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 #include <asm/msidef.h>
24 
25 #include "../irq_remapping.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_ir(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 	return NULL;
215 }
216 
217 static struct intel_iommu *map_ioapic_to_ir(int apic)
218 {
219 	int i;
220 
221 	for (i = 0; i < MAX_IO_APICS; i++)
222 		if (ir_ioapic[i].id == apic && ir_ioapic[i].iommu)
223 			return ir_ioapic[i].iommu;
224 	return NULL;
225 }
226 
227 static struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
228 {
229 	struct dmar_drhd_unit *drhd;
230 
231 	drhd = dmar_find_matched_drhd_unit(dev);
232 	if (!drhd)
233 		return NULL;
234 
235 	return drhd->iommu;
236 }
237 
238 static int clear_entries(struct irq_2_iommu *irq_iommu)
239 {
240 	struct irte *start, *entry, *end;
241 	struct intel_iommu *iommu;
242 	int index;
243 
244 	if (irq_iommu->sub_handle)
245 		return 0;
246 
247 	iommu = irq_iommu->iommu;
248 	index = irq_iommu->irte_index;
249 
250 	start = iommu->ir_table->base + index;
251 	end = start + (1 << irq_iommu->irte_mask);
252 
253 	for (entry = start; entry < end; entry++) {
254 		set_64bit(&entry->low, 0);
255 		set_64bit(&entry->high, 0);
256 	}
257 	bitmap_release_region(iommu->ir_table->bitmap, index,
258 			      irq_iommu->irte_mask);
259 
260 	return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
261 }
262 
263 /*
264  * source validation type
265  */
266 #define SVT_NO_VERIFY		0x0  /* no verification is required */
267 #define SVT_VERIFY_SID_SQ	0x1  /* verify using SID and SQ fields */
268 #define SVT_VERIFY_BUS		0x2  /* verify bus of request-id */
269 
270 /*
271  * source-id qualifier
272  */
273 #define SQ_ALL_16	0x0  /* verify all 16 bits of request-id */
274 #define SQ_13_IGNORE_1	0x1  /* verify most significant 13 bits, ignore
275 			      * the third least significant bit
276 			      */
277 #define SQ_13_IGNORE_2	0x2  /* verify most significant 13 bits, ignore
278 			      * the second and third least significant bits
279 			      */
280 #define SQ_13_IGNORE_3	0x3  /* verify most significant 13 bits, ignore
281 			      * the least three significant bits
282 			      */
283 
284 /*
285  * set SVT, SQ and SID fields of irte to verify
286  * source ids of interrupt requests
287  */
288 static void set_irte_sid(struct irte *irte, unsigned int svt,
289 			 unsigned int sq, unsigned int sid)
290 {
291 	if (disable_sourceid_checking)
292 		svt = SVT_NO_VERIFY;
293 	irte->svt = svt;
294 	irte->sq = sq;
295 	irte->sid = sid;
296 }
297 
298 /*
299  * Set an IRTE to match only the bus number. Interrupt requests that reference
300  * this IRTE must have a requester-id whose bus number is between or equal
301  * to the start_bus and end_bus arguments.
302  */
303 static void set_irte_verify_bus(struct irte *irte, unsigned int start_bus,
304 				unsigned int end_bus)
305 {
306 	set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
307 		     (start_bus << 8) | end_bus);
308 }
309 
310 static int set_ioapic_sid(struct irte *irte, int apic)
311 {
312 	int i;
313 	u16 sid = 0;
314 
315 	if (!irte)
316 		return -1;
317 
318 	down_read(&dmar_global_lock);
319 	for (i = 0; i < MAX_IO_APICS; i++) {
320 		if (ir_ioapic[i].iommu && ir_ioapic[i].id == apic) {
321 			sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
322 			break;
323 		}
324 	}
325 	up_read(&dmar_global_lock);
326 
327 	if (sid == 0) {
328 		pr_warn("Failed to set source-id of IOAPIC (%d)\n", apic);
329 		return -1;
330 	}
331 
332 	set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, sid);
333 
334 	return 0;
335 }
336 
337 static int set_hpet_sid(struct irte *irte, u8 id)
338 {
339 	int i;
340 	u16 sid = 0;
341 
342 	if (!irte)
343 		return -1;
344 
345 	down_read(&dmar_global_lock);
346 	for (i = 0; i < MAX_HPET_TBS; i++) {
347 		if (ir_hpet[i].iommu && ir_hpet[i].id == id) {
348 			sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
349 			break;
350 		}
351 	}
352 	up_read(&dmar_global_lock);
353 
354 	if (sid == 0) {
355 		pr_warn("Failed to set source-id of HPET block (%d)\n", id);
356 		return -1;
357 	}
358 
359 	/*
360 	 * Should really use SQ_ALL_16. Some platforms are broken.
361 	 * While we figure out the right quirks for these broken platforms, use
362 	 * SQ_13_IGNORE_3 for now.
363 	 */
364 	set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
365 
366 	return 0;
367 }
368 
369 struct set_msi_sid_data {
370 	struct pci_dev *pdev;
371 	u16 alias;
372 	int count;
373 	int busmatch_count;
374 };
375 
376 static int set_msi_sid_cb(struct pci_dev *pdev, u16 alias, void *opaque)
377 {
378 	struct set_msi_sid_data *data = opaque;
379 
380 	if (data->count == 0 || PCI_BUS_NUM(alias) == PCI_BUS_NUM(data->alias))
381 		data->busmatch_count++;
382 
383 	data->pdev = pdev;
384 	data->alias = alias;
385 	data->count++;
386 
387 	return 0;
388 }
389 
390 static int set_msi_sid(struct irte *irte, struct pci_dev *dev)
391 {
392 	struct set_msi_sid_data data;
393 
394 	if (!irte || !dev)
395 		return -1;
396 
397 	data.count = 0;
398 	data.busmatch_count = 0;
399 	pci_for_each_dma_alias(dev, set_msi_sid_cb, &data);
400 
401 	/*
402 	 * DMA alias provides us with a PCI device and alias.  The only case
403 	 * where the it will return an alias on a different bus than the
404 	 * device is the case of a PCIe-to-PCI bridge, where the alias is for
405 	 * the subordinate bus.  In this case we can only verify the bus.
