1 // SPDX-License-Identifier: GPL-2.0
2 
3 #define pr_fmt(fmt)     "DMAR-IR: " fmt
4 
5 #include <linux/interrupt.h>
6 #include <linux/dmar.h>
7 #include <linux/spinlock.h>
8 #include <linux/slab.h>
9 #include <linux/jiffies.h>
10 #include <linux/hpet.h>
11 #include <linux/pci.h>
12 #include <linux/irq.h>
13 #include <linux/intel-iommu.h>
14 #include <linux/acpi.h>
15 #include <linux/irqdomain.h>
16 #include <linux/crash_dump.h>
17 #include <asm/io_apic.h>
18 #include <asm/apic.h>
19 #include <asm/smp.h>
20 #include <asm/cpu.h>
21 #include <asm/irq_remapping.h>
22 #include <asm/pci-direct.h>
23 
24 #include "../irq_remapping.h"
25 #include "cap_audit.h"
26 
27 enum irq_mode {
28 	IRQ_REMAPPING,
29 	IRQ_POSTING,
30 };
31 
32 struct ioapic_scope {
33 	struct intel_iommu *iommu;
34 	unsigned int id;
35 	unsigned int bus;	/* PCI bus number */
36 	unsigned int devfn;	/* PCI devfn number */
37 };
38 
39 struct hpet_scope {
40 	struct intel_iommu *iommu;
41 	u8 id;
42 	unsigned int bus;
43 	unsigned int devfn;
44 };
45 
46 struct irq_2_iommu {
47 	struct intel_iommu *iommu;
48 	u16 irte_index;
49 	u16 sub_handle;
50 	u8  irte_mask;
51 	enum irq_mode mode;
52 };
53 
54 struct intel_ir_data {
55 	struct irq_2_iommu			irq_2_iommu;
56 	struct irte				irte_entry;
57 	union {
58 		struct msi_msg			msi_entry;
59 	};
60 };
61 
62 #define IR_X2APIC_MODE(mode) (mode ? (1 << 11) : 0)
63 #define IRTE_DEST(dest) ((eim_mode) ? dest : dest << 8)
64 
65 static int __read_mostly eim_mode;
66 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
67 static struct hpet_scope ir_hpet[MAX_HPET_TBS];
68 
69 /*
70  * Lock ordering:
71  * ->dmar_global_lock
72  *	->irq_2_ir_lock
73  *		->qi->q_lock
74  *	->iommu->register_lock
75  * Note:
76  * intel_irq_remap_ops.{supported,prepare,enable,disable,reenable} are called
77  * in single-threaded environment with interrupt disabled, so no need to tabke
78  * the dmar_global_lock.
79  */
80 DEFINE_RAW_SPINLOCK(irq_2_ir_lock);
81 static const struct irq_domain_ops intel_ir_domain_ops;
82 
83 static void iommu_disable_irq_remapping(struct intel_iommu *iommu);
84 static int __init parse_ioapics_under_ir(void);
85 
86 static bool ir_pre_enabled(struct intel_iommu *iommu)
87 {
88 	return (iommu->flags & VTD_FLAG_IRQ_REMAP_PRE_ENABLED);
89 }
90 
91 static void clear_ir_pre_enabled(struct intel_iommu *iommu)
92 {
93 	iommu->flags &= ~VTD_FLAG_IRQ_REMAP_PRE_ENABLED;
94 }
95 
96 static void init_ir_status(struct intel_iommu *iommu)
97 {
98 	u32 gsts;
99 
100 	gsts = readl(iommu->reg + DMAR_GSTS_REG);
101 	if (gsts & DMA_GSTS_IRES)
102 		iommu->flags |= VTD_FLAG_IRQ_REMAP_PRE_ENABLED;
103 }
104 
105 static int alloc_irte(struct intel_iommu *iommu,
106 		      struct irq_2_iommu *irq_iommu, u16 count)
107 {
108 	struct ir_table *table = iommu->ir_table;
109 	unsigned int mask = 0;
110 	unsigned long flags;
111 	int index;
112 
113 	if (!count || !irq_iommu)
114 		return -1;
115 
116 	if (count > 1) {
117 		count = __roundup_pow_of_two(count);
118 		mask = ilog2(count);
119 	}
120 
121 	if (mask > ecap_max_handle_mask(iommu->ecap)) {
122 		pr_err("Requested mask %x exceeds the max invalidation handle"
123 		       " mask value %Lx\n", mask,
124 		       ecap_max_handle_mask(iommu->ecap));
125 		return -1;
126 	}
127 
128 	raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
129 	index = bitmap_find_free_region(table->bitmap,
130 					INTR_REMAP_TABLE_ENTRIES, mask);
131 	if (index < 0) {
132 		pr_warn("IR%d: can't allocate an IRTE\n", iommu->seq_id);
133 	} else {
134 		irq_iommu->iommu = iommu;
135 		irq_iommu->irte_index =  index;
136 		irq_iommu->sub_handle = 0;
137 		irq_iommu->irte_mask = mask;
138 		irq_iommu->mode = IRQ_REMAPPING;
139 	}
140 	raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
141 
142 	return index;
143 }
144 
145 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
146 {
147 	struct qi_desc desc;
148 
149 	desc.qw0 = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
150 		   | QI_IEC_SELECTIVE;
151 	desc.qw1 = 0;
152 	desc.qw2 = 0;
153 	desc.qw3 = 0;
154 
155 	return qi_submit_sync(iommu, &desc, 1, 0);
156 }
157 
158 static int modify_irte(struct irq_2_iommu *irq_iommu,
159 		       struct irte *irte_modified)
160 {
161 	struct intel_iommu *iommu;
162 	unsigned long flags;
163 	struct irte *irte;
164 	int rc, index;
165 
166 	if (!irq_iommu)
167 		return -1;
168 
169 	raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
170 
171 	iommu = irq_iommu->iommu;
172 
173 	index = irq_iommu->irte_index + irq_iommu->sub_handle;
174 	irte = &iommu->ir_table->base[index];
175 
176 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE)
177 	if ((irte->pst == 1) || (irte_modified->pst == 1)) {
178 		bool ret;
179 
180 		ret = cmpxchg_double(&irte->low, &irte->high,
181 				     irte->low, irte->high,
182 				     irte_modified->low, irte_modified->high);
183 		/*
184 		 * We use cmpxchg16 to atomically update the 128-bit IRTE,
185 		 * and it cannot be updated by the hardware or other processors
186 		 * behind us, so the return value of cmpxchg16 should be the
187 		 * same as the old value.
