xref: /linux/drivers/iommu/intel/dmar.c (revision 6a4aee277740d04ac0fd54cfa17cc28261932ddc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2006, Intel Corporation.
4  *
5  * Copyright (C) 2006-2008 Intel Corporation
6  * Author: Ashok Raj <ashok.raj@intel.com>
7  * Author: Shaohua Li <shaohua.li@intel.com>
8  * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
9  *
10  * This file implements early detection/parsing of Remapping Devices
11  * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
12  * tables.
13  *
14  * These routines are used by both DMA-remapping and Interrupt-remapping
15  */
16 
17 #define pr_fmt(fmt)     "DMAR: " fmt
18 
19 #include <linux/pci.h>
20 #include <linux/dmar.h>
21 #include <linux/iova.h>
22 #include <linux/timer.h>
23 #include <linux/irq.h>
24 #include <linux/interrupt.h>
25 #include <linux/tboot.h>
26 #include <linux/dmi.h>
27 #include <linux/slab.h>
28 #include <linux/iommu.h>
29 #include <linux/numa.h>
30 #include <linux/limits.h>
31 #include <asm/irq_remapping.h>
32 
33 #include "iommu.h"
34 #include "../irq_remapping.h"
35 #include "perf.h"
36 #include "trace.h"
37 #include "perfmon.h"
38 
39 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
40 struct dmar_res_callback {
41 	dmar_res_handler_t	cb[ACPI_DMAR_TYPE_RESERVED];
42 	void			*arg[ACPI_DMAR_TYPE_RESERVED];
43 	bool			ignore_unhandled;
44 	bool			print_entry;
45 };
46 
47 /*
48  * Assumptions:
49  * 1) The hotplug framework guarentees that DMAR unit will be hot-added
50  *    before IO devices managed by that unit.
51  * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
52  *    after IO devices managed by that unit.
53  * 3) Hotplug events are rare.
54  *
55  * Locking rules for DMA and interrupt remapping related global data structures:
56  * 1) Use dmar_global_lock in process context
57  * 2) Use RCU in interrupt context
58  */
59 DECLARE_RWSEM(dmar_global_lock);
60 LIST_HEAD(dmar_drhd_units);
61 
62 struct acpi_table_header * __initdata dmar_tbl;
63 static int dmar_dev_scope_status = 1;
64 static DEFINE_IDA(dmar_seq_ids);
65 
66 static int alloc_iommu(struct dmar_drhd_unit *drhd);
67 static void free_iommu(struct intel_iommu *iommu);
68 
69 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
70 {
71 	/*
72 	 * add INCLUDE_ALL at the tail, so scan the list will find it at
73 	 * the very end.
74 	 */
75 	if (drhd->include_all)
76 		list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
77 	else
78 		list_add_rcu(&drhd->list, &dmar_drhd_units);
79 }
80 
81 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
82 {
83 	struct acpi_dmar_device_scope *scope;
84 
85 	*cnt = 0;
86 	while (start < end) {
87 		scope = start;
88 		if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
89 		    scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
90 		    scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
91 			(*cnt)++;
92 		else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
93 			scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
94 			pr_warn("Unsupported device scope\n");
95 		}
96 		start += scope->length;
97 	}
98 	if (*cnt == 0)
99 		return NULL;
100 
101 	return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
102 }
103 
104 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
105 {
106 	int i;
107 	struct device *tmp_dev;
108 
109 	if (*devices && *cnt) {
110 		for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
111 			put_device(tmp_dev);
112 		kfree(*devices);
113 	}
114 
115 	*devices = NULL;
116 	*cnt = 0;
117 }
118 
119 /* Optimize out kzalloc()/kfree() for normal cases */
120 static char dmar_pci_notify_info_buf[64];
121 
122 static struct dmar_pci_notify_info *
123 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
124 {
125 	int level = 0;
126 	size_t size;
127 	struct pci_dev *tmp;
128 	struct dmar_pci_notify_info *info;
129 
130 	/*
131 	 * Ignore devices that have a domain number higher than what can
132 	 * be looked up in DMAR, e.g. VMD subdevices with domain 0x10000
133 	 */
134 	if (pci_domain_nr(dev->bus) > U16_MAX)
135 		return NULL;
136 
137 	/* Only generate path[] for device addition event */
138 	if (event == BUS_NOTIFY_ADD_DEVICE)
139 		for (tmp = dev; tmp; tmp = tmp->bus->self)
140 			level++;
141 
142 	size = struct_size(info, path, level);
143 	if (size <= sizeof(dmar_pci_notify_info_buf)) {
144 		info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
145 	} else {
146 		info = kzalloc(size, GFP_KERNEL);
147 		if (!info) {
148 			if (dmar_dev_scope_status == 0)
149 				dmar_dev_scope_status = -ENOMEM;
150 			return NULL;
151 		}
152 	}
153 
154 	info->event = event;
155 	info->dev = dev;
156 	info->seg = pci_domain_nr(dev->bus);
157 	info->level = level;
158 	if (event == BUS_NOTIFY_ADD_DEVICE) {
159 		for (tmp = dev; tmp; tmp = tmp->bus->self) {
160 			level--;
161 			info->path[level].bus = tmp->bus->number;
162 			info->path[level].device = PCI_SLOT(tmp->devfn);
163 			info->path[level].function = PCI_FUNC(tmp->devfn);
164 			if (pci_is_root_bus(tmp->bus))
165 				info->bus = tmp->bus->number;
166 		}
167 	}
168 
169 	return info;
170 }
171 
172 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
173 {
174 	if ((void *)info != dmar_pci_notify_info_buf)
175 		kfree(info);
176 }
177 
178 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
179 				struct acpi_dmar_pci_path *path, int count)
180 {
181 	int i;
182 
183 	if (info->bus != bus)
184 		goto fallback;
185 	if (info->level != count)
186 		goto fallback;
187 
188 	for (i = 0; i < count; i++) {
189 		if (path[i].device != info->path[i].device ||
190 		    path[i].function != info->path[i].function)
191 			goto fallback;
192 	}
193 
194 	return true;
195 
196 fallback:
197 
198 	if (count != 1)
199 		return false;
200 
201 	i = info->level - 1;
202 	if (bus              == info->path[i].bus &&
203 	    path[0].device   == info->path[i].device &&
204 	    path[0].function == info->path[i].function) {
205 		pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
206 			bus, path[0].device, path[0].function);
207 		return true;
208 	}
209 
210 	return false;
211 }
212 
213 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
214 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
215 			  void *start, void*end, u16 segment,
216 			  struct dmar_dev_scope *devices,
217 			  int devices_cnt)
218 {
219 	int i, level;
220 	struct device *tmp, *dev = &info->dev->dev;
221 	struct acpi_dmar_device_scope *scope;
222 	struct acpi_dmar_pci_path *path;
223 
224 	if (segment != info->seg)
225 		return 0;
226 
227 	for (; start < end; start += scope->length) {
228 		scope = start;
229 		if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
230 		    scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
231 			continue;
232 
233 		path = (struct acpi_dmar_pci_path *)(scope + 1);
234 		level = (scope->length - sizeof(*scope)) / sizeof(*path);
235 		if (!dmar_match_pci_path(info, scope->bus, path, level))
236 			continue;
237 
238 		/*
239 		 * We expect devices with endpoint scope to have normal PCI
240 		 * headers, and devices with bridge scope to have bridge PCI
241 		 * headers.  However PCI NTB devices may be listed in the
242 		 * DMAR table with bridge scope, even though they have a
243 		 * normal PCI header.  NTB devices are identified by class
244 		 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
245 		 * for this special case.