406 	 *
407 	 * If there are multiple aliases, all with the same bus number,
408 	 * then all we can do is verify the bus. This is typical in NTB
409 	 * hardware which use proxy IDs where the device will generate traffic
410 	 * from multiple devfn numbers on the same bus.
411 	 *
412 	 * If the alias device is on a different bus than our source device
413 	 * then we have a topology based alias, use it.
414 	 *
415 	 * Otherwise, the alias is for a device DMA quirk and we cannot
416 	 * assume that MSI uses the same requester ID.  Therefore use the
417 	 * original device.
418 	 */
419 	if (PCI_BUS_NUM(data.alias) != data.pdev->bus->number)
420 		set_irte_verify_bus(irte, PCI_BUS_NUM(data.alias),
421 				    dev->bus->number);
422 	else if (data.count >= 2 && data.busmatch_count == data.count)
423 		set_irte_verify_bus(irte, dev->bus->number, dev->bus->number);
424 	else if (data.pdev->bus->number != dev->bus->number)
425 		set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, data.alias);
426 	else
427 		set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
428 			     pci_dev_id(dev));
429 
430 	return 0;
431 }
432 
433 static int iommu_load_old_irte(struct intel_iommu *iommu)
434 {
435 	struct irte *old_ir_table;
436 	phys_addr_t irt_phys;
437 	unsigned int i;
438 	size_t size;
439 	u64 irta;
440 
441 	/* Check whether the old ir-table has the same size as ours */
442 	irta = dmar_readq(iommu->reg + DMAR_IRTA_REG);
443 	if ((irta & INTR_REMAP_TABLE_REG_SIZE_MASK)
444 	     != INTR_REMAP_TABLE_REG_SIZE)
445 		return -EINVAL;
446 
447 	irt_phys = irta & VTD_PAGE_MASK;
448 	size     = INTR_REMAP_TABLE_ENTRIES*sizeof(struct irte);
449 
450 	/* Map the old IR table */
451 	old_ir_table = memremap(irt_phys, size, MEMREMAP_WB);
452 	if (!old_ir_table)
453 		return -ENOMEM;
454 
455 	/* Copy data over */
456 	memcpy(iommu->ir_table->base, old_ir_table, size);
457 
458 	__iommu_flush_cache(iommu, iommu->ir_table->base, size);
459 
460 	/*
461 	 * Now check the table for used entries and mark those as
462 	 * allocated in the bitmap
463 	 */
464 	for (i = 0; i < INTR_REMAP_TABLE_ENTRIES; i++) {
465 		if (iommu->ir_table->base[i].present)
466 			bitmap_set(iommu->ir_table->bitmap, i, 1);
467 	}
468 
469 	memunmap(old_ir_table);
470 
471 	return 0;
472 }
473 
474 
475 static void iommu_set_irq_remapping(struct intel_iommu *iommu, int mode)
476 {
477 	unsigned long flags;
478 	u64 addr;
479 	u32 sts;
480 
481 	addr = virt_to_phys((void *)iommu->ir_table->base);
482 
483 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
484 
485 	dmar_writeq(iommu->reg + DMAR_IRTA_REG,
486 		    (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
487 
488 	/* Set interrupt-remapping table pointer */
489 	writel(iommu->gcmd | DMA_GCMD_SIRTP, iommu->reg + DMAR_GCMD_REG);
490 
491 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
492 		      readl, (sts & DMA_GSTS_IRTPS), sts);
493 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
494 
495 	/*
496 	 * Global invalidation of interrupt entry cache to make sure the
497 	 * hardware uses the new irq remapping table.
498 	 */
499 	qi_global_iec(iommu);
500 }
501 
502 static void iommu_enable_irq_remapping(struct intel_iommu *iommu)
503 {
504 	unsigned long flags;
505 	u32 sts;
506 
507 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
508 
509 	/* Enable interrupt-remapping */
510 	iommu->gcmd |= DMA_GCMD_IRE;
511 	iommu->gcmd &= ~DMA_GCMD_CFI;  /* Block compatibility-format MSIs */
512 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
513 
514 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
515 		      readl, (sts & DMA_GSTS_IRES), sts);
516 
517 	/*
518 	 * With CFI clear in the Global Command register, we should be
519 	 * protected from dangerous (i.e. compatibility) interrupts
520 	 * regardless of x2apic status.  Check just to be sure.
521 	 */
522 	if (sts & DMA_GSTS_CFIS)
523 		WARN(1, KERN_WARNING
524 			"Compatibility-format IRQs enabled despite intr remapping;\n"
525 			"you are vulnerable to IRQ injection.\n");
526 
527 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
528 }
529 
530 static int intel_setup_irq_remapping(struct intel_iommu *iommu)
531 {
532 	struct ir_table *ir_table;
533 	struct fwnode_handle *fn;
534 	unsigned long *bitmap;
535 	struct page *pages;
536 
537 	if (iommu->ir_table)
538 		return 0;
539 
540 	ir_table = kzalloc(sizeof(struct ir_table), GFP_KERNEL);
541 	if (!ir_table)
542 		return -ENOMEM;
543 
544 	pages = alloc_pages_node(iommu->node, GFP_KERNEL | __GFP_ZERO,
545 				 INTR_REMAP_PAGE_ORDER);
546 	if (!pages) {
547 		pr_err("IR%d: failed to allocate pages of order %d\n",
548 		       iommu->seq_id, INTR_REMAP_PAGE_ORDER);
549 		goto out_free_table;
550 	}
551 
552 	bitmap = bitmap_zalloc(INTR_REMAP_TABLE_ENTRIES, GFP_ATOMIC);
553 	if (bitmap == NULL) {
554 		pr_err("IR%d: failed to allocate bitmap\n", iommu->seq_id);
555 		goto out_free_pages;
556 	}
557 
558 	fn = irq_domain_alloc_named_id_fwnode("INTEL-IR", iommu->seq_id);
559 	if (!fn)
560 		goto out_free_bitmap;
561 
562 	iommu->ir_domain =
563 		irq_domain_create_hierarchy(arch_get_ir_parent_domain(),
564 					    0, INTR_REMAP_TABLE_ENTRIES,
565 					    fn, &intel_ir_domain_ops,
566 					    iommu);
567 	if (!iommu->ir_domain) {
568 		irq_domain_free_fwnode(fn);
569 		pr_err("IR%d: failed to allocate irqdomain\n", iommu->seq_id);
570 		goto out_free_bitmap;
571 	}
572 	iommu->ir_msi_domain =
573 		arch_create_remap_msi_irq_domain(iommu->ir_domain,
574 						 "INTEL-IR-MSI",
575 						 iommu->seq_id);
576 
577 	ir_table->base = page_address(pages);
578 	ir_table->bitmap = bitmap;
579 	iommu->ir_table = ir_table;
580 
581 	/*
582 	 * If the queued invalidation is already initialized,
583 	 * shouldn't disable it.