188 		 */
189 		WARN_ON(!ret);
190 	} else
191 #endif
192 	{
193 		set_64bit(&irte->low, irte_modified->low);
194 		set_64bit(&irte->high, irte_modified->high);
195 	}
196 	__iommu_flush_cache(iommu, irte, sizeof(*irte));
197 
198 	rc = qi_flush_iec(iommu, index, 0);
199 
200 	/* Update iommu mode according to the IRTE mode */
201 	irq_iommu->mode = irte->pst ? IRQ_POSTING : IRQ_REMAPPING;
202 	raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
203 
204 	return rc;
205 }
206 
207 static struct intel_iommu *map_hpet_to_iommu(u8 hpet_id)
208 {
209 	int i;
210 
211 	for (i = 0; i < MAX_HPET_TBS; i++) {
212 		if (ir_hpet[i].id == hpet_id && ir_hpet[i].iommu)
213 			return ir_hpet[i].iommu;
214 	}
215 	return NULL;
216 }
217 
218 static struct intel_iommu *map_ioapic_to_iommu(int apic)
219 {
220 	int i;
221 
222 	for (i = 0; i < MAX_IO_APICS; i++) {
223 		if (ir_ioapic[i].id == apic && ir_ioapic[i].iommu)
224 			return ir_ioapic[i].iommu;
225 	}
226 	return NULL;
227 }
228 
229 static struct irq_domain *map_dev_to_ir(struct pci_dev *dev)
230 {
231 	struct dmar_drhd_unit *drhd = dmar_find_matched_drhd_unit(dev);
232 
233 	return drhd ? drhd->iommu->ir_msi_domain : NULL;
234 }
235 
236 static int clear_entries(struct irq_2_iommu *irq_iommu)
237 {
238 	struct irte *start, *entry, *end;
239 	struct intel_iommu *iommu;
240 	int index;
241 
242 	if (irq_iommu->sub_handle)
243 		return 0;
244 
245 	iommu = irq_iommu->iommu;
246 	index = irq_iommu->irte_index;
247 
248 	start = iommu->ir_table->base + index;
249 	end = start + (1 << irq_iommu->irte_mask);
250 
251 	for (entry = start; entry < end; entry++) {
252 		set_64bit(&entry->low, 0);
253 		set_64bit(&entry->high, 0);
254 	}
255 	bitmap_release_region(iommu->ir_table->bitmap, index,
256 			      irq_iommu->irte_mask);
257 
258 	return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
259 }
260 
261 /*
262  * source validation type
263  */
264 #define SVT_NO_VERIFY		0x0  /* no verification is required */
265 #define SVT_VERIFY_SID_SQ	0x1  /* verify using SID and SQ fields */
266 #define SVT_VERIFY_BUS		0x2  /* verify bus of request-id */
267 
268 /*
269  * source-id qualifier
270  */
271 #define SQ_ALL_16	0x0  /* verify all 16 bits of request-id */
272 #define SQ_13_IGNORE_1	0x1  /* verify most significant 13 bits, ignore
273 			      * the third least significant bit
274 			      */
275 #define SQ_13_IGNORE_2	0x2  /* verify most significant 13 bits, ignore
276 			      * the second and third least significant bits
277 			      */
278 #define SQ_13_IGNORE_3	0x3  /* verify most significant 13 bits, ignore
279 			      * the least three significant bits
280 			      */
281 
282 /*
283  * set SVT, SQ and SID fields of irte to verify
284  * source ids of interrupt requests
285  */
286 static void set_irte_sid(struct irte *irte, unsigned int svt,
287 			 unsigned int sq, unsigned int sid)
288 {
289 	if (disable_sourceid_checking)
290 		svt = SVT_NO_VERIFY;
291 	irte->svt = svt;
292 	irte->sq = sq;
293 	irte->sid = sid;
294 }
295 
296 /*
297  * Set an IRTE to match only the bus number. Interrupt requests that reference
298  * this IRTE must have a requester-id whose bus number is between or equal
299  * to the start_bus and end_bus arguments.
300  */
301 static void set_irte_verify_bus(struct irte *irte, unsigned int start_bus,
302 				unsigned int end_bus)
303 {
304 	set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
305 		     (start_bus << 8) | end_bus);
306 }
307 
308 static int set_ioapic_sid(struct irte *irte, int apic)
309 {
310 	int i;
311 	u16 sid = 0;
312 
313 	if (!irte)
314 		return -1;
315 
316 	down_read(&dmar_global_lock);
317 	for (i = 0; i < MAX_IO_APICS; i++) {
318 		if (ir_ioapic[i].iommu && ir_ioapic[i].id == apic) {
319 			sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
320 			break;
321 		}
322 	}
323 	up_read(&dmar_global_lock);
324 
325 	if (sid == 0) {
326 		pr_warn("Failed to set source-id of IOAPIC (%d)\n", apic);
327 		return -1;
328 	}
329 
330 	set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, sid);
331 
332 	return 0;
333 }
334 
335 static int set_hpet_sid(struct irte *irte, u8 id)
336 {
337 	int i;
338 	u16 sid = 0;
339 
340 	if (!irte)
341 		return -1;
342 
343 	down_read(&dmar_global_lock);
344 	for (i = 0; i < MAX_HPET_TBS; i++) {
345 		if (ir_hpet[i].iommu && ir_hpet[i].id == id) {
346 			sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
347 			break;
348 		}
349 	}
350 	up_read(&dmar_global_lock);
351 
352 	if (sid == 0) {
353 		pr_warn("Failed to set source-id of HPET block (%d)\n", id);
354 		return -1;
355 	}
356 
357 	/*
358 	 * Should really use SQ_ALL_16. Some platforms are broken.
359 	 * While we figure out the right quirks for these broken platforms, use
360 	 * SQ_13_IGNORE_3 for now.
361 	 */
362 	set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
363 
364 	return 0;
365 }
366 
367 struct set_msi_sid_data {
368 	struct pci_dev *pdev;
369 	u16 alias;
370 	int count;
371 	int busmatch_count;
372 };
373 
374 static int set_msi_sid_cb(struct pci_dev *pdev, u16 alias, void *opaque)
375 {
376 	struct set_msi_sid_data *data = opaque;
377 
378 	if (data->count == 0 || PCI_BUS_NUM(alias) == PCI_BUS_NUM(data->alias))
379 		data->busmatch_count++;
380 
381 	data->pdev = pdev;
382 	data->alias = alias;
383 	data->count++;
384 
385 	return 0;
386 }
387 
388 static int set_msi_sid(struct irte *irte, struct pci_dev *dev)
389 {
390 	struct set_msi_sid_data data;
391 
392 	if (!irte || !dev)
393 		return -1;
394 
395 	data.count = 0;
396 	data.busmatch_count = 0;
397 	pci_for_each_dma_alias(dev, set_msi_sid_cb, &data);
398 
399 	/*
400 	 * DMA alias provides us with a PCI device and alias.  The only case
401 	 * where the it will return an alias on a different bus than the
402 	 * device is the case of a PCIe-to-PCI bridge, where the alias is for
403 	 * the subordinate bus.  In this case we can only verify the bus.
404 	 *
405 	 * If there are multiple aliases, all with the same bus number,
406 	 * then all we can do is verify the bus. This is typical in NTB
407 	 * hardware which use proxy IDs where the device will generate traffic
408 	 * from multiple devfn numbers on the same bus.
