Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Local APIC related interfaces to support IOAPIC, MSI, etc.
   4 *
   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
   9 */
  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>
  19#include <asm/apic.h>
  20#include <asm/i8259.h>
  21#include <asm/desc.h>
  22#include <asm/irq_remapping.h>
  23
  24#include <asm/trace/irq_vectors.h>
  25
  26struct apic_chip_data {
  27	struct irq_cfg		hw_irq_cfg;
  28	unsigned int		vector;
  29	unsigned int		prev_vector;
  30	unsigned int		cpu;
  31	unsigned int		prev_cpu;
  32	unsigned int		irq;
  33	struct hlist_node	clist;
  34	unsigned int		move_in_progress	: 1,
  35				is_managed		: 1,
  36				can_reserve		: 1,
  37				has_reserved		: 1;
  38};
  39
  40struct irq_domain *x86_vector_domain;
  41EXPORT_SYMBOL_GPL(x86_vector_domain);
  42static DEFINE_RAW_SPINLOCK(vector_lock);
  43static cpumask_var_t vector_searchmask;
  44static struct irq_chip lapic_controller;
  45static struct irq_matrix *vector_matrix;
  46#ifdef CONFIG_SMP
  47static DEFINE_PER_CPU(struct hlist_head, cleanup_list);
 
 
 
 
 
 
 
 
 
 
 
  48#endif
  49
  50void lock_vector_lock(void)
  51{
  52	/* Used to the online set of cpus does not change
  53	 * during assign_irq_vector.
  54	 */
  55	raw_spin_lock(&vector_lock);
  56}
  57
  58void unlock_vector_lock(void)
  59{
  60	raw_spin_unlock(&vector_lock);
  61}
  62
  63void init_irq_alloc_info(struct irq_alloc_info *info,
  64			 const struct cpumask *mask)
  65{
  66	memset(info, 0, sizeof(*info));
  67	info->mask = mask;
  68}
  69
  70void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
  71{
  72	if (src)
  73		*dst = *src;
  74	else
  75		memset(dst, 0, sizeof(*dst));
  76}
  77
  78static struct apic_chip_data *apic_chip_data(struct irq_data *irqd)
  79{
  80	if (!irqd)
  81		return NULL;
  82
  83	while (irqd->parent_data)
  84		irqd = irqd->parent_data;
  85
  86	return irqd->chip_data;
  87}
  88
  89struct irq_cfg *irqd_cfg(struct irq_data *irqd)
  90{
  91	struct apic_chip_data *apicd = apic_chip_data(irqd);
  92
  93	return apicd ? &apicd->hw_irq_cfg : NULL;
  94}
  95EXPORT_SYMBOL_GPL(irqd_cfg);
  96
  97struct irq_cfg *irq_cfg(unsigned int irq)
  98{
  99	return irqd_cfg(irq_get_irq_data(irq));
 100}
 101
 102static struct apic_chip_data *alloc_apic_chip_data(int node)
 103{
 104	struct apic_chip_data *apicd;
 105
 106	apicd = kzalloc_node(sizeof(*apicd), GFP_KERNEL, node);
 107	if (apicd)
 108		INIT_HLIST_NODE(&apicd->clist);
 109	return apicd;
 110}
 111
 112static void free_apic_chip_data(struct apic_chip_data *apicd)
 113{
 114	kfree(apicd);
 115}
 116
 117static void apic_update_irq_cfg(struct irq_data *irqd, unsigned int vector,
 118				unsigned int cpu)
 119{
 120	struct apic_chip_data *apicd = apic_chip_data(irqd);
 121
 122	lockdep_assert_held(&vector_lock);
 123
 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);
 129}
 130
 131static void apic_update_vector(struct irq_data *irqd, unsigned int newvec,
 132			       unsigned int newcpu)
 133{
 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);
 137
 138	lockdep_assert_held(&vector_lock);
 139
 140	trace_vector_update(irqd->irq, newvec, newcpu, apicd->vector,
 141			    apicd->cpu);
 142
 143	/*
 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.
 148	 */
 149	apicd->prev_vector = 0;
 150	if (!apicd->vector || apicd->vector == MANAGED_IRQ_SHUTDOWN_VECTOR)
 151		goto setnew;
 152	/*
 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.
 159	 */
 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	} else {
 165		irq_matrix_free(vector_matrix, apicd->cpu, apicd->vector,
 166				managed);
 167	}
 168
 169setnew:
 170	apicd->vector = newvec;
 171	apicd->cpu = newcpu;
 172	BUG_ON(!IS_ERR_OR_NULL(per_cpu(vector_irq, newcpu)[newvec]));
 173	per_cpu(vector_irq, newcpu)[newvec] = desc;
 174}
 175
 176static void vector_assign_managed_shutdown(struct irq_data *irqd)
 177{
 178	unsigned int cpu = cpumask_first(cpu_online_mask);
 179
 180	apic_update_irq_cfg(irqd, MANAGED_IRQ_SHUTDOWN_VECTOR, cpu);
 181}
 182
 183static int reserve_managed_vector(struct irq_data *irqd)
 184{
 185	const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
 186	struct apic_chip_data *apicd = apic_chip_data(irqd);
 187	unsigned long flags;
 188	int ret;
 189
 190	raw_spin_lock_irqsave(&vector_lock, flags);
 191	apicd->is_managed = true;
 192	ret = irq_matrix_reserve_managed(vector_matrix, affmsk);
 193	raw_spin_unlock_irqrestore(&vector_lock, flags);
 194	trace_vector_reserve_managed(irqd->irq, ret);
 195	return ret;
 196}
 197
 198static void reserve_irq_vector_locked(struct irq_data *irqd)
 199{
 200	struct apic_chip_data *apicd = apic_chip_data(irqd);
 201
 202	irq_matrix_reserve(vector_matrix);
 203	apicd->can_reserve = true;
 204	apicd->has_reserved = true;
 205	irqd_set_can_reserve(irqd);
 206	trace_vector_reserve(irqd->irq, 0);
 207	vector_assign_managed_shutdown(irqd);
 208}
 209
 210static int reserve_irq_vector(struct irq_data *irqd)
 211{
 212	unsigned long flags;
 213
 214	raw_spin_lock_irqsave(&vector_lock, flags);
 215	reserve_irq_vector_locked(irqd);
 216	raw_spin_unlock_irqrestore(&vector_lock, flags);
 217	return 0;
 218}
 219
 220static int
 221assign_vector_locked(struct irq_data *irqd, const struct cpumask *dest)
 222{
 223	struct apic_chip_data *apicd = apic_chip_data(irqd);
 224	bool resvd = apicd->has_reserved;
 225	unsigned int cpu = apicd->cpu;
 226	int vector = apicd->vector;
 227
 228	lockdep_assert_held(&vector_lock);
 229
 230	/*
 231	 * If the current target CPU is online and in the new requested
 232	 * affinity mask, there is no point in moving the interrupt from
 233	 * one CPU to another.
 234	 */
 235	if (vector && cpu_online(cpu) && cpumask_test_cpu(cpu, dest))
 236		return 0;
 237
 238	/*
 239	 * Careful here. @apicd might either have move_in_progress set or
 240	 * be enqueued for cleanup. Assigning a new vector would either
 241	 * leave a stale vector on some CPU around or in case of a pending
 242	 * cleanup corrupt the hlist.
 243	 */
 244	if (apicd->move_in_progress || !hlist_unhashed(&apicd->clist))
 245		return -EBUSY;
 246
 247	vector = irq_matrix_alloc(vector_matrix, dest, resvd, &cpu);
 248	trace_vector_alloc(irqd->irq, vector, resvd, vector);
 249	if (vector < 0)
 250		return vector;
 251	apic_update_vector(irqd, vector, cpu);
 252	apic_update_irq_cfg(irqd, vector, cpu);
 253
 254	return 0;
 255}
 256
 257static int assign_irq_vector(struct irq_data *irqd, const struct cpumask *dest)
 258{
 259	unsigned long flags;
 260	int ret;
 261
 262	raw_spin_lock_irqsave(&vector_lock, flags);
 263	cpumask_and(vector_searchmask, dest, cpu_online_mask);
 264	ret = assign_vector_locked(irqd, vector_searchmask);
 265	raw_spin_unlock_irqrestore(&vector_lock, flags);
 266	return ret;
 267}
 268
 269static int assign_irq_vector_any_locked(struct irq_data *irqd)
 270{
 271	/* Get the affinity mask - either irq_default_affinity or (user) set */
 272	const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
 273	int node = irq_data_get_node(irqd);
 274
 275	if (node == NUMA_NO_NODE)
 276		goto all;
 277	/* Try the intersection of @affmsk and node mask */
 278	cpumask_and(vector_searchmask, cpumask_of_node(node), affmsk);
 279	if (!assign_vector_locked(irqd, vector_searchmask))
 280		return 0;
 281	/* Try the node mask */
 282	if (!assign_vector_locked(irqd, cpumask_of_node(node)))
 283		return 0;
 284all:
 285	/* Try the full affinity mask */
 286	cpumask_and(vector_searchmask, affmsk, cpu_online_mask);
 287	if (!assign_vector_locked(irqd, vector_searchmask))
 288		return 0;
 
 
 
 
 
 
 
