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);
v4.10.11
 
  1/*
  2 * Local APIC related interfaces to support IOAPIC, MSI, HT_IRQ etc.
  3 *
  4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
  5 *	Moved from arch/x86/kernel/apic/io_apic.c.
  6 * Jiang Liu <jiang.liu@linux.intel.com>
  7 *	Enable support of hierarchical irqdomains
  8 *
  9 * This program is free software; you can redistribute it and/or modify
 10 * it under the terms of the GNU General Public License version 2 as
 11 * published by the Free Software Foundation.
 12 */
 13#include <linux/interrupt.h>
 
 
 14#include <linux/init.h>
 15#include <linux/compiler.h>
 16#include <linux/slab.h>
 17#include <asm/irqdomain.h>
 18#include <asm/hw_irq.h>
 
 19#include <asm/apic.h>
 20#include <asm/i8259.h>
 21#include <asm/desc.h>
 22#include <asm/irq_remapping.h>
 23
 
 
 24struct apic_chip_data {
 25	struct irq_cfg		cfg;
 26	cpumask_var_t		domain;
 27	cpumask_var_t		old_domain;
 28	u8			move_in_progress : 1;
 
 
 
 
 
 
 
 29};
 30
 31struct irq_domain *x86_vector_domain;
 32EXPORT_SYMBOL_GPL(x86_vector_domain);
 33static DEFINE_RAW_SPINLOCK(vector_lock);
 34static cpumask_var_t vector_cpumask, vector_searchmask, searched_cpumask;
 35static struct irq_chip lapic_controller;
 36#ifdef	CONFIG_X86_IO_APIC
 37static struct apic_chip_data *legacy_irq_data[NR_IRQS_LEGACY];
 
 38#endif
 39
 40void lock_vector_lock(void)
 41{
 42	/* Used to the online set of cpus does not change
 43	 * during assign_irq_vector.
 44	 */
 45	raw_spin_lock(&vector_lock);
 46}
 47
 48void unlock_vector_lock(void)
 49{
 50	raw_spin_unlock(&vector_lock);
 51}
 52
 53static struct apic_chip_data *apic_chip_data(struct irq_data *irq_data)
 
 54{
 55	if (!irq_data)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 56		return NULL;
 57
 58	while (irq_data->parent_data)
 59		irq_data = irq_data->parent_data;
 60
 61	return irq_data->chip_data;
 62}
 63
 64struct irq_cfg *irqd_cfg(struct irq_data *irq_data)
 65{
 66	struct apic_chip_data *data = apic_chip_data(irq_data);
 67
 68	return data ? &data->cfg : NULL;
 69}
 70EXPORT_SYMBOL_GPL(irqd_cfg);
 71
 72struct irq_cfg *irq_cfg(unsigned int irq)
 73{
 74	return irqd_cfg(irq_get_irq_data(irq));
 75}
 76
 77static struct apic_chip_data *alloc_apic_chip_data(int node)
 78{
 79	struct apic_chip_data *data;
 80
 81	data = kzalloc_node(sizeof(*data), GFP_KERNEL, node);
 82	if (!data)
 83		return NULL;
 84	if (!zalloc_cpumask_var_node(&data->domain, GFP_KERNEL, node))
 85		goto out_data;
 86	if (!zalloc_cpumask_var_node(&data->old_domain, GFP_KERNEL, node))
 87		goto out_domain;
 88	return data;
 89out_domain:
 90	free_cpumask_var(data->domain);
 91out_data:
 92	kfree(data);
 93	return NULL;
 94}
 95
 96static void free_apic_chip_data(struct apic_chip_data *data)
 97{
 98	if (data) {
 99		free_cpumask_var(data->domain);
100		free_cpumask_var(data->old_domain);
101		kfree(data);
102	}
103}
104
105static int __assign_irq_vector(int irq, struct apic_chip_data *d,
106			       const struct cpumask *mask)
107{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
108	/*
109	 * NOTE! The local APIC isn't very good at handling
110	 * multiple interrupts at the same interrupt level.
111	 * As the interrupt level is determined by taking the
112	 * vector number and shifting that right by 4, we
113	 * want to spread these out a bit so that they don't
114	 * all fall in the same interrupt level.
115	 *
116	 * Also, we've got to be careful not to trash gate
117	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
118	 */
119	static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
120	static int current_offset = VECTOR_OFFSET_START % 16;
121	int cpu, vector;
122
123	/*
124	 * If there is still a move in progress or the previous move has not
125	 * been cleaned up completely, tell the caller to come back later.
 