584 	 */
585 	if (!iommu->qi) {
586 		/*
587 		 * Clear previous faults.
588 		 */
589 		dmar_fault(-1, iommu);
590 		dmar_disable_qi(iommu);
591 
592 		if (dmar_enable_qi(iommu)) {
593 			pr_err("Failed to enable queued invalidation\n");
594 			goto out_free_bitmap;
595 		}
596 	}
597 
598 	init_ir_status(iommu);
599 
600 	if (ir_pre_enabled(iommu)) {
601 		if (!is_kdump_kernel()) {
602 			pr_warn("IRQ remapping was enabled on %s but we are not in kdump mode\n",
603 				iommu->name);
604 			clear_ir_pre_enabled(iommu);
605 			iommu_disable_irq_remapping(iommu);
606 		} else if (iommu_load_old_irte(iommu))
607 			pr_err("Failed to copy IR table for %s from previous kernel\n",
608 			       iommu->name);
609 		else
610 			pr_info("Copied IR table for %s from previous kernel\n",
611 				iommu->name);
612 	}
613 
614 	iommu_set_irq_remapping(iommu, eim_mode);
615 
616 	return 0;
617 
618 out_free_bitmap:
619 	bitmap_free(bitmap);
620 out_free_pages:
621 	__free_pages(pages, INTR_REMAP_PAGE_ORDER);
622 out_free_table:
623 	kfree(ir_table);
624 
625 	iommu->ir_table  = NULL;
626 
627 	return -ENOMEM;
628 }
629 
630 static void intel_teardown_irq_remapping(struct intel_iommu *iommu)
631 {
632 	struct fwnode_handle *fn;
633 
634 	if (iommu && iommu->ir_table) {
635 		if (iommu->ir_msi_domain) {
636 			fn = iommu->ir_msi_domain->fwnode;
637 
638 			irq_domain_remove(iommu->ir_msi_domain);
639 			irq_domain_free_fwnode(fn);
640 			iommu->ir_msi_domain = NULL;
641 		}
642 		if (iommu->ir_domain) {
643 			fn = iommu->ir_domain->fwnode;
644 
645 			irq_domain_remove(iommu->ir_domain);
646 			irq_domain_free_fwnode(fn);
647 			iommu->ir_domain = NULL;
648 		}
649 		free_pages((unsigned long)iommu->ir_table->base,
650 			   INTR_REMAP_PAGE_ORDER);
651 		bitmap_free(iommu->ir_table->bitmap);
652 		kfree(iommu->ir_table);
653 		iommu->ir_table = NULL;
654 	}
655 }
656 
657 /*
658  * Disable Interrupt Remapping.
659  */
660 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
661 {
662 	unsigned long flags;
663 	u32 sts;
664 
665 	if (!ecap_ir_support(iommu->ecap))
666 		return;
667 
668 	/*
669 	 * global invalidation of interrupt entry cache before disabling
670 	 * interrupt-remapping.
671 	 */
672 	qi_global_iec(iommu);
673 
674 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
675 
676 	sts = readl(iommu->reg + DMAR_GSTS_REG);
677 	if (!(sts & DMA_GSTS_IRES))
678 		goto end;
679 
680 	iommu->gcmd &= ~DMA_GCMD_IRE;
681 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
682 
683 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
684 		      readl, !(sts & DMA_GSTS_IRES), sts);
685 
686 end:
687 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
688 }
689 
690 static int __init dmar_x2apic_optout(void)
691 {
692 	struct acpi_table_dmar *dmar;
693 	dmar = (struct acpi_table_dmar *)dmar_tbl;
694 	if (!dmar || no_x2apic_optout)
695 		return 0;
696 	return dmar->flags & DMAR_X2APIC_OPT_OUT;
697 }
698 
699 static void __init intel_cleanup_irq_remapping(void)
700 {
701 	struct dmar_drhd_unit *drhd;
702 	struct intel_iommu *iommu;
703 
704 	for_each_iommu(iommu, drhd) {
705 		if (ecap_ir_support(iommu->ecap)) {
706 			iommu_disable_irq_remapping(iommu);
707 			intel_teardown_irq_remapping(iommu);
708 		}
709 	}
710 
711 	if (x2apic_supported())
712 		pr_warn("Failed to enable irq remapping. You are vulnerable to irq-injection attacks.\n");
713 }
714 
715 static int __init intel_prepare_irq_remapping(void)
716 {
717 	struct dmar_drhd_unit *drhd;
718 	struct intel_iommu *iommu;
719 	int eim = 0;
720 
721 	if (irq_remap_broken) {
722 		pr_warn("This system BIOS has enabled interrupt remapping\n"
723 			"on a chipset that contains an erratum making that\n"
724 			"feature unstable.  To maintain system stability\n"
725 			"interrupt remapping is being disabled.  Please\n"
726 			"contact your BIOS vendor for an update\n");
727 		add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
728 		return -ENODEV;
729 	}
730 
731 	if (dmar_table_init() < 0)
732 		return -ENODEV;
733 
734 	if (!dmar_ir_support())
735 		return -ENODEV;
736 
737 	if (parse_ioapics_under_ir()) {
738 		pr_info("Not enabling interrupt remapping\n");
739 		goto error;
740 	}
741 
742 	/* First make sure all IOMMUs support IRQ remapping */
743 	for_each_iommu(iommu, drhd)
744 		if (!ecap_ir_support(iommu->ecap))
745 			goto error;
746 
747 	/* Detect remapping mode: lapic or x2apic */
748 	if (x2apic_supported()) {
749 		eim = !dmar_x2apic_optout();
750 		if (!eim) {
751 			pr_info("x2apic is disabled because BIOS sets x2apic opt out bit.");
752 			pr_info("Use 'intremap=no_x2apic_optout' to override the BIOS setting.\n");
753 		}
754 	}
755 
756 	for_each_iommu(iommu, drhd) {
757 		if (eim && !ecap_eim_support(iommu->ecap)) {
758 			pr_info("%s does not support EIM\n", iommu->name);
759 			eim = 0;
760 		}
761 	}
762 
763 	eim_mode = eim;
764 	if (eim)
765 		pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
766 
767 	/* Do the initializations early */
768 	for_each_iommu(iommu, drhd) {
769 		if (intel_setup_irq_remapping(iommu)) {
770 			pr_err("Failed to setup irq remapping for %s\n",
771 			       iommu->name);
772 			goto error;
773 		}
774 	}
775 
776 	return 0;
777 
778 error:
779 	intel_cleanup_irq_remapping();
780 	return -ENODEV;
781 }
782 
783 /*
784  * Set Posted-Interrupts capability.