 289	/* Try the full online mask */
 290	return assign_vector_locked(irqd, cpu_online_mask);
 291}
 292
 293static int
 294assign_irq_vector_policy(struct irq_data *irqd, struct irq_alloc_info *info)
 295{
 296	if (irqd_affinity_is_managed(irqd))
 297		return reserve_managed_vector(irqd);
 298	if (info->mask)
 299		return assign_irq_vector(irqd, info->mask);
 300	/*
 301	 * Make only a global reservation with no guarantee. A real vector
 302	 * is associated at activation time.
 303	 */
 304	return reserve_irq_vector(irqd);
 305}
 306
 307static int
 308assign_managed_vector(struct irq_data *irqd, const struct cpumask *dest)
 309{
 310	const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
 311	struct apic_chip_data *apicd = apic_chip_data(irqd);
 312	int vector, cpu;
 313
 314	cpumask_and(vector_searchmask, dest, affmsk);
 315
 316	/* set_affinity might call here for nothing */
 317	if (apicd->vector && cpumask_test_cpu(apicd->cpu, vector_searchmask))
 318		return 0;
 319	vector = irq_matrix_alloc_managed(vector_matrix, vector_searchmask,
 320					  &cpu);
 321	trace_vector_alloc_managed(irqd->irq, vector, vector);
 322	if (vector < 0)
 323		return vector;
 324	apic_update_vector(irqd, vector, cpu);
 325	apic_update_irq_cfg(irqd, vector, cpu);
 326	return 0;
 327}
 328
 329static void clear_irq_vector(struct irq_data *irqd)
 330{
 331	struct apic_chip_data *apicd = apic_chip_data(irqd);
 332	bool managed = irqd_affinity_is_managed(irqd);
 333	unsigned int vector = apicd->vector;
 334
 335	lockdep_assert_held(&vector_lock);
 336
 337	if (!vector)
 338		return;
 339
 340	trace_vector_clear(irqd->irq, vector, apicd->cpu, apicd->prev_vector,
 341			   apicd->prev_cpu);
 342
 343	per_cpu(vector_irq, apicd->cpu)[vector] = VECTOR_SHUTDOWN;
 344	irq_matrix_free(vector_matrix, apicd->cpu, vector, managed);
 345	apicd->vector = 0;
 346
 347	/* Clean up move in progress */
 348	vector = apicd->prev_vector;
 349	if (!vector)
 350		return;
 351
 352	per_cpu(vector_irq, apicd->prev_cpu)[vector] = VECTOR_SHUTDOWN;
 353	irq_matrix_free(vector_matrix, apicd->prev_cpu, vector, managed);
 354	apicd->prev_vector = 0;
 355	apicd->move_in_progress = 0;
 356	hlist_del_init(&apicd->clist);
 357}
 358
 359static void x86_vector_deactivate(struct irq_domain *dom, struct irq_data *irqd)
 360{
 361	struct apic_chip_data *apicd = apic_chip_data(irqd);
 362	unsigned long flags;
 363
 364	trace_vector_deactivate(irqd->irq, apicd->is_managed,
 365				apicd->can_reserve, false);
 366
 367	/* Regular fixed assigned interrupt */
 368	if (!apicd->is_managed && !apicd->can_reserve)
 369		return;
 370	/* If the interrupt has a global reservation, nothing to do */
 371	if (apicd->has_reserved)
 372		return;
 373
 374	raw_spin_lock_irqsave(&vector_lock, flags);
 375	clear_irq_vector(irqd);
 376	if (apicd->can_reserve)
 377		reserve_irq_vector_locked(irqd);
 378	else
 379		vector_assign_managed_shutdown(irqd);
 380	raw_spin_unlock_irqrestore(&vector_lock, flags);
 381}
 382
 383static int activate_reserved(struct irq_data *irqd)
 384{
 385	struct apic_chip_data *apicd = apic_chip_data(irqd);
 386	int ret;
 387
 388	ret = assign_irq_vector_any_locked(irqd);
 389	if (!ret) {
 390		apicd->has_reserved = false;
 391		/*
 392		 * Core might have disabled reservation mode after
 393		 * allocating the irq descriptor. Ideally this should
 394		 * happen before allocation time, but that would require
 395		 * completely convoluted ways of transporting that
 396		 * information.
 397		 */
 398		if (!irqd_can_reserve(irqd))
 399			apicd->can_reserve = false;
 400	}
 401
 402	/*
 403	 * Check to ensure that the effective affinity mask is a subset
 404	 * the user supplied affinity mask, and warn the user if it is not
 405	 */
 406	if (!cpumask_subset(irq_data_get_effective_affinity_mask(irqd),
 407			    irq_data_get_affinity_mask(irqd))) {
 408		pr_warn("irq %u: Affinity broken due to vector space exhaustion.\n",
 409			irqd->irq);
 410	}
 411
 412	return ret;
 413}
 414
 415static int activate_managed(struct irq_data *irqd)
 416{
 417	const struct cpumask *dest = irq_data_get_affinity_mask(irqd);
 418	int ret;
 419
 420	cpumask_and(vector_searchmask, dest, cpu_online_mask);
 421	if (WARN_ON_ONCE(cpumask_empty(vector_searchmask))) {
 422		/* Something in the core code broke! Survive gracefully */
 423		pr_err("Managed startup for irq %u, but no CPU\n", irqd->irq);
 424		return -EINVAL;
 425	}
 426
 427	ret = assign_managed_vector(irqd, vector_searchmask);
 428	/*
 429	 * This should not happen. The vector reservation got buggered.  Handle
 430	 * it gracefully.
 431	 */
 432	if (WARN_ON_ONCE(ret < 0)) {
 433		pr_err("Managed startup irq %u, no vector available\n",
 434		       irqd->irq);
 435	}
 436	return ret;
 437}
 438
 439static int x86_vector_activate(struct irq_domain *dom, struct irq_data *irqd,
 440			       bool reserve)
 441{
 442	struct apic_chip_data *apicd = apic_chip_data(irqd);
 443	unsigned long flags;
 444	int ret = 0;
 445
 446	trace_vector_activate(irqd->irq, apicd->is_managed,
 447			      apicd->can_reserve, reserve);
 448
 449	/* Nothing to do for fixed assigned vectors */
 450	if (!apicd->can_reserve && !apicd->is_managed)
 451		return 0;
 452
 453	raw_spin_lock_irqsave(&vector_lock, flags);
 454	if (reserve || irqd_is_managed_and_shutdown(irqd))
 
 
 455		vector_assign_managed_shutdown(irqd);
 456	else if (apicd->is_managed)
 457		ret = activate_managed(irqd);
 458	else if (apicd->has_reserved)
 459		ret = activate_reserved(irqd);
 460	raw_spin_unlock_irqrestore(&vector_lock, flags);
 461	return ret;
 462}
 463
 464static void vector_free_reserved_and_managed(struct irq_data *irqd)
 465{
 466	const struct cpumask *dest = irq_data_get_affinity_mask(irqd);
 467	struct apic_chip_data *apicd = apic_chip_data(irqd);
 468
 469	trace_vector_teardown(irqd->irq, apicd->is_managed,
 470			      apicd->has_reserved);
 471
 472	if (apicd->has_reserved)
 473		irq_matrix_remove_reserved(vector_matrix);
 474	if (apicd->is_managed)
 475		irq_matrix_remove_managed(vector_matrix, dest);
 476}
 477
 478static void x86_vector_free_irqs(struct irq_domain *domain,
 479				 unsigned int virq, unsigned int nr_irqs)
 480{
 481	struct apic_chip_data *apicd;
 482	struct irq_data *irqd;
 483	unsigned long flags;
 484	int i;
 485
 486	for (i = 0; i < nr_irqs; i++) {
 487		irqd = irq_domain_get_irq_data(x86_vector_domain, virq + i);
 488		if (irqd && irqd->chip_data) {
 489			raw_spin_lock_irqsave(&vector_lock, flags);
 490			clear_irq_vector(irqd);
 491			vector_free_reserved_and_managed(irqd);
 492			apicd = irqd->chip_data;
 493			irq_domain_reset_irq_data(irqd);
 494			raw_spin_unlock_irqrestore(&vector_lock, flags);
 495			free_apic_chip_data(apicd);
 496		}
 497	}
 498}
 499
 500static bool vector_configure_legacy(unsigned int virq, struct irq_data *irqd,
 501				    struct apic_chip_data *apicd)
 502{
 503	unsigned long flags;
 504	bool realloc = false;
 505
 506	apicd->vector = ISA_IRQ_VECTOR(virq);
 507	apicd->cpu = 0;
 508
 509	raw_spin_lock_irqsave(&vector_lock, flags);
 510	/*
 511	 * If the interrupt is activated, then it must stay at this vector
 512	 * position. That's usually the timer interrupt (0).
 513	 */
 514	if (irqd_is_activated(irqd)) {
 515		trace_vector_setup(virq, true, 0);
 516		apic_update_irq_cfg(irqd, apicd->vector, apicd->cpu);
 517	} else {
 518		/* Release the vector */
 519		apicd->can_reserve = true;
 520		irqd_set_can_reserve(irqd);
 521		clear_irq_vector(irqd);
 522		realloc = true;
 523	}
 524	raw_spin_unlock_irqrestore(&vector_lock, flags);
 525	return realloc;
 526}
 527
 528static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
 529				 unsigned int nr_irqs, void *arg)
 530{
 531	struct irq_alloc_info *info = arg;
 532	struct apic_chip_data *apicd;
 533	struct irq_data *irqd;
 534	int i, err, node;
 535
 536	if (disable_apic)
 537		return -ENXIO;
 538
 539	/* Currently vector allocator can't guarantee contiguous allocations */
 540	if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
 541		return -ENOSYS;
 
 
 
 
 542
 543	for (i = 0; i < nr_irqs; i++) {
 544		irqd = irq_domain_get_irq_data(domain, virq + i);
 545		BUG_ON(!irqd);
 546		node = irq_data_get_node(irqd);
 547		WARN_ON_ONCE(irqd->chip_data);
 548		apicd = alloc_apic_chip_data(node);
 549		if (!apicd) {
 550			err = -ENOMEM;
 551			goto error;
 552		}
 553
 554		apicd->irq = virq + i;
 555		irqd->chip = &lapic_controller;
 556		irqd->chip_data = apicd;
 557		irqd->hwirq = virq + i;
 558		irqd_set_single_target(irqd);
 559		/*
 
 
 
 
 
 
 
 
 
 
 560		 * Legacy vectors are already assigned when the IOAPIC
 561		 * takes them over. They stay on the same vector. This is
 562		 * required for check_timer() to work correctly as it might
 563		 * switch back to legacy mode. Only update the hardware
 564		 * config.
 565		 */
 566		if (info->flags & X86_IRQ_ALLOC_LEGACY) {
 567			if (!vector_configure_legacy(virq + i, irqd, apicd))
 568				continue;
 569		}
 570
 571		err = assign_irq_vector_policy(irqd, info);
 572		trace_vector_setup(virq + i, false, err);
 573		if (err) {
 574			irqd->chip_data = NULL;
 575			free_apic_chip_data(apicd);
 576			goto error;
 577		}
 578	}
 579
 580	return 0;
 581
 582error:
 583	x86_vector_free_irqs(domain, virq, i);
 584	return err;
 585}
 586
 587#ifdef CONFIG_GENERIC_IRQ_DEBUGFS
 588static void x86_vector_debug_show(struct seq_file *m, struct irq_domain *d,
 589				  struct irq_data *irqd, int ind)
 590{
 591	struct apic_chip_data apicd;
 592	unsigned long flags;
 593	int irq;
 594
 595	if (!irqd) {
 596		irq_matrix_debug_show(m, vector_matrix, ind);
 597		return;
 598	}
 599
 600	irq = irqd->irq;
 601	if (irq < nr_legacy_irqs() && !test_bit(irq, &io_apic_irqs)) {
 602		seq_printf(m, "%*sVector: %5d\n", ind, "", ISA_IRQ_VECTOR(irq));
 603		seq_printf(m, "%*sTarget: Legacy PIC all CPUs\n", ind, "");
 604		return;
 605	}
 606
 607	if (!irqd->chip_data) {
 608		seq_printf(m, "%*sVector: Not assigned\n", ind, "");
 609		return;
 610	}
 611
 612	raw_spin_lock_irqsave(&vector_lock, flags);
 613	memcpy(&apicd, irqd->chip_data, sizeof(apicd));
 614	raw_spin_unlock_irqrestore(&vector_lock, flags);
 615
 616	seq_printf(m, "%*sVector: %5u\n", ind, "", apicd.vector);
 617	seq_printf(m, "%*sTarget: %5u\n", ind, "", apicd.cpu);
 618	if (apicd.prev_vector) {
 619		seq_printf(m, "%*sPrevious vector: %5u\n", ind, "", apicd.prev_vector);
 620		seq_printf(m, "%*sPrevious target: %5u\n", ind, "", apicd.prev_cpu);
 621	}
 622	seq_printf(m, "%*smove_in_progress: %u\n", ind, "", apicd.move_in_progress ? 1 : 0);
 623	seq_printf(m, "%*sis_managed:       %u\n", ind, "", apicd.is_managed ? 1 : 0);
 624	seq_printf(m, "%*scan_reserve:      %u\n", ind, "", apicd.can_reserve ? 1 : 0);
 625	seq_printf(m, "%*shas_reserved:     %u\n", ind, "", apicd.has_reserved ? 1 : 0);
 626	seq_printf(m, "%*scleanup_pending:  %u\n", ind, "", !hlist_unhashed(&apicd.clist));
 627}
 628#endif
 629
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 630static const struct irq_domain_ops x86_vector_domain_ops = {
 
 631	.alloc		= x86_vector_alloc_irqs,
 632	.free		= x86_vector_free_irqs,
 633	.activate	= x86_vector_activate,
 634	.deactivate	= x86_vector_deactivate,
 635#ifdef CONFIG_GENERIC_IRQ_DEBUGFS
 636	.debug_show	= x86_vector_debug_show,
 637#endif
 638};
 639
 640int __init arch_probe_nr_irqs(void)
 641{
 642	int nr;
 643
 644	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
 645		nr_irqs = NR_VECTORS * nr_cpu_ids;
 646
 647	nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
 648#if defined(CONFIG_PCI_MSI)
 649	/*
 650	 * for MSI and HT dyn irq
 651	 */
 652	if (gsi_top <= NR_IRQS_LEGACY)
 653		nr +=  8 * nr_cpu_ids;
 654	else
 655		nr += gsi_top * 16;
 656#endif
 657	if (nr < nr_irqs)
 658		nr_irqs = nr;
 659
 660	/*
 661	 * We don't know if PIC is present at this point so we need to do
 662	 * probe() to get the right number of legacy IRQs.
 663	 */
 664	return legacy_pic->probe();
 665}
 666
 667void lapic_assign_legacy_vector(unsigned int irq, bool replace)
 668{
 669	/*
 670	 * Use assign system here so it wont get accounted as allocated
 671	 * and moveable in the cpu hotplug check and it prevents managed
 672	 * irq reservation from touching it.
 673	 */
 674	irq_matrix_assign_system(vector_matrix, ISA_IRQ_VECTOR(irq), replace);
 675}
 676
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 677void __init lapic_assign_system_vectors(void)
 678{
 679	unsigned int i, vector = 0;
 680
 681	for_each_set_bit_from(vector, system_vectors, NR_VECTORS)
 682		irq_matrix_assign_system(vector_matrix, vector, false);
 683
 684	if (nr_legacy_irqs() > 1)
 685		lapic_assign_legacy_vector(PIC_CASCADE_IR, false);
 686
 687	/* System vectors are reserved, online it */
 688	irq_matrix_online(vector_matrix);
 689
 690	/* Mark the preallocated legacy interrupts */
 691	for (i = 0; i < nr_legacy_irqs(); i++) {
 