246 		 */
247 		if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
248 		     info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
249 		    (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
250 		     (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
251 		      info->dev->class >> 16 != PCI_BASE_CLASS_BRIDGE))) {
252 			pr_warn("Device scope type does not match for %s\n",
253 				pci_name(info->dev));
254 			return -EINVAL;
255 		}
256 
257 		for_each_dev_scope(devices, devices_cnt, i, tmp)
258 			if (tmp == NULL) {
259 				devices[i].bus = info->dev->bus->number;
260 				devices[i].devfn = info->dev->devfn;
261 				rcu_assign_pointer(devices[i].dev,
262 						   get_device(dev));
263 				return 1;
264 			}
265 		if (WARN_ON(i >= devices_cnt))
266 			return -EINVAL;
267 	}
268 
269 	return 0;
270 }
271 
272 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
273 			  struct dmar_dev_scope *devices, int count)
274 {
275 	int index;
276 	struct device *tmp;
277 
278 	if (info->seg != segment)
279 		return 0;
280 
281 	for_each_active_dev_scope(devices, count, index, tmp)
282 		if (tmp == &info->dev->dev) {
283 			RCU_INIT_POINTER(devices[index].dev, NULL);
284 			synchronize_rcu();
285 			put_device(tmp);
286 			return 1;
287 		}
288 
289 	return 0;
290 }
291 
292 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
293 {
294 	int ret = 0;
295 	struct dmar_drhd_unit *dmaru;
296 	struct acpi_dmar_hardware_unit *drhd;
297 
298 	for_each_drhd_unit(dmaru) {
299 		if (dmaru->include_all)
300 			continue;
301 
302 		drhd = container_of(dmaru->hdr,
303 				    struct acpi_dmar_hardware_unit, header);
304 		ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
305 				((void *)drhd) + drhd->header.length,
306 				dmaru->segment,
307 				dmaru->devices, dmaru->devices_cnt);
308 		if (ret)
309 			break;
310 	}
311 	if (ret >= 0)
312 		ret = dmar_iommu_notify_scope_dev(info);
313 	if (ret < 0 && dmar_dev_scope_status == 0)
314 		dmar_dev_scope_status = ret;
315 
316 	if (ret >= 0)
317 		intel_irq_remap_add_device(info);
318 
319 	return ret;
320 }
321 
322 static void  dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
323 {
324 	struct dmar_drhd_unit *dmaru;
325 
326 	for_each_drhd_unit(dmaru)
327 		if (dmar_remove_dev_scope(info, dmaru->segment,
328 			dmaru->devices, dmaru->devices_cnt))
329 			break;
330 	dmar_iommu_notify_scope_dev(info);
331 }
332 
333 static inline void vf_inherit_msi_domain(struct pci_dev *pdev)
334 {
335 	struct pci_dev *physfn = pci_physfn(pdev);
336 
337 	dev_set_msi_domain(&pdev->dev, dev_get_msi_domain(&physfn->dev));
338 }
339 
340 static int dmar_pci_bus_notifier(struct notifier_block *nb,
341 				 unsigned long action, void *data)
342 {
343 	struct pci_dev *pdev = to_pci_dev(data);
344 	struct dmar_pci_notify_info *info;
345 
346 	/* Only care about add/remove events for physical functions.
347 	 * For VFs we actually do the lookup based on the corresponding
348 	 * PF in device_to_iommu() anyway. */
349 	if (pdev->is_virtfn) {
350 		/*
351 		 * Ensure that the VF device inherits the irq domain of the
352 		 * PF device. Ideally the device would inherit the domain
353 		 * from the bus, but DMAR can have multiple units per bus
354 		 * which makes this impossible. The VF 'bus' could inherit
355 		 * from the PF device, but that's yet another x86'sism to
356 		 * inflict on everybody else.
