb(0x4d1) << 8 | inb(0x4d0);
1266 pr_debug("... PIC ELCR: %04x\n", v);
1269 static int show_lapic __initdata = 1;
1270 static __init int setup_show_lapic(char *arg)
1274 if (strcmp(arg, "all") == 0) {
1275 show_lapic = CONFIG_NR_CPUS;
1277 get_option(&arg, &num);
1284 __setup("show_lapic=", setup_show_lapic);
1286 static int __init print_ICs(void)
1288 if (apic_verbosity == APIC_QUIET)
1293 /* don't print out if apic is not there */
1294 if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1297 print_local_APICs(show_lapic);
1303 late_initcall(print_ICs);