 
 
 
 
 692		if (i != PIC_CASCADE_IR)
 693			irq_matrix_assign(vector_matrix, ISA_IRQ_VECTOR(i));
 694	}
 695}
 696
 697int __init arch_early_irq_init(void)
 698{
 699	struct fwnode_handle *fn;
 700
 701	fn = irq_domain_alloc_named_fwnode("VECTOR");
 702	BUG_ON(!fn);
 703	x86_vector_domain = irq_domain_create_tree(fn, &x86_vector_domain_ops,
 704						   NULL);
 705	BUG_ON(x86_vector_domain == NULL);
 706	irq_domain_free_fwnode(fn);
 707	irq_set_default_host(x86_vector_domain);
 708
 709	arch_init_msi_domain(x86_vector_domain);
 710
 711	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
 712
 713	/*
 714	 * Allocate the vector matrix allocator data structure and limit the
 715	 * search area.
 716	 */
 717	vector_matrix = irq_alloc_matrix(NR_VECTORS, FIRST_EXTERNAL_VECTOR,
 718					 FIRST_SYSTEM_VECTOR);
 719	BUG_ON(!vector_matrix);
 720
 721	return arch_early_ioapic_init();
 722}
 723
 724#ifdef CONFIG_SMP
 725
 726static struct irq_desc *__setup_vector_irq(int vector)
 727{
 728	int isairq = vector - ISA_IRQ_VECTOR(0);
 729
 730	/* Check whether the irq is in the legacy space */
 731	if (isairq < 0 || isairq >= nr_legacy_irqs())
 732		return VECTOR_UNUSED;
 733	/* Check whether the irq is handled by the IOAPIC */
 734	if (test_bit(isairq, &io_apic_irqs))
 735		return VECTOR_UNUSED;
 736	return irq_to_desc(isairq);
 737}
 738
 739/* Online the local APIC infrastructure and initialize the vectors */
 740void lapic_online(void)
 741{
 742	unsigned int vector;
 743
 744	lockdep_assert_held(&vector_lock);
 745
 746	/* Online the vector matrix array for this CPU */
 747	irq_matrix_online(vector_matrix);
 748
 749	/*
 750	 * The interrupt affinity logic never targets interrupts to offline
 751	 * CPUs. The exception are the legacy PIC interrupts. In general
 752	 * they are only targeted to CPU0, but depending on the platform
 753	 * they can be distributed to any online CPU in hardware. The
 754	 * kernel has no influence on that. So all active legacy vectors
 755	 * must be installed on all CPUs. All non legacy interrupts can be
 756	 * cleared.
 757	 */
 758	for (vector = 0; vector < NR_VECTORS; vector++)
 759		this_cpu_write(vector_irq[vector], __setup_vector_irq(vector));
 760}
 761
 
 
 762void lapic_offline(void)
 763{
 
 
 764	lock_vector_lock();
 
 
 
 
 765	irq_matrix_offline(vector_matrix);
 
 
 
 766	unlock_vector_lock();
 767}
 768
 769static int apic_set_affinity(struct irq_data *irqd,
 770			     const struct cpumask *dest, bool force)
 771{
 772	struct apic_chip_data *apicd = apic_chip_data(irqd);
 773	int err;
 774
 775	/*
 776	 * Core code can call here for inactive interrupts. For inactive
 777	 * interrupts which use managed or reservation mode there is no
 778	 * point in going through the vector assignment right now as the
 779	 * activation will assign a vector which fits the destination
 780	 * cpumask. Let the core code store the destination mask and be
 781	 * done with it.
 782	 */
 783	if (!irqd_is_activated(irqd) &&
 784	    (apicd->is_managed || apicd->can_reserve))
 785		return IRQ_SET_MASK_OK;
 786
 787	raw_spin_lock(&vector_lock);
 788	cpumask_and(vector_searchmask, dest, cpu_online_mask);
 789	if (irqd_affinity_is_managed(irqd))
 790		err = assign_managed_vector(irqd, vector_searchmask);
 791	else
 792		err = assign_vector_locked(irqd, vector_searchmask);
 793	raw_spin_unlock(&vector_lock);
 794	return err ? err : IRQ_SET_MASK_OK;
 795}
 796
 797#else
 798# define apic_set_affinity	NULL
 799#endif
 800
 801static int apic_retrigger_irq(struct irq_data *irqd)
 802{
 803	struct apic_chip_data *apicd = apic_chip_data(irqd);
 804	unsigned long flags;
 805
 806	raw_spin_lock_irqsave(&vector_lock, flags);
 807	apic->send_IPI(apicd->cpu, apicd->vector);
 808	raw_spin_unlock_irqrestore(&vector_lock, flags);
 809
 810	return 1;
 811}
 812
 813void apic_ack_irq(struct irq_data *irqd)
 814{
 815	irq_move_irq(irqd);
 816	ack_APIC_irq();
 817}
 818
 819void apic_ack_edge(struct irq_data *irqd)
 820{
 821	irq_complete_move(irqd_cfg(irqd));
 822	apic_ack_irq(irqd);
 823}
 824
 
 
 
 
 
 
 825static struct irq_chip lapic_controller = {
 826	.name			= "APIC",
 827	.irq_ack		= apic_ack_edge,
 828	.irq_set_affinity	= apic_set_affinity,
 
 829	.irq_retrigger		= apic_retrigger_irq,
 830};
 831
 832#ifdef CONFIG_SMP
 833
 834static void free_moved_vector(struct apic_chip_data *apicd)
 835{
 836	unsigned int vector = apicd->prev_vector;
 837	unsigned int cpu = apicd->prev_cpu;
 838	bool managed = apicd->is_managed;
 839
 840	/*
 841	 * This should never happen. Managed interrupts are not
 842	 * migrated except on CPU down, which does not involve the
 843	 * cleanup vector. But try to keep the accounting correct
 844	 * nevertheless.
 
 
 
 
 845	 */
 846	WARN_ON_ONCE(managed);
 847
 848	trace_vector_free_moved(apicd->irq, cpu, vector, managed);
 849	irq_matrix_free(vector_matrix, cpu, vector, managed);
 850	per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
 851	hlist_del_init(&apicd->clist);
 852	apicd->prev_vector = 0;
 853	apicd->move_in_progress = 0;
 854}
 855
 856asmlinkage __visible void __irq_entry smp_irq_move_cleanup_interrupt(void)
 857{
 858	struct hlist_head *clhead = this_cpu_ptr(&cleanup_list);
 859	struct apic_chip_data *apicd;
 860	struct hlist_node *tmp;
 
 861
 862	entering_ack_irq();
 863	/* Prevent vectors vanishing under us */
 864	raw_spin_lock(&vector_lock);
 865
 866	hlist_for_each_entry_safe(apicd, tmp, clhead, clist) {
 867		unsigned int irr, vector = apicd->prev_vector;
 868
 869		/*
 870		 * Paranoia: Check if the vector that needs to be cleaned
 871		 * up is registered at the APICs IRR. If so, then this is
 872		 * not the best time to clean it up. Clean it up in the
 873		 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
 874		 * to this CPU. IRQ_MOVE_CLEANUP_VECTOR is the lowest
 875		 * priority external vector, so on return from this
 876		 * interrupt the device interrupt will happen first.
 
 
 
 877		 */
 878		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
 879		if (irr & (1U << (vector % 32))) {
 880			apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
 
 881			continue;
 882		}
 883		free_moved_vector(apicd);
 884	}
 885
 886	raw_spin_unlock(&vector_lock);
 887	exiting_irq();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 888}
 889
 890static void __send_cleanup_vector(struct apic_chip_data *apicd)
 891{
 892	unsigned int cpu;
 893
 894	raw_spin_lock(&vector_lock);
 895	apicd->move_in_progress = 0;
 896	cpu = apicd->prev_cpu;
 897	if (cpu_online(cpu)) {
 898		hlist_add_head(&apicd->clist, per_cpu_ptr(&cleanup_list, cpu));
 899		apic->send_IPI(cpu, IRQ_MOVE_CLEANUP_VECTOR);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 900	} else {
 901		apicd->prev_vector = 0;
 
 902	}
 903	raw_spin_unlock(&vector_lock);
 904}
 905
 906void send_cleanup_vector(struct irq_cfg *cfg)
 907{
 908	struct apic_chip_data *apicd;
 909
 910	apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg);
 911	if (apicd->move_in_progress)
 912		__send_cleanup_vector(apicd);
 913}
 914
 915static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
 916{
 917	struct apic_chip_data *apicd;
 918
 919	apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg);
 920	if (likely(!apicd->move_in_progress))
 921		return;
 922
 923	if (vector == apicd->vector && apicd->cpu == smp_processor_id())
 924		__send_cleanup_vector(apicd);
 925}
 926
 927void irq_complete_move(struct irq_cfg *cfg)
 928{
 929	__irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
 
 930}
 931
 932/*
 933 * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
 934 */
 935void irq_force_complete_move(struct irq_desc *desc)
 936{
 
 937	struct apic_chip_data *apicd;
 938	struct irq_data *irqd;
 939	unsigned int vector;
 940
 941	/*
 942	 * The function is called for all descriptors regardless of which
 943	 * irqdomain they belong to. For example if an IRQ is provided by
 944	 * an irq_chip as part of a GPIO driver, the chip data for that
 945	 * descriptor is specific to the irq_chip in question.
 946	 *
 947	 * Check first that the chip_data is what we expect
 948	 * (apic_chip_data) before touching it any further.
 949	 */
 950	irqd = irq_domain_get_irq_data(x86_vector_domain,
 951				       irq_desc_get_irq(desc));
 952	if (!irqd)
 953		return;
 954
 955	raw_spin_lock(&vector_lock);
 956	apicd = apic_chip_data(irqd);
 957	if (!apicd)
 958		goto unlock;
 959
 960	/*
 961	 * If prev_vector is empty, no action required.
 