 
 
 
126	 */
127	if (d->move_in_progress ||
128	    cpumask_intersects(d->old_domain, cpu_online_mask))
129		return -EBUSY;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
130
131	/* Only try and allocate irqs on cpus that are present */
132	cpumask_clear(d->old_domain);
133	cpumask_clear(searched_cpumask);
134	cpu = cpumask_first_and(mask, cpu_online_mask);
135	while (cpu < nr_cpu_ids) {
136		int new_cpu, offset;
137
138		/* Get the possible target cpus for @mask/@cpu from the apic */
139		apic->vector_allocation_domain(cpu, vector_cpumask, mask);
 
 
 
140
141		/*
142		 * Clear the offline cpus from @vector_cpumask for searching
143		 * and verify whether the result overlaps with @mask. If true,
144		 * then the call to apic->cpu_mask_to_apicid_and() will
145		 * succeed as well. If not, no point in trying to find a
146		 * vector in this mask.
147		 */
148		cpumask_and(vector_searchmask, vector_cpumask, cpu_online_mask);
149		if (!cpumask_intersects(vector_searchmask, mask))
150			goto next_cpu;
151
152		if (cpumask_subset(vector_cpumask, d->domain)) {
153			if (cpumask_equal(vector_cpumask, d->domain))
154				goto success;
155			/*
156			 * Mark the cpus which are not longer in the mask for
157			 * cleanup.
158			 */
159			cpumask_andnot(d->old_domain, d->domain, vector_cpumask);
160			vector = d->cfg.vector;
161			goto update;
162		}
163
164		vector = current_vector;
165		offset = current_offset;
166next:
167		vector += 16;
168		if (vector >= first_system_vector) {
169			offset = (offset + 1) % 16;
170			vector = FIRST_EXTERNAL_VECTOR + offset;
171		}
172
173		/* If the search wrapped around, try the next cpu */
174		if (unlikely(current_vector == vector))
175			goto next_cpu;
176
177		if (test_bit(vector, used_vectors))
178			goto next;
179
180		for_each_cpu(new_cpu, vector_searchmask) {
181			if (!IS_ERR_OR_NULL(per_cpu(vector_irq, new_cpu)[vector]))
182				goto next;
183		}
184		/* Found one! */
185		current_vector = vector;
186		current_offset = offset;
187		/* Schedule the old vector for cleanup on all cpus */
188		if (d->cfg.vector)
189			cpumask_copy(d->old_domain, d->domain);
190		for_each_cpu(new_cpu, vector_searchmask)
191			per_cpu(vector_irq, new_cpu)[vector] = irq_to_desc(irq);
192		goto update;
193
194next_cpu:
195		/*
196		 * We exclude the current @vector_cpumask from the requested
197		 * @mask and try again with the next online cpu in the
198		 * result. We cannot modify @mask, so we use @vector_cpumask
199		 * as a temporary buffer here as it will be reassigned when
200		 * calling apic->vector_allocation_domain() above.
201		 */
202		cpumask_or(searched_cpumask, searched_cpumask, vector_cpumask);
203		cpumask_andnot(vector_cpumask, mask, searched_cpumask);
204		cpu = cpumask_first_and(vector_cpumask, cpu_online_mask);
205		continue;
206	}
207	return -ENOSPC;
208
209update:
210	/*
211	 * Exclude offline cpus from the cleanup mask and set the
212	 * move_in_progress flag when the result is not empty.
213	 */
214	cpumask_and(d->old_domain, d->old_domain, cpu_online_mask);
215	d->move_in_progress = !cpumask_empty(d->old_domain);
216	d->cfg.old_vector = d->move_in_progress ? d->cfg.vector : 0;
217	d->cfg.vector = vector;
218	cpumask_copy(d->domain, vector_cpumask);
219success:
220	/*
221	 * Cache destination APIC IDs into cfg->dest_apicid. This cannot fail
222	 * as we already established, that mask & d->domain & cpu_online_mask
223	 * is not empty.
224	 */
225	BUG_ON(apic->cpu_mask_to_apicid_and(mask, d->domain,
226					    &d->cfg.dest_apicid));
 