785  */
786 static inline void set_irq_posting_cap(void)
787 {
788 	struct dmar_drhd_unit *drhd;
789 	struct intel_iommu *iommu;
790 
791 	if (!disable_irq_post) {
792 		/*
793 		 * If IRTE is in posted format, the 'pda' field goes across the
794 		 * 64-bit boundary, we need use cmpxchg16b to atomically update
795 		 * it. We only expose posted-interrupt when X86_FEATURE_CX16
796 		 * is supported. Actually, hardware platforms supporting PI
797 		 * should have X86_FEATURE_CX16 support, this has been confirmed
798 		 * with Intel hardware guys.
799 		 */
800 		if (boot_cpu_has(X86_FEATURE_CX16))
801 			intel_irq_remap_ops.capability |= 1 << IRQ_POSTING_CAP;
802 
803 		for_each_iommu(iommu, drhd)
804 			if (!cap_pi_support(iommu->cap)) {
805 				intel_irq_remap_ops.capability &=
806 						~(1 << IRQ_POSTING_CAP);
807 				break;
808 			}
809 	}
810 }
811 
812 static int __init intel_enable_irq_remapping(void)
813 {
814 	struct dmar_drhd_unit *drhd;
815 	struct intel_iommu *iommu;
816 	bool setup = false;
817 
818 	/*
819 	 * Setup Interrupt-remapping for all the DRHD's now.
820 	 */
821 	for_each_iommu(iommu, drhd) {
822 		if (!ir_pre_enabled(iommu))
823 			iommu_enable_irq_remapping(iommu);
824 		setup = true;
825 	}
826 
827 	if (!setup)
828 		goto error;
829 
830 	irq_remapping_enabled = 1;
831 
832 	set_irq_posting_cap();
833 
834 	pr_info("Enabled IRQ remapping in %s mode\n", eim_mode ? "x2apic" : "xapic");
835 
836 	return eim_mode ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
837 
838 error:
839 	intel_cleanup_irq_remapping();
840 	return -1;
841 }
842 
843 static int ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
844 				   struct intel_iommu *iommu,
845 				   struct acpi_dmar_hardware_unit *drhd)
846 {
847 	struct acpi_dmar_pci_path *path;
848 	u8 bus;
849 	int count, free = -1;
850 
851 	bus = scope->bus;
852 	path = (struct acpi_dmar_pci_path *)(scope + 1);
853 	count = (scope->length - sizeof(struct acpi_dmar_device_scope))
854 		/ sizeof(struct acpi_dmar_pci_path);
855 
856 	while (--count > 0) {
857 		/*
858 		 * Access PCI directly due to the PCI
859 		 * subsystem isn't initialized yet.
860 		 */
861 		bus = read_pci_config_byte(bus, path->device, path->function,
862 					   PCI_SECONDARY_BUS);
863 		path++;
864 	}
865 
866 	for (count = 0; count < MAX_HPET_TBS; count++) {
867 		if (ir_hpet[count].iommu == iommu &&
868 		    ir_hpet[count].id == scope->enumeration_id)
869 			return 0;
870 		else if (ir_hpet[count].iommu == NULL && free == -1)
871 			free = count;
872 	}
873 	if (free == -1) {
874 		pr_warn("Exceeded Max HPET blocks\n");
875 		return -ENOSPC;
876 	}
877 
878 	ir_hpet[free].iommu = iommu;
879 	ir_hpet[free].id    = scope->enumeration_id;
880 	ir_hpet[free].bus   = bus;
881 	ir_hpet[free].devfn = PCI_DEVFN(path->device, path->function);
882 	pr_info("HPET id %d under DRHD base 0x%Lx\n",
883 		scope->enumeration_id, drhd->address);
884 
885 	return 0;
886 }
887 
888 static int ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
889 				     struct intel_iommu *iommu,
890 				     struct acpi_dmar_hardware_unit *drhd)
891 {
892 	struct acpi_dmar_pci_path *path;
893 	u8 bus;
894 	int count, free = -1;
895 
896 	bus = scope->bus;
897 	path = (struct acpi_dmar_pci_path *)(scope + 1);
898 	count = (scope->length - sizeof(struct acpi_dmar_device_scope))
899 		/ sizeof(struct acpi_dmar_pci_path);
900 
901 	while (--count > 0) {
902 		/*
903 		 * Access PCI directly due to the PCI
904 		 * subsystem isn't initialized yet.
905 		 */
906 		bus = read_pci_config_byte(bus, path->device, path->function,
907 					   PCI_SECONDARY_BUS);
908 		path++;
909 	}
910 
911 	for (count = 0; count < MAX_IO_APICS; count++) {
912 		if (ir_ioapic[count].iommu == iommu &&
913 		    ir_ioapic[count].id == scope->enumeration_id)
914 			return 0;
915 		else if (ir_ioapic[count].iommu == NULL && free == -1)
916 			free = count;
917 	}
918 	if (free == -1) {
919 		pr_warn("Exceeded Max IO APICS\n");
920 		return -ENOSPC;
921 	}
922 
923 	ir_ioapic[free].bus   = bus;
924 	ir_ioapic[free].devfn = PCI_DEVFN(path->device, path->function);
925 	ir_ioapic[free].iommu = iommu;
926 	ir_ioapic[free].id    = scope->enumeration_id;
927 	pr_info("IOAPIC id %d under DRHD base  0x%Lx IOMMU %d\n",
928 		scope->enumeration_id, drhd->address, iommu->seq_id);
929 
930 	return 0;
931 }
932 
933 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
934 				      struct intel_iommu *iommu)
935 {
936 	int ret = 0;
937 	struct acpi_dmar_hardware_unit *drhd;
938 	struct acpi_dmar_device_scope *scope;
939 	void *start, *end;
940 
941 	drhd = (struct acpi_dmar_hardware_unit *)header;
942 	start = (void *)(drhd + 1);
943 	end = ((void *)drhd) + header->length;
944 
945 	while (start < end && ret == 0) {
946 		scope = start;
947 		if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC)
948 			ret = ir_parse_one_ioapic_scope(scope, iommu, drhd);
949 		else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET)
950 			ret = ir_parse_one_hpet_scope(scope, iommu, drhd);
951 		start += scope->length;
952 	}
953 
954 	return ret;
955 }
956 
957 static void ir_remove_ioapic_hpet_scope(struct intel_iommu *iommu)
958 {
959 	int i;
960 
961 	for (i = 0; i < MAX_HPET_TBS; i++)
962 		if (ir_hpet[i].iommu == iommu)
963 			ir_hpet[i].iommu = NULL;
964 
965 	for (i = 0; i < MAX_IO_APICS; i++)
966 		if (ir_ioapic[i].iommu == iommu)
967 			ir_ioapic[i].iommu = NULL;
968 }
969 
970 /*
971  * Finds the assocaition between IOAPIC's and its Interrupt-remapping
972  * hardware unit.
973  */
974 static int __init parse_ioapics_under_ir(void)
975 {
976 	struct dmar_drhd_unit *drhd;
977 	struct intel_iommu *iommu;
978 	bool ir_supported = false;
979 	int ioapic_idx;
980 
981 	for_each_iommu(iommu, drhd) {
982 		int ret;
983 
984 		if (!ecap_ir_support(iommu->ecap))
985 			continue;
986 
987 		ret = ir_parse_ioapic_hpet_scope(drhd->hdr, iommu);
988 		if (ret)
989 			return ret;
990 
991 		ir_supported = true;
992 	}
993 
994 	if (!ir_supported)
995 		return -ENODEV;
996 
997 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
998 		int ioapic_id = mpc_ioapic_id(ioapic_idx);
999 		if (!map_ioapic_to_ir(ioapic_id)) {
1000 			pr_err(FW_BUG "ioapic %d has no mapping iommu, "
1001 			       "interrupt remapping will be disabled\n",
1002 			       ioapic_id);
1003 			return -1;
1004 		}
1005 	}
1006 
1007 	return 0;
1008 }
1009 
1010 static int __init ir_dev_scope_init(void)
1011 {
1012 	int ret;
1013 
1014 	if (!irq_remapping_enabled)
1015 		return 0;
1016 
1017 	down_write(&dmar_global_lock);
1018 	ret = dmar_dev_scope_init();
1019 	up_write(&dmar_global_lock);
1020 
1021 	return ret;
1022 }
1023 rootfs_initcall(ir_dev_scope_init);
1024 
1025 static void disable_irq_remapping(void)
1026 {
1027 	struct dmar_drhd_unit *drhd;
1028 	struct intel_iommu *iommu = NULL;
1029 
1030 	/*
1031 	 * Disable Interrupt-remapping for all the DRHD's now.
1032 	 */
1033 	for_each_iommu(iommu, drhd) {
1034 		if (!ecap_ir_support(iommu->ecap))
1035 			continue;
1036 
1037 		iommu_disable_irq_remapping(iommu);
1038 	}
1039 
1040 	/*
1041 	 * Clear Posted-Interrupts capability.
1042 	 */
1043 	if (!disable_irq_post)
1044 		intel_irq_remap_ops.capability &= ~(1 << IRQ_POSTING_CAP);
1045 }
1046 
1047 static int reenable_irq_remapping(int eim)
1048 {
1049 	struct dmar_drhd_unit *drhd;
1050 	bool setup = false;
1051 	struct intel_iommu *iommu = NULL;
1052 
1053 	for_each_iommu(iommu, drhd)
1054 		if (iommu->qi)
1055 			dmar_reenable_qi(iommu);
1056 
1057 	/*
1058 	 * Setup Interrupt-remapping for all the DRHD's now.
1059 	 */
1060 	for_each_iommu(iommu, drhd) {
1061 		if (!ecap_ir_support(iommu->ecap))
1062 			continue;
1063 
1064 		/* Set up interrupt remapping for iommu.*/
1065 		iommu_set_irq_remapping(iommu, eim);
1066 		iommu_enable_irq_remapping(iommu);
1067 		setup = true;
1068 	}
1069 
1070 	if (!setup)
1071 		goto error;
1072 
1073 	set_irq_posting_cap();
1074 
1075 	return 0;
1076 
1077 error:
1078 	/*
1079 	 * handle error condition gracefully here!
1080 	 */
1081 	return -1;
1082 }
1083 
1084 static void prepare_irte(struct irte *irte, int vector, unsigned int dest)
1085 {
1086 	memset(irte, 0, sizeof(*irte));
1087 
1088 	irte->present = 1;
1089 	irte->dst_mode = apic->irq_dest_mode;
1090 	/*
1091 	 * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
1092 	 * actual level or edge trigger will be setup in the IO-APIC
1093 	 * RTE. This will help simplify level triggered irq migration.