 962	 */
 963	vector = apicd->prev_vector;
 964	if (!vector)
 965		goto unlock;
 966
 967	/*
 968	 * This is tricky. If the cleanup of the old vector has not been
 969	 * done yet, then the following setaffinity call will fail with
 970	 * -EBUSY. This can leave the interrupt in a stale state.
 971	 *
 972	 * All CPUs are stuck in stop machine with interrupts disabled so
 973	 * calling __irq_complete_move() would be completely pointless.
 974	 *
 975	 * 1) The interrupt is in move_in_progress state. That means that we
 976	 *    have not seen an interrupt since the io_apic was reprogrammed to
 977	 *    the new vector.
 978	 *
 979	 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
 980	 *    have not been processed yet.
 981	 */
 982	if (apicd->move_in_progress) {
 983		/*
 984		 * In theory there is a race:
 985		 *
 986		 * set_ioapic(new_vector) <-- Interrupt is raised before update
 987		 *			      is effective, i.e. it's raised on
 988		 *			      the old vector.
 989		 *
 990		 * So if the target cpu cannot handle that interrupt before
 991		 * the old vector is cleaned up, we get a spurious interrupt
 992		 * and in the worst case the ioapic irq line becomes stale.
 993		 *
 994		 * But in case of cpu hotplug this should be a non issue
 995		 * because if the affinity update happens right before all
 996		 * cpus rendevouz in stop machine, there is no way that the
 997		 * interrupt can be blocked on the target cpu because all cpus
 998		 * loops first with interrupts enabled in stop machine, so the
 999		 * old vector is not yet cleaned up when the interrupt fires.
1000		 *
1001		 * So the only way to run into this issue is if the delivery
1002		 * of the interrupt on the apic/system bus would be delayed
1003		 * beyond the point where the target cpu disables interrupts
1004		 * in stop machine. I doubt that it can happen, but at least
1005		 * there is a theroretical chance. Virtualization might be
1006		 * able to expose this, but AFAICT the IOAPIC emulation is not
1007		 * as stupid as the real hardware.
1008		 *
1009		 * Anyway, there is nothing we can do about that at this point
1010		 * w/o refactoring the whole fixup_irq() business completely.
1011		 * We print at least the irq number and the old vector number,
1012		 * so we have the necessary information when a problem in that
1013		 * area arises.
1014		 */
1015		pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
1016			irqd->irq, vector);
1017	}
1018	free_moved_vector(apicd);
1019unlock:
1020	raw_spin_unlock(&vector_lock);
1021}
1022
1023#ifdef CONFIG_HOTPLUG_CPU
1024/*
1025 * Note, this is not accurate accounting, but at least good enough to
1026 * prevent that the actual interrupt move will run out of vectors.
1027 */
1028int lapic_can_unplug_cpu(void)
1029{
1030	unsigned int rsvd, avl, tomove, cpu = smp_processor_id();
1031	int ret = 0;
1032
1033	raw_spin_lock(&vector_lock);
1034	tomove = irq_matrix_allocated(vector_matrix);
1035	avl = irq_matrix_available(vector_matrix, true);
1036	if (avl < tomove) {
1037		pr_warn("CPU %u has %u vectors, %u available. Cannot disable CPU\n",
1038			cpu, tomove, avl);
1039		ret = -ENOSPC;
1040		goto out;
1041	}
1042	rsvd = irq_matrix_reserved(vector_matrix);
1043	if (avl < rsvd) {
1044		pr_warn("Reserved vectors %u > available %u. IRQ request may fail\n",
1045			rsvd, avl);
1046	}
1047out:
1048	raw_spin_unlock(&vector_lock);
1049	return ret;
1050}
1051#endif /* HOTPLUG_CPU */
1052#endif /* SMP */
1053
1054static void __init print_APIC_field(int base)
1055{
1056	int i;
1057
1058	printk(KERN_DEBUG);
1059
1060	for (i = 0; i < 8; i++)
1061		pr_cont("%08x", apic_read(base + i*0x10));
1062
1063	pr_cont("\n");
1064}
1065
1066static void __init print_local_APIC(void *dummy)
1067{
1068	unsigned int i, v, ver, maxlvt;
1069	u64 icr;
1070
1071	pr_debug("printing local APIC contents on CPU#%d/%d:\n",
1072		 smp_processor_id(), hard_smp_processor_id());
1073	v = apic_read(APIC_ID);
1074	pr_info("... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1075	v = apic_read(APIC_LVR);
1076	pr_info("... APIC VERSION: %08x\n", v);
1077	ver = GET_APIC_VERSION(v);
1078	maxlvt = lapic_get_maxlvt();
1079
1080	v = apic_read(APIC_TASKPRI);
1081	pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1082
1083	/* !82489DX */
1084	if (APIC_INTEGRATED(ver)) {
1085		if (!APIC_XAPIC(ver)) {
1086			v = apic_read(APIC_ARBPRI);
1087			pr_debug("... APIC ARBPRI: %08x (%02x)\n",
1088				 v, v & APIC_ARBPRI_MASK);
1089		}
1090		v = apic_read(APIC_PROCPRI);
1091		pr_debug("... APIC PROCPRI: %08x\n", v);
1092	}
1093
1094	/*
1095	 * Remote read supported only in the 82489DX and local APIC for
1096	 * Pentium processors.
1097	 */
1098	if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1099		v = apic_read(APIC_RRR);
1100		pr_debug("... APIC RRR: %08x\n", v);
1101	}
1102
1103	v = apic_read(APIC_LDR);
1104	pr_debug("... APIC LDR: %08x\n", v);
1105	if (!x2apic_enabled()) {
1106		v = apic_read(APIC_DFR);
1107		pr_debug("... APIC DFR: %08x\n", v);
1108	}
1109	v = apic_read(APIC_SPIV);
1110	pr_debug("... APIC SPIV: %08x\n", v);
1111
1112	pr_debug("... APIC ISR field:\n");
1113	print_APIC_field(APIC_ISR);
1114	pr_debug("... APIC TMR field:\n");
1115	print_APIC_field(APIC_TMR);
1116	pr_debug("... APIC IRR field:\n");
1117	print_APIC_field(APIC_IRR);
1118
1119	/* !82489DX */
1120	if (APIC_INTEGRATED(ver)) {
1121		/* Due to the Pentium erratum 3AP. */
1122		if (maxlvt > 3)
1123			apic_write(APIC_ESR, 0);
1124
1125		v = apic_read(APIC_ESR);
1126		pr_debug("... APIC ESR: %08x\n", v);
1127	}
1128
1129	icr = apic_icr_read();
1130	pr_debug("... APIC ICR: %08x\n", (u32)icr);
1131	pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
1132
1133	v = apic_read(APIC_LVTT);
1134	pr_debug("... APIC LVTT: %08x\n", v);
1135
1136	if (maxlvt > 3) {
1137		/* PC is LVT#4. */
1138		v = apic_read(APIC_LVTPC);
1139		pr_debug("... APIC LVTPC: %08x\n", v);
1140	}
1141	v = apic_read(APIC_LVT0);
1142	pr_debug("... APIC LVT0: %08x\n", v);
1143	v = apic_read(APIC_LVT1);
1144	pr_debug("... APIC LVT1: %08x\n", v);
1145
1146	if (maxlvt > 2) {
1147		/* ERR is LVT#3. */
1148		v = apic_read(APIC_LVTERR);
1149		pr_debug("... APIC LVTERR: %08x\n", v);
1150	}
1151
1152	v = apic_read(APIC_TMICT);
1153	pr_debug("... APIC TMICT: %08x\n", v);
1154	v = apic_read(APIC_TMCCT);
1155	pr_debug("... APIC TMCCT: %08x\n", v);
1156	v = apic_read(APIC_TDCR);
1157	pr_debug("... APIC TDCR: %08x\n", v);
1158
1159	if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1160		v = apic_read(APIC_EFEAT);
1161		maxlvt = (v >> 16) & 0xff;
1162		pr_debug("... APIC EFEAT: %08x\n", v);
1163		v = apic_read(APIC_ECTRL);
1164		pr_debug("... APIC ECTRL: %08x\n", v);
1165		for (i = 0; i < maxlvt; i++) {
1166			v = apic_read(APIC_EILVTn(i));
1167			pr_debug("... APIC EILVT%d: %08x\n", i, v);
1168		}
1169	}
1170	pr_cont("\n");
1171}
1172
1173static void __init print_local_APICs(int maxcpu)
1174{
1175	int cpu;
1176
1177	if (!maxcpu)
1178		return;
1179
1180	preempt_disable();
1181	for_each_online_cpu(cpu) {
1182		if (cpu >= maxcpu)
1183			break;
1184		smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1185	}
1186	preempt_enable();
1187}
1188
1189static void __init print_PIC(void)
1190{
1191	unsigned int v;
1192	unsigned long flags;
1193
1194	if (!nr_legacy_irqs())
1195		return;
1196
1197	pr_debug("\nprinting PIC contents\n");
1198
1199	raw_spin_lock_irqsave(&i8259A_lock, flags);
1200
1201	v = inb(0xa1) << 8 | inb(0x21);
1202	pr_debug("... PIC  IMR: %04x\n", v);
1203
1204	v = inb(0xa0) << 8 | inb(0x20);
1205	pr_debug("... PIC  IRR: %04x\n", v);
1206
1207	outb(0x0b, 0xa0);
1208	outb(0x0b, 0x20);
1209	v = inb(0xa0) << 8 | inb(0x20);
1210	outb(0x0a, 0xa0);
1211	outb(0x0a, 0x20);
1212
1213	raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1214
1215	pr_debug("... PIC  ISR: %04x\n", v);
1216
1217	v = inb(0x4d1) << 8 | inb(0x4d0);
1218	pr_debug("... PIC ELCR: %04x\n", v);
1219}
1220
1221static int show_lapic __initdata = 1;
1222static __init int setup_show_lapic(char *arg)
1223{
1224	int num = -1;
1225
1226	if (strcmp(arg, "all") == 0) {
1227		show_lapic = CONFIG_NR_CPUS;
1228	} else {
1229		get_option(&arg, &num);
1230		if (num >= 0)
1231			show_lapic = num;
1232	}
1233
1234	return 1;
1235}
1236__setup("show_lapic=", setup_show_lapic);
1237
1238static int __init print_ICs(void)
1239{
1240	if (apic_verbosity == APIC_QUIET)
1241		return 0;
1242
1243	print_PIC();
1244
1245	/* don't print out if apic is not there */
1246	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1247		return 0;
1248
1249	print_local_APICs(show_lapic);
1250	print_IO_APICs();
1251
1252	return 0;
1253}
1254
1255late_initcall(print_ICs);
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Local APIC related interfaces to support IOAPIC, MSI, etc.
   4 *
   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
   9 */
  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>
  19#include <asm/apic.h>
  20#include <asm/i8259.h>
  21#include <asm/desc.h>
  22#include <asm/irq_remapping.h>
  23
  24#include <asm/trace/irq_vectors.h>
  25
  26struct apic_chip_data {
  27	struct irq_cfg		hw_irq_cfg;
  28	unsigned int		vector;
  29	unsigned int		prev_vector;
  30	unsigned int		cpu;
  31	unsigned int		prev_cpu;
  32	unsigned int		irq;
  33	struct hlist_node	clist;
  34	unsigned int		move_in_progress	: 1,
  35				is_managed		: 1,
  36				can_reserve		: 1,
  37				has_reserved		: 1;
  38};
  39
  40struct irq_domain *x86_vector_domain;
  41EXPORT_SYMBOL_GPL(x86_vector_domain);
  42static DEFINE_RAW_SPINLOCK(vector_lock);
  43static cpumask_var_t vector_searchmask;
  44static struct irq_chip lapic_controller;
  45static struct irq_matrix *vector_matrix;
  46#ifdef CONFIG_SMP
  47
  48static void vector_cleanup_callback(struct timer_list *tmr);
  49
  50struct vector_cleanup {
  51	struct hlist_head	head;
  52	struct timer_list	timer;
  53};
  54
  55static DEFINE_PER_CPU(struct vector_cleanup, vector_cleanup) = {
  56	.head	= HLIST_HEAD_INIT,
  57	.timer	= __TIMER_INITIALIZER(vector_cleanup_callback, TIMER_PINNED),
  58};
  59#endif
  60
  61void lock_vector_lock(void)
  62{
  63	/* Used to the online set of cpus does not change
  64	 * during assign_irq_vector.
  65	 */
  66	raw_spin_lock(&vector_lock);
  67}
  68
  69void unlock_vector_lock(void)
  70{
  71	raw_spin_unlock(&vector_lock);
  72}
  73
  74void init_irq_alloc_info(struct irq_alloc_info *info,
  75			 const struct cpumask *mask)
  76{
  77	memset(info, 0, sizeof(*info));
  78	info->mask = mask;
  79}
  80
  81void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
  82{
  83	if (src)
  84		*dst = *src;
  85	else
  86		memset(dst, 0, sizeof(*dst));
  87}
  88
  89static struct apic_chip_data *apic_chip_data(struct irq_data *irqd)
  90{
  91	if (!irqd)
  92		return NULL;
  93
  94	while (irqd->parent_data)
  95		irqd = irqd->parent_data;
  96
  97	return irqd->chip_data;
  98}
  99
 100struct irq_cfg *irqd_cfg(struct irq_data *irqd)
 101{
 102	struct apic_chip_data *apicd = apic_chip_data(irqd);
 103
 104	return apicd ? &apicd->hw_irq_cfg : NULL;
 105}
 106EXPORT_SYMBOL_GPL(irqd_cfg);
 107
 108struct irq_cfg *irq_cfg(unsigned int irq)
 109{
 110	return irqd_cfg(irq_get_irq_data(irq));
 111}
 112
 113static struct apic_chip_data *alloc_apic_chip_data(int node)
 114{
 115	struct apic_chip_data *apicd;
 116
 117	apicd = kzalloc_node(sizeof(*apicd), GFP_KERNEL, node);
 118	if (apicd)
 119		INIT_HLIST_NODE(&apicd->clist);
 120	return apicd;
 121}
 122
 123static void free_apic_chip_data(struct apic_chip_data *apicd)
 124{
 125	kfree(apicd);
 126}
 127
 128static void apic_update_irq_cfg(struct irq_data *irqd, unsigned int vector,
 129				unsigned int cpu)
 130{
 131	struct apic_chip_data *apicd = apic_chip_data(irqd);
 132
 133	lockdep_assert_held(&vector_lock);
 134
 135	apicd->hw_irq_cfg.vector = vector;
 136	apicd->hw_irq_cfg.dest_apicid = apic->calc_dest_apicid(cpu);
 137	irq_data_update_effective_affinity(irqd, cpumask_of(cpu));
 138	trace_vector_config(irqd->irq, vector, cpu,
 139			    apicd->hw_irq_cfg.dest_apicid);
 140}
 141
 142static void apic_update_vector(struct irq_data *irqd, unsigned int newvec,
 143			       unsigned int newcpu)
 144{
 145	struct apic_chip_data *apicd = apic_chip_data(irqd);
 146	struct irq_desc *desc = irq_data_to_desc(irqd);
 147	bool managed = irqd_affinity_is_managed(irqd);
 148
 149	lockdep_assert_held(&vector_lock);
 150
 151	trace_vector_update(irqd->irq, newvec, newcpu, apicd->vector,
 152			    apicd->cpu);
 153
 154	/*
 155	 * If there is no vector associated or if the associated vector is
 156	 * the shutdown vector, which is associated to make PCI/MSI
 157	 * shutdown mode work, then there is nothing to release. Clear out
 158	 * prev_vector for this and the offlined target case.
 159	 */
 160	apicd->prev_vector = 0;
 161	if (!apicd->vector || apicd->vector == MANAGED_IRQ_SHUTDOWN_VECTOR)
 162		goto setnew;
 163	/*
 164	 * If the target CPU of the previous vector is online, then mark
 165	 * the vector as move in progress and store it for cleanup when the
 166	 * first interrupt on the new vector arrives. If the target CPU is
 167	 * offline then the regular release mechanism via the cleanup
 168	 * vector is not possible and the vector can be immediately freed
 169	 * in the underlying matrix allocator.
 170	 */
 171	if (cpu_online(apicd->cpu)) {
 172		apicd->move_in_progress = true;
 173		apicd->prev_vector = apicd->vector;
 174		apicd->prev_cpu = apicd->cpu;
 175		WARN_ON_ONCE(apicd->cpu == newcpu);
 176	} else {
 177		irq_matrix_free(vector_matrix, apicd->cpu, apicd->vector,
 178				managed);
 179	}
 180
 181setnew:
 182	apicd->vector = newvec;
 183	apicd->cpu = newcpu;
 184	BUG_ON(!IS_ERR_OR_NULL(per_cpu(vector_irq, newcpu)[newvec]));
 185	per_cpu(vector_irq, newcpu)[newvec] = desc;
 186}
 187
 188static void vector_assign_managed_shutdown(struct irq_data *irqd)
 189{
 190	unsigned int cpu = cpumask_first(cpu_online_mask);
 191
 192	apic_update_irq_cfg(irqd, MANAGED_IRQ_SHUTDOWN_VECTOR, cpu);
 193}
 194
 195static int reserve_managed_vector(struct irq_data *irqd)
 196{
 197	const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
 198	struct apic_chip_data *apicd = apic_chip_data(irqd);
 199	unsigned long flags;
 200	int ret;
 201
 202	raw_spin_lock_irqsave(&vector_lock, flags);
 203	apicd->is_managed = true;
 204	ret = irq_matrix_reserve_managed(vector_matrix, affmsk);
 205	raw_spin_unlock_irqrestore(&vector_lock, flags);
 206	trace_vector_reserve_managed(irqd->irq, ret);
 207	return ret;
 208}
 209
 210static void reserve_irq_vector_locked(struct irq_data *irqd)
 211{
 212	struct apic_chip_data *apicd = apic_chip_data(irqd);
 213
 214	irq_matrix_reserve(vector_matrix);
 215	apicd->can_reserve = true;
 216	apicd->has_reserved = true;
 217	irqd_set_can_reserve(irqd);
 218	trace_vector_reserve(irqd->irq, 0);
 219	vector_assign_managed_shutdown(irqd);
 220}
 221
 222static int reserve_irq_vector(struct irq_data *irqd)
 223{
 224	unsigned long flags;
 225
 226	raw_spin_lock_irqsave(&vector_lock, flags);
 227	reserve_irq_vector_locked(irqd);
 228	raw_spin_unlock_irqrestore(&vector_lock, flags);
 229	return 0;
 230}
 231
 232static int
 233assign_vector_locked(struct irq_data *irqd, const struct cpumask *dest)
 234{
 235	struct apic_chip_data *apicd = apic_chip_data(irqd);
 236	bool resvd = apicd->has_reserved;
 237	unsigned int cpu = apicd->cpu;
 238	int vector = apicd->vector;
 239
 240	lockdep_assert_held(&vector_lock);
 241
 242	/*
 243	 * If the current target CPU is online and in the new requested
 244	 * affinity mask, there is no point in moving the interrupt from
 245	 * one CPU to another.
 246	 */
 247	if (vector && cpu_online(cpu) && cpumask_test_cpu(cpu, dest))
 248		return 0;
 249
 250	/*
 251	 * Careful here. @apicd might either have move_in_progress set or
 252	 * be enqueued for cleanup. Assigning a new vector would either
 253	 * leave a stale vector on some CPU around or in case of a pending
 254	 * cleanup corrupt the hlist.
 255	 */
 256	if (apicd->move_in_progress || !hlist_unhashed(&apicd->clist))
 257		return -EBUSY;
 258
 259	vector = irq_matrix_alloc(vector_matrix, dest, resvd, &cpu);
 260	trace_vector_alloc(irqd->irq, vector, resvd, vector);
 261	if (vector < 0)
 262		return vector;
 263	apic_update_vector(irqd, vector, cpu);
 264	apic_update_irq_cfg(irqd, vector, cpu);
 265
 266	return 0;
 267}
 268
 269static int assign_irq_vector(struct irq_data *irqd, const struct cpumask *dest)
 270{
 271	unsigned long flags;
 272	int ret;
 273
 274	raw_spin_lock_irqsave(&vector_lock, flags);
 275	cpumask_and(vector_searchmask, dest, cpu_online_mask);
 276	ret = assign_vector_locked(irqd, vector_searchmask);
 277	raw_spin_unlock_irqrestore(&vector_lock, flags);
 278	return ret;
 279}
 280
 281static int assign_irq_vector_any_locked(struct irq_data *irqd)
 282{
 283	/* Get the affinity mask - either irq_default_affinity or (user) set */
 284	const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
 285	int node = irq_data_get_node(irqd);
 286
 287	if (node != NUMA_NO_NODE) {
 288		/* Try the intersection of @affmsk and node mask */
 289		cpumask_and(vector_searchmask, cpumask_of_node(node), affmsk);
 290		if (!assign_vector_locked(irqd, vector_searchmask))
 291			return 0;
 292	}
 293
 