 
 
 
 
 
 
 
227	return 0;
228}
229
230static int assign_irq_vector(int irq, struct apic_chip_data *data,
231			     const struct cpumask *mask)
232{
233	int err;
234	unsigned long flags;
 
235
236	raw_spin_lock_irqsave(&vector_lock, flags);
237	err = __assign_irq_vector(irq, data, mask);
 
238	raw_spin_unlock_irqrestore(&vector_lock, flags);
239	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
240}
241
242static int assign_irq_vector_policy(int irq, int node,
243				    struct apic_chip_data *data,
244				    struct irq_alloc_info *info)
245{
246	if (info && info->mask)
247		return assign_irq_vector(irq, data, info->mask);
248	if (node != NUMA_NO_NODE &&
249	    assign_irq_vector(irq, data, cpumask_of_node(node)) == 0)
 
 
 
250		return 0;
251	return assign_irq_vector(irq, data, apic->target_cpus());
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
252}
253
254static void clear_irq_vector(int irq, struct apic_chip_data *data)
255{
256	struct irq_desc *desc;
257	int cpu, vector;
 
 
 
258
259	if (!data->cfg.vector)
 
 
 
 
260		return;
261
262	vector = data->cfg.vector;
263	for_each_cpu_and(cpu, data->domain, cpu_online_mask)
264		per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
 
 
 
 
 
265
266	data->cfg.vector = 0;
267	cpumask_clear(data->domain);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
268
269	/*
270	 * If move is in progress or the old_domain mask is not empty,
271	 * i.e. the cleanup IPI has not been processed yet, we need to remove
272	 * the old references to desc from all cpus vector tables.
273	 */
274	if (!data->move_in_progress && cpumask_empty(data->old_domain))
275		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
276
277	desc = irq_to_desc(irq);
278	for_each_cpu_and(cpu, data->old_domain, cpu_online_mask) {
279		for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
280		     vector++) {
281			if (per_cpu(vector_irq, cpu)[vector] != desc)
282				continue;
283			per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
284			break;
285		}
286	}
287	data->move_in_progress = 0;
288}
289
290void init_irq_alloc_info(struct irq_alloc_info *info,
291			 const struct cpumask *mask)
292{
293	memset(info, 0, sizeof(*info));
294	info->mask = mask;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
295}
296
297void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
298{
299	if (src)
300		*dst = *src;
301	else
302		memset(dst, 0, sizeof(*dst));
 
 
 
 
 
 
303}
304
305static void x86_vector_free_irqs(struct irq_domain *domain,
306				 unsigned int virq, unsigned int nr_irqs)
307{
308	struct apic_chip_data *apic_data;
309	struct irq_data *irq_data;
310	unsigned long flags;
311	int i;
312
313	for (i = 0; i < nr_irqs; i++) {
314		irq_data = irq_domain_get_irq_data(x86_vector_domain, virq + i);
315		if (irq_data && irq_data->chip_data) {
316			raw_spin_lock_irqsave(&vector_lock, flags);
317			clear_irq_vector(virq + i, irq_data->chip_data);
318			apic_data = irq_data->chip_data;
319			irq_domain_reset_irq_data(irq_data);
 
320			raw_spin_unlock_irqrestore(&vector_lock, flags);
321			free_apic_chip_data(apic_data);
322#ifdef	CONFIG_X86_IO_APIC
323			if (virq + i < nr_legacy_irqs())
324				legacy_irq_data[virq + i] = NULL;
325#endif
326		}
327	}
328}
329
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
330static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
331				 unsigned int nr_irqs, void *arg)
332{
333	struct irq_alloc_info *info = arg;
334	struct apic_chip_data *data;
335	struct irq_data *irq_data;
336	int i, err, node;
337
338	if (disable_apic)
339		return -ENXIO;
340
341	/* Currently vector allocator can't guarantee contiguous allocations */
342	if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
343		return -ENOSYS;
344
345	for (i = 0; i < nr_irqs; i++) {
346		irq_data = irq_domain_get_irq_data(domain, virq + i);
347		BUG_ON(!irq_data);
348		node = irq_data_get_node(irq_data);
349#ifdef	CONFIG_X86_IO_APIC
350		if (virq + i < nr_legacy_irqs() && legacy_irq_data[virq + i])
351			data = legacy_irq_data[virq + i];
352		else
353#endif
354			data = alloc_apic_chip_data(node);
355		if (!data) {
356			err = -ENOMEM;
357			goto error;
358		}
359
360		irq_data->chip = &lapic_controller;
361		irq_data->chip_data = data;
362		irq_data->hwirq = virq + i;
363		err = assign_irq_vector_policy(virq + i, node, data, info);
364		if (err)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
365			goto error;
 