1094 	 * For more details, see the comments (in io_apic.c) explainig IO-APIC
1095 	 * irq migration in the presence of interrupt-remapping.
1096 	*/
1097 	irte->trigger_mode = 0;
1098 	irte->dlvry_mode = apic->irq_delivery_mode;
1099 	irte->vector = vector;
1100 	irte->dest_id = IRTE_DEST(dest);
1101 	irte->redir_hint = 1;
1102 }
1103 
1104 static struct irq_domain *intel_get_ir_irq_domain(struct irq_alloc_info *info)
1105 {
1106 	struct intel_iommu *iommu = NULL;
1107 
1108 	if (!info)
1109 		return NULL;
1110 
1111 	switch (info->type) {
1112 	case X86_IRQ_ALLOC_TYPE_IOAPIC:
1113 		iommu = map_ioapic_to_ir(info->ioapic_id);
1114 		break;
1115 	case X86_IRQ_ALLOC_TYPE_HPET:
1116 		iommu = map_hpet_to_ir(info->hpet_id);
1117 		break;
1118 	case X86_IRQ_ALLOC_TYPE_MSI:
1119 	case X86_IRQ_ALLOC_TYPE_MSIX:
1120 		iommu = map_dev_to_ir(info->msi_dev);
1121 		break;
1122 	default:
1123 		BUG_ON(1);
1124 		break;
1125 	}
1126 
1127 	return iommu ? iommu->ir_domain : NULL;
1128 }
1129 
1130 static struct irq_domain *intel_get_irq_domain(struct irq_alloc_info *info)
1131 {
1132 	struct intel_iommu *iommu;
1133 
1134 	if (!info)
1135 		return NULL;
1136 
1137 	switch (info->type) {
1138 	case X86_IRQ_ALLOC_TYPE_MSI:
1139 	case X86_IRQ_ALLOC_TYPE_MSIX:
1140 		iommu = map_dev_to_ir(info->msi_dev);
1141 		if (iommu)
1142 			return iommu->ir_msi_domain;
1143 		break;
1144 	default:
1145 		break;
1146 	}
1147 
1148 	return NULL;
1149 }
1150 
1151 struct irq_remap_ops intel_irq_remap_ops = {
1152 	.prepare		= intel_prepare_irq_remapping,
1153 	.enable			= intel_enable_irq_remapping,
1154 	.disable		= disable_irq_remapping,
1155 	.reenable		= reenable_irq_remapping,
1156 	.enable_faulting	= enable_drhd_fault_handling,
1157 	.get_ir_irq_domain	= intel_get_ir_irq_domain,
1158 	.get_irq_domain		= intel_get_irq_domain,
1159 };
1160 
1161 static void intel_ir_reconfigure_irte(struct irq_data *irqd, bool force)
1162 {
1163 	struct intel_ir_data *ir_data = irqd->chip_data;
1164 	struct irte *irte = &ir_data->irte_entry;
1165 	struct irq_cfg *cfg = irqd_cfg(irqd);
1166 
1167 	/*
1168 	 * Atomically updates the IRTE with the new destination, vector
1169 	 * and flushes the interrupt entry cache.
1170 	 */
1171 	irte->vector = cfg->vector;
1172 	irte->dest_id = IRTE_DEST(cfg->dest_apicid);
1173 
1174 	/* Update the hardware only if the interrupt is in remapped mode. */
1175 	if (force || ir_data->irq_2_iommu.mode == IRQ_REMAPPING)
1176 		modify_irte(&ir_data->irq_2_iommu, irte);
1177 }
1178 
1179 /*
1180  * Migrate the IO-APIC irq in the presence of intr-remapping.
1181  *
1182  * For both level and edge triggered, irq migration is a simple atomic
1183  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
1184  *
1185  * For level triggered, we eliminate the io-apic RTE modification (with the
1186  * updated vector information), by using a virtual vector (io-apic pin number).
1187  * Real vector that is used for interrupting cpu will be coming from
1188  * the interrupt-remapping table entry.
1189  *
1190  * As the migration is a simple atomic update of IRTE, the same mechanism
1191  * is used to migrate MSI irq's in the presence of interrupt-remapping.
1192  */
1193 static int
1194 intel_ir_set_affinity(struct irq_data *data, const struct cpumask *mask,
1195 		      bool force)
1196 {
1197 	struct irq_data *parent = data->parent_data;
1198 	struct irq_cfg *cfg = irqd_cfg(data);
1199 	int ret;
1200 
1201 	ret = parent->chip->irq_set_affinity(parent, mask, force);
1202 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
1203 		return ret;
1204 
1205 	intel_ir_reconfigure_irte(data, false);
1206 	/*
1207 	 * After this point, all the interrupts will start arriving
1208 	 * at the new destination. So, time to cleanup the previous
1209 	 * vector allocation.
1210 	 */
1211 	send_cleanup_vector(cfg);
1212 
1213 	return IRQ_SET_MASK_OK_DONE;
1214 }
1215 
1216 static void intel_ir_compose_msi_msg(struct irq_data *irq_data,
1217 				     struct msi_msg *msg)
1218 {
1219 	struct intel_ir_data *ir_data = irq_data->chip_data;
1220 
1221 	*msg = ir_data->msi_entry;
1222 }
1223 
1224 static int intel_ir_set_vcpu_affinity(struct irq_data *data, void *info)
1225 {
1226 	struct intel_ir_data *ir_data = data->chip_data;
1227 	struct vcpu_data *vcpu_pi_info = info;
1228 
1229 	/* stop posting interrupts, back to remapping mode */
1230 	if (!vcpu_pi_info) {
1231 		modify_irte(&ir_data->irq_2_iommu, &ir_data->irte_entry);
1232 	} else {
1233 		struct irte irte_pi;
1234 
1235 		/*
1236 		 * We are not caching the posted interrupt entry. We
1237 		 * copy the data from the remapped entry and modify
1238 		 * the fields which are relevant for posted mode. The
1239 		 * cached remapped entry is used for switching back to
1240 		 * remapped mode.