357 		 */
358 		if (action == BUS_NOTIFY_ADD_DEVICE)
359 			vf_inherit_msi_domain(pdev);
360 		return NOTIFY_DONE;
361 	}
362 
363 	if (action != BUS_NOTIFY_ADD_DEVICE &&
364 	    action != BUS_NOTIFY_REMOVED_DEVICE)
365 		return NOTIFY_DONE;
366 
367 	info = dmar_alloc_pci_notify_info(pdev, action);
368 	if (!info)
369 		return NOTIFY_DONE;
370 
371 	down_write(&dmar_global_lock);
372 	if (action == BUS_NOTIFY_ADD_DEVICE)
373 		dmar_pci_bus_add_dev(info);
374 	else if (action == BUS_NOTIFY_REMOVED_DEVICE)
375 		dmar_pci_bus_del_dev(info);
376 	up_write(&dmar_global_lock);
377 
378 	dmar_free_pci_notify_info(info);
379 
380 	return NOTIFY_OK;
381 }
382 
383 static struct notifier_block dmar_pci_bus_nb = {
384 	.notifier_call = dmar_pci_bus_notifier,
385 	.priority = 1,
386 };
387 
388 static struct dmar_drhd_unit *
389 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
390 {
391 	struct dmar_drhd_unit *dmaru;
392 
393 	list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list,
394 				dmar_rcu_check())
395 		if (dmaru->segment == drhd->segment &&
396 		    dmaru->reg_base_addr == drhd->address)
397 			return dmaru;
398 
399 	return NULL;
400 }
401 
402 /*
403  * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
404  * structure which uniquely represent one DMA remapping hardware unit
405  * present in the platform
406  */
407 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
408 {
409 	struct acpi_dmar_hardware_unit *drhd;
410 	struct dmar_drhd_unit *dmaru;
411 	int ret;
412 
413 	drhd = (struct acpi_dmar_hardware_unit *)header;
414 	dmaru = dmar_find_dmaru(drhd);
415 	if (dmaru)
416 		goto out;
417 
418 	dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
419 	if (!dmaru)
420 		return -ENOMEM;
421 
422 	/*
423 	 * If header is allocated from slab by ACPI _DSM method, we need to
424 	 * copy the content because the memory buffer will be freed on return.
425 	 */
426 	dmaru->hdr = (void *)(dmaru + 1);
427 	memcpy(dmaru->hdr, header, header->length);
428 	dmaru->reg_base_addr = drhd->address;
429 	dmaru->segment = drhd->segment;
430 	/* The size of the register set is 2 ^ N 4 KB pages. */
431 	dmaru->reg_size = 1UL << (drhd->size + 12);
432 	dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
433 	dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
434 					      ((void *)drhd) + drhd->header.length,
435 					      &dmaru->devices_cnt);
436 	if (dmaru->devices_cnt && dmaru->devices == NULL) {
437 		kfree(dmaru);
438 		return -ENOMEM;
439 	}
440 
441 	ret = alloc_iommu(dmaru);
442 	if (ret) {
443 		dmar_free_dev_scope(&dmaru->devices,
444 				    &dmaru->devices_cnt);
445 		kfree(dmaru);
446 		return ret;
447 	}
448 	dmar_register_drhd_unit(dmaru);
449 
450 out:
451 	if (arg)
452 		(*(int *)arg)++;
453 
454 	return 0;
455 }
456 
457 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
458 {
459 	if (dmaru->devices && dmaru->devices_cnt)
460 		dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
461 	if (dmaru->iommu)
462 		free_iommu(dmaru->iommu);
463 	kfree(dmaru);
464 }
465 
466 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
467 				      void *arg)
468 {
469 	struct acpi_dmar_andd *andd = (void *)header;
470 
471 	/* Check for NUL termination within the designated length */
472 	if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
473 		pr_warn(FW_BUG
474 			   "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
475 			   "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
476 			   dmi_get_system_info(DMI_BIOS_VENDOR),
477 			   dmi_get_system_info(DMI_BIOS_VERSION),
478 			   dmi_get_system_info(DMI_PRODUCT_VERSION));
479 		add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
480 		return -EINVAL;
481 	}
482 	pr_info("ANDD device: %x name: %s\n", andd->device_number,
483 		andd->device_name);
484 
485 	return 0;
486 }
487 
488 #ifdef CONFIG_ACPI_NUMA
489 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
490 {
491 	struct acpi_dmar_rhsa *rhsa;
492 	struct dmar_drhd_unit *drhd;
493 
494 	rhsa = (struct acpi_dmar_rhsa *)header;
495 	for_each_drhd_unit(drhd) {
496 		if (drhd->reg_base_addr == rhsa->base_address) {
497 			int node = pxm_to_node(rhsa->proximity_domain);
498 
499 			if (node != NUMA_NO_NODE && !node_online(node))
500 				node = NUMA_NO_NODE;
501 			drhd->iommu->node = node;
502 			return 0;
503 		}
504 	}
505 	pr_warn(FW_BUG
506 		"Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
507 		"BIOS vendor: %s; Ver: %s; Product Version: %s\n",
508 		rhsa->base_address,
509 		dmi_get_system_info(DMI_BIOS_VENDOR),
510 		dmi_get_system_info(DMI_BIOS_VERSION),
511 		dmi_get_system_info(DMI_PRODUCT_VERSION));
512 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
513 
514 	return 0;
515 }
516 #else
517 #define	dmar_parse_one_rhsa		dmar_res_noop
518 #endif
519 
520 static void
521 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
522 {
523 	struct acpi_dmar_hardware_unit *drhd;
524 	struct acpi_dmar_reserved_memory *rmrr;
525 	struct acpi_dmar_atsr *atsr;
526 	struct acpi_dmar_rhsa *rhsa;
527 	struct acpi_dmar_satc *satc;
528 
529 	switch (header->type) {
530 	case ACPI_DMAR_TYPE_HARDWARE_UNIT:
531 		drhd = container_of(header, struct acpi_dmar_hardware_unit,
532 				    header);
533 		pr_info("DRHD base: %#016Lx flags: %#x\n",
534 			(unsigned long long)drhd->address, drhd->flags);
535 		break;
536 	case ACPI_DMAR_TYPE_RESERVED_MEMORY:
537 		rmrr = container_of(header, struct acpi_dmar_reserved_memory,
538 				    header);
539 		pr_info("RMRR base: %#016Lx end: %#016Lx\n",
540 			(unsigned long long)rmrr->base_address,
541 			(unsigned long long)rmrr->end_address);
542 		break;
543 	case ACPI_DMAR_TYPE_ROOT_ATS:
544 		atsr = container_of(header, struct acpi_dmar_atsr, header);
545 		pr_info("ATSR flags: %#x\n", atsr->flags);
546 		break;
547 	case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