366	}
367
368	return 0;
369
370error:
371	x86_vector_free_irqs(domain, virq, i + 1);
372	return err;
373}
374
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
375static const struct irq_domain_ops x86_vector_domain_ops = {
376	.alloc	= x86_vector_alloc_irqs,
377	.free	= x86_vector_free_irqs,
 
 
 
 
 
378};
379
380int __init arch_probe_nr_irqs(void)
381{
382	int nr;
383
384	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
385		nr_irqs = NR_VECTORS * nr_cpu_ids;
386
387	nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
388#if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
389	/*
390	 * for MSI and HT dyn irq
391	 */
392	if (gsi_top <= NR_IRQS_LEGACY)
393		nr +=  8 * nr_cpu_ids;
394	else
395		nr += gsi_top * 16;
396#endif
397	if (nr < nr_irqs)
398		nr_irqs = nr;
399
400	/*
401	 * We don't know if PIC is present at this point so we need to do
402	 * probe() to get the right number of legacy IRQs.
403	 */
404	return legacy_pic->probe();
405}
406
407#ifdef	CONFIG_X86_IO_APIC
408static void init_legacy_irqs(void)
409{
410	int i, node = cpu_to_node(0);
411	struct apic_chip_data *data;
412
413	/*
414	 * For legacy IRQ's, start with assigning irq0 to irq15 to
415	 * ISA_IRQ_VECTOR(i) for all cpu's.
 
416	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
417	for (i = 0; i < nr_legacy_irqs(); i++) {
418		data = legacy_irq_data[i] = alloc_apic_chip_data(node);
419		BUG_ON(!data);
420
421		data->cfg.vector = ISA_IRQ_VECTOR(i);
422		cpumask_setall(data->domain);
423		irq_set_chip_data(i, data);
424	}
425}
426#else
427static void init_legacy_irqs(void) { }
428#endif
429
430int __init arch_early_irq_init(void)
431{
432	init_legacy_irqs();
433
434	x86_vector_domain = irq_domain_add_tree(NULL, &x86_vector_domain_ops,
435						NULL);
 
 
436	BUG_ON(x86_vector_domain == NULL);
 
437	irq_set_default_host(x86_vector_domain);
438
439	arch_init_msi_domain(x86_vector_domain);
440	arch_init_htirq_domain(x86_vector_domain);
441
442	BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
443	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
444	BUG_ON(!alloc_cpumask_var(&searched_cpumask, GFP_KERNEL));
 
 
 
 
 
 
 
445
446	return arch_early_ioapic_init();
447}
448
449/* Initialize vector_irq on a new cpu */
450static void __setup_vector_irq(int cpu)
 
451{
452	struct apic_chip_data *data;
453	struct irq_desc *desc;
454	int irq, vector;
455
456	/* Mark the inuse vectors */
457	for_each_irq_desc(irq, desc) {
458		struct irq_data *idata = irq_desc_get_irq_data(desc);
459
460		data = apic_chip_data(idata);
461		if (!data || !cpumask_test_cpu(cpu, data->domain))
462			continue;
463		vector = data->cfg.vector;
464		per_cpu(vector_irq, cpu)[vector] = desc;
465	}
466	/* Mark the free vectors */
467	for (vector = 0; vector < NR_VECTORS; ++vector) {
468		desc = per_cpu(vector_irq, cpu)[vector];
469		if (IS_ERR_OR_NULL(desc))
470			continue;
471
472		data = apic_chip_data(irq_desc_get_irq_data(desc));
473		if (!cpumask_test_cpu(cpu, data->domain))
474			per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
475	}
 
 
 
476}
477
478/*
479 * Setup the vector to irq mappings. Must be called with vector_lock held.
480 */
481void setup_vector_irq(int cpu)
482{
483	int irq;
484
485	lockdep_assert_held(&vector_lock);
 
 
 
 
486	/*
487	 * On most of the platforms, legacy PIC delivers the interrupts on the
488	 * boot cpu. But there are certain platforms where PIC interrupts are
489	 * delivered to multiple cpu's. If the legacy IRQ is handled by the
490	 * legacy PIC, for the new cpu that is coming online, setup the static
491	 * legacy vector to irq mapping:
 
 
492	 */
493	for (irq = 0; irq < nr_legacy_irqs(); irq++)
494		per_cpu(vector_irq, cpu)[ISA_IRQ_VECTOR(irq)] = irq_to_desc(irq);
 
495
496	__setup_vector_irq(cpu);
 
 
 
 
497}
498
499static int apic_retrigger_irq(struct irq_data *irq_data)
 
500{
501	struct apic_chip_data *data = apic_chip_data(irq_data);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
502	unsigned long flags;
503	int cpu;
504
505	raw_spin_lock_irqsave(&vector_lock, flags);
506	cpu = cpumask_first_and(data->domain, cpu_online_mask);
507	apic->send_IPI_mask(cpumask_of(cpu), data->cfg.vector);
508	raw_spin_unlock_irqrestore(&vector_lock, flags);
509
510	return 1;
511}
512
513void apic_ack_edge(struct irq_data *data)
514{
515	irq_complete_move(irqd_cfg(data));
516	irq_move_irq(data);
517	ack_APIC_irq();
518}
519
520static int apic_set_affinity(struct irq_data *irq_data,
521			     const struct cpumask *dest, bool force)
522{
523	struct apic_chip_data *data = irq_data->chip_data;
524	int err, irq = irq_data->irq;
525
526	if (!IS_ENABLED(CONFIG_SMP))
527		return -EPERM;
528
529	if (!cpumask_intersects(dest, cpu_online_mask))
530		return -EINVAL;
531
532	err = assign_irq_vector(irq, data, dest);
533	return err ? err : IRQ_SET_MASK_OK;
534}
535
536static struct irq_chip lapic_controller = {
 
537	.irq_ack		= apic_ack_edge,
538	.irq_set_affinity	= apic_set_affinity,
539	.irq_retrigger		= apic_retrigger_irq,
540};
541
542#ifdef CONFIG_SMP
543static void __send_cleanup_vector(struct apic_chip_data *data)
 
544{
545	raw_spin_lock(&vector_lock);
546	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
547	data->move_in_progress = 0;
548	if (!cpumask_empty(data->old_domain))
549		apic->send_IPI_mask(data->old_domain, IRQ_MOVE_CLEANUP_VECTOR);
550	raw_spin_unlock(&vector_lock);
551}
552
553void send_cleanup_vector(struct irq_cfg *cfg)
554{
555	struct apic_chip_data *data;
 
 
 
 
556
557	data = container_of(cfg, struct apic_chip_data, cfg);
558	if (data->move_in_progress)
559		__send_cleanup_vector(data);
 
 
 
560}
561
562asmlinkage __visible void smp_irq_move_cleanup_interrupt(void)
563{
564	unsigned vector, me;
 
 
565
566	entering_ack_irq();
567
568	/* Prevent vectors vanishing under us */
569	raw_spin_lock(&vector_lock);
570
571	me = smp_processor_id();
572	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
573		struct apic_chip_data *data;
574		struct irq_desc *desc;
575		unsigned int irr;
576
577	retry:
578		desc = __this_cpu_read(vector_irq[vector]);
579		if (IS_ERR_OR_NULL(desc))
580			continue;
581
582		if (!raw_spin_trylock(&desc->lock)) {
583			raw_spin_unlock(&vector_lock);
584			cpu_relax();
585			raw_spin_lock(&vector_lock);
586			goto retry;
587		}
588
589		data = apic_chip_data(irq_desc_get_irq_data(desc));
590		if (!data)
591			goto unlock;
592
593		/*
594		 * Nothing to cleanup if irq migration is in progress
595		 * or this cpu is not set in the cleanup mask.
596		 */
597		if (data->move_in_progress ||
598		    !cpumask_test_cpu(me, data->old_domain))
599			goto unlock;
600
601		/*
602		 * We have two cases to handle here:
603		 * 1) vector is unchanged but the target mask got reduced
604		 * 2) vector and the target mask has changed
605		 *
606		 * #1 is obvious, but in #2 we have two vectors with the same
607		 * irq descriptor: the old and the new vector. So we need to
608		 * make sure that we only cleanup the old vector. The new
609		 * vector has the current @vector number in the config and
610		 * this cpu is part of the target mask. We better leave that
611		 * one alone.
612		 */
613		if (vector == data->cfg.vector &&
614		    cpumask_test_cpu(me, data->domain))
615			goto unlock;
616
617		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
618		/*
619		 * Check if the vector that needs to be cleanedup is
620		 * registered at the cpu's IRR. If so, then this is not
621		 * the best time to clean it up. Lets clean it up in the
622		 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
623		 * to myself.
624		 */
625		if (irr  & (1 << (vector % 32))) {
626			apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
627			goto unlock;
628		}
629		__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
630		cpumask_clear_cpu(me, data->old_domain);
631unlock:
632		raw_spin_unlock(&desc->lock);
633	}
634
635	raw_spin_unlock(&vector_lock);
 