1241 		 */
1242 		memset(&irte_pi, 0, sizeof(irte_pi));
1243 		dmar_copy_shared_irte(&irte_pi, &ir_data->irte_entry);
1244 
1245 		/* Update the posted mode fields */
1246 		irte_pi.p_pst = 1;
1247 		irte_pi.p_urgent = 0;
1248 		irte_pi.p_vector = vcpu_pi_info->vector;
1249 		irte_pi.pda_l = (vcpu_pi_info->pi_desc_addr >>
1250 				(32 - PDA_LOW_BIT)) & ~(-1UL << PDA_LOW_BIT);
1251 		irte_pi.pda_h = (vcpu_pi_info->pi_desc_addr >> 32) &
1252 				~(-1UL << PDA_HIGH_BIT);
1253 
1254 		modify_irte(&ir_data->irq_2_iommu, &irte_pi);
1255 	}
1256 
1257 	return 0;
1258 }
1259 
1260 static struct irq_chip intel_ir_chip = {
1261 	.name			= "INTEL-IR",
1262 	.irq_ack		= apic_ack_irq,
1263 	.irq_set_affinity	= intel_ir_set_affinity,
1264 	.irq_compose_msi_msg	= intel_ir_compose_msi_msg,
1265 	.irq_set_vcpu_affinity	= intel_ir_set_vcpu_affinity,
1266 };
1267 
1268 static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,
1269 					     struct irq_cfg *irq_cfg,
1270 					     struct irq_alloc_info *info,
1271 					     int index, int sub_handle)
1272 {
1273 	struct IR_IO_APIC_route_entry *entry;
1274 	struct irte *irte = &data->irte_entry;
1275 	struct msi_msg *msg = &data->msi_entry;
1276 
1277 	prepare_irte(irte, irq_cfg->vector, irq_cfg->dest_apicid);
1278 	switch (info->type) {
1279 	case X86_IRQ_ALLOC_TYPE_IOAPIC:
1280 		/* Set source-id of interrupt request */
1281 		set_ioapic_sid(irte, info->ioapic_id);
1282 		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",
1283 			info->ioapic_id, irte->present, irte->fpd,
1284 			irte->dst_mode, irte->redir_hint,
1285 			irte->trigger_mode, irte->dlvry_mode,
1286 			irte->avail, irte->vector, irte->dest_id,
1287 			irte->sid, irte->sq, irte->svt);
1288 
1289 		entry = (struct IR_IO_APIC_route_entry *)info->ioapic_entry;
1290 		info->ioapic_entry = NULL;
1291 		memset(entry, 0, sizeof(*entry));
1292 		entry->index2	= (index >> 15) & 0x1;
1293 		entry->zero	= 0;
1294 		entry->format	= 1;
1295 		entry->index	= (index & 0x7fff);
1296 		/*
1297 		 * IO-APIC RTE will be configured with virtual vector.
1298 		 * irq handler will do the explicit EOI to the io-apic.
1299 		 */
1300 		entry->vector	= info->ioapic_pin;
1301 		entry->mask	= 0;			/* enable IRQ */
1302 		entry->trigger	= info->ioapic_trigger;
1303 		entry->polarity	= info->ioapic_polarity;
1304 		if (info->ioapic_trigger)
1305 			entry->mask = 1; /* Mask level triggered irqs. */
1306 		break;
1307 
1308 	case X86_IRQ_ALLOC_TYPE_HPET:
1309 	case X86_IRQ_ALLOC_TYPE_MSI:
1310 	case X86_IRQ_ALLOC_TYPE_MSIX:
1311 		if (info->type == X86_IRQ_ALLOC_TYPE_HPET)
1312 			set_hpet_sid(irte, info->hpet_id);
1313 		else
1314 			set_msi_sid(irte, info->msi_dev);
1315 
1316 		msg->address_hi = MSI_ADDR_BASE_HI;
1317 		msg->data = sub_handle;
1318 		msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
1319 				  MSI_ADDR_IR_SHV |
1320 				  MSI_ADDR_IR_INDEX1(index) |
1321 				  MSI_ADDR_IR_INDEX2(index);
1322 		break;
1323 
1324 	default:
1325 		BUG_ON(1);
1326 		break;
1327 	}
1328 }
1329 
1330 static void intel_free_irq_resources(struct irq_domain *domain,
1331 				     unsigned int virq, unsigned int nr_irqs)
1332 {
1333 	struct irq_data *irq_data;
1334 	struct intel_ir_data *data;
1335 	struct irq_2_iommu *irq_iommu;
1336 	unsigned long flags;
1337 	int i;
1338 	for (i = 0; i < nr_irqs; i++) {
1339 		irq_data = irq_domain_get_irq_data(domain, virq  + i);
1340 		if (irq_data && irq_data->chip_data) {
1341 			data = irq_data->chip_data;
1342 			irq_iommu = &data->irq_2_iommu;
1343 			raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
1344 			clear_entries(irq_iommu);
1345 			raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
1346 			irq_domain_reset_irq_data(irq_data);
1347 			kfree(data);
1348 		}
1349 	}
1350 }
1351 
1352 static int intel_irq_remapping_alloc(struct irq_domain *domain,
1353 				     unsigned int virq, unsigned int nr_irqs,
1354 				     void *arg)
1355 {
1356 	struct intel_iommu *iommu = domain->host_data;
1357 	struct irq_alloc_info *info = arg;
1358 	struct intel_ir_data *data, *ird;
1359 	struct irq_data *irq_data;
1360 	struct irq_cfg *irq_cfg;
1361 	int i, ret, index;
1362 
1363 	if (!info || !iommu)
1364 		return -EINVAL;
1365 	if (nr_irqs > 1 && info->type != X86_IRQ_ALLOC_TYPE_MSI &&
1366 	    info->type != X86_IRQ_ALLOC_TYPE_MSIX)
1367 		return -EINVAL;
1368 
1369 	/*
1370 	 * With IRQ remapping enabled, don't need contiguous CPU vectors
1371 	 * to support multiple MSI interrupts.
1372 	 */
1373 	if (info->type == X86_IRQ_ALLOC_TYPE_MSI)
1374 		info->flags &= ~X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
1375 
1376 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
1377 	if (ret < 0)
1378 		return ret;
1379 
1380 	ret = -ENOMEM;
1381 	data = kzalloc(sizeof(*data), GFP_KERNEL);
1382 	if (!data)
1383 		goto out_free_parent;
1384 
1385 	down_read(&dmar_global_lock);
1386 	index = alloc_irte(iommu, &data->irq_2_iommu, nr_irqs);
1387 	up_read(&dmar_global_lock);
1388 	if (index < 0) {
1389 		pr_warn("Failed to allocate IRTE\n");
1390 		kfree(data);
1391 		goto out_free_parent;
1392 	}
1393 
1394 	for (i = 0; i < nr_irqs; i++) {
1395 		irq_data = irq_domain_get_irq_data(domain, virq + i);
1396 		irq_cfg = irqd_cfg(irq_data);
1397 		if (!irq_data || !irq_cfg) {
1398 			ret = -EINVAL;
1399 			goto out_free_data;
1400 		}
1401 
1402 		if (i > 0) {
1403 			ird = kzalloc(sizeof(*ird), GFP_KERNEL);
1404 			if (!ird)
1405 				goto out_free_data;
1406 			/* Initialize the common data */
1407 			ird->irq_2_iommu = data->irq_2_iommu;
1408 			ird->irq_2_iommu.sub_handle = i;
1409 		} else {
1410 			ird = data;
1411 		}
1412 
1413 		irq_data->hwirq = (index << 16) + i;
1414 		irq_data->chip_data = ird;
1415 		irq_data->chip = &intel_ir_chip;
1416 		intel_irq_remapping_prepare_irte(ird, irq_cfg, info, index, i);
1417 		irq_set_status_flags(virq + i, IRQ_MOVE_PCNTXT);
1418 	}
1419 	return 0;
1420 
1421 out_free_data:
1422 	intel_free_irq_resources(domain, virq, i);
1423 out_free_parent:
1424 	irq_domain_free_irqs_common(domain, virq, nr_irqs);
1425 	return ret;
1426 }
1427 
1428 static void intel_irq_remapping_free(struct irq_domain *domain,
1429 				     unsigned int virq, unsigned int nr_irqs)
1430 {
1431 	intel_free_irq_resources(domain, virq, nr_irqs);
1432 	irq_domain_free_irqs_common(domain, virq, nr_irqs);
1433 }
1434 
1435 static int intel_irq_remapping_activate(struct irq_domain *domain,
1436 					struct irq_data *irq_data, bool reserve)
1437 {
1438 	intel_ir_reconfigure_irte(irq_data, true);
1439 	return 0;
1440 }
1441 
1442 static void intel_irq_remapping_deactivate(struct irq_domain *domain,
1443 					   struct irq_data *irq_data)
1444 {
1445 	struct intel_ir_data *data = irq_data->chip_data;
1446 	struct irte entry;
1447 
1448 	memset(&entry, 0, sizeof(entry));
1449 	modify_irte(&data->irq_2_iommu, &entry);
1450 }
1451 
1452 static const struct irq_domain_ops intel_ir_domain_ops = {
1453 	.alloc = intel_irq_remapping_alloc,
1454 	.free = intel_irq_remapping_free,
1455 	.activate = intel_irq_remapping_activate,
1456 	.deactivate = intel_irq_remapping_deactivate,
1457 };
1458 
1459 /*
1460  * Support of Interrupt Remapping Unit Hotplug
1461  */
1462 static int dmar_ir_add(struct dmar_drhd_unit *dmaru, struct intel_iommu *iommu)
1463 {
1464 	int ret;
1465 	int eim = x2apic_enabled();
1466 
1467 	if (eim && !ecap_eim_support(iommu->ecap)) {
1468 		pr_info("DRHD %Lx: EIM not supported by DRHD, ecap %Lx\n",
1469 			iommu->reg_phys, iommu->ecap);
1470 		return -ENODEV;
1471 	}
1472 
1473 	if (ir_parse_ioapic_hpet_scope(dmaru->hdr, iommu)) {
1474 		pr_warn("DRHD %Lx: failed to parse managed IOAPIC/HPET\n",
1475 			iommu->reg_phys);
1476 		return -ENODEV;
1477 	}
1478 
1479 	/* TODO: check all IOAPICs are covered by IOMMU */
1480 
1481 	/* Setup Interrupt-remapping now. */
1482 	ret = intel_setup_irq_remapping(iommu);
1483 	if (ret) {
1484 		pr_err("Failed to setup irq remapping for %s\n",
1485 		       iommu->name);
1486 		intel_teardown_irq_remapping(iommu);
1487 		ir_remove_ioapic_hpet_scope(iommu);
1488 	} else {
1489 		iommu_enable_irq_remapping(iommu);
1490 	}
1491 
1492 	return ret;
1493 }
1494 
1495 int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert)
1496 {
1497 	int ret = 0;
1498 	struct intel_iommu *iommu = dmaru->iommu;
1499 
1500 	if (!irq_remapping_enabled)
1501 		return 0;
1502 	if (iommu == NULL)
1503 		return -EINVAL;
1504 	if (!ecap_ir_support(iommu->ecap))
1505 		return 0;
1506 	if (irq_remapping_cap(IRQ_POSTING_CAP) &&
1507 	    !cap_pi_support(iommu->cap))
1508 		return -EBUSY;
1509 
1510 	if (insert) {
1511 		if (!iommu->ir_table)
1512 			ret = dmar_ir_add(dmaru, iommu);
1513 	} else {
1514 		if (iommu->ir_table) {
1515 			if (!bitmap_empty(iommu->ir_table->bitmap,
1516 					  INTR_REMAP_TABLE_ENTRIES)) {
1517 				ret = -EBUSY;
1518 			} else {
1519 				iommu_disable_irq_remapping(iommu);
1520 				intel_teardown_irq_remapping(iommu);
1521 				ir_remove_ioapic_hpet_scope(iommu);
1522 			}
1523 		}
1524 	}
1525 
1526 	return ret;
1527 }
1528