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/acpi.h>
14 #include <linux/irqdomain.h>
15 #include <linux/crash_dump.h>
16 #include <asm/io_apic.h>
17 #include <asm/apic.h>
18 #include <asm/smp.h>
19 #include <asm/cpu.h>
20 #include <asm/irq_remapping.h>
21 #include <asm/pci-direct.h>
22 
23 #include "iommu.h"
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 	if (!cap_esirtps(iommu->cap))
498 		qi_global_iec(iommu);
499 }
500 
501 static void iommu_enable_irq_remapping(struct intel_iommu *iommu)
502 {
503 	unsigned long flags;
504 	u32 sts;
505 
506 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
507 
508 	/* Enable interrupt-remapping */
509 	iommu->gcmd |= DMA_GCMD_IRE;
510 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
511 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
512 		      readl, (sts & DMA_GSTS_IRES), sts);
513 
514 	/* Block compatibility-format MSIs */
515 	if (sts & DMA_GSTS_CFIS) {
516 		iommu->gcmd &= ~DMA_GCMD_CFI;
517 		writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
518 		IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
519 			      readl, !(sts & DMA_GSTS_CFIS), sts);
520 	}
521 
522 	/*
523 	 * With CFI clear in the Global Command register, we should be
524 	 * protected from dangerous (i.e. compatibility) interrupts
525 	 * regardless of x2apic status.  Check just to be sure.
526 	 */
527 	if (sts & DMA_GSTS_CFIS)
528 		WARN(1, KERN_WARNING
529 			"Compatibility-format IRQs enabled despite intr remapping;\n"
530 			"you are vulnerable to IRQ injection.\n");
531 
532 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
533 }
534 
535 static int intel_setup_irq_remapping(struct intel_iommu *iommu)
536 {
537 	struct ir_table *ir_table;
538 	struct fwnode_handle *fn;
539 	unsigned long *bitmap;
540 	struct page *pages;
541 
542 	if (iommu->ir_table)
543 		return 0;
544 
545 	ir_table = kzalloc(sizeof(struct ir_table), GFP_KERNEL);
546 	if (!ir_table)
547 		return -ENOMEM;
548 
549 	pages = alloc_pages_node(iommu->node, GFP_KERNEL | __GFP_ZERO,
550 				 INTR_REMAP_PAGE_ORDER);
551 	if (!pages) {
552 		pr_err("IR%d: failed to allocate pages of order %d\n",
553 		       iommu->seq_id, INTR_REMAP_PAGE_ORDER);
554 		goto out_free_table;
555 	}
556 
557 	bitmap = bitmap_zalloc(INTR_REMAP_TABLE_ENTRIES, GFP_ATOMIC);
558 	if (bitmap == NULL) {
559 		pr_err("IR%d: failed to allocate bitmap\n", iommu->seq_id);
560 		goto out_free_pages;
561 	}
562 
563 	fn = irq_domain_alloc_named_id_fwnode("INTEL-IR", iommu->seq_id);
564 	if (!fn)
565 		goto out_free_bitmap;
566 
567 	iommu->ir_domain =
568 		irq_domain_create_hierarchy(arch_get_ir_parent_domain(),
569 					    0, INTR_REMAP_TABLE_ENTRIES,
570 					    fn, &intel_ir_domain_ops,
571 					    iommu);
572 	if (!iommu->ir_domain) {
573 		pr_err("IR%d: failed to allocate irqdomain\n", iommu->seq_id);
574 		goto out_free_fwnode;
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_ir_domain;
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_ir_domain:
623 	if (iommu->ir_msi_domain)
624 		irq_domain_remove(iommu->ir_msi_domain);
625 	iommu->ir_msi_domain = NULL;
626 	irq_domain_remove(iommu->ir_domain);
627 	iommu->ir_domain = NULL;
628 out_free_fwnode:
629 	irq_domain_free_fwnode(fn);
630 out_free_bitmap:
631 	bitmap_free(bitmap);
632 out_free_pages:
633 	__free_pages(pages, INTR_REMAP_PAGE_ORDER);
634 out_free_table:
635 	kfree(ir_table);
636 
637 	iommu->ir_table  = NULL;
638 
639 	return -ENOMEM;
640 }
641 
642 static void intel_teardown_irq_remapping(struct intel_iommu *iommu)
643 {
644 	struct fwnode_handle *fn;
645 
646 	if (iommu && iommu->ir_table) {
647 		if (iommu->ir_msi_domain) {
648 			fn = iommu->ir_msi_domain->fwnode;
649 
650 			irq_domain_remove(iommu->ir_msi_domain);
651 			irq_domain_free_fwnode(fn);
652 			iommu->ir_msi_domain = NULL;
653 		}
654 		if (iommu->ir_domain) {
655 			fn = iommu->ir_domain->fwnode;
656 
657 			irq_domain_remove(iommu->ir_domain);
658 			irq_domain_free_fwnode(fn);
659 			iommu->ir_domain = NULL;
660 		}
661 		free_pages((unsigned long)iommu->ir_table->base,
662 			   INTR_REMAP_PAGE_ORDER);
663 		bitmap_free(iommu->ir_table->bitmap);
664 		kfree(iommu->ir_table);
665 		iommu->ir_table = NULL;
666 	}
667 }
668 
669 /*
670  * Disable Interrupt Remapping.