 
636
637	exiting_irq();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
638}
639
640static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
641{
642	unsigned me;
643	struct apic_chip_data *data;
644
645	data = container_of(cfg, struct apic_chip_data, cfg);
646	if (likely(!data->move_in_progress))
647		return;
648
649	me = smp_processor_id();
650	if (vector == data->cfg.vector && cpumask_test_cpu(me, data->domain))
651		__send_cleanup_vector(data);
652}
653
654void irq_complete_move(struct irq_cfg *cfg)
655{
656	__irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
657}
658
659/*
660 * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
661 */
662void irq_force_complete_move(struct irq_desc *desc)
663{
664	struct irq_data *irqdata;
665	struct apic_chip_data *data;
666	struct irq_cfg *cfg;
667	unsigned int cpu;
668
669	/*
670	 * The function is called for all descriptors regardless of which
671	 * irqdomain they belong to. For example if an IRQ is provided by
672	 * an irq_chip as part of a GPIO driver, the chip data for that
673	 * descriptor is specific to the irq_chip in question.
674	 *
675	 * Check first that the chip_data is what we expect
676	 * (apic_chip_data) before touching it any further.
677	 */
678	irqdata = irq_domain_get_irq_data(x86_vector_domain,
679					  irq_desc_get_irq(desc));
680	if (!irqdata)
681		return;
682
683	data = apic_chip_data(irqdata);
684	cfg = data ? &data->cfg : NULL;
 
 
685
686	if (!cfg)
687		return;
 
 
 
 
688
689	/*
690	 * This is tricky. If the cleanup of @data->old_domain has not been
691	 * done yet, then the following setaffinity call will fail with
692	 * -EBUSY. This can leave the interrupt in a stale state.
693	 *
694	 * All CPUs are stuck in stop machine with interrupts disabled so
695	 * calling __irq_complete_move() would be completely pointless.
696	 */
697	raw_spin_lock(&vector_lock);
698	/*
699	 * Clean out all offline cpus (including the outgoing one) from the
700	 * old_domain mask.
701	 */
702	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
703
704	/*
705	 * If move_in_progress is cleared and the old_domain mask is empty,
706	 * then there is nothing to cleanup. fixup_irqs() will take care of
707	 * the stale vectors on the outgoing cpu.
708	 */
709	if (!data->move_in_progress && cpumask_empty(data->old_domain)) {
710		raw_spin_unlock(&vector_lock);
711		return;
712	}
713
714	/*
715	 * 1) The interrupt is in move_in_progress state. That means that we
716	 *    have not seen an interrupt since the io_apic was reprogrammed to
717	 *    the new vector.
718	 *
719	 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
720	 *    have not been processed yet.
721	 */
722	if (data->move_in_progress) {
723		/*
724		 * In theory there is a race:
725		 *
726		 * set_ioapic(new_vector) <-- Interrupt is raised before update
727		 *			      is effective, i.e. it's raised on
728		 *			      the old vector.
729		 *
730		 * So if the target cpu cannot handle that interrupt before
731		 * the old vector is cleaned up, we get a spurious interrupt
732		 * and in the worst case the ioapic irq line becomes stale.
733		 *
734		 * But in case of cpu hotplug this should be a non issue
735		 * because if the affinity update happens right before all
736		 * cpus rendevouz in stop machine, there is no way that the
737		 * interrupt can be blocked on the target cpu because all cpus
738		 * loops first with interrupts enabled in stop machine, so the
739		 * old vector is not yet cleaned up when the interrupt fires.
740		 *
741		 * So the only way to run into this issue is if the delivery
742		 * of the interrupt on the apic/system bus would be delayed
743		 * beyond the point where the target cpu disables interrupts
744		 * in stop machine. I doubt that it can happen, but at least
745		 * there is a theroretical chance. Virtualization might be
746		 * able to expose this, but AFAICT the IOAPIC emulation is not
747		 * as stupid as the real hardware.
748		 *
749		 * Anyway, there is nothing we can do about that at this point
750		 * w/o refactoring the whole fixup_irq() business completely.
751		 * We print at least the irq number and the old vector number,
752		 * so we have the necessary information when a problem in that
753		 * area arises.
754		 */
755		pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
756			irqdata->irq, cfg->old_vector);
757	}
758	/*
759	 * If old_domain is not empty, then other cpus still have the irq
760	 * descriptor set in their vector array. Clean it up.
761	 */
762	for_each_cpu(cpu, data->old_domain)
763		per_cpu(vector_irq, cpu)[cfg->old_vector] = VECTOR_UNUSED;
 
 
 
 
 
 
 
 
764
765	/* Cleanup the left overs of the (half finished) move */
766	cpumask_clear(data->old_domain);
767	data->move_in_progress = 0;
 
 
 
 
 
 
 
 
 
 
 
 
768	raw_spin_unlock(&vector_lock);
 
769}
770#endif
 
771
772static void __init print_APIC_field(int base)
773{
774	int i;
775
776	printk(KERN_DEBUG);
777
778	for (i = 0; i < 8; i++)
779		pr_cont("%08x", apic_read(base + i*0x10));
780
781	pr_cont("\n");
782}
783
784static void __init print_local_APIC(void *dummy)
785{
786	unsigned int i, v, ver, maxlvt;
787	u64 icr;
788
789	pr_debug("printing local APIC contents on CPU#%d/%d:\n",
790		 smp_processor_id(), hard_smp_processor_id());
791	v = apic_read(APIC_ID);
792	pr_info("... APIC ID:      %08x (%01x)\n", v, read_apic_id());
793	v = apic_read(APIC_LVR);
794	pr_info("... APIC VERSION: %08x\n", v);
795	ver = GET_APIC_VERSION(v);
796	maxlvt = lapic_get_maxlvt();
797
798	v = apic_read(APIC_TASKPRI);
799	pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
800
801	/* !82489DX */
802	if (APIC_INTEGRATED(ver)) {
803		if (!APIC_XAPIC(ver)) {
804			v = apic_read(APIC_ARBPRI);
805			pr_debug("... APIC ARBPRI: %08x (%02x)\n",
806				 v, v & APIC_ARBPRI_MASK);
807		}
808		v = apic_read(APIC_PROCPRI);
809		pr_debug("... APIC PROCPRI: %08x\n", v);
810	}
811
812	/*
813	 * Remote read supported only in the 82489DX and local APIC for
814	 * Pentium processors.
815	 */
816	if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
817		v = apic_read(APIC_RRR);
818		pr_debug("... APIC RRR: %08x\n", v);
819	}
820
821	v = apic_read(APIC_LDR);
822	pr_debug("... APIC LDR: %08x\n", v);
823	if (!x2apic_enabled()) {
824		v = apic_read(APIC_DFR);
825		pr_debug("... APIC DFR: %08x\n", v);
826	}
827	v = apic_read(APIC_SPIV);
828	pr_debug("... APIC SPIV: %08x\n", v);
829
830	pr_debug("... APIC ISR field:\n");
831	print_APIC_field(APIC_ISR);
832	pr_debug("... APIC TMR field:\n");
833	print_APIC_field(APIC_TMR);
834	pr_debug("... APIC IRR field:\n");
835	print_APIC_field(APIC_IRR);
836
837	/* !82489DX */
838	if (APIC_INTEGRATED(ver)) {
839		/* Due to the Pentium erratum 3AP. */
840		if (maxlvt > 3)
841			apic_write(APIC_ESR, 0);
842
843		v = apic_read(APIC_ESR);
844		pr_debug("... APIC ESR: %08x\n", v);
845	}
846
847	icr = apic_icr_read();
848	pr_debug("... APIC ICR: %08x\n", (u32)icr);
849	pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
850
851	v = apic_read(APIC_LVTT);
852	pr_debug("... APIC LVTT: %08x\n", v);
853
854	if (maxlvt > 3) {
855		/* PC is LVT#4. */
856		v = apic_read(APIC_LVTPC);
857		pr_debug("... APIC LVTPC: %08x\n", v);
858	}
859	v = apic_read(APIC_LVT0);
860	pr_debug("... APIC LVT0: %08x\n", v);
861	v = apic_read(APIC_LVT1);
862	pr_debug("... APIC LVT1: %08x\n", v);
863
864	if (maxlvt > 2) {
865		/* ERR is LVT#3. */
866		v = apic_read(APIC_LVTERR);
867		pr_debug("... APIC LVTERR: %08x\n", v);
868	}
869
870	v = apic_read(APIC_TMICT);
871	pr_debug("... APIC TMICT: %08x\n", v);
872	v = apic_read(APIC_TMCCT);
873	pr_debug("... APIC TMCCT: %08x\n", v);
874	v = apic_read(APIC_TDCR);
875	pr_debug("... APIC TDCR: %08x\n", v);
876
877	if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
878		v = apic_read(APIC_EFEAT);
879		maxlvt = (v >> 16) & 0xff;
880		pr_debug("... APIC EFEAT: %08x\n", v);
881		v = apic_read(APIC_ECTRL);
882		pr_debug("... APIC ECTRL: %08x\n", v);
883		for (i = 0; i < maxlvt; i++) {
884			v = apic_read(APIC_EILVTn(i));
885			pr_debug("... APIC EILVT%d: %08x\n", i, v);
886		}
887	}
888	pr_cont("\n");
889}
890
891static void __init print_local_APICs(int maxcpu)
892{
893	int cpu;
894
895	if (!maxcpu)
896		return;
897
898	preempt_disable();
899	for_each_online_cpu(cpu) {
900		if (cpu >= maxcpu)
901			break;
902		smp_call_function_single(cpu, print_local_APIC, NULL, 1);
903	}
904	preempt_enable();
905}
906
907static void __init print_PIC(void)
908{
909	unsigned int v;
910	unsigned long flags;
911
912	if (!nr_legacy_irqs())
913		return;
914
915	pr_debug("\nprinting PIC contents\n");
916
917	raw_spin_lock_irqsave(&i8259A_lock, flags);
918
919	v = inb(0xa1) << 8 | inb(0x21);
920	pr_debug("... PIC  IMR: %04x\n", v);
921
922	v = inb(0xa0) << 8 | inb(0x20);
923	pr_debug("... PIC  IRR: %04x\n", v);
924
925	outb(0x0b, 0xa0);
926	outb(0x0b, 0x20);
927	v = inb(0xa0) << 8 | inb(0x20);
928	outb(0x0a, 0xa0);
929	outb(0x0a, 0x20);
930
931	raw_spin_unlock_irqrestore(&i8259A_lock, flags);
932
933	pr_debug("... PIC  ISR: %04x\n", v);
934
935	v = inb(0x4d1) << 8 | inb(0x4d0);
936	pr_debug("... PIC ELCR: %04x\n", v);
937}
938
939static int show_lapic __initdata = 1;
940static __init int setup_show_lapic(char *arg)
941{
942	int num = -1;
943
944	if (strcmp(arg, "all") == 0) {
945		show_lapic = CONFIG_NR_CPUS;
946	} else {
947		get_option(&arg, &num);
948		if (num >= 0)
949			show_lapic = num;
950	}
951
952	return 1;
953}
954__setup("show_lapic=", setup_show_lapic);
955
956static int __init print_ICs(void)
957{
958	if (apic_verbosity == APIC_QUIET)
959		return 0;
960
961	print_PIC();
962
963	/* don't print out if apic is not there */
964	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
965		return 0;
966
967	print_local_APICs(show_lapic);
968	print_IO_APICs();
969
970	return 0;
971}
972
973late_initcall(print_ICs);