 
 
 294	/* Try the full affinity mask */
 295	cpumask_and(vector_searchmask, affmsk, cpu_online_mask);
 296	if (!assign_vector_locked(irqd, vector_searchmask))
 297		return 0;
 298
 299	if (node != NUMA_NO_NODE) {
 300		/* Try the node mask */
 301		if (!assign_vector_locked(irqd, cpumask_of_node(node)))
 302			return 0;
 303	}
 304
 305	/* Try the full online mask */
 306	return assign_vector_locked(irqd, cpu_online_mask);
 307}
 308
 309static int
 310assign_irq_vector_policy(struct irq_data *irqd, struct irq_alloc_info *info)
 311{
 312	if (irqd_affinity_is_managed(irqd))
 313		return reserve_managed_vector(irqd);
 314	if (info->mask)
 315		return assign_irq_vector(irqd, info->mask);
 316	/*
 317	 * Make only a global reservation with no guarantee. A real vector
 318	 * is associated at activation time.
 319	 */
 320	return reserve_irq_vector(irqd);
 321}
 322
 323static int
 324assign_managed_vector(struct irq_data *irqd, const struct cpumask *dest)
 325{
 326	const struct cpumask *affmsk = irq_data_get_affinity_mask(irqd);
 327	struct apic_chip_data *apicd = apic_chip_data(irqd);
 328	int vector, cpu;
 329
 330	cpumask_and(vector_searchmask, dest, affmsk);
 331
 332	/* set_affinity might call here for nothing */
 333	if (apicd->vector && cpumask_test_cpu(apicd->cpu, vector_searchmask))
 334		return 0;
 335	vector = irq_matrix_alloc_managed(vector_matrix, vector_searchmask,
 336					  &cpu);
 337	trace_vector_alloc_managed(irqd->irq, vector, vector);
 338	if (vector < 0)
 339		return vector;
 340	apic_update_vector(irqd, vector, cpu);
 341	apic_update_irq_cfg(irqd, vector, cpu);
 342	return 0;
 343}
 344
 345static void clear_irq_vector(struct irq_data *irqd)
 346{
 347	struct apic_chip_data *apicd = apic_chip_data(irqd);
 348	bool managed = irqd_affinity_is_managed(irqd);
 349	unsigned int vector = apicd->vector;
 350
 351	lockdep_assert_held(&vector_lock);
 352
 353	if (!vector)
 354		return;
 355
 356	trace_vector_clear(irqd->irq, vector, apicd->cpu, apicd->prev_vector,
 357			   apicd->prev_cpu);
 358
 359	per_cpu(vector_irq, apicd->cpu)[vector] = VECTOR_SHUTDOWN;
 360	irq_matrix_free(vector_matrix, apicd->cpu, vector, managed);
 361	apicd->vector = 0;
 362
 363	/* Clean up move in progress */
 364	vector = apicd->prev_vector;
 365	if (!vector)
 366		return;
 367
 368	per_cpu(vector_irq, apicd->prev_cpu)[vector] = VECTOR_SHUTDOWN;
 369	irq_matrix_free(vector_matrix, apicd->prev_cpu, vector, managed);
 370	apicd->prev_vector = 0;
 371	apicd->move_in_progress = 0;
 372	hlist_del_init(&apicd->clist);
 373}
 374
 375static void x86_vector_deactivate(struct irq_domain *dom, struct irq_data *irqd)
 376{
 377	struct apic_chip_data *apicd = apic_chip_data(irqd);
 378	unsigned long flags;
 379
 380	trace_vector_deactivate(irqd->irq, apicd->is_managed,
 381				apicd->can_reserve, false);
 382
 383	/* Regular fixed assigned interrupt */
 384	if (!apicd->is_managed && !apicd->can_reserve)
 385		return;
 386	/* If the interrupt has a global reservation, nothing to do */
 387	if (apicd->has_reserved)
 388		return;
 389
 390	raw_spin_lock_irqsave(&vector_lock, flags);
 391	clear_irq_vector(irqd);
 392	if (apicd->can_reserve)
 393		reserve_irq_vector_locked(irqd);
 394	else
 395		vector_assign_managed_shutdown(irqd);
 396	raw_spin_unlock_irqrestore(&vector_lock, flags);
 397}
 398
 399static int activate_reserved(struct irq_data *irqd)
 400{
 401	struct apic_chip_data *apicd = apic_chip_data(irqd);
 402	int ret;
 403
 404	ret = assign_irq_vector_any_locked(irqd);
 405	if (!ret) {
 406		apicd->has_reserved = false;
 407		/*
 408		 * Core might have disabled reservation mode after
 409		 * allocating the irq descriptor. Ideally this should
 410		 * happen before allocation time, but that would require
 411		 * completely convoluted ways of transporting that
 412		 * information.
 413		 */
 414		if (!irqd_can_reserve(irqd))
 415			apicd->can_reserve = false;
 416	}
 417
 418	/*
 419	 * Check to ensure that the effective affinity mask is a subset
 420	 * the user supplied affinity mask, and warn the user if it is not
 421	 */
 422	if (!cpumask_subset(irq_data_get_effective_affinity_mask(irqd),
 423			    irq_data_get_affinity_mask(irqd))) {
 424		pr_warn("irq %u: Affinity broken due to vector space exhaustion.\n",
 425			irqd->irq);
 426	}
 427
 428	return ret;
 429}
 430
 431static int activate_managed(struct irq_data *irqd)
 432{
 433	const struct cpumask *dest = irq_data_get_affinity_mask(irqd);
 434	int ret;
 435
 436	cpumask_and(vector_searchmask, dest, cpu_online_mask);
 437	if (WARN_ON_ONCE(cpumask_empty(vector_searchmask))) {
 438		/* Something in the core code broke! Survive gracefully */
 439		pr_err("Managed startup for irq %u, but no CPU\n", irqd->irq);
 440		return -EINVAL;
 441	}
 442
 443	ret = assign_managed_vector(irqd, vector_searchmask);
 444	/*
 445	 * This should not happen. The vector reservation got buggered.  Handle
 446	 * it gracefully.
 447	 */
 448	if (WARN_ON_ONCE(ret < 0)) {
 449		pr_err("Managed startup irq %u, no vector available\n",
 450		       irqd->irq);
 451	}
 452	return ret;
 453}
 454
 455static int x86_vector_activate(struct irq_domain *dom, struct irq_data *irqd,
 456			       bool reserve)
 457{
 458	struct apic_chip_data *apicd = apic_chip_data(irqd);
 459	unsigned long flags;
 460	int ret = 0;
 461
 462	trace_vector_activate(irqd->irq, apicd->is_managed,
 463			      apicd->can_reserve, reserve);
 464
 
 
 
 
 465	raw_spin_lock_irqsave(&vector_lock, flags);
 466	if (!apicd->can_reserve && !apicd->is_managed)
 467		assign_irq_vector_any_locked(irqd);
 468	else if (reserve || irqd_is_managed_and_shutdown(irqd))
 469		vector_assign_managed_shutdown(irqd);
 470	else if (apicd->is_managed)
 471		ret = activate_managed(irqd);
 472	else if (apicd->has_reserved)
 473		ret = activate_reserved(irqd);
 474	raw_spin_unlock_irqrestore(&vector_lock, flags);
 475	return ret;
 476}
 477
 478static void vector_free_reserved_and_managed(struct irq_data *irqd)
 479{
 480	const struct cpumask *dest = irq_data_get_affinity_mask(irqd);
 481	struct apic_chip_data *apicd = apic_chip_data(irqd);
 482
 483	trace_vector_teardown(irqd->irq, apicd->is_managed,
 484			      apicd->has_reserved);
 485
 486	if (apicd->has_reserved)
 487		irq_matrix_remove_reserved(vector_matrix);
 488	if (apicd->is_managed)
 489		irq_matrix_remove_managed(vector_matrix, dest);
 490}
 491
 492static void x86_vector_free_irqs(struct irq_domain *domain,
 493				 unsigned int virq, unsigned int nr_irqs)
 494{
 495	struct apic_chip_data *apicd;
 496	struct irq_data *irqd;
 497	unsigned long flags;
 498	int i;
 499
 500	for (i = 0; i < nr_irqs; i++) {
 501		irqd = irq_domain_get_irq_data(x86_vector_domain, virq + i);
 502		if (irqd && irqd->chip_data) {
 503			raw_spin_lock_irqsave(&vector_lock, flags);
 504			clear_irq_vector(irqd);
 505			vector_free_reserved_and_managed(irqd);
 506			apicd = irqd->chip_data;
 507			irq_domain_reset_irq_data(irqd);
 508			raw_spin_unlock_irqrestore(&vector_lock, flags);
 509			free_apic_chip_data(apicd);
 510		}
 511	}
 512}
 513
 514static bool vector_configure_legacy(unsigned int virq, struct irq_data *irqd,
 515				    struct apic_chip_data *apicd)
 516{
 517	unsigned long flags;
 518	bool realloc = false;
 519
 520	apicd->vector = ISA_IRQ_VECTOR(virq);
 521	apicd->cpu = 0;
 522
 523	raw_spin_lock_irqsave(&vector_lock, flags);
 524	/*
 525	 * If the interrupt is activated, then it must stay at this vector
 526	 * position. That's usually the timer interrupt (0).
 527	 */
 528	if (irqd_is_activated(irqd)) {
 529		trace_vector_setup(virq, true, 0);
 530		apic_update_irq_cfg(irqd, apicd->vector, apicd->cpu);
 531	} else {
 532		/* Release the vector */
 533		apicd->can_reserve = true;
 534		irqd_set_can_reserve(irqd);
 535		clear_irq_vector(irqd);
 536		realloc = true;
 537	}
 538	raw_spin_unlock_irqrestore(&vector_lock, flags);
 539	return realloc;
 540}
 541
 542static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
 543				 unsigned int nr_irqs, void *arg)
 544{
 545	struct irq_alloc_info *info = arg;
 546	struct apic_chip_data *apicd;
 547	struct irq_data *irqd;
 548	int i, err, node;
 549
 550	if (apic_is_disabled)
 551		return -ENXIO;
 552
 553	/*
 554	 * Catch any attempt to touch the cascade interrupt on a PIC
 555	 * equipped system.
 556	 */
 557	if (WARN_ON_ONCE(info->flags & X86_IRQ_ALLOC_LEGACY &&
 558			 virq == PIC_CASCADE_IR))
 559		return -EINVAL;
 560
 561	for (i = 0; i < nr_irqs; i++) {
 562		irqd = irq_domain_get_irq_data(domain, virq + i);
 563		BUG_ON(!irqd);
 564		node = irq_data_get_node(irqd);
 565		WARN_ON_ONCE(irqd->chip_data);
 566		apicd = alloc_apic_chip_data(node);
 567		if (!apicd) {
 568			err = -ENOMEM;
 569			goto error;
 570		}
 571
 572		apicd->irq = virq + i;
 573		irqd->chip = &lapic_controller;
 574		irqd->chip_data = apicd;
 575		irqd->hwirq = virq + i;
 576		irqd_set_single_target(irqd);
 577		/*
 578		 * Prevent that any of these interrupts is invoked in
 579		 * non interrupt context via e.g. generic_handle_irq()
 580		 * as that can corrupt the affinity move state.
 581		 */
 582		irqd_set_handle_enforce_irqctx(irqd);
 583
 584		/* Don't invoke affinity setter on deactivated interrupts */
 585		irqd_set_affinity_on_activate(irqd);
 586
 587		/*
 588		 * Legacy vectors are already assigned when the IOAPIC
 589		 * takes them over. They stay on the same vector. This is
 590		 * required for check_timer() to work correctly as it might
 591		 * switch back to legacy mode. Only update the hardware
 592		 * config.
 593		 */
 594		if (info->flags & X86_IRQ_ALLOC_LEGACY) {
 595			if (!vector_configure_legacy(virq + i, irqd, apicd))
 596				continue;
 597		}
 598
 599		err = assign_irq_vector_policy(irqd, info);
 600		trace_vector_setup(virq + i, false, err);
 601		if (err) {
 602			irqd->chip_data = NULL;
 603			free_apic_chip_data(apicd);
 604			goto error;
 605		}
 606	}
 607
 608	return 0;
 609
 610error:
 611	x86_vector_free_irqs(domain, virq, i);
 612	return err;
 613}
 614
 615#ifdef CONFIG_GENERIC_IRQ_DEBUGFS
 616static void x86_vector_debug_show(struct seq_file *m, struct irq_domain *d,
 617				  struct irq_data *irqd, int ind)
 618{
 619	struct apic_chip_data apicd;
 620	unsigned long flags;
 621	int irq;
 622
 623	if (!irqd) {
 624		irq_matrix_debug_show(m, vector_matrix, ind);
 625		return;
 626	}
 627
 628	irq = irqd->irq;
 629	if (irq < nr_legacy_irqs() && !test_bit(irq, &io_apic_irqs)) {
 630		seq_printf(m, "%*sVector: %5d\n", ind, "", ISA_IRQ_VECTOR(irq));
 631		seq_printf(m, "%*sTarget: Legacy PIC all CPUs\n", ind, "");
 632		return;
 633	}
 634
 635	if (!irqd->chip_data) {
 636		seq_printf(m, "%*sVector: Not assigned\n", ind, "");
 637		return;
 638	}
 639
 640	raw_spin_lock_irqsave(&vector_lock, flags);
 641	memcpy(&apicd, irqd->chip_data, sizeof(apicd));
 642	raw_spin_unlock_irqrestore(&vector_lock, flags);
 643
 644	seq_printf(m, "%*sVector: %5u\n", ind, "", apicd.vector);
 645	seq_printf(m, "%*sTarget: %5u\n", ind, "", apicd.cpu);
 646	if (apicd.prev_vector) {
 647		seq_printf(m, "%*sPrevious vector: %5u\n", ind, "", apicd.prev_vector);
 648		seq_printf(m, "%*sPrevious target: %5u\n", ind, "", apicd.prev_cpu);
 649	}
 650	seq_printf(m, "%*smove_in_progress: %u\n", ind, "", apicd.move_in_progress ? 1 : 0);
 651	seq_printf(m, "%*sis_managed:       %u\n", ind, "", apicd.is_managed ? 1 : 0);
 652	seq_printf(m, "%*scan_reserve:      %u\n", ind, "", apicd.can_reserve ? 1 : 0);
 653	seq_printf(m, "%*shas_reserved:     %u\n", ind, "", apicd.has_reserved ? 1 : 0);
 654	seq_printf(m, "%*scleanup_pending:  %u\n", ind, "", !hlist_unhashed(&apicd.clist));
 655}
 656#endif
 657
 658int x86_fwspec_is_ioapic(struct irq_fwspec *fwspec)
 659{
 660	if (fwspec->param_count != 1)
 661		return 0;
 662
 663	if (is_fwnode_irqchip(fwspec->fwnode)) {
 664		const char *fwname = fwnode_get_name(fwspec->fwnode);
 665		return fwname && !strncmp(fwname, "IO-APIC-", 8) &&
 666			simple_strtol(fwname+8, NULL, 10) == fwspec->param[0];
 667	}
 668	return to_of_node(fwspec->fwnode) &&
 669		of_device_is_compatible(to_of_node(fwspec->fwnode),
 670					"intel,ce4100-ioapic");
 671}
 672
 673int x86_fwspec_is_hpet(struct irq_fwspec *fwspec)
 674{
 675	if (fwspec->param_count != 1)
 676		return 0;
 677
 678	if (is_fwnode_irqchip(fwspec->fwnode)) {
 679		const char *fwname = fwnode_get_name(fwspec->fwnode);
 680		return fwname && !strncmp(fwname, "HPET-MSI-", 9) &&
 681			simple_strtol(fwname+9, NULL, 10) == fwspec->param[0];
 682	}
 683	return 0;
 684}
 685
 686static int x86_vector_select(struct irq_domain *d, struct irq_fwspec *fwspec,
 687			     enum irq_domain_bus_token bus_token)
 688{
 689	/*
 690	 * HPET and I/OAPIC cannot be parented in the vector domain
 691	 * if IRQ remapping is enabled. APIC IDs above 15 bits are
 692	 * only permitted if IRQ remapping is enabled, so check that.
 693	 */
 694	if (apic_id_valid(32768))
 695		return 0;
 696
 697	return x86_fwspec_is_ioapic(fwspec) || x86_fwspec_is_hpet(fwspec);
 698}
 699
 700static const struct irq_domain_ops x86_vector_domain_ops = {
 701	.select		= x86_vector_select,
 702	.alloc		= x86_vector_alloc_irqs,
 703	.free		= x86_vector_free_irqs,
 704	.activate	= x86_vector_activate,
 705	.deactivate	= x86_vector_deactivate,
 706#ifdef CONFIG_GENERIC_IRQ_DEBUGFS
 707	.debug_show	= x86_vector_debug_show,
 708#endif
 709};
 710
 711int __init arch_probe_nr_irqs(void)
 712{
 713	int nr;
 714
 715	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
 716		nr_irqs = NR_VECTORS * nr_cpu_ids;
 717
 718	nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
 719#if defined(CONFIG_PCI_MSI)
 720	/*
 721	 * for MSI and HT dyn irq
 722	 */
 723	if (gsi_top <= NR_IRQS_LEGACY)
 724		nr +=  8 * nr_cpu_ids;
 725	else
 726		nr += gsi_top * 16;
 727#endif
 728	if (nr < nr_irqs)
 729		nr_irqs = nr;
 730
 731	/*
 732	 * We don't know if PIC is present at this point so we need to do
 733	 * probe() to get the right number of legacy IRQs.
 734	 */
 735	return legacy_pic->probe();
 736}
 737
 738void lapic_assign_legacy_vector(unsigned int irq, bool replace)
 739{
 740	/*
 741	 * Use assign system here so it won't get accounted as allocated
 742	 * and movable in the cpu hotplug check and it prevents managed
 743	 * irq reservation from touching it.
 744	 */
 745	irq_matrix_assign_system(vector_matrix, ISA_IRQ_VECTOR(irq), replace);
 746}
 747
 748void __init lapic_update_legacy_vectors(void)
 749{
 750	unsigned int i;
 751
 752	if (IS_ENABLED(CONFIG_X86_IO_APIC) && nr_ioapics > 0)
 753		return;
 754
 755	/*
 756	 * If the IO/APIC is disabled via config, kernel command line or
 757	 * lack of enumeration then all legacy interrupts are routed
 758	 * through the PIC. Make sure that they are marked as legacy
 759	 * vectors. PIC_CASCADE_IRQ has already been marked in
 760	 * lapic_assign_system_vectors().
 761	 */
 762	for (i = 0; i < nr_legacy_irqs(); i++) {
 763		if (i != PIC_CASCADE_IR)
 764			lapic_assign_legacy_vector(i, true);
 765	}
 766}
 767
 768void __init lapic_assign_system_vectors(void)
 769{
 770	unsigned int i, vector;
 771
 772	for_each_set_bit(vector, system_vectors, NR_VECTORS)
 773		irq_matrix_assign_system(vector_matrix, vector, false);
 774
 775	if (nr_legacy_irqs() > 1)
 776		lapic_assign_legacy_vector(PIC_CASCADE_IR, false);
 777
 778	/* System vectors are reserved, online it */
 779	irq_matrix_online(vector_matrix);
 780
 781	/* Mark the preallocated legacy interrupts */
 782	for (i = 0; i < nr_legacy_irqs(); i++) {
 783		/*
 784		 * Don't touch the cascade interrupt. It's unusable
 785		 * on PIC equipped machines. See the large comment
 786		 * in the IO/APIC code.
 787		 */
 788		if (i != PIC_CASCADE_IR)
 789			irq_matrix_assign(vector_matrix, ISA_IRQ_VECTOR(i));
 790	}
 791}
 792
 793int __init arch_early_irq_init(void)
 794{
 795	struct fwnode_handle *fn;
 796
 797	fn = irq_domain_alloc_named_fwnode("VECTOR");
 798	BUG_ON(!fn);
 799	x86_vector_domain = irq_domain_create_tree(fn, &x86_vector_domain_ops,
 800						   NULL);
 801	BUG_ON(x86_vector_domain == NULL);
 
 802	irq_set_default_host(x86_vector_domain);
 803
 
 
 804	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
 805
 806	/*
 807	 * Allocate the vector matrix allocator data structure and limit the
 808	 * search area.
 809	 */
 810	vector_matrix = irq_alloc_matrix(NR_VECTORS, FIRST_EXTERNAL_VECTOR,
 811					 FIRST_SYSTEM_VECTOR);
 812	BUG_ON(!vector_matrix);
 813
 814	return arch_early_ioapic_init();
 815}
 816
 817#ifdef CONFIG_SMP
 818
 819static struct irq_desc *__setup_vector_irq(int vector)
 820{
 821	int isairq = vector - ISA_IRQ_VECTOR(0);
 822
 823	/* Check whether the irq is in the legacy space */
 824	if (isairq < 0 || isairq >= nr_legacy_irqs())
 825		return VECTOR_UNUSED;
 826	/* Check whether the irq is handled by the IOAPIC */
 827	if (test_bit(isairq, &io_apic_irqs))
 828		return VECTOR_UNUSED;
 829	return irq_to_desc(isairq);
 830}
 831
 832/* Online the local APIC infrastructure and initialize the vectors */
 833void lapic_online(void)
 834{
 835	unsigned int vector;
 836
 837	lockdep_assert_held(&vector_lock);
 838
 839	/* Online the vector matrix array for this CPU */
 840	irq_matrix_online(vector_matrix);
 841
 842	/*
 843	 * The interrupt affinity logic never targets interrupts to offline
 844	 * CPUs. The exception are the legacy PIC interrupts. In general
 845	 * they are only targeted to CPU0, but depending on the platform
 846	 * they can be distributed to any online CPU in hardware. The
 847	 * kernel has no influence on that. So all active legacy vectors
 848	 * must be installed on all CPUs. All non legacy interrupts can be
 849	 * cleared.
 850	 */
 851	for (vector = 0; vector < NR_VECTORS; vector++)
 852		this_cpu_write(vector_irq[vector], __setup_vector_irq(vector));
 853}
 854
 855static void __vector_cleanup(struct vector_cleanup *cl, bool check_irr);
 856
 857void lapic_offline(void)
 858{
 859	struct vector_cleanup *cl = this_cpu_ptr(&vector_cleanup);
 860
 861	lock_vector_lock();
 862
 863	/* In case the vector cleanup timer has not expired */
 864	__vector_cleanup(cl, false);
 865
 866	irq_matrix_offline(vector_matrix);
 867	WARN_ON_ONCE(try_to_del_timer_sync(&cl->timer) < 0);
 868	WARN_ON_ONCE(!hlist_empty(&cl->head));
 869
 870	unlock_vector_lock();
 871}
 872
 873static int apic_set_affinity(struct irq_data *irqd,
 874			     const struct cpumask *dest, bool force)
 875{
 
 876	int err;
 877
 878	if (WARN_ON_ONCE(!irqd_is_activated(irqd)))
 879		return -EIO;
 
 
 
 
 
 
 
 
 
 880
 881	raw_spin_lock(&vector_lock);
 882	cpumask_and(vector_searchmask, dest, cpu_online_mask);
 883	if (irqd_affinity_is_managed(irqd))
 884		err = assign_managed_vector(irqd, vector_searchmask);
 885	else
 886		err = assign_vector_locked(irqd, vector_searchmask);
 887	raw_spin_unlock(&vector_lock);
 888	return err ? err : IRQ_SET_MASK_OK;
 889}
 890
 891#else
 892# define apic_set_affinity	NULL
 893#endif
 894
 895static int apic_retrigger_irq(struct irq_data *irqd)
 896{
 897	struct apic_chip_data *apicd = apic_chip_data(irqd);
 898	unsigned long flags;
 899
 900	raw_spin_lock_irqsave(&vector_lock, flags);
 901	__apic_send_IPI(apicd->cpu, apicd->vector);
 902	raw_spin_unlock_irqrestore(&vector_lock, flags);
 903
 904	return 1;
 905}
 906
 907void apic_ack_irq(struct irq_data *irqd)
 908{
 909	irq_move_irq(irqd);
 910	apic_eoi();
 911}
 912
 913void apic_ack_edge(struct irq_data *irqd)
 914{
 915	irq_complete_move(irqd_cfg(irqd));
 916	apic_ack_irq(irqd);
 917}
 918
 919static void x86_vector_msi_compose_msg(struct irq_data *data,
 920				       struct msi_msg *msg)
 921{
 922       __irq_msi_compose_msg(irqd_cfg(data), msg, false);
 923}
 924
 925static struct irq_chip lapic_controller = {
 926	.name			= "APIC",
 927	.irq_ack		= apic_ack_edge,
 928	.irq_set_affinity	= apic_set_affinity,
 929	.irq_compose_msi_msg	= x86_vector_msi_compose_msg,
 930	.irq_retrigger		= apic_retrigger_irq,
 931};
 932
 933#ifdef CONFIG_SMP
 934
 935static void free_moved_vector(struct apic_chip_data *apicd)
 936{
 937	unsigned int vector = apicd->prev_vector;
 938	unsigned int cpu = apicd->prev_cpu;
 939	bool managed = apicd->is_managed;
 940
 941	/*
 942	 * Managed interrupts are usually not migrated away
 943	 * from an online CPU, but CPU isolation 'managed_irq'
 944	 * can make that happen.
 945	 * 1) Activation does not take the isolation into account
 946	 *    to keep the code simple
 947	 * 2) Migration away from an isolated CPU can happen when
 948	 *    a non-isolated CPU which is in the calculated
 949	 *    affinity mask comes online.
 950	 */
 
 
 951	trace_vector_free_moved(apicd->irq, cpu, vector, managed);
 952	irq_matrix_free(vector_matrix, cpu, vector, managed);
 953	per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
 954	hlist_del_init(&apicd->clist);
 955	apicd->prev_vector = 0;
 956	apicd->move_in_progress = 0;
 957}
 958
 959static void __vector_cleanup(struct vector_cleanup *cl, bool check_irr)
 960{
 
 961	struct apic_chip_data *apicd;
 962	struct hlist_node *tmp;
 963	bool rearm = false;
 964
 965	lockdep_assert_held(&vector_lock);
 
 
 966
 967	hlist_for_each_entry_safe(apicd, tmp, &cl->head, clist) {
 968		unsigned int irr, vector = apicd->prev_vector;
 969
 970		/*
 971		 * Paranoia: Check if the vector that needs to be cleaned
 972		 * up is registered at the APICs IRR. That's clearly a
 973		 * hardware issue if the vector arrived on the old target
 974		 * _after_ interrupts were disabled above. Keep @apicd
 975		 * on the list and schedule the timer again to give the CPU
 976		 * a chance to handle the pending interrupt.
 977		 *
 978		 * Do not check IRR when called from lapic_offline(), because
 979		 * fixup_irqs() was just called to scan IRR for set bits and
 980		 * forward them to new destination CPUs via IPIs.
 981		 */
 982		irr = check_irr ? apic_read(APIC_IRR + (vector / 32 * 0x10)) : 0;
 983		if (irr & (1U << (vector % 32))) {
 984			pr_warn_once("Moved interrupt pending in old target APIC %u\n", apicd->irq);
 985			rearm = true;
 986			continue;
 987		}
 988		free_moved_vector(apicd);
 989	}
 990
 991	/*
 992	 * Must happen under vector_lock to make the timer_pending() check
 993	 * in __vector_schedule_cleanup() race free against the rearm here.
 994	 */
 995	if (rearm)
 996		mod_timer(&cl->timer, jiffies + 1);
 997}
 998
 999static void vector_cleanup_callback(struct timer_list *tmr)
1000{
1001	struct vector_cleanup *cl = container_of(tmr, typeof(*cl), timer);
1002
1003	/* Prevent vectors vanishing under us */
1004	raw_spin_lock_irq(&vector_lock);
1005	__vector_cleanup(cl, true);
1006	raw_spin_unlock_irq(&vector_lock);
1007}
1008
1009static void __vector_schedule_cleanup(struct apic_chip_data *apicd)
1010{
1011	unsigned int cpu = apicd->prev_cpu;
1012
1013	raw_spin_lock(&vector_lock);
1014	apicd->move_in_progress = 0;
 
1015	if (cpu_online(cpu)) {
1016		struct vector_cleanup *cl = per_cpu_ptr(&vector_cleanup, cpu);
1017
1018		hlist_add_head(&apicd->clist, &cl->head);
1019
1020		/*
1021		 * The lockless timer_pending() check is safe here. If it
1022		 * returns true, then the callback will observe this new
1023		 * apic data in the hlist as everything is serialized by
1024		 * vector lock.
1025		 *
1026		 * If it returns false then the timer is either not armed
1027		 * or the other CPU executes the callback, which again
1028		 * would be blocked on vector lock. Rearming it in the
1029		 * latter case makes it fire for nothing.
1030		 *
1031		 * This is also safe against the callback rearming the timer
1032		 * because that's serialized via vector lock too.
1033		 */
1034		if (!timer_pending(&cl->timer)) {
1035			cl->timer.expires = jiffies + 1;
1036			add_timer_on(&cl->timer, cpu);
1037		}
1038	} else {
1039		pr_warn("IRQ %u schedule cleanup for offline CPU %u\n", apicd->irq, cpu);
1040		free_moved_vector(apicd);
1041	}
1042	raw_spin_unlock(&vector_lock);
1043}
1044
1045void vector_schedule_cleanup(struct irq_cfg *cfg)
1046{
1047	struct apic_chip_data *apicd;
1048
1049	apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg);
1050	if (apicd->move_in_progress)
1051		__vector_schedule_cleanup(apicd);
1052}
1053
1054void irq_complete_move(struct irq_cfg *cfg)
1055{
1056	struct apic_chip_data *apicd;
1057
1058	apicd = container_of(cfg, struct apic_chip_data, hw_irq_cfg);
1059	if (likely(!apicd->move_in_progress))
1060		return;
1061
1062	/*
1063	 * If the interrupt arrived on the new target CPU, cleanup the
1064	 * vector on the old target CPU. A vector check is not required
1065	 * because an interrupt can never move from one vector to another
1066	 * on the same CPU.
1067	 */
1068	if (apicd->cpu == smp_processor_id())
1069		__vector_schedule_cleanup(apicd);
1070}
1071
1072/*
1073 * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
1074 */
1075void irq_force_complete_move(struct irq_desc *desc)
1076{
1077	unsigned int cpu = smp_processor_id();
1078	struct apic_chip_data *apicd;
1079	struct irq_data *irqd;
1080	unsigned int vector;
1081
1082	/*
1083	 * The function is called for all descriptors regardless of which
1084	 * irqdomain they belong to. For example if an IRQ is provided by
1085	 * an irq_chip as part of a GPIO driver, the chip data for that
1086	 * descriptor is specific to the irq_chip in question.
1087	 *
1088	 * Check first that the chip_data is what we expect
1089	 * (apic_chip_data) before touching it any further.
1090	 */
1091	irqd = irq_domain_get_irq_data(x86_vector_domain,
1092				       irq_desc_get_irq(desc));
1093	if (!irqd)
1094		return;
1095
1096	raw_spin_lock(&vector_lock);
1097	apicd = apic_chip_data(irqd);
1098	if (!apicd)
1099		goto unlock;
1100
1101	/*
1102	 * If prev_vector is empty or the descriptor is neither currently
1103	 * nor previously on the outgoing CPU no action required.
1104	 */
1105	vector = apicd->prev_vector;
1106	if (!vector || (apicd->cpu != cpu && apicd->prev_cpu != cpu))
1107		goto unlock;
1108
1109	/*
1110	 * This is tricky. If the cleanup of the old vector has not been
1111	 * done yet, then the following setaffinity call will fail with
1112	 * -EBUSY. This can leave the interrupt in a stale state.
1113	 *
1114	 * All CPUs are stuck in stop machine with interrupts disabled so
1115	 * calling __irq_complete_move() would be completely pointless.
1116	 *
1117	 * 1) The interrupt is in move_in_progress state. That means that we
1118	 *    have not seen an interrupt since the io_apic was reprogrammed to
1119	 *    the new vector.
1120	 *
1121	 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
1122	 *    have not been processed yet.
1123	 */
1124	if (apicd->move_in_progress) {
1125		/*
1126		 * In theory there is a race:
1127		 *
1128		 * set_ioapic(new_vector) <-- Interrupt is raised before update
1129		 *			      is effective, i.e. it's raised on
1130		 *			      the old vector.
1131		 *
1132		 * So if the target cpu cannot handle that interrupt before
1133		 * the old vector is cleaned up, we get a spurious interrupt
1134		 * and in the worst case the ioapic irq line becomes stale.
1135		 *
1136		 * But in case of cpu hotplug this should be a non issue
1137		 * because if the affinity update happens right before all
1138		 * cpus rendezvous in stop machine, there is no way that the
1139		 * interrupt can be blocked on the target cpu because all cpus
1140		 * loops first with interrupts enabled in stop machine, so the
1141		 * old vector is not yet cleaned up when the interrupt fires.
1142		 *
1143		 * So the only way to run into this issue is if the delivery
1144		 * of the interrupt on the apic/system bus would be delayed
1145		 * beyond the point where the target cpu disables interrupts
1146		 * in stop machine. I doubt that it can happen, but at least
1147		 * there is a theoretical chance. Virtualization might be
1148		 * able to expose this, but AFAICT the IOAPIC emulation is not
1149		 * as stupid as the real hardware.
1150		 *
1151		 * Anyway, there is nothing we can do about that at this point
1152		 * w/o refactoring the whole fixup_irq() business completely.
1153		 * We print at least the irq number and the old vector number,
1154		 * so we have the necessary information when a problem in that
1155		 * area arises.
1156		 */
1157		pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
1158			irqd->irq, vector);
1159	}
1160	free_moved_vector(apicd);
1161unlock:
1162	raw_spin_unlock(&vector_lock);
1163}
1164
1165#ifdef CONFIG_HOTPLUG_CPU
1166/*
1167 * Note, this is not accurate accounting, but at least good enough to
1168 * prevent that the actual interrupt move will run out of vectors.
1169 */
1170int lapic_can_unplug_cpu(void)
1171{
1172	unsigned int rsvd, avl, tomove, cpu = smp_processor_id();
1173	int ret = 0;
1174
1175	raw_spin_lock(&vector_lock);
1176	tomove = irq_matrix_allocated(vector_matrix);
1177	avl = irq_matrix_available(vector_matrix, true);
1178	if (avl < tomove) {
1179		pr_warn("CPU %u has %u vectors, %u available. Cannot disable CPU\n",
1180			cpu, tomove, avl);
1181		ret = -ENOSPC;
1182		goto out;
1183	}
1184	rsvd = irq_matrix_reserved(vector_matrix);
1185	if (avl < rsvd) {
1186		pr_warn("Reserved vectors %u > available %u. IRQ request may fail\n",
1187			rsvd, avl);
1188	}
1189out:
1190	raw_spin_unlock(&vector_lock);
1191	return ret;
1192}
1193#endif /* HOTPLUG_CPU */
1194#endif /* SMP */
1195
1196static void __init print_APIC_field(int base)
1197{
1198	int i;
1199
1200	printk(KERN_DEBUG);
1201
1202	for (i = 0; i < 8; i++)
1203		pr_cont("%08x", apic_read(base + i*0x10));
1204
1205	pr_cont("\n");
1206}
1207
1208static void __init print_local_APIC(void *dummy)
1209{
1210	unsigned int i, v, ver, maxlvt;
1211	u64 icr;
1212
1213	pr_debug("printing local APIC contents on CPU#%d/%d:\n",
1214		 smp_processor_id(), read_apic_id());
1215	v = apic_read(APIC_ID);
1216	pr_info("... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1217	v = apic_read(APIC_LVR);
1218	pr_info("... APIC VERSION: %08x\n", v);
1219	ver = GET_APIC_VERSION(v);
1220	maxlvt = lapic_get_maxlvt();
1221
1222	v = apic_read(APIC_TASKPRI);
1223	pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1224
1225	/* !82489DX */
1226	if (APIC_INTEGRATED(ver)) {
1227		if (!APIC_XAPIC(ver)) {
1228			v = apic_read(APIC_ARBPRI);
1229			pr_debug("... APIC ARBPRI: %08x (%02x)\n",
1230				 v, v & APIC_ARBPRI_MASK);
1231		}
1232		v = apic_read(APIC_PROCPRI);
1233		pr_debug("... APIC PROCPRI: %08x\n", v);
1234	}
1235
1236	/*
1237	 * Remote read supported only in the 82489DX and local APIC for
1238	 * Pentium processors.
1239	 */
1240	if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1241		v = apic_read(APIC_RRR);
1242		pr_debug("... APIC RRR: %08x\n", v);
1243	}
1244
1245	v = apic_read(APIC_LDR);
1246	pr_debug("... APIC LDR: %08x\n", v);
1247	if (!x2apic_enabled()) {
1248		v = apic_read(APIC_DFR);
1249		pr_debug("... APIC DFR: %08x\n", v);
1250	}
1251	v = apic_read(APIC_SPIV);
1252	pr_debug("... APIC SPIV: %08x\n", v);
1253
1254	pr_debug("... APIC ISR field:\n");
1255	print_APIC_field(APIC_ISR);
1256	pr_debug("... APIC TMR field:\n");
1257	print_APIC_field(APIC_TMR);
1258	pr_debug("... APIC IRR field:\n");
1259	print_APIC_field(APIC_IRR);
1260
1261	/* !82489DX */
1262	if (APIC_INTEGRATED(ver)) {
1263		/* Due to the Pentium erratum 3AP. */
1264		if (maxlvt > 3)
1265			apic_write(APIC_ESR, 0);
1266
1267		v = apic_read(APIC_ESR);
1268		pr_debug("... APIC ESR: %08x\n", v);
1269	}
1270
1271	icr = apic_icr_read();
1272	pr_debug("... APIC ICR: %08x\n", (u32)icr);
1273	pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
1274
1275	v = apic_read(APIC_LVTT);
1276	pr_debug("... APIC LVTT: %08x\n", v);
1277
1278	if (maxlvt > 3) {
1279		/* PC is LVT#4. */
1280		v = apic_read(APIC_LVTPC);
1281		pr_debug("... APIC LVTPC: %08x\n", v);
1282	}
1283	v = apic_read(APIC_LVT0);
1284	pr_debug("... APIC LVT0: %08x\n", v);
1285	v = apic_read(APIC_LVT1);
1286	pr_debug("... APIC LVT1: %08x\n", v);
1287
1288	if (maxlvt > 2) {
1289		/* ERR is LVT#3. */
1290		v = apic_read(APIC_LVTERR);
1291		pr_debug("... APIC LVTERR: %08x\n", v);
1292	}
1293
1294	v = apic_read(APIC_TMICT);
1295	pr_debug("... APIC TMICT: %08x\n", v);
1296	v = apic_read(APIC_TMCCT);
1297	pr_debug("... APIC TMCCT: %08x\n", v);
1298	v = apic_read(APIC_TDCR);
1299	pr_debug("... APIC TDCR: %08x\n", v);
1300
1301	if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1302		v = apic_read(APIC_EFEAT);
1303		maxlvt = (v >> 16) & 0xff;
1304		pr_debug("... APIC EFEAT: %08x\n", v);
1305		v = apic_read(APIC_ECTRL);
1306		pr_debug("... APIC ECTRL: %08x\n", v);
1307		for (i = 0; i < maxlvt; i++) {
1308			v = apic_read(APIC_EILVTn(i));
1309			pr_debug("... APIC EILVT%d: %08x\n", i, v);
1310		}
1311	}
1312	pr_cont("\n");
1313}
1314
1315static void __init print_local_APICs(int maxcpu)
1316{
1317	int cpu;
1318
1319	if (!maxcpu)
1320		return;
1321
1322	preempt_disable();
1323	for_each_online_cpu(cpu) {
1324		if (cpu >= maxcpu)
1325			break;
1326		smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1327	}
1328	preempt_enable();
1329}
1330
1331static void __init print_PIC(void)
1332{
1333	unsigned int v;
1334	unsigned long flags;
1335
1336	if (!nr_legacy_irqs())
1337		return;
1338
1339	pr_debug("\nprinting PIC contents\n");
1340
1341	raw_spin_lock_irqsave(&i8259A_lock, flags);
1342
1343	v = inb(0xa1) << 8 | inb(0x21);
1344	pr_debug("... PIC  IMR: %04x\n", v);
1345
1346	v = inb(0xa0) << 8 | inb(0x20);
1347	pr_debug("... PIC  IRR: %04x\n", v);
1348
1349	outb(0x0b, 0xa0);
1350	outb(0x0b, 0x20);
1351	v = inb(0xa0) << 8 | inb(0x20);
1352	outb(0x0a, 0xa0);
1353	outb(0x0a, 0x20);
1354
1355	raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1356
1357	pr_debug("... PIC  ISR: %04x\n", v);
1358
1359	v = inb(PIC_ELCR2) << 8 | inb(PIC_ELCR1);
1360	pr_debug("... PIC ELCR: %04x\n", v);
1361}
1362
1363static int show_lapic __initdata = 1;
1364static __init int setup_show_lapic(char *arg)
1365{
1366	int num = -1;
1367
1368	if (strcmp(arg, "all") == 0) {
1369		show_lapic = CONFIG_NR_CPUS;
1370	} else {
1371		get_option(&arg, &num);
1372		if (num >= 0)
1373			show_lapic = num;
1374	}
1375
1376	return 1;
1377}
1378__setup("show_lapic=", setup_show_lapic);
1379
1380static int __init print_ICs(void)
1381{
1382	if (apic_verbosity == APIC_QUIET)
1383		return 0;
1384
1385	print_PIC();
1386
1387	/* don't print out if apic is not there */
1388	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
1389		return 0;
1390
1391	print_local_APICs(show_lapic);
1392	print_IO_APICs();
1393
1394	return 0;
1395}
1396
1397late_initcall(print_ICs);