671  */
672 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
673 {
674 	unsigned long flags;
675 	u32 sts;
676 
677 	if (!ecap_ir_support(iommu->ecap))
678 		return;
679 
680 	/*
681 	 * global invalidation of interrupt entry cache before disabling
682 	 * interrupt-remapping.
683 	 */
684 	if (!cap_esirtps(iommu->cap))
685 		qi_global_iec(iommu);
686 
687 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
688 
689 	sts = readl(iommu->reg + DMAR_GSTS_REG);
690 	if (!(sts & DMA_GSTS_IRES))
691 		goto end;
692 
693 	iommu->gcmd &= ~DMA_GCMD_IRE;
694 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
695 
696 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
697 		      readl, !(sts & DMA_GSTS_IRES), sts);
698 
699 end:
700 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
701 }
702 
703 static int __init dmar_x2apic_optout(void)
704 {
705 	struct acpi_table_dmar *dmar;
706 	dmar = (struct acpi_table_dmar *)dmar_tbl;
707 	if (!dmar || no_x2apic_optout)
708 		return 0;
709 	return dmar->flags & DMAR_X2APIC_OPT_OUT;
710 }
711 
712 static void __init intel_cleanup_irq_remapping(void)
713 {
714 	struct dmar_drhd_unit *drhd;
715 	struct intel_iommu *iommu;
716 
717 	for_each_iommu(iommu, drhd) {
718 		if (ecap_ir_support(iommu->ecap)) {
719 			iommu_disable_irq_remapping(iommu);
720 			intel_teardown_irq_remapping(iommu);
721 		}
722 	}
723 
724 	if (x2apic_supported())
725 		pr_warn("Failed to enable irq remapping. You are vulnerable to irq-injection attacks.\n");
726 }
727 
728 static int __init intel_prepare_irq_remapping(void)
729 {
730 	struct dmar_drhd_unit *drhd;
731 	struct intel_iommu *iommu;
732 	int eim = 0;
733 
734 	if (irq_remap_broken) {
735 		pr_warn("This system BIOS has enabled interrupt remapping\n"
736 			"on a chipset that contains an erratum making that\n"
737 			"feature unstable.  To maintain system stability\n"
738 			"interrupt remapping is being disabled.  Please\n"
739 			"contact your BIOS vendor for an update\n");
740 		add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
741 		return -ENODEV;
742 	}
743 
744 	if (dmar_table_init() < 0)
745 		return -ENODEV;
746 
747 	if (intel_cap_audit(CAP_AUDIT_STATIC_IRQR, NULL))
748 		return -ENODEV;
749 
750 	if (!dmar_ir_support())
751 		return -ENODEV;
752 
753 	if (parse_ioapics_under_ir()) {
754 		pr_info("Not enabling interrupt remapping\n");
755 		goto error;
756 	}
757 
758 	/* First make sure all IOMMUs support IRQ remapping */
759 	for_each_iommu(iommu, drhd)
760 		if (!ecap_ir_support(iommu->ecap))
761 			goto error;
762 
763 	/* Detect remapping mode: lapic or x2apic */
764 	if (x2apic_supported()) {
765 		eim = !dmar_x2apic_optout();
766 		if (!eim) {
767 			pr_info("x2apic is disabled because BIOS sets x2apic opt out bit.");
768 			pr_info("Use 'intremap=no_x2apic_optout' to override the BIOS setting.\n");
769 		}
770 	}
771 
772 	for_each_iommu(iommu, drhd) {
773 		if (eim && !ecap_eim_support(iommu->ecap)) {
774 			pr_info("%s does not support EIM\n", iommu->name);
775 			eim = 0;
776 		}
777 	}
778 
779 	eim_mode = eim;
780 	if (eim)
781 		pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
782 
783 	/* Do the initializations early */
784 	for_each_iommu(iommu, drhd) {
785 		if (intel_setup_irq_remapping(iommu)) {
786 			pr_err("Failed to setup irq remapping for %s\n",
787 			       iommu->name);
788 			goto error;
789 		}
790 	}
791 
792 	return 0;
793 
794 error:
795 	intel_cleanup_irq_remapping();
796 	return -ENODEV;
797 }
798 
799 /*
800  * Set Posted-Interrupts capability.