548 		rhsa = container_of(header, struct acpi_dmar_rhsa, header);
549 		pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
550 		       (unsigned long long)rhsa->base_address,
551 		       rhsa->proximity_domain);
552 		break;
553 	case ACPI_DMAR_TYPE_NAMESPACE:
554 		/* We don't print this here because we need to sanity-check
555 		   it first. So print it in dmar_parse_one_andd() instead. */
556 		break;
557 	case ACPI_DMAR_TYPE_SATC:
558 		satc = container_of(header, struct acpi_dmar_satc, header);
559 		pr_info("SATC flags: 0x%x\n", satc->flags);
560 		break;
561 	}
562 }
563 
564 /**
565  * dmar_table_detect - checks to see if the platform supports DMAR devices
566  */
567 static int __init dmar_table_detect(void)
568 {
569 	acpi_status status = AE_OK;
570 
571 	/* if we could find DMAR table, then there are DMAR devices */
572 	status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
573 
574 	if (ACPI_SUCCESS(status) && !dmar_tbl) {
575 		pr_warn("Unable to map DMAR\n");
576 		status = AE_NOT_FOUND;
577 	}
578 
579 	return ACPI_SUCCESS(status) ? 0 : -ENOENT;
580 }
581 
582 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
583 				       size_t len, struct dmar_res_callback *cb)
584 {
585 	struct acpi_dmar_header *iter, *next;
586 	struct acpi_dmar_header *end = ((void *)start) + len;
587 
588 	for (iter = start; iter < end; iter = next) {
589 		next = (void *)iter + iter->length;
590 		if (iter->length == 0) {
591 			/* Avoid looping forever on bad ACPI tables */
592 			pr_debug(FW_BUG "Invalid 0-length structure\n");
593 			break;
594 		} else if (next > end) {
595 			/* Avoid passing table end */
596 			pr_warn(FW_BUG "Record passes table end\n");
597 			return -EINVAL;
598 		}
599 
600 		if (cb->print_entry)
601 			dmar_table_print_dmar_entry(iter);
602 
603 		if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
604 			/* continue for forward compatibility */
605 			pr_debug("Unknown DMAR structure type %d\n",
606 				 iter->type);
607 		} else if (cb->cb[iter->type]) {
608 			int ret;
609 
610 			ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
611 			if (ret)
612 				return ret;
613 		} else if (!cb->ignore_unhandled) {
614 			pr_warn("No handler for DMAR structure type %d\n",
615 				iter->type);
616 			return -EINVAL;
617 		}
618 	}
619 
620 	return 0;
621 }
622 
623 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
624 				       struct dmar_res_callback *cb)
625 {
626 	return dmar_walk_remapping_entries((void *)(dmar + 1),
627 			dmar->header.length - sizeof(*dmar), cb);
628 }
629 
630 /**
631  * parse_dmar_table - parses the DMA reporting table
632  */
633 static int __init
634 parse_dmar_table(void)
635 {
636 	struct acpi_table_dmar *dmar;
637 	int drhd_count = 0;
638 	int ret;
639 	struct dmar_res_callback cb = {
640 		.print_entry = true,
641 		.ignore_unhandled = true,
642 		.arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
643 		.cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
644 		.cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
645 		.cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
646 		.cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
647 		.cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
648 		.cb[ACPI_DMAR_TYPE_SATC] = &dmar_parse_one_satc,
649 	};
650 
651 	/*
652 	 * Do it again, earlier dmar_tbl mapping could be mapped with
653 	 * fixed map.
654 	 */
655 	dmar_table_detect();
656 
657 	/*
658 	 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
659 	 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
660 	 */
661 	dmar_tbl = tboot_get_dmar_table(dmar_tbl);
662 
663 	dmar = (struct acpi_table_dmar *)dmar_tbl;
664 	if (!dmar)
665 		return -ENODEV;
666 
667 	if (dmar->width < PAGE_SHIFT - 1) {
668 		pr_warn("Invalid DMAR haw\n");
669 		return -EINVAL;
670 	}
671 
672 	pr_info("Host address width %d\n", dmar->width + 1);
673 	ret = dmar_walk_dmar_table(dmar, &cb);
674 	if (ret == 0 && drhd_count == 0)
675 		pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
676 
677 	return ret;
678 }
679 
680 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
681 				 int cnt, struct pci_dev *dev)
682 {
683 	int index;
684 	struct device *tmp;
685 
686 	while (dev) {
687 		for_each_active_dev_scope(devices, cnt, index, tmp)
688 			if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
689 				return 1;
690 
691 		/* Check our parent */
692 		dev = dev->bus->self;
693 	}
694 
695 	return 0;
696 }
697 
698 struct dmar_drhd_unit *
699 dmar_find_matched_drhd_unit(struct pci_dev *dev)
700 {
701 	struct dmar_drhd_unit *dmaru;
702 	struct acpi_dmar_hardware_unit *drhd;
703 
704 	dev = pci_physfn(dev);
705 
706 	rcu_read_lock();
707 	for_each_drhd_unit(dmaru) {
708 		drhd = container_of(dmaru->hdr,
709 				    struct acpi_dmar_hardware_unit,
710 				    header);
711 
712 		if (dmaru->include_all &&
713 		    drhd->segment == pci_domain_nr(dev->bus))
714 			goto out;
715 
716 		if (dmar_pci_device_match(dmaru->devices,
717 					  dmaru->devices_cnt, dev))
718 			goto out;
719 	}
720 	dmaru = NULL;
721 out:
722 	rcu_read_unlock();
723 
724 	return dmaru;
725 }
726 
727 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
728 					      struct acpi_device *adev)
729 {
730 	struct dmar_drhd_unit *dmaru;
731 	struct acpi_dmar_hardware_unit *drhd;
732 	struct acpi_dmar_device_scope *scope;
733 	struct device *tmp;
734 	int i;
735 	struct acpi_dmar_pci_path *path;
736 
737 	for_each_drhd_unit(dmaru) {
738 		drhd = container_of(dmaru->hdr,
739 				    struct acpi_dmar_hardware_unit,
740 				    header);
741 
742 		for (scope = (void *)(drhd + 1);
743 		     (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
744 		     scope = ((void *)scope) + scope->length) {
745 			if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
746 				continue;
747 			if (scope->enumeration_id != device_number)
748 				continue;
749 