409 	 *
410 	 * If the alias device is on a different bus than our source device
411 	 * then we have a topology based alias, use it.
412 	 *
413 	 * Otherwise, the alias is for a device DMA quirk and we cannot
414 	 * assume that MSI uses the same requester ID.  Therefore use the
415 	 * original device.
416 	 */
417 	if (PCI_BUS_NUM(data.alias) != data.pdev->bus->number)
418 		set_irte_verify_bus(irte, PCI_BUS_NUM(data.alias),
419 				    dev->bus->number);
420 	else if (data.count >= 2 && data.busmatch_count == data.count)
421 		set_irte_verify_bus(irte, dev->bus->number, dev->bus->number);
422 	else if (data.pdev->bus->number != dev->bus->number)
423 		set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, data.alias);
424 	else
425 		set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
426 			     pci_dev_id(dev));
427 
428 	return 0;
429 }
430 
431 static int iommu_load_old_irte(struct intel_iommu *iommu)
432 {
433 	struct irte *old_ir_table;
434 	phys_addr_t irt_phys;
435 	unsigned int i;
436 	size_t size;
437 	u64 irta;
438 
439 	/* Check whether the old ir-table has the same size as ours */
440 	irta = dmar_readq(iommu->reg + DMAR_IRTA_REG);
441 	if ((irta & INTR_REMAP_TABLE_REG_SIZE_MASK)
442 	     != INTR_REMAP_TABLE_REG_SIZE)
443 		return -EINVAL;
444 
445 	irt_phys = irta & VTD_PAGE_MASK;
446 	size     = INTR_REMAP_TABLE_ENTRIES*sizeof(struct irte);
447 
448 	/* Map the old IR table */
449 	old_ir_table = memremap(irt_phys, size, MEMREMAP_WB);
450 	if (!old_ir_table)
451 		return -ENOMEM;
452 
453 	/* Copy data over */
454 	memcpy(iommu->ir_table->base, old_ir_table, size);
455 
456 	__iommu_flush_cache(iommu, iommu->ir_table->base, size);
457 
458 	/*
459 	 * Now check the table for used entries and mark those as
460 	 * allocated in the bitmap
461 	 */
462 	for (i = 0; i < INTR_REMAP_TABLE_ENTRIES; i++) {
463 		if (iommu->ir_table->base[i].present)
464 			bitmap_set(iommu->ir_table->bitmap, i, 1);
465 	}
466 
467 	memunmap(old_ir_table);
468 
469 	return 0;
470 }
471 
472 
473 static void iommu_set_irq_remapping(struct intel_iommu *iommu, int mode)
474 {
475 	unsigned long flags;
476 	u64 addr;
477 	u32 sts;
478 
479 	addr = virt_to_phys((void *)iommu->ir_table->base);
480 
481 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
482 
483 	dmar_writeq(iommu->reg + DMAR_IRTA_REG,
484 		    (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
485 
486 	/* Set interrupt-remapping table pointer */
487 	writel(iommu->gcmd | DMA_GCMD_SIRTP, iommu->reg + DMAR_GCMD_REG);
488 
489 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
490 		      readl, (sts & DMA_GSTS_IRTPS), sts);
491 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
492 
493 	/*
494 	 * Global invalidation of interrupt entry cache to make sure the
495 	 * hardware uses the new irq remapping table.
496 	 */
497 	qi_global_iec(iommu);
498 }
499 
500 static void iommu_enable_irq_remapping(struct intel_iommu *iommu)
501 {
502 	unsigned long flags;
503 	u32 sts;
504 
505 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
506 
507 	/* Enable interrupt-remapping */
508 	iommu->gcmd |= DMA_GCMD_IRE;
509 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
510 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
511 		      readl, (sts & DMA_GSTS_IRES), sts);
512 
513 	/* Block compatibility-format MSIs */
514 	if (sts & DMA_GSTS_CFIS) {
515 		iommu->gcmd &= ~DMA_GCMD_CFI;
516 		writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
517 		IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
518 			      readl, !(sts & DMA_GSTS_CFIS), sts);
519 	}
520 
521 	/*
522 	 * With CFI clear in the Global Command register, we should be
523 	 * protected from dangerous (i.e. compatibility) interrupts
524 	 * regardless of x2apic status.  Check just to be sure.
525 	 */
526 	if (sts & DMA_GSTS_CFIS)
527 		WARN(1, KERN_WARNING
528 			"Compatibility-format IRQs enabled despite intr remapping;\n"
529 			"you are vulnerable to IRQ injection.\n");
530 
531 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
532 }
533 
534 static int intel_setup_irq_remapping(struct intel_iommu *iommu)
535 {
536 	struct ir_table *ir_table;
537 	struct fwnode_handle *fn;
538 	unsigned long *bitmap;
539 	struct page *pages;
540 
541 	if (iommu->ir_table)
542 		return 0;
543 
544 	ir_table = kzalloc(sizeof(struct ir_table), GFP_KERNEL);
545 	if (!ir_table)
546 		return -ENOMEM;
547 
548 	pages = alloc_pages_node(iommu->node, GFP_KERNEL | __GFP_ZERO,
549 				 INTR_REMAP_PAGE_ORDER);
550 	if (!pages) {
551 		pr_err("IR%d: failed to allocate pages of order %d\n",
552 		       iommu->seq_id, INTR_REMAP_PAGE_ORDER);
553 		goto out_free_table;
554 	}
555 
556 	bitmap = bitmap_zalloc(INTR_REMAP_TABLE_ENTRIES, GFP_ATOMIC);
557 	if (bitmap == NULL) {
558 		pr_err("IR%d: failed to allocate bitmap\n", iommu->seq_id);
559 		goto out_free_pages;
560 	}
561 
562 	fn = irq_domain_alloc_named_id_fwnode("INTEL-IR", iommu->seq_id);
563 	if (!fn)
564 		goto out_free_bitmap;
565 
566 	iommu->ir_domain =
567 		irq_domain_create_hierarchy(arch_get_ir_parent_domain(),
568 					    0, INTR_REMAP_TABLE_ENTRIES,
569 					    fn, &intel_ir_domain_ops,
570 					    iommu);
571 	if (!iommu->ir_domain) {
572 		irq_domain_free_fwnode(fn);
573 		pr_err("IR%d: failed to allocate irqdomain\n", iommu->seq_id);
574 		goto out_free_bitmap;
575 	}
576 	iommu->ir_msi_domain =
577 		arch_create_remap_msi_irq_domain(iommu->ir_domain,
578 						 "INTEL-IR-MSI",
579 						 iommu->seq_id);
580 
581 	ir_table->base = page_address(pages);
582 	ir_table->bitmap = bitmap;
583 	iommu->ir_table = ir_table;
584 
585 	/*
586 	 * If the queued invalidation is already initialized,
587 	 * shouldn't disable it.
588 	 */
589 	if (!iommu->qi) {
590 		/*
591 		 * Clear previous faults.
592 		 */
593 		dmar_fault(-1, iommu);
594 		dmar_disable_qi(iommu);
595 
596 		if (dmar_enable_qi(iommu)) {
597 			pr_err("Failed to enable queued invalidation\n");
598 			goto out_free_bitmap;
599 		}
600 	}
601 
602 	init_ir_status(iommu);
603 
604 	if (ir_pre_enabled(iommu)) {
605 		if (!is_kdump_kernel()) {
606 			pr_warn("IRQ remapping was enabled on %s but we are not in kdump mode\n",
607 				iommu->name);
608 			clear_ir_pre_enabled(iommu);
609 			iommu_disable_irq_remapping(iommu);
610 		} else if (iommu_load_old_irte(iommu))
611 			pr_err("Failed to copy IR table for %s from previous kernel\n",
612 			       iommu->name);
613 		else
614 			pr_info("Copied IR table for %s from previous kernel\n",
615 				iommu->name);
616 	}
617 
618 	iommu_set_irq_remapping(iommu, eim_mode);
619 
620 	return 0;
621 
622 out_free_bitmap:
623 	bitmap_free(bitmap);
624 out_free_pages:
625 	__free_pages(pages, INTR_REMAP_PAGE_ORDER);
626 out_free_table:
627 	kfree(ir_table);
628 
629 	iommu->ir_table  = NULL;
630 
631 	return -ENOMEM;
632 }
633 
634 static void intel_teardown_irq_remapping(struct intel_iommu *iommu)
635 {
636 	struct fwnode_handle *fn;
637 
638 	if (iommu && iommu->ir_table) {
639 		if (iommu->ir_msi_domain) {
640 			fn = iommu->ir_msi_domain->fwnode;
641 
642 			irq_domain_remove(iommu->ir_msi_domain);
643 			irq_domain_free_fwnode(fn);
644 			iommu->ir_msi_domain = NULL;
645 		}
646 		if (iommu->ir_domain) {
647 			fn = iommu->ir_domain->fwnode;
648 
649 			irq_domain_remove(iommu->ir_domain);
650 			irq_domain_free_fwnode(fn);
651 			iommu->ir_domain = NULL;
652 		}
653 		free_pages((unsigned long)iommu->ir_table->base,
654 			   INTR_REMAP_PAGE_ORDER);
655 		bitmap_free(iommu->ir_table->bitmap);
656 		kfree(iommu->ir_table);
657 		iommu->ir_table = NULL;
658 	}
659 }
660 
661 /*
662  * Disable Interrupt Remapping.
663  */
664 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
665 {
666 	unsigned long flags;
667 	u32 sts;
668 
669 	if (!ecap_ir_support(iommu->ecap))
670 		return;
671 
672 	/*
673 	 * global invalidation of interrupt entry cache before disabling
674 	 * interrupt-remapping.
675 	 */
676 	qi_global_iec(iommu);
677 
678 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
679 
680 	sts = readl(iommu->reg + DMAR_GSTS_REG);
681 	if (!(sts & DMA_GSTS_IRES))
682 		goto end;
683 
684 	iommu->gcmd &= ~DMA_GCMD_IRE;
685 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
686 
687 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
688 		      readl, !(sts & DMA_GSTS_IRES), sts);
689 
690 end:
691 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
692 }
693 
694 static int __init dmar_x2apic_optout(void)
695 {
696 	struct acpi_table_dmar *dmar;
697 	dmar = (struct acpi_table_dmar *)dmar_tbl;
698 	if (!dmar || no_x2apic_optout)
699 		return 0;
700 	return dmar->flags & DMAR_X2APIC_OPT_OUT;
701 }
702 
703 static void __init intel_cleanup_irq_remapping(void)
704 {
705 	struct dmar_drhd_unit *drhd;
706 	struct intel_iommu *iommu;
707 
708 	for_each_iommu(iommu, drhd) {
709 		if (ecap_ir_support(iommu->ecap)) {
710 			iommu_disable_irq_remapping(iommu);
711 			intel_teardown_irq_remapping(iommu);
712 		}
713 	}
714 
715 	if (x2apic_supported())
716 		pr_warn("Failed to enable irq remapping. You are vulnerable to irq-injection attacks.\n");
717 }
718 
719 static int __init intel_prepare_irq_remapping(void)
720 {
721 	struct dmar_drhd_unit *drhd;
722 	struct intel_iommu *iommu;
723 	int eim = 0;
724 
725 	if (irq_remap_broken) {
726 		pr_warn("This system BIOS has enabled interrupt remapping\n"
727 			"on a chipset that contains an erratum making that\n"
728 			"feature unstable.  To maintain system stability\n"
729 			"interrupt remapping is being disabled.  Please\n"
730 			"contact your BIOS vendor for an update\n");
731 		add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
732 		return -ENODEV;
733 	}
734 
735 	if (dmar_table_init() < 0)
736 		return -ENODEV;
737 
738 	if (intel_cap_audit(CAP_AUDIT_STATIC_IRQR, NULL))
739 		goto error;
740 
741 	if (!dmar_ir_support())
742 		return -ENODEV;
743 
744 	if (parse_ioapics_under_ir()) {
745 		pr_info("Not enabling interrupt remapping\n");
746 		goto error;
747 	}
748 
749 	/* First make sure all IOMMUs support IRQ remapping */
750 	for_each_iommu(iommu, drhd)
751 		if (!ecap_ir_support(iommu->ecap))
752 			goto error;
753 
754 	/* Detect remapping mode: lapic or x2apic */
755 	if (x2apic_supported()) {
756 		eim = !dmar_x2apic_optout();
757 		if (!eim) {
758 			pr_info("x2apic is disabled because BIOS sets x2apic opt out bit.");
759 			pr_info("Use 'intremap=no_x2apic_optout' to override the BIOS setting.\n");
760 		}
761 	}
762 
763 	for_each_iommu(iommu, drhd) {
764 		if (eim && !ecap_eim_support(iommu->ecap)) {
765 			pr_info("%s does not support EIM\n", iommu->name);
766 			eim = 0;
767 		}
768 	}
769 
770 	eim_mode = eim;
771 	if (eim)
772 		pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
773 
774 	/* Do the initializations early */
775 	for_each_iommu(iommu, drhd) {
776 		if (intel_setup_irq_remapping(iommu)) {
777 			pr_err("Failed to setup irq remapping for %s\n",
778 			       iommu->name);
779 			goto error;
780 		}
781 	}
782 
783 	return 0;
784 
785 error:
786 	intel_cleanup_irq_remapping();
787 	return -ENODEV;
788 }
789 
790 /*
791  * Set Posted-Interrupts capability.
792  */
793 static inline void set_irq_posting_cap(void)
794 {
795 	struct dmar_drhd_unit *drhd;
796 	struct intel_iommu *iommu;
797 
798 	if (!disable_irq_post) {
799 		/*
800 		 * If IRTE is in posted format, the 'pda' field goes across the
801 		 * 64-bit boundary, we need use cmpxchg16b to atomically update
802 		 * it. We only expose posted-interrupt when X86_FEATURE_CX16
803 		 * is supported. Actually, hardware platforms supporting PI
804 		 * should have X86_FEATURE_CX16 support, this has been confirmed
805 		 * with Intel hardware guys.
806 		 */
807 		if (boot_cpu_has(X86_FEATURE_CX16))
808 			intel_irq_remap_ops.capability |= 1 << IRQ_POSTING_CAP;
809 
810 		for_each_iommu(iommu, drhd)
811 			if (!cap_pi_support(iommu->cap)) {
812 				intel_irq_remap_ops.capability &=
813 						~(1 << IRQ_POSTING_CAP);
814 				break;
815 			}
816 	}
817 }
818 
819 static int __init intel_enable_irq_remapping(void)
820 {
821 	struct dmar_drhd_unit *drhd;
822 	struct intel_iommu *iommu;
823 	bool setup = false;
824 
825 	/*
826 	 * Setup Interrupt-remapping for all the DRHD's now.
827 	 */
828 	for_each_iommu(iommu, drhd) {
829 		if (!ir_pre_enabled(iommu))
830 			iommu_enable_irq_remapping(iommu);
831 		setup = true;
832 	}
833 
834 	if (!setup)
835 		goto error;
836 
837 	irq_remapping_enabled = 1;
838 
839 	set_irq_posting_cap();
840 
841 	pr_info("Enabled IRQ remapping in %s mode\n", eim_mode ? "x2apic" : "xapic");
842 
843 	return eim_mode ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
844 
845 error:
846 	intel_cleanup_irq_remapping();
847 	return -1;
848 }
849 
850 static int ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
851 				   struct intel_iommu *iommu,
852 				   struct acpi_dmar_hardware_unit *drhd)
853 {
854 	struct acpi_dmar_pci_path *path;
855 	u8 bus;
856 	int count, free = -1;
857 
858 	bus = scope->bus;
859 	path = (struct acpi_dmar_pci_path *)(scope + 1);
860 	count = (scope->length - sizeof(struct acpi_dmar_device_scope))
861 		/ sizeof(struct acpi_dmar_pci_path);
862 
863 	while (--count > 0) {
864 		/*
865 		 * Access PCI directly due to the PCI
866 		 * subsystem isn't initialized yet.
867 		 */
868 		bus = read_pci_config_byte(bus, path->device, path->function,
869 					   PCI_SECONDARY_BUS);
870 		path++;
871 	}
872 
873 	for (count = 0; count < MAX_HPET_TBS; count++) {
874 		if (ir_hpet[count].iommu == iommu &&
875 		    ir_hpet[count].id == scope->enumeration_id)
876 			return 0;
877 		else if (ir_hpet[count].iommu == NULL && free == -1)
878 			free = count;
879 	}
880 	if (free == -1) {
881 		pr_warn("Exceeded Max HPET blocks\n");
882 		return -ENOSPC;
883 	}
884 
885 	ir_hpet[free].iommu = iommu;
886 	ir_hpet[free].id    = scope->enumeration_id;
887 	ir_hpet[free].bus   = bus;
888 	ir_hpet[free].devfn = PCI_DEVFN(path->device, path->function);
889 	pr_info("HPET id %d under DRHD base 0x%Lx\n",
890 		scope->enumeration_id, drhd->address);
891 
892 	return 0;
893 }
894 
895 static int ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
896 				     struct intel_iommu *iommu,
897 				     struct acpi_dmar_hardware_unit *drhd)
898 {
899 	struct acpi_dmar_pci_path *path;
900 	u8 bus;
901 	int count, free = -1;
902 
903 	bus = scope->bus;
904 	path = (struct acpi_dmar_pci_path *)(scope + 1);
905 	count = (scope->length - sizeof(struct acpi_dmar_device_scope))
906 		/ sizeof(struct acpi_dmar_pci_path);
907 
908 	while (--count > 0) {
909 		/*
910 		 * Access PCI directly due to the PCI
911 		 * subsystem isn't initialized yet.
912 		 */
913 		bus = read_pci_config_byte(bus, path->device, path->function,
914 					   PCI_SECONDARY_BUS);
915 		path++;
916 	}
917 
918 	for (count = 0; count < MAX_IO_APICS; count++) {
919 		if (ir_ioapic[count].iommu == iommu &&
920 		    ir_ioapic[count].id == scope->enumeration_id)
921 			return 0;
922 		else if (ir_ioapic[count].iommu == NULL && free == -1)
923 			free = count;
924 	}
925 	if (free == -1) {
926 		pr_warn("Exceeded Max IO APICS\n");
927 		return -ENOSPC;
928 	}
929 
930 	ir_ioapic[free].bus   = bus;
931 	ir_ioapic[free].devfn = PCI_DEVFN(path->device, path->function);
932 	ir_ioapic[free].iommu = iommu;
933 	ir_ioapic[free].id    = scope->enumeration_id;
934 	pr_info("IOAPIC id %d under DRHD base  0x%Lx IOMMU %d\n",
935 		scope->enumeration_id, drhd->address, iommu->seq_id);
936 
937 	return 0;
938 }
939 
940 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
941 				      struct intel_iommu *iommu)
942 {
943 	int ret = 0;
944 	struct acpi_dmar_hardware_unit *drhd;
945 	struct acpi_dmar_device_scope *scope;
946 	void *start, *end;
947 
948 	drhd = (struct acpi_dmar_hardware_unit *)header;
949 	start = (void *)(drhd + 1);
950 	end = ((void *)drhd) + header->length;
951 
952 	while (start < end && ret == 0) {
953 		scope = start;
954 		if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC)
955 			ret = ir_parse_one_ioapic_scope(scope, iommu, drhd);
956 		else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET)
957 			ret = ir_parse_one_hpet_scope(scope, iommu, drhd);
958 		start += scope->length;
959 	}
960 
961 	return ret;
962 }
963 
964 static void ir_remove_ioapic_hpet_scope(struct intel_iommu *iommu)
965 {
966 	int i;
967 
968 	for (i = 0; i < MAX_HPET_TBS; i++)
969 		if (ir_hpet[i].iommu == iommu)
970 			ir_hpet[i].iommu = NULL;
971 
972 	for (i = 0; i < MAX_IO_APICS; i++)
973 		if (ir_ioapic[i].iommu == iommu)
974 			ir_ioapic[i].iommu = NULL;
975 }
976 
977 /*
978  * Finds the assocaition between IOAPIC's and its Interrupt-remapping
979  * hardware unit.
980  */
981 static int __init parse_ioapics_under_ir(void)
982 {
983 	struct dmar_drhd_unit *drhd;
984 	struct intel_iommu *iommu;
985 	bool ir_supported = false;
986 	int ioapic_idx;
987 
988 	for_each_iommu(iommu, drhd) {
989 		int ret;
990 
991 		if (!ecap_ir_support(iommu->ecap))
992 			continue;
993 
994 		ret = ir_parse_ioapic_hpet_scope(drhd->hdr, iommu);
995 		if (ret)
996 			return ret;
997 
998 		ir_supported = true;
999 	}
1000 
1001 	if (!ir_supported)
1002 		return -ENODEV;
1003 
1004 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
1005 		int ioapic_id = mpc_ioapic_id(ioapic_idx);
1006 		if (!map_ioapic_to_iommu(ioapic_id)) {
1007 			pr_err(FW_BUG "ioapic %d has no mapping iommu, "
1008 			       "interrupt remapping will be disabled\n",
1009 			       ioapic_id);
1010 			return -1;
1011 		}
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 static int __init ir_dev_scope_init(void)
1018 {
1019 	int ret;
1020 
1021 	if (!irq_remapping_enabled)
1022 		return 0;
1023 
1024 	down_write(&dmar_global_lock);
1025 	ret = dmar_dev_scope_init();
1026 	up_write(&dmar_global_lock);
1027 
1028 	return ret;
1029 }
1030 rootfs_initcall(ir_dev_scope_init);
1031 
1032 static void disable_irq_remapping(void)
1033 {
1034 	struct dmar_drhd_unit *drhd;
1035 	struct intel_iommu *iommu = NULL;
1036 
1037 	/*
1038 	 * Disable Interrupt-remapping for all the DRHD's now.
1039 	 */
1040 	for_each_iommu(iommu, drhd) {
1041 		if (!ecap_ir_support(iommu->ecap))
1042 			continue;
1043 
1044 		iommu_disable_irq_remapping(iommu);
1045 	}
1046 
1047 	/*
1048 	 * Clear Posted-Interrupts capability.
1049 	 */
1050 	if (!disable_irq_post)
1051 		intel_irq_remap_ops.capability &= ~(1 << IRQ_POSTING_CAP);
1052 }
1053 
1054 static int reenable_irq_remapping(int eim)
1055 {
1056 	struct dmar_drhd_unit *drhd;
1057 	bool setup = false;
1058 	struct intel_iommu *iommu = NULL;
1059 
1060 	for_each_iommu(iommu, drhd)
1061 		if (iommu->qi)
1062 			dmar_reenable_qi(iommu);
1063 
1064 	/*
1065 	 * Setup Interrupt-remapping for all the DRHD's now.
1066 	 */
1067 	for_each_iommu(iommu, drhd) {
1068 		if (!ecap_ir_support(iommu->ecap))
1069 			continue;
1070 
1071 		/* Set up interrupt remapping for iommu.*/
1072 		iommu_set_irq_remapping(iommu, eim);
1073 		iommu_enable_irq_remapping(iommu);
1074 		setup = true;
1075 	}
1076 
1077 	if (!setup)
1078 		goto error;
1079 
1080 	set_irq_posting_cap();
1081 
1082 	return 0;
1083 
1084 error:
1085 	/*
1086 	 * handle error condition gracefully here!
1087 	 */
1088 	return -1;
1089 }
1090 
1091 /*
1092  * Store the MSI remapping domain pointer in the device if enabled.
1093  *
1094  * This is called from dmar_pci_bus_add_dev() so it works even when DMA
1095  * remapping is disabled. Only update the pointer if the device is not
1096  * already handled by a non default PCI/MSI interrupt domain. This protects
1097  * e.g. VMD devices.
1098  */
1099 void intel_irq_remap_add_device(struct dmar_pci_notify_info *info)
1100 {
1101 	if (!irq_remapping_enabled || pci_dev_has_special_msi_domain(info->dev))
1102 		return;
1103 
1104 	dev_set_msi_domain(&info->dev->dev, map_dev_to_ir(info->dev));
1105 }
1106 
1107 static void prepare_irte(struct irte *irte, int vector, unsigned int dest)
1108 {
1109 	memset(irte, 0, sizeof(*irte));
1110 
1111 	irte->present = 1;
1112 	irte->dst_mode = apic->dest_mode_logical;
1113 	/*
1114 	 * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
1115 	 * actual level or edge trigger will be setup in the IO-APIC
1116 	 * RTE. This will help simplify level triggered irq migration.
1117 	 * For more details, see the comments (in io_apic.c) explainig IO-APIC
1118 	 * irq migration in the presence of interrupt-remapping.
1119 	*/
1120 	irte->trigger_mode = 0;
1121 	irte->dlvry_mode = apic->delivery_mode;
1122 	irte->vector = vector;
1123 	irte->dest_id = IRTE_DEST(dest);
1124 	irte->redir_hint = 1;
1125 }
1126 
1127 struct irq_remap_ops intel_irq_remap_ops = {
1128 	.prepare		= intel_prepare_irq_remapping,
1129 	.enable			= intel_enable_irq_remapping,
1130 	.disable		= disable_irq_remapping,
1131 	.reenable		= reenable_irq_remapping,
1132 	.enable_faulting	= enable_drhd_fault_handling,
1133 };
1134 
1135 static void intel_ir_reconfigure_irte(struct irq_data *irqd, bool force)
1136 {
1137 	struct intel_ir_data *ir_data = irqd->chip_data;
1138 	struct irte *irte = &ir_data->irte_entry;
1139 	struct irq_cfg *cfg = irqd_cfg(irqd);
1140 
1141 	/*
1142 	 * Atomically updates the IRTE with the new destination, vector
1143 	 * and flushes the interrupt entry cache.
1144 	 */
1145 	irte->vector = cfg->vector;
1146 	irte->dest_id = IRTE_DEST(cfg->dest_apicid);
1147 
1148 	/* Update the hardware only if the interrupt is in remapped mode. */
1149 	if (force || ir_data->irq_2_iommu.mode == IRQ_REMAPPING)
1150 		modify_irte(&ir_data->irq_2_iommu, irte);
1151 }
1152 
1153 /*
1154  * Migrate the IO-APIC irq in the presence of intr-remapping.
1155  *
1156  * For both level and edge triggered, irq migration is a simple atomic
1157  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
1158  *
1159  * For level triggered, we eliminate the io-apic RTE modification (with the
1160  * updated vector information), by using a virtual vector (io-apic pin number).
1161  * Real vector that is used for interrupting cpu will be coming from
1162  * the interrupt-remapping table entry.
1163  *
1164  * As the migration is a simple atomic update of IRTE, the same mechanism
1165  * is used to migrate MSI irq's in the presence of interrupt-remapping.
1166  */
1167 static int
1168 intel_ir_set_affinity(struct irq_data *data, const struct cpumask *mask,
1169 		      bool force)
1170 {
1171 	struct irq_data *parent = data->parent_data;
1172 	struct irq_cfg *cfg = irqd_cfg(data);
1173 	int ret;
1174 
1175 	ret = parent->chip->irq_set_affinity(parent, mask, force);
1176 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
1177 		return ret;
1178 
1179 	intel_ir_reconfigure_irte(data, false);
1180 	/*
1181 	 * After this point, all the interrupts will start arriving
1182 	 * at the new destination. So, time to cleanup the previous
1183 	 * vector allocation.
1184 	 */
1185 	send_cleanup_vector(cfg);
1186 
1187 	return IRQ_SET_MASK_OK_DONE;
1188 }
1189 
1190 static void intel_ir_compose_msi_msg(struct irq_data *irq_data,
1191 				     struct msi_msg *msg)
1192 {
1193 	struct intel_ir_data *ir_data = irq_data->chip_data;
1194 
1195 	*msg = ir_data->msi_entry;
1196 }
1197 
1198 static int intel_ir_set_vcpu_affinity(struct irq_data *data, void *info)
1199 {
1200 	struct intel_ir_data *ir_data = data->chip_data;
1201 	struct vcpu_data *vcpu_pi_info = info;
1202 
1203 	/* stop posting interrupts, back to remapping mode */
1204 	if (!vcpu_pi_info) {
1205 		modify_irte(&ir_data->irq_2_iommu, &ir_data->irte_entry);
1206 	} else {
1207 		struct irte irte_pi;
1208 
1209 		/*
1210 		 * We are not caching the posted interrupt entry. We
1211 		 * copy the data from the remapped entry and modify
1212 		 * the fields which are relevant for posted mode. The
1213 		 * cached remapped entry is used for switching back to
1214 		 * remapped mode.
1215 		 */
1216 		memset(&irte_pi, 0, sizeof(irte_pi));
1217 		dmar_copy_shared_irte(&irte_pi, &ir_data->irte_entry);
1218 
1219 		/* Update the posted mode fields */
1220 		irte_pi.p_pst = 1;
1221 		irte_pi.p_urgent = 0;
1222 		irte_pi.p_vector = vcpu_pi_info->vector;
1223 		irte_pi.pda_l = (vcpu_pi_info->pi_desc_addr >>
1224 				(32 - PDA_LOW_BIT)) & ~(-1UL << PDA_LOW_BIT);
1225 		irte_pi.pda_h = (vcpu_pi_info->pi_desc_addr >> 32) &
1226 				~(-1UL << PDA_HIGH_BIT);
1227 
1228 		modify_irte(&ir_data->irq_2_iommu, &irte_pi);
1229 	}
1230 
1231 	return 0;
1232 }
1233 
1234 static struct irq_chip intel_ir_chip = {
1235 	.name			= "INTEL-IR",
1236 	.irq_ack		= apic_ack_irq,
1237 	.irq_set_affinity	= intel_ir_set_affinity,
1238 	.irq_compose_msi_msg	= intel_ir_compose_msi_msg,
1239 	.irq_set_vcpu_affinity	= intel_ir_set_vcpu_affinity,
1240 };
1241 
1242 static void fill_msi_msg(struct msi_msg *msg, u32 index, u32 subhandle)
1243 {
1244 	memset(msg, 0, sizeof(*msg));
1245 
1246 	msg->arch_addr_lo.dmar_base_address = X86_MSI_BASE_ADDRESS_LOW;
1247 	msg->arch_addr_lo.dmar_subhandle_valid = true;
1248 	msg->arch_addr_lo.dmar_format = true;
1249 	msg->arch_addr_lo.dmar_index_0_14 = index & 0x7FFF;
1250 	msg->arch_addr_lo.dmar_index_15 = !!(index & 0x8000);
1251 
1252 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
1253 
1254 	msg->arch_data.dmar_subhandle = subhandle;
1255 }
1256 
1257 static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,
1258 					     struct irq_cfg *irq_cfg,
1259 					     struct irq_alloc_info *info,
1260 					     int index, int sub_handle)
1261 {
1262 	struct irte *irte = &data->irte_entry;
1263 
1264 	prepare_irte(irte, irq_cfg->vector, irq_cfg->dest_apicid);
1265 
1266 	switch (info->type) {
1267 	case X86_IRQ_ALLOC_TYPE_IOAPIC:
1268 		/* Set source-id of interrupt request */
1269 		set_ioapic_sid(irte, info->devid);
1270 		apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: Set IRTE entry (P:%d FPD:%d Dst_Mode:%d Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X Avail:%X Vector:%02X Dest:%08X SID:%04X SQ:%X SVT:%X)\n",
1271 			info->devid, irte->present, irte->fpd,
1272 			irte->dst_mode, irte->redir_hint,
1273 			irte->trigger_mode, irte->dlvry_mode,
1274 			irte->avail, irte->vector, irte->dest_id,
1275 			irte->sid, irte->sq, irte->svt);
1276 		sub_handle = info->ioapic.pin;
1277 		break;
1278 	case X86_IRQ_ALLOC_TYPE_HPET:
1279 		set_hpet_sid(irte, info->devid);
1280 		break;
1281 	case X86_IRQ_ALLOC_TYPE_PCI_MSI:
1282 	case X86_IRQ_ALLOC_TYPE_PCI_MSIX:
1283 		set_msi_sid(irte, msi_desc_to_pci_dev(info->desc));
1284 		break;
1285 	default:
1286 		BUG_ON(1);
1287 		break;
1288 	}
1289 	fill_msi_msg(&data->msi_entry, index, sub_handle);
1290 }
1291 
1292 static void intel_free_irq_resources(struct irq_domain *domain,
1293 				     unsigned int virq, unsigned int nr_irqs)
1294 {
1295 	struct irq_data *irq_data;
1296 	struct intel_ir_data *data;
1297 	struct irq_2_iommu *irq_iommu;
1298 	unsigned long flags;
1299 	int i;
1300 	for (i = 0; i < nr_irqs; i++) {
1301 		irq_data = irq_domain_get_irq_data(domain, virq  + i);
1302 		if (irq_data && irq_data->chip_data) {
1303 			data = irq_data->chip_data;
1304 			irq_iommu = &data->irq_2_iommu;
1305 			raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
1306 			clear_entries(irq_iommu);
1307 			raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
1308 			irq_domain_reset_irq_data(irq_data);
1309 			kfree(data);
1310 		}
1311 	}
1312 }
1313 
1314 static int intel_irq_remapping_alloc(struct irq_domain *domain,
1315 				     unsigned int virq, unsigned int nr_irqs,
1316 				     void *arg)
1317 {
1318 	struct intel_iommu *iommu = domain->host_data;
1319 	struct irq_alloc_info *info = arg;
1320 	struct intel_ir_data *data, *ird;
1321 	struct irq_data *irq_data;
1322 	struct irq_cfg *irq_cfg;
1323 	int i, ret, index;
1324 
1325 	if (!info || !iommu)
1326 		return -EINVAL;
1327 	if (nr_irqs > 1 && info->type != X86_IRQ_ALLOC_TYPE_PCI_MSI &&
1328 	    info->type != X86_IRQ_ALLOC_TYPE_PCI_MSIX)
1329 		return -EINVAL;
1330 
1331 	/*
1332 	 * With IRQ remapping enabled, don't need contiguous CPU vectors
1333 	 * to support multiple MSI interrupts.
1334 	 */
1335 	if (info->type == X86_IRQ_ALLOC_TYPE_PCI_MSI)
1336 		info->flags &= ~X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
1337 
1338 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
1339 	if (ret < 0)
1340 		return ret;
1341 
1342 	ret = -ENOMEM;
1343 	data = kzalloc(sizeof(*data), GFP_KERNEL);
1344 	if (!data)
1345 		goto out_free_parent;
1346 
1347 	down_read(&dmar_global_lock);
1348 	index = alloc_irte(iommu, &data->irq_2_iommu, nr_irqs);
1349 	up_read(&dmar_global_lock);
1350 	if (index < 0) {
1351 		pr_warn("Failed to allocate IRTE\n");
1352 		kfree(data);
1353 		goto out_free_parent;
1354 	}
1355 
1356 	for (i = 0; i < nr_irqs; i++) {
1357 		irq_data = irq_domain_get_irq_data(domain, virq + i);
1358 		irq_cfg = irqd_cfg(irq_data);
1359 		if (!irq_data || !irq_cfg) {
1360 			if (!i)
1361 				kfree(data);
1362 			ret = -EINVAL;
1363 			goto out_free_data;
1364 		}
1365 
1366 		if (i > 0) {
1367 			ird = kzalloc(sizeof(*ird), GFP_KERNEL);
1368 			if (!ird)
1369 				goto out_free_data;
1370 			/* Initialize the common data */
1371 			ird->irq_2_iommu = data->irq_2_iommu;
1372 			ird->irq_2_iommu.sub_handle = i;
1373 		} else {
1374 			ird = data;
1375 		}
1376 
1377 		irq_data->hwirq = (index << 16) + i;
1378 		irq_data->chip_data = ird;
1379 		irq_data->chip = &intel_ir_chip;
1380 		intel_irq_remapping_prepare_irte(ird, irq_cfg, info, index, i);
1381 		irq_set_status_flags(virq + i, IRQ_MOVE_PCNTXT);
1382 	}
1383 	return 0;
1384 
1385 out_free_data:
1386 	intel_free_irq_resources(domain, virq, i);
1387 out_free_parent:
1388 	irq_domain_free_irqs_common(domain, virq, nr_irqs);
1389 	return ret;
1390 }
1391 
1392 static void intel_irq_remapping_free(struct irq_domain *domain,
1393 				     unsigned int virq, unsigned int nr_irqs)
1394 {
1395 	intel_free_irq_resources(domain, virq, nr_irqs);
1396 	irq_domain_free_irqs_common(domain, virq, nr_irqs);
1397 }
1398 
1399 static int intel_irq_remapping_activate(struct irq_domain *domain,
1400 					struct irq_data *irq_data, bool reserve)
1401 {
1402 	intel_ir_reconfigure_irte(irq_data, true);
1403 	return 0;
1404 }
1405 
1406 static void intel_irq_remapping_deactivate(struct irq_domain *domain,
1407 					   struct irq_data *irq_data)
1408 {
1409 	struct intel_ir_data *data = irq_data->chip_data;
1410 	struct irte entry;
1411 
1412 	memset(&entry, 0, sizeof(entry));
1413 	modify_irte(&data->irq_2_iommu, &entry);
1414 }
1415 
1416 static int intel_irq_remapping_select(struct irq_domain *d,
1417 				      struct irq_fwspec *fwspec,
1418 				      enum irq_domain_bus_token bus_token)
1419 {
1420 	struct intel_iommu *iommu = NULL;
1421 
1422 	if (x86_fwspec_is_ioapic(fwspec))
1423 		iommu = map_ioapic_to_iommu(fwspec->param[0]);
1424 	else if (x86_fwspec_is_hpet(fwspec))
1425 		iommu = map_hpet_to_iommu(fwspec->param[0]);
1426 
1427 	return iommu && d == iommu->ir_domain;
1428 }
1429 
1430 static const struct irq_domain_ops intel_ir_domain_ops = {
1431 	.select = intel_irq_remapping_select,
1432 	.alloc = intel_irq_remapping_alloc,
1433 	.free = intel_irq_remapping_free,
1434 	.activate = intel_irq_remapping_activate,
1435 	.deactivate = intel_irq_remapping_deactivate,
1436 };
1437 
1438 /*
1439  * Support of Interrupt Remapping Unit Hotplug
1440  */
1441 static int dmar_ir_add(struct dmar_drhd_unit *dmaru, struct intel_iommu *iommu)
1442 {
1443 	int ret;
1444 	int eim = x2apic_enabled();
1445 
1446 	ret = intel_cap_audit(CAP_AUDIT_HOTPLUG_IRQR, iommu);
1447 	if (ret)
1448 		return ret;
1449 
1450 	if (eim && !ecap_eim_support(iommu->ecap)) {
1451 		pr_info("DRHD %Lx: EIM not supported by DRHD, ecap %Lx\n",
1452 			iommu->reg_phys, iommu->ecap);
1453 		return -ENODEV;
1454 	}
1455 
1456 	if (ir_parse_ioapic_hpet_scope(dmaru->hdr, iommu)) {
1457 		pr_warn("DRHD %Lx: failed to parse managed IOAPIC/HPET\n",
1458 			iommu->reg_phys);
1459 		return -ENODEV;
1460 	}
1461 
1462 	/* TODO: check all IOAPICs are covered by IOMMU */
1463 
1464 	/* Setup Interrupt-remapping now. */
1465 	ret = intel_setup_irq_remapping(iommu);
1466 	if (ret) {
1467 		pr_err("Failed to setup irq remapping for %s\n",
1468 		       iommu->name);
1469 		intel_teardown_irq_remapping(iommu);
1470 		ir_remove_ioapic_hpet_scope(iommu);
1471 	} else {
1472 		iommu_enable_irq_remapping(iommu);
1473 	}
1474 
1475 	return ret;
1476 }
1477 
1478 int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert)
1479 {
1480 	int ret = 0;
1481 	struct intel_iommu *iommu = dmaru->iommu;
1482 
1483 	if (!irq_remapping_enabled)
1484 		return 0;
1485 	if (iommu == NULL)
1486 		return -EINVAL;
1487 	if (!ecap_ir_support(iommu->ecap))
1488 		return 0;
1489 	if (irq_remapping_cap(IRQ_POSTING_CAP) &&
1490 	    !cap_pi_support(iommu->cap))
1491 		return -EBUSY;
1492 
1493 	if (insert) {
1494 		if (!iommu->ir_table)
1495 			ret = dmar_ir_add(dmaru, iommu);
1496 	} else {
1497 		if (iommu->ir_table) {
1498 			if (!bitmap_empty(iommu->ir_table->bitmap,
1499 					  INTR_REMAP_TABLE_ENTRIES)) {
1500 				ret = -EBUSY;
1501 			} else {
1502 				iommu_disable_irq_remapping(iommu);
1503 				intel_teardown_irq_remapping(iommu);
1504 				ir_remove_ioapic_hpet_scope(iommu);
1505 			}
1506 		}
1507 	}
1508 
1509 	return ret;
1510 }
1511