801  */
802 static inline void set_irq_posting_cap(void)
803 {
804 	struct dmar_drhd_unit *drhd;
805 	struct intel_iommu *iommu;
806 
807 	if (!disable_irq_post) {
808 		/*
809 		 * If IRTE is in posted format, the 'pda' field goes across the
810 		 * 64-bit boundary, we need use cmpxchg16b to atomically update
811 		 * it. We only expose posted-interrupt when X86_FEATURE_CX16
812 		 * is supported. Actually, hardware platforms supporting PI
813 		 * should have X86_FEATURE_CX16 support, this has been confirmed
814 		 * with Intel hardware guys.
815 		 */
816 		if (boot_cpu_has(X86_FEATURE_CX16))
817 			intel_irq_remap_ops.capability |= 1 << IRQ_POSTING_CAP;
818 
819 		for_each_iommu(iommu, drhd)
820 			if (!cap_pi_support(iommu->cap)) {
821 				intel_irq_remap_ops.capability &=
822 						~(1 << IRQ_POSTING_CAP);
823 				break;
824 			}
825 	}
826 }
827 
828 static int __init intel_enable_irq_remapping(void)
829 {
830 	struct dmar_drhd_unit *drhd;
831 	struct intel_iommu *iommu;
832 	bool setup = false;
833 
834 	/*
835 	 * Setup Interrupt-remapping for all the DRHD's now.
836 	 */
837 	for_each_iommu(iommu, drhd) {
838 		if (!ir_pre_enabled(iommu))
839 			iommu_enable_irq_remapping(iommu);
840 		setup = true;
841 	}
842 
843 	if (!setup)
844 		goto error;
845 
846 	irq_remapping_enabled = 1;
847 
848 	set_irq_posting_cap();
849 
850 	pr_info("Enabled IRQ remapping in %s mode\n", eim_mode ? "x2apic" : "xapic");
851 
852 	return eim_mode ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
853 
854 error:
855 	intel_cleanup_irq_remapping();
856 	return -1;
857 }
858 
859 static int ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
860 				   struct intel_iommu *iommu,
861 				   struct acpi_dmar_hardware_unit *drhd)
862 {
863 	struct acpi_dmar_pci_path *path;
864 	u8 bus;
865 	int count, free = -1;
866 
867 	bus = scope->bus;
868 	path = (struct acpi_dmar_pci_path *)(scope + 1);
869 	count = (scope->length - sizeof(struct acpi_dmar_device_scope))
870 		/ sizeof(struct acpi_dmar_pci_path);
871 
872 	while (--count > 0) {
873 		/*
874 		 * Access PCI directly due to the PCI
875 		 * subsystem isn't initialized yet.
876 		 */
877 		bus = read_pci_config_byte(bus, path->device, path->function,
878 					   PCI_SECONDARY_BUS);
879 		path++;
880 	}
881 
882 	for (count = 0; count < MAX_HPET_TBS; count++) {
883 		if (ir_hpet[count].iommu == iommu &&
884 		    ir_hpet[count].id == scope->enumeration_id)
885 			return 0;
886 		else if (ir_hpet[count].iommu == NULL && free == -1)
887 			free = count;
888 	}
889 	if (free == -1) {
890 		pr_warn("Exceeded Max HPET blocks\n");
891 		return -ENOSPC;
892 	}
893 
894 	ir_hpet[free].iommu = iommu;
895 	ir_hpet[free].id    = scope->enumeration_id;
896 	ir_hpet[free].bus   = bus;
897 	ir_hpet[free].devfn = PCI_DEVFN(path->device, path->function);
898 	pr_info("HPET id %d under DRHD base 0x%Lx\n",
899 		scope->enumeration_id, drhd->address);
900 
901 	return 0;
902 }
903 
904 static int ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
905 				     struct intel_iommu *iommu,
906 				     struct acpi_dmar_hardware_unit *drhd)
907 {
908 	struct acpi_dmar_pci_path *path;
909 	u8 bus;
910 	int count, free = -1;
911 
912 	bus = scope->bus;
913 	path = (struct acpi_dmar_pci_path *)(scope + 1);
914 	count = (scope->length - sizeof(struct acpi_dmar_device_scope))
915 		/ sizeof(struct acpi_dmar_pci_path);
916 
917 	while (--count > 0) {
918 		/*
919 		 * Access PCI directly due to the PCI
920 		 * subsystem isn't initialized yet.
921 		 */
922 		bus = read_pci_config_byte(bus, path->device, path->function,
923 					   PCI_SECONDARY_BUS);
924 		path++;
925 	}
926 
927 	for (count = 0; count < MAX_IO_APICS; count++) {
928 		if (ir_ioapic[count].iommu == iommu &&
929 		    ir_ioapic[count].id == scope->enumeration_id)
930 			return 0;
931 		else if (ir_ioapic[count].iommu == NULL && free == -1)
932 			free = count;
933 	}
934 	if (free == -1) {
935 		pr_warn("Exceeded Max IO APICS\n");
936 		return -ENOSPC;
937 	}
938 
939 	ir_ioapic[free].bus   = bus;
940 	ir_ioapic[free].devfn = PCI_DEVFN(path->device, path->function);
941 	ir_ioapic[free].iommu = iommu;
942 	ir_ioapic[free].id    = scope->enumeration_id;
943 	pr_info("IOAPIC id %d under DRHD base  0x%Lx IOMMU %d\n",
944 		scope->enumeration_id, drhd->address, iommu->seq_id);
945 
946 	return 0;
947 }
948 
949 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
950 				      struct intel_iommu *iommu)
951 {
952 	int ret = 0;
953 	struct acpi_dmar_hardware_unit *drhd;
954 	struct acpi_dmar_device_scope *scope;
955 	void *start, *end;
956 
957 	drhd = (struct acpi_dmar_hardware_unit *)header;
958 	start = (void *)(drhd + 1);
959 	end = ((void *)drhd) + header->length;
960 
961 	while (start < end && ret == 0) {
962 		scope = start;
963 		if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC)
964 			ret = ir_parse_one_ioapic_scope(scope, iommu, drhd);
965 		else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET)
966 			ret = ir_parse_one_hpet_scope(scope, iommu, drhd);
967 		start += scope->length;
968 	}
969 
970 	return ret;
971 }
972 
973 static void ir_remove_ioapic_hpet_scope(struct intel_iommu *iommu)
974 {
975 	int i;
976 
977 	for (i = 0; i < MAX_HPET_TBS; i++)
978 		if (ir_hpet[i].iommu == iommu)
979 			ir_hpet[i].iommu = NULL;
980 
981 	for (i = 0; i < MAX_IO_APICS; i++)
982 		if (ir_ioapic[i].iommu == iommu)
983 			ir_ioapic[i].iommu = NULL;
984 }
985 
986 /*
987  * Finds the assocaition between IOAPIC's and its Interrupt-remapping
988  * hardware unit.
989  */
990 static int __init parse_ioapics_under_ir(void)
991 {
992 	struct dmar_drhd_unit *drhd;
993 	struct intel_iommu *iommu;
994 	bool ir_supported = false;
995 	int ioapic_idx;
996 
997 	for_each_iommu(iommu, drhd) {
998 		int ret;
999 
1000 		if (!ecap_ir_support(iommu->ecap))
1001 			continue;
1002 
1003 		ret = ir_parse_ioapic_hpet_scope(drhd->hdr, iommu);
1004 		if (ret)
1005 			return ret;
1006 
1007 		ir_supported = true;
1008 	}
1009 
1010 	if (!ir_supported)
1011 		return -ENODEV;
1012 
1013 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
1014 		int ioapic_id = mpc_ioapic_id(ioapic_idx);
1015 		if (!map_ioapic_to_iommu(ioapic_id)) {
1016 			pr_err(FW_BUG "ioapic %d has no mapping iommu, "
1017 			       "interrupt remapping will be disabled\n",
1018 			       ioapic_id);
1019 			return -1;
1020 		}
1021 	}
1022 
1023 	return 0;
1024 }
1025 
1026 static int __init ir_dev_scope_init(void)
1027 {
1028 	int ret;
1029 
1030 	if (!irq_remapping_enabled)
1031 		return 0;
1032 
1033 	down_write(&dmar_global_lock);
1034 	ret = dmar_dev_scope_init();
1035 	up_write(&dmar_global_lock);
1036 
1037 	return ret;
1038 }
1039 rootfs_initcall(ir_dev_scope_init);
1040 
1041 static void disable_irq_remapping(void)
1042 {
1043 	struct dmar_drhd_unit *drhd;
1044 	struct intel_iommu *iommu = NULL;
1045 
1046 	/*
1047 	 * Disable Interrupt-remapping for all the DRHD's now.
1048 	 */
1049 	for_each_iommu(iommu, drhd) {
1050 		if (!ecap_ir_support(iommu->ecap))
1051 			continue;
1052 
1053 		iommu_disable_irq_remapping(iommu);
1054 	}
1055 
1056 	/*
1057 	 * Clear Posted-Interrupts capability.
1058 	 */
1059 	if (!disable_irq_post)
1060 		intel_irq_remap_ops.capability &= ~(1 << IRQ_POSTING_CAP);
1061 }
1062 
1063 static int reenable_irq_remapping(int eim)
1064 {
1065 	struct dmar_drhd_unit *drhd;
1066 	bool setup = false;
1067 	struct intel_iommu *iommu = NULL;
1068 
1069 	for_each_iommu(iommu, drhd)
1070 		if (iommu->qi)
1071 			dmar_reenable_qi(iommu);
1072 
1073 	/*
1074 	 * Setup Interrupt-remapping for all the DRHD's now.
1075 	 */
1076 	for_each_iommu(iommu, drhd) {
1077 		if (!ecap_ir_support(iommu->ecap))
1078 			continue;
1079 
1080 		/* Set up interrupt remapping for iommu.*/
1081 		iommu_set_irq_remapping(iommu, eim);
1082 		iommu_enable_irq_remapping(iommu);
1083 		setup = true;
1084 	}
1085 
1086 	if (!setup)
1087 		goto error;
1088 
1089 	set_irq_posting_cap();
1090 
1091 	return 0;
1092 
1093 error:
1094 	/*
1095 	 * handle error condition gracefully here!
1096 	 */
1097 	return -1;
1098 }
1099 
1100 /*
1101  * Store the MSI remapping domain pointer in the device if enabled.
1102  *
1103  * This is called from dmar_pci_bus_add_dev() so it works even when DMA
1104  * remapping is disabled. Only update the pointer if the device is not
1105  * already handled by a non default PCI/MSI interrupt domain. This protects
1106  * e.g. VMD devices.
1107  */
1108 void intel_irq_remap_add_device(struct dmar_pci_notify_info *info)
1109 {
1110 	if (!irq_remapping_enabled || pci_dev_has_special_msi_domain(info->dev))
1111 		return;
1112 
1113 	dev_set_msi_domain(&info->dev->dev, map_dev_to_ir(info->dev));
1114 }
1115 
1116 static void prepare_irte(struct irte *irte, int vector, unsigned int dest)
1117 {
1118 	memset(irte, 0, sizeof(*irte));
1119 
1120 	irte->present = 1;
1121 	irte->dst_mode = apic->dest_mode_logical;
1122 	/*
1123 	 * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
1124 	 * actual level or edge trigger will be setup in the IO-APIC
1125 	 * RTE. This will help simplify level triggered irq migration.
1126 	 * For more details, see the comments (in io_apic.c) explainig IO-APIC
1127 	 * irq migration in the presence of interrupt-remapping.
1128 	*/
1129 	irte->trigger_mode = 0;
1130 	irte->dlvry_mode = apic->delivery_mode;
1131 	irte->vector = vector;
1132 	irte->dest_id = IRTE_DEST(dest);
1133 	irte->redir_hint = 1;
1134 }
1135 
1136 struct irq_remap_ops intel_irq_remap_ops = {
1137 	.prepare		= intel_prepare_irq_remapping,
1138 	.enable			= intel_enable_irq_remapping,
1139 	.disable		= disable_irq_remapping,
1140 	.reenable		= reenable_irq_remapping,
1141 	.enable_faulting	= enable_drhd_fault_handling,
1142 };
1143 
1144 static void intel_ir_reconfigure_irte(struct irq_data *irqd, bool force)
1145 {
1146 	struct intel_ir_data *ir_data = irqd->chip_data;
1147 	struct irte *irte = &ir_data->irte_entry;
1148 	struct irq_cfg *cfg = irqd_cfg(irqd);
1149 
1150 	/*
1151 	 * Atomically updates the IRTE with the new destination, vector
1152 	 * and flushes the interrupt entry cache.
1153 	 */
1154 	irte->vector = cfg->vector;
1155 	irte->dest_id = IRTE_DEST(cfg->dest_apicid);
1156 
1157 	/* Update the hardware only if the interrupt is in remapped mode. */
1158 	if (force || ir_data->irq_2_iommu.mode == IRQ_REMAPPING)
1159 		modify_irte(&ir_data->irq_2_iommu, irte);
1160 }
1161 
1162 /*
1163  * Migrate the IO-APIC irq in the presence of intr-remapping.
1164  *
1165  * For both level and edge triggered, irq migration is a simple atomic
1166  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
1167  *
1168  * For level triggered, we eliminate the io-apic RTE modification (with the
1169  * updated vector information), by using a virtual vector (io-apic pin number).
1170  * Real vector that is used for interrupting cpu will be coming from
1171  * the interrupt-remapping table entry.
1172  *
1173  * As the migration is a simple atomic update of IRTE, the same mechanism
1174  * is used to migrate MSI irq's in the presence of interrupt-remapping.
1175  */
1176 static int
1177 intel_ir_set_affinity(struct irq_data *data, const struct cpumask *mask,
1178 		      bool force)
1179 {
1180 	struct irq_data *parent = data->parent_data;
1181 	struct irq_cfg *cfg = irqd_cfg(data);
1182 	int ret;
1183 
1184 	ret = parent->chip->irq_set_affinity(parent, mask, force);
1185 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
1186 		return ret;
1187 
1188 	intel_ir_reconfigure_irte(data, false);
1189 	/*
1190 	 * After this point, all the interrupts will start arriving
1191 	 * at the new destination. So, time to cleanup the previous
1192 	 * vector allocation.
1193 	 */
1194 	send_cleanup_vector(cfg);
1195 
1196 	return IRQ_SET_MASK_OK_DONE;
1197 }
1198 
1199 static void intel_ir_compose_msi_msg(struct irq_data *irq_data,
1200 				     struct msi_msg *msg)
1201 {
1202 	struct intel_ir_data *ir_data = irq_data->chip_data;
1203 
1204 	*msg = ir_data->msi_entry;
1205 }
1206 
1207 static int intel_ir_set_vcpu_affinity(struct irq_data *data, void *info)
1208 {
1209 	struct intel_ir_data *ir_data = data->chip_data;
1210 	struct vcpu_data *vcpu_pi_info = info;
1211 
1212 	/* stop posting interrupts, back to remapping mode */
1213 	if (!vcpu_pi_info) {
1214 		modify_irte(&ir_data->irq_2_iommu, &ir_data->irte_entry);
1215 	} else {
1216 		struct irte irte_pi;
1217 
1218 		/*
1219 		 * We are not caching the posted interrupt entry. We
1220 		 * copy the data from the remapped entry and modify
1221 		 * the fields which are relevant for posted mode. The
1222 		 * cached remapped entry is used for switching back to
1223 		 * remapped mode.
1224 		 */
1225 		memset(&irte_pi, 0, sizeof(irte_pi));
1226 		dmar_copy_shared_irte(&irte_pi, &ir_data->irte_entry);
1227 
1228 		/* Update the posted mode fields */
1229 		irte_pi.p_pst = 1;
1230 		irte_pi.p_urgent = 0;
1231 		irte_pi.p_vector = vcpu_pi_info->vector;
1232 		irte_pi.pda_l = (vcpu_pi_info->pi_desc_addr >>
1233 				(32 - PDA_LOW_BIT)) & ~(-1UL << PDA_LOW_BIT);
1234 		irte_pi.pda_h = (vcpu_pi_info->pi_desc_addr >> 32) &
1235 				~(-1UL << PDA_HIGH_BIT);
1236 
1237 		modify_irte(&ir_data->irq_2_iommu, &irte_pi);
1238 	}
1239 
1240 	return 0;
1241 }
1242 
1243 static struct irq_chip intel_ir_chip = {
1244 	.name			= "INTEL-IR",
1245 	.irq_ack		= apic_ack_irq,
1246 	.irq_set_affinity	= intel_ir_set_affinity,
1247 	.irq_compose_msi_msg	= intel_ir_compose_msi_msg,
1248 	.irq_set_vcpu_affinity	= intel_ir_set_vcpu_affinity,
1249 };
1250 
1251 static void fill_msi_msg(struct msi_msg *msg, u32 index, u32 subhandle)
1252 {
1253 	memset(msg, 0, sizeof(*msg));
1254 
1255 	msg->arch_addr_lo.dmar_base_address = X86_MSI_BASE_ADDRESS_LOW;
1256 	msg->arch_addr_lo.dmar_subhandle_valid = true;
1257 	msg->arch_addr_lo.dmar_format = true;
1258 	msg->arch_addr_lo.dmar_index_0_14 = index & 0x7FFF;
1259 	msg->arch_addr_lo.dmar_index_15 = !!(index & 0x8000);
1260 
1261 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
1262 
1263 	msg->arch_data.dmar_subhandle = subhandle;
1264 }
1265 
1266 static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,
1267 					     struct irq_cfg *irq_cfg,
1268 					     struct irq_alloc_info *info,
1269 					     int index, int sub_handle)
1270 {
1271 	struct irte *irte = &data->irte_entry;
1272 
1273 	prepare_irte(irte, irq_cfg->vector, irq_cfg->dest_apicid);
1274 
1275 	switch (info->type) {
1276 	case X86_IRQ_ALLOC_TYPE_IOAPIC:
1277 		/* Set source-id of interrupt request */
1278 		set_ioapic_sid(irte, info->devid);
1279 		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",
1280 			info->devid, irte->present, irte->fpd,
1281 			irte->dst_mode, irte->redir_hint,
1282 			irte->trigger_mode, irte->dlvry_mode,
1283 			irte->avail, irte->vector, irte->dest_id,
1284 			irte->sid, irte->sq, irte->svt);
1285 		sub_handle = info->ioapic.pin;
1286 		break;
1287 	case X86_IRQ_ALLOC_TYPE_HPET:
1288 		set_hpet_sid(irte, info->devid);
1289 		break;
1290 	case X86_IRQ_ALLOC_TYPE_PCI_MSI:
1291 	case X86_IRQ_ALLOC_TYPE_PCI_MSIX:
1292 		set_msi_sid(irte,
1293 			    pci_real_dma_dev(msi_desc_to_pci_dev(info->desc)));
1294 		break;
1295 	default:
1296 		BUG_ON(1);
1297 		break;
1298 	}
1299 	fill_msi_msg(&data->msi_entry, index, sub_handle);
1300 }
1301 
1302 static void intel_free_irq_resources(struct irq_domain *domain,
1303 				     unsigned int virq, unsigned int nr_irqs)
1304 {
1305 	struct irq_data *irq_data;
1306 	struct intel_ir_data *data;
1307 	struct irq_2_iommu *irq_iommu;
1308 	unsigned long flags;
1309 	int i;
1310 	for (i = 0; i < nr_irqs; i++) {
1311 		irq_data = irq_domain_get_irq_data(domain, virq  + i);
1312 		if (irq_data && irq_data->chip_data) {
1313 			data = irq_data->chip_data;
1314 			irq_iommu = &data->irq_2_iommu;
1315 			raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
1316 			clear_entries(irq_iommu);
1317 			raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
1318 			irq_domain_reset_irq_data(irq_data);
1319 			kfree(data);
1320 		}
1321 	}
1322 }
1323 
1324 static int intel_irq_remapping_alloc(struct irq_domain *domain,
1325 				     unsigned int virq, unsigned int nr_irqs,
1326 				     void *arg)
1327 {
1328 	struct intel_iommu *iommu = domain->host_data;
1329 	struct irq_alloc_info *info = arg;
1330 	struct intel_ir_data *data, *ird;
1331 	struct irq_data *irq_data;
1332 	struct irq_cfg *irq_cfg;
1333 	int i, ret, index;
1334 
1335 	if (!info || !iommu)
1336 		return -EINVAL;
1337 	if (nr_irqs > 1 && info->type != X86_IRQ_ALLOC_TYPE_PCI_MSI &&
1338 	    info->type != X86_IRQ_ALLOC_TYPE_PCI_MSIX)
1339 		return -EINVAL;
1340 
1341 	/*
1342 	 * With IRQ remapping enabled, don't need contiguous CPU vectors
1343 	 * to support multiple MSI interrupts.
1344 	 */
1345 	if (info->type == X86_IRQ_ALLOC_TYPE_PCI_MSI)
1346 		info->flags &= ~X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
1347 
1348 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
1349 	if (ret < 0)
1350 		return ret;
1351 
1352 	ret = -ENOMEM;
1353 	data = kzalloc(sizeof(*data), GFP_KERNEL);
1354 	if (!data)
1355 		goto out_free_parent;
1356 
1357 	down_read(&dmar_global_lock);
1358 	index = alloc_irte(iommu, &data->irq_2_iommu, nr_irqs);
1359 	up_read(&dmar_global_lock);
1360 	if (index < 0) {
1361 		pr_warn("Failed to allocate IRTE\n");
1362 		kfree(data);
1363 		goto out_free_parent;
1364 	}
1365 
1366 	for (i = 0; i < nr_irqs; i++) {
1367 		irq_data = irq_domain_get_irq_data(domain, virq + i);
1368 		irq_cfg = irqd_cfg(irq_data);
1369 		if (!irq_data || !irq_cfg) {
1370 			if (!i)
1371 				kfree(data);
1372 			ret = -EINVAL;
1373 			goto out_free_data;
1374 		}
1375 
1376 		if (i > 0) {
1377 			ird = kzalloc(sizeof(*ird), GFP_KERNEL);
1378 			if (!ird)
1379 				goto out_free_data;
1380 			/* Initialize the common data */
1381 			ird->irq_2_iommu = data->irq_2_iommu;
1382 			ird->irq_2_iommu.sub_handle = i;
1383 		} else {
1384 			ird = data;
1385 		}
1386 
1387 		irq_data->hwirq = (index << 16) + i;
1388 		irq_data->chip_data = ird;
1389 		irq_data->chip = &intel_ir_chip;
1390 		intel_irq_remapping_prepare_irte(ird, irq_cfg, info, index, i);
1391 		irq_set_status_flags(virq + i, IRQ_MOVE_PCNTXT);
1392 	}
1393 	return 0;
1394 
1395 out_free_data:
1396 	intel_free_irq_resources(domain, virq, i);
1397 out_free_parent:
1398 	irq_domain_free_irqs_common(domain, virq, nr_irqs);
1399 	return ret;
1400 }
1401 
1402 static void intel_irq_remapping_free(struct irq_domain *domain,
1403 				     unsigned int virq, unsigned int nr_irqs)
1404 {
1405 	intel_free_irq_resources(domain, virq, nr_irqs);
1406 	irq_domain_free_irqs_common(domain, virq, nr_irqs);
1407 }
1408 
1409 static int intel_irq_remapping_activate(struct irq_domain *domain,
1410 					struct irq_data *irq_data, bool reserve)
1411 {
1412 	intel_ir_reconfigure_irte(irq_data, true);
1413 	return 0;
1414 }
1415 
1416 static void intel_irq_remapping_deactivate(struct irq_domain *domain,
1417 					   struct irq_data *irq_data)
1418 {
1419 	struct intel_ir_data *data = irq_data->chip_data;
1420 	struct irte entry;
1421 
1422 	memset(&entry, 0, sizeof(entry));
1423 	modify_irte(&data->irq_2_iommu, &entry);
1424 }
1425 
1426 static int intel_irq_remapping_select(struct irq_domain *d,
1427 				      struct irq_fwspec *fwspec,
1428 				      enum irq_domain_bus_token bus_token)
1429 {
1430 	struct intel_iommu *iommu = NULL;
1431 
1432 	if (x86_fwspec_is_ioapic(fwspec))
1433 		iommu = map_ioapic_to_iommu(fwspec->param[0]);
1434 	else if (x86_fwspec_is_hpet(fwspec))
1435 		iommu = map_hpet_to_iommu(fwspec->param[0]);
1436 
1437 	return iommu && d == iommu->ir_domain;
1438 }
1439 
1440 static const struct irq_domain_ops intel_ir_domain_ops = {
1441 	.select = intel_irq_remapping_select,
1442 	.alloc = intel_irq_remapping_alloc,
1443 	.free = intel_irq_remapping_free,
1444 	.activate = intel_irq_remapping_activate,
1445 	.deactivate = intel_irq_remapping_deactivate,
1446 };
1447 
1448 /*
1449  * Support of Interrupt Remapping Unit Hotplug
1450  */
1451 static int dmar_ir_add(struct dmar_drhd_unit *dmaru, struct intel_iommu *iommu)
1452 {
1453 	int ret;
1454 	int eim = x2apic_enabled();
1455 
1456 	ret = intel_cap_audit(CAP_AUDIT_HOTPLUG_IRQR, iommu);
1457 	if (ret)
1458 		return ret;
1459 
1460 	if (eim && !ecap_eim_support(iommu->ecap)) {
1461 		pr_info("DRHD %Lx: EIM not supported by DRHD, ecap %Lx\n",
1462 			iommu->reg_phys, iommu->ecap);
1463 		return -ENODEV;
1464 	}
1465 
1466 	if (ir_parse_ioapic_hpet_scope(dmaru->hdr, iommu)) {
1467 		pr_warn("DRHD %Lx: failed to parse managed IOAPIC/HPET\n",
1468 			iommu->reg_phys);
1469 		return -ENODEV;
1470 	}
1471 
1472 	/* TODO: check all IOAPICs are covered by IOMMU */
1473 
1474 	/* Setup Interrupt-remapping now. */
1475 	ret = intel_setup_irq_remapping(iommu);
1476 	if (ret) {
1477 		pr_err("Failed to setup irq remapping for %s\n",
1478 		       iommu->name);
1479 		intel_teardown_irq_remapping(iommu);
1480 		ir_remove_ioapic_hpet_scope(iommu);
1481 	} else {
1482 		iommu_enable_irq_remapping(iommu);
1483 	}
1484 
1485 	return ret;
1486 }
1487 
1488 int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert)
1489 {
1490 	int ret = 0;
1491 	struct intel_iommu *iommu = dmaru->iommu;
1492 
1493 	if (!irq_remapping_enabled)
1494 		return 0;
1495 	if (iommu == NULL)
1496 		return -EINVAL;
1497 	if (!ecap_ir_support(iommu->ecap))
1498 		return 0;
1499 	if (irq_remapping_cap(IRQ_POSTING_CAP) &&
1500 	    !cap_pi_support(iommu->cap))
1501 		return -EBUSY;
1502 
1503 	if (insert) {
1504 		if (!iommu->ir_table)
1505 			ret = dmar_ir_add(dmaru, iommu);
1506 	} else {
1507 		if (iommu->ir_table) {
1508 			if (!bitmap_empty(iommu->ir_table->bitmap,
1509 					  INTR_REMAP_TABLE_ENTRIES)) {
1510 				ret = -EBUSY;
1511 			} else {
1512 				iommu_disable_irq_remapping(iommu);
1513 				intel_teardown_irq_remapping(iommu);
1514 				ir_remove_ioapic_hpet_scope(iommu);
1515 			}
1516 		}
1517 	}
1518 
1519 	return ret;
1520 }
1521