750 			path = (void *)(scope + 1);
751 			pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
752 				dev_name(&adev->dev), dmaru->reg_base_addr,
753 				scope->bus, path->device, path->function);
754 			for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
755 				if (tmp == NULL) {
756 					dmaru->devices[i].bus = scope->bus;
757 					dmaru->devices[i].devfn = PCI_DEVFN(path->device,
758 									    path->function);
759 					rcu_assign_pointer(dmaru->devices[i].dev,
760 							   get_device(&adev->dev));
761 					return;
762 				}
763 			BUG_ON(i >= dmaru->devices_cnt);
764 		}
765 	}
766 	pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
767 		device_number, dev_name(&adev->dev));
768 }
769 
770 static int __init dmar_acpi_dev_scope_init(void)
771 {
772 	struct acpi_dmar_andd *andd;
773 
774 	if (dmar_tbl == NULL)
775 		return -ENODEV;
776 
777 	for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
778 	     ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
779 	     andd = ((void *)andd) + andd->header.length) {
780 		if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
781 			acpi_handle h;
782 			struct acpi_device *adev;
783 
784 			if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
785 							  andd->device_name,
786 							  &h))) {
787 				pr_err("Failed to find handle for ACPI object %s\n",
788 				       andd->device_name);
789 				continue;
790 			}
791 			adev = acpi_fetch_acpi_dev(h);
792 			if (!adev) {
793 				pr_err("Failed to get device for ACPI object %s\n",
794 				       andd->device_name);
795 				continue;
796 			}
797 			dmar_acpi_insert_dev_scope(andd->device_number, adev);
798 		}
799 	}
800 	return 0;
801 }
802 
803 int __init dmar_dev_scope_init(void)
804 {
805 	struct pci_dev *dev = NULL;
806 	struct dmar_pci_notify_info *info;
807 
808 	if (dmar_dev_scope_status != 1)
809 		return dmar_dev_scope_status;
810 
811 	if (list_empty(&dmar_drhd_units)) {
812 		dmar_dev_scope_status = -ENODEV;
813 	} else {
814 		dmar_dev_scope_status = 0;
815 
816 		dmar_acpi_dev_scope_init();
817 
818 		for_each_pci_dev(dev) {
819 			if (dev->is_virtfn)
820 				continue;
821 
822 			info = dmar_alloc_pci_notify_info(dev,
823 					BUS_NOTIFY_ADD_DEVICE);
824 			if (!info) {
825 				pci_dev_put(dev);
826 				return dmar_dev_scope_status;
827 			} else {
828 				dmar_pci_bus_add_dev(info);
829 				dmar_free_pci_notify_info(info);
830 			}
831 		}
832 	}
833 
834 	return dmar_dev_scope_status;
835 }
836 
837 void __init dmar_register_bus_notifier(void)
838 {
839 	bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
840 }
841 
842 
843 int __init dmar_table_init(void)
844 {
845 	static int dmar_table_initialized;
846 	int ret;
847 
848 	if (dmar_table_initialized == 0) {
849 		ret = parse_dmar_table();
850 		if (ret < 0) {
851 			if (ret != -ENODEV)
852 				pr_info("Parse DMAR table failure.\n");
853 		} else  if (list_empty(&dmar_drhd_units)) {
854 			pr_info("No DMAR devices found\n");
855 			ret = -ENODEV;
856 		}
857 
858 		if (ret < 0)
859 			dmar_table_initialized = ret;
860 		else
861 			dmar_table_initialized = 1;
862 	}
863 
864 	return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
865 }
866 
867 static void warn_invalid_dmar(u64 addr, const char *message)
868 {
869 	pr_warn_once(FW_BUG
870 		"Your BIOS is broken; DMAR reported at address %llx%s!\n"
871 		"BIOS vendor: %s; Ver: %s; Product Version: %s\n",
872 		addr, message,
873 		dmi_get_system_info(DMI_BIOS_VENDOR),
874 		dmi_get_system_info(DMI_BIOS_VERSION),
875 		dmi_get_system_info(DMI_PRODUCT_VERSION));
876 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
877 }
878 
879 static int __ref
880 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
881 {
882 	struct acpi_dmar_hardware_unit *drhd;
883 	void __iomem *addr;
884 	u64 cap, ecap;
885 
886 	drhd = (void *)entry;
887 	if (!drhd->address) {
888 		warn_invalid_dmar(0, "");
889 		return -EINVAL;
890 	}
891 
892 	if (arg)
893 		addr = ioremap(drhd->address, VTD_PAGE_SIZE);
894 	else
895 		addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
896 	if (!addr) {
897 		pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
898 		return -EINVAL;
899 	}
900 
901 	cap = dmar_readq(addr + DMAR_CAP_REG);
902 	ecap = dmar_readq(addr + DMAR_ECAP_REG);
903 
904 	if (arg)
905 		iounmap(addr);
906 	else
907 		early_iounmap(addr, VTD_PAGE_SIZE);
908 
909 	if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
910 		warn_invalid_dmar(drhd->address, " returns all ones");
911 		return -EINVAL;
912 	}
913 
914 	return 0;
915 }
916 
917 void __init detect_intel_iommu(void)
918 {
919 	int ret;
920 	struct dmar_res_callback validate_drhd_cb = {
921 		.cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
922 		.ignore_unhandled = true,
923 	};
924 
925 	down_write(&dmar_global_lock);
926 	ret = dmar_table_detect();
927 	if (!ret)
928 		ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
929 					   &validate_drhd_cb);
930 	if (!ret && !no_iommu && !iommu_detected &&
931 	    (!dmar_disabled || dmar_platform_optin())) {
932 		iommu_detected = 1;
933 		/* Make sure ACS will be enabled */
934 		pci_request_acs();
935 	}
936 
937 #ifdef CONFIG_X86
938 	if (!ret) {
939 		x86_init.iommu.iommu_init = intel_iommu_init;
940 		x86_platform.iommu_shutdown = intel_iommu_shutdown;
941 	}
942 
943 #endif
944 
945 	if (dmar_tbl) {
946 		acpi_put_table(dmar_tbl);
947 		dmar_tbl = NULL;
948 	}
949 	up_write(&dmar_global_lock);
950 }
951 
952 static void unmap_iommu(struct intel_iommu *iommu)
953 {
954 	iounmap(iommu->reg);
955 	release_mem_region(iommu->reg_phys, iommu->reg_size);
956 }
957 
958 /**
959  * map_iommu: map the iommu's registers
960  * @iommu: the iommu to map
961  * @drhd: DMA remapping hardware definition structure
962  *
963  * Memory map the iommu's registers.  Start w/ a single page, and
964  * possibly expand if that turns out to be insufficent.
965  */
966 static int map_iommu(struct intel_iommu *iommu, struct dmar_drhd_unit *drhd)
967 {
968 	u64 phys_addr = drhd->reg_base_addr;
969 	int map_size, err=0;
970 
971 	iommu->reg_phys = phys_addr;
972 	iommu->reg_size = drhd->reg_size;
973 
974 	if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
975 		pr_err("Can't reserve memory\n");
976 		err = -EBUSY;
977 		goto out;
978 	}
979 
980 	iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
981 	if (!iommu->reg) {
982 		pr_err("Can't map the region\n");
983 		err = -ENOMEM;
984 		goto release;
985 	}
986 
987 	iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
988 	iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
989 
990 	if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
991 		err = -EINVAL;
992 		warn_invalid_dmar(phys_addr, " returns all ones");
993 		goto unmap;
994 	}
995 
996 	/* the registers might be more than one page */
997 	map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
998 			 cap_max_fault_reg_offset(iommu->cap));
999 	map_size = VTD_PAGE_ALIGN(map_size);
1000 	if (map_size > iommu->reg_size) {
1001 		iounmap(iommu->reg);
1002 		release_mem_region(iommu->reg_phys, iommu->reg_size);
1003 		iommu->reg_size = map_size;
1004 		if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
1005 					iommu->name)) {
1006 			pr_err("Can't reserve memory\n");
1007 			err = -EBUSY;
1008 			goto out;
1009 		}
1010 		iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
1011 		if (!iommu->reg) {
1012 			pr_err("Can't map the region\n");
1013 			err = -ENOMEM;
1014 			goto release;
1015 		}
1016 	}
1017 
1018 	if (cap_ecmds(iommu->cap)) {
1019 		int i;
1020 
1021 		for (i = 0; i < DMA_MAX_NUM_ECMDCAP; i++) {
1022 			iommu->ecmdcap[i] = dmar_readq(iommu->reg + DMAR_ECCAP_REG +
1023 						       i * DMA_ECMD_REG_STEP);
1024 		}
1025 	}
1026 
1027 	err = 0;
1028 	goto out;
1029 
1030 unmap:
1031 	iounmap(iommu->reg);
1032 release:
1033 	release_mem_region(iommu->reg_phys, iommu->reg_size);
1034 out:
1035 	return err;
1036 }
1037 
1038 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1039 {
1040 	struct intel_iommu *iommu;
1041 	u32 ver, sts;
1042 	int agaw = -1;
1043 	int msagaw = -1;
1044 	int err;
1045 
1046 	if (!drhd->reg_base_addr) {
1047 		warn_invalid_dmar(0, "");
1048 		return -EINVAL;
1049 	}
1050 
1051 	iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1052 	if (!iommu)
1053 		return -ENOMEM;
1054 
1055 	iommu->seq_id = ida_alloc_range(&dmar_seq_ids, 0,
1056 					DMAR_UNITS_SUPPORTED - 1, GFP_KERNEL);
1057 	if (iommu->seq_id < 0) {
1058 		pr_err("Failed to allocate seq_id\n");
1059 		err = iommu->seq_id;
1060 		goto error;
1061 	}
1062 	sprintf(iommu->name, "dmar%d", iommu->seq_id);
1063 
1064 	err = map_iommu(iommu, drhd);
1065 	if (err) {
1066 		pr_err("Failed to map %s\n", iommu->name);
1067 		goto error_free_seq_id;
1068 	}
1069 
1070 	err = -EINVAL;
1071 	if (!cap_sagaw(iommu->cap) &&
1072 	    (!ecap_smts(iommu->ecap) || ecap_slts(iommu->ecap))) {
1073 		pr_info("%s: No supported address widths. Not attempting DMA translation.\n",
1074 			iommu->name);
1075 		drhd->ignored = 1;
1076 	}
1077 
1078 	if (!drhd->ignored) {
1079 		agaw = iommu_calculate_agaw(iommu);
1080 		if (agaw < 0) {
1081 			pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1082 			       iommu->seq_id);
1083 			drhd->ignored = 1;
1084 		}
1085 	}
1086 	if (!drhd->ignored) {
1087 		msagaw = iommu_calculate_max_sagaw(iommu);
1088 		if (msagaw < 0) {
1089 			pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1090 			       iommu->seq_id);
1091 			drhd->ignored = 1;
1092 			agaw = -1;
1093 		}
1094 	}
1095 	iommu->agaw = agaw;
1096 	iommu->msagaw = msagaw;
1097 	iommu->segment = drhd->segment;
1098 	iommu->device_rbtree = RB_ROOT;
1099 	spin_lock_init(&iommu->device_rbtree_lock);
1100 	mutex_init(&iommu->iopf_lock);
1101 	iommu->node = NUMA_NO_NODE;
1102 
1103 	ver = readl(iommu->reg + DMAR_VER_REG);
1104 	pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1105 		iommu->name,
1106 		(unsigned long long)drhd->reg_base_addr,
1107 		DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1108 		(unsigned long long)iommu->cap,
1109 		(unsigned long long)iommu->ecap);
1110 
1111 	/* Reflect status in gcmd */
1112 	sts = readl(iommu->reg + DMAR_GSTS_REG);
1113 	if (sts & DMA_GSTS_IRES)
1114 		iommu->gcmd |= DMA_GCMD_IRE;
1115 	if (sts & DMA_GSTS_TES)
1116 		iommu->gcmd |= DMA_GCMD_TE;
1117 	if (sts & DMA_GSTS_QIES)
1118 		iommu->gcmd |= DMA_GCMD_QIE;
1119 
1120 	if (alloc_iommu_pmu(iommu))
1121 		pr_debug("Cannot alloc PMU for iommu (seq_id = %d)\n", iommu->seq_id);
1122 
1123 	raw_spin_lock_init(&iommu->register_lock);
1124 
1125 	/*
1126 	 * A value of N in PSS field of eCap register indicates hardware
1127 	 * supports PASID field of N+1 bits.
1128 	 */
1129 	if (pasid_supported(iommu))
1130 		iommu->iommu.max_pasids = 2UL << ecap_pss(iommu->ecap);
1131 
1132 	/*
1133 	 * This is only for hotplug; at boot time intel_iommu_enabled won't
1134 	 * be set yet. When intel_iommu_init() runs, it registers the units
1135 	 * present at boot time, then sets intel_iommu_enabled.
1136 	 */
1137 	if (intel_iommu_enabled && !drhd->ignored) {
1138 		err = iommu_device_sysfs_add(&iommu->iommu, NULL,
1139 					     intel_iommu_groups,
1140 					     "%s", iommu->name);
1141 		if (err)
1142 			goto err_unmap;
1143 
1144 		err = iommu_device_register(&iommu->iommu, &intel_iommu_ops, NULL);
1145 		if (err)
1146 			goto err_sysfs;
1147 
1148 		iommu_pmu_register(iommu);
1149 	}
1150 
1151 	drhd->iommu = iommu;
1152 	iommu->drhd = drhd;
1153 
1154 	return 0;
1155 
1156 err_sysfs:
1157 	iommu_device_sysfs_remove(&iommu->iommu);
1158 err_unmap:
1159 	free_iommu_pmu(iommu);
1160 	unmap_iommu(iommu);
1161 error_free_seq_id:
1162 	ida_free(&dmar_seq_ids, iommu->seq_id);
1163 error:
1164 	kfree(iommu);
1165 	return err;
1166 }
1167 
1168 static void free_iommu(struct intel_iommu *iommu)
1169 {
1170 	if (intel_iommu_enabled && !iommu->drhd->ignored) {
1171 		iommu_pmu_unregister(iommu);
1172 		iommu_device_unregister(&iommu->iommu);
1173 		iommu_device_sysfs_remove(&iommu->iommu);
1174 	}
1175 
1176 	free_iommu_pmu(iommu);
1177 
1178 	if (iommu->irq) {
1179 		if (iommu->pr_irq) {
1180 			free_irq(iommu->pr_irq, iommu);
1181 			dmar_free_hwirq(iommu->pr_irq);
1182 			iommu->pr_irq = 0;
1183 		}
1184 		free_irq(iommu->irq, iommu);
1185 		dmar_free_hwirq(iommu->irq);
1186 		iommu->irq = 0;
1187 	}
1188 
1189 	if (iommu->qi) {
1190 		free_page((unsigned long)iommu->qi->desc);
1191 		kfree(iommu->qi->desc_status);
1192 		kfree(iommu->qi);
1193 	}
1194 
1195 	if (iommu->reg)
1196 		unmap_iommu(iommu);
1197 
1198 	ida_free(&dmar_seq_ids, iommu->seq_id);
1199 	kfree(iommu);
1200 }
1201 
1202 /*
1203  * Reclaim all the submitted descriptors which have completed its work.
1204  */
1205 static inline void reclaim_free_desc(struct q_inval *qi)
1206 {
1207 	while (qi->desc_status[qi->free_tail] == QI_DONE ||
1208 	       qi->desc_status[qi->free_tail] == QI_ABORT) {
1209 		qi->desc_status[qi->free_tail] = QI_FREE;
1210 		qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1211 		qi->free_cnt++;
1212 	}
1213 }
1214 
1215 static const char *qi_type_string(u8 type)
1216 {
1217 	switch (type) {
1218 	case QI_CC_TYPE:
1219 		return "Context-cache Invalidation";
1220 	case QI_IOTLB_TYPE:
1221 		return "IOTLB Invalidation";
1222 	case QI_DIOTLB_TYPE:
1223 		return "Device-TLB Invalidation";
1224 	case QI_IEC_TYPE:
1225 		return "Interrupt Entry Cache Invalidation";
1226 	case QI_IWD_TYPE:
1227 		return "Invalidation Wait";
1228 	case QI_EIOTLB_TYPE:
1229 		return "PASID-based IOTLB Invalidation";
1230 	case QI_PC_TYPE:
1231 		return "PASID-cache Invalidation";
1232 	case QI_DEIOTLB_TYPE:
1233 		return "PASID-based Device-TLB Invalidation";
1234 	case QI_PGRP_RESP_TYPE:
1235 		return "Page Group Response";
1236 	default:
1237 		return "UNKNOWN";
1238 	}
1239 }
1240 
1241 static void qi_dump_fault(struct intel_iommu *iommu, u32 fault)
1242 {
1243 	unsigned int head = dmar_readl(iommu->reg + DMAR_IQH_REG);
1244 	u64 iqe_err = dmar_readq(iommu->reg + DMAR_IQER_REG);
1245 	struct qi_desc *desc = iommu->qi->desc + head;
1246 
1247 	if (fault & DMA_FSTS_IQE)
1248 		pr_err("VT-d detected Invalidation Queue Error: Reason %llx",
1249 		       DMAR_IQER_REG_IQEI(iqe_err));
1250 	if (fault & DMA_FSTS_ITE)
1251 		pr_err("VT-d detected Invalidation Time-out Error: SID %llx",
1252 		       DMAR_IQER_REG_ITESID(iqe_err));
1253 	if (fault & DMA_FSTS_ICE)
1254 		pr_err("VT-d detected Invalidation Completion Error: SID %llx",
1255 		       DMAR_IQER_REG_ICESID(iqe_err));
1256 
1257 	pr_err("QI HEAD: %s qw0 = 0x%llx, qw1 = 0x%llx\n",
1258 	       qi_type_string(desc->qw0 & 0xf),
1259 	       (unsigned long long)desc->qw0,
1260 	       (unsigned long long)desc->qw1);
1261 
1262 	head = ((head >> qi_shift(iommu)) + QI_LENGTH - 1) % QI_LENGTH;
1263 	head <<= qi_shift(iommu);
1264 	desc = iommu->qi->desc + head;
1265 
1266 	pr_err("QI PRIOR: %s qw0 = 0x%llx, qw1 = 0x%llx\n",
1267 	       qi_type_string(desc->qw0 & 0xf),
1268 	       (unsigned long long)desc->qw0,
1269 	       (unsigned long long)desc->qw1);
1270 }
1271 
1272 static int qi_check_fault(struct intel_iommu *iommu, int index, int wait_index)
1273 {
1274 	u32 fault;
1275 	int head, tail;
1276 	struct device *dev;
1277 	u64 iqe_err, ite_sid;
1278 	struct q_inval *qi = iommu->qi;
1279 	int shift = qi_shift(iommu);
1280 
1281 	if (qi->desc_status[wait_index] == QI_ABORT)
1282 		return -EAGAIN;
1283 
1284 	fault = readl(iommu->reg + DMAR_FSTS_REG);
1285 	if (fault & (DMA_FSTS_IQE | DMA_FSTS_ITE | DMA_FSTS_ICE))
1286 		qi_dump_fault(iommu, fault);
1287 
1288 	/*
1289 	 * If IQE happens, the head points to the descriptor associated
1290 	 * with the error. No new descriptors are fetched until the IQE
1291 	 * is cleared.
1292 	 */
1293 	if (fault & DMA_FSTS_IQE) {
1294 		head = readl(iommu->reg + DMAR_IQH_REG);
1295 		if ((head >> shift) == index) {
1296 			struct qi_desc *desc = qi->desc + head;
1297 
1298 			/*
1299 			 * desc->qw2 and desc->qw3 are either reserved or
1300 			 * used by software as private data. We won't print
1301 			 * out these two qw's for security consideration.
1302 			 */
1303 			memcpy(desc, qi->desc + (wait_index << shift),
1304 			       1 << shift);
1305 			writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1306 			pr_info("Invalidation Queue Error (IQE) cleared\n");
1307 			return -EINVAL;
1308 		}
1309 	}
1310 
1311 	/*
1312 	 * If ITE happens, all pending wait_desc commands are aborted.
1313 	 * No new descriptors are fetched until the ITE is cleared.
1314 	 */
1315 	if (fault & DMA_FSTS_ITE) {
1316 		head = readl(iommu->reg + DMAR_IQH_REG);
1317 		head = ((head >> shift) - 1 + QI_LENGTH) % QI_LENGTH;
1318 		head |= 1;
1319 		tail = readl(iommu->reg + DMAR_IQT_REG);
1320 		tail = ((tail >> shift) - 1 + QI_LENGTH) % QI_LENGTH;
1321 
1322 		/*
1323 		 * SID field is valid only when the ITE field is Set in FSTS_REG
1324 		 * see Intel VT-d spec r4.1, section 11.4.9.9
1325 		 */
1326 		iqe_err = dmar_readq(iommu->reg + DMAR_IQER_REG);
1327 		ite_sid = DMAR_IQER_REG_ITESID(iqe_err);
1328 
1329 		writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1330 		pr_info("Invalidation Time-out Error (ITE) cleared\n");
1331 
1332 		do {
1333 			if (qi->desc_status[head] == QI_IN_USE)
1334 				qi->desc_status[head] = QI_ABORT;
1335 			head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1336 		} while (head != tail);
1337 
1338 		/*
1339 		 * If device was released or isn't present, no need to retry
1340 		 * the ATS invalidate request anymore.
1341 		 *
1342 		 * 0 value of ite_sid means old VT-d device, no ite_sid value.
1343 		 * see Intel VT-d spec r4.1, section 11.4.9.9
1344 		 */
1345 		if (ite_sid) {
1346 			dev = device_rbtree_find(iommu, ite_sid);
1347 			if (!dev || !dev_is_pci(dev) ||
1348 			    !pci_device_is_present(to_pci_dev(dev)))
1349 				return -ETIMEDOUT;
1350 		}
1351 		if (qi->desc_status[wait_index] == QI_ABORT)
1352 			return -EAGAIN;
1353 	}
1354 
1355 	if (fault & DMA_FSTS_ICE) {
1356 		writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1357 		pr_info("Invalidation Completion Error (ICE) cleared\n");
1358 	}
1359 
1360 	return 0;
1361 }
1362 
1363 /*
1364  * Function to submit invalidation descriptors of all types to the queued
1365  * invalidation interface(QI). Multiple descriptors can be submitted at a
1366  * time, a wait descriptor will be appended to each submission to ensure
1367  * hardware has completed the invalidation before return. Wait descriptors
1368  * can be part of the submission but it will not be polled for completion.
1369  */
1370 int qi_submit_sync(struct intel_iommu *iommu, struct qi_desc *desc,
1371 		   unsigned int count, unsigned long options)
1372 {
1373 	struct q_inval *qi = iommu->qi;
1374 	s64 devtlb_start_ktime = 0;
1375 	s64 iotlb_start_ktime = 0;
1376 	s64 iec_start_ktime = 0;
1377 	struct qi_desc wait_desc;
1378 	int wait_index, index;
1379 	unsigned long flags;
1380 	int offset, shift;
1381 	int rc, i;
1382 	u64 type;
1383 
1384 	if (!qi)
1385 		return 0;
1386 
1387 	type = desc->qw0 & GENMASK_ULL(3, 0);
1388 
1389 	if ((type == QI_IOTLB_TYPE || type == QI_EIOTLB_TYPE) &&
1390 	    dmar_latency_enabled(iommu, DMAR_LATENCY_INV_IOTLB))
1391 		iotlb_start_ktime = ktime_to_ns(ktime_get());
1392 
1393 	if ((type == QI_DIOTLB_TYPE || type == QI_DEIOTLB_TYPE) &&
1394 	    dmar_latency_enabled(iommu, DMAR_LATENCY_INV_DEVTLB))
1395 		devtlb_start_ktime = ktime_to_ns(ktime_get());
1396 
1397 	if (type == QI_IEC_TYPE &&
1398 	    dmar_latency_enabled(iommu, DMAR_LATENCY_INV_IEC))
1399 		iec_start_ktime = ktime_to_ns(ktime_get());
1400 
1401 restart:
1402 	rc = 0;
1403 
1404 	raw_spin_lock_irqsave(&qi->q_lock, flags);
1405 	/*
1406 	 * Check if we have enough empty slots in the queue to submit,
1407 	 * the calculation is based on:
1408 	 * # of desc + 1 wait desc + 1 space between head and tail
1409 	 */
1410 	while (qi->free_cnt < count + 2) {
1411 		raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1412 		cpu_relax();
1413 		raw_spin_lock_irqsave(&qi->q_lock, flags);
1414 	}
1415 
1416 	index = qi->free_head;
1417 	wait_index = (index + count) % QI_LENGTH;
1418 	shift = qi_shift(iommu);
1419 
1420 	for (i = 0; i < count; i++) {
1421 		offset = ((index + i) % QI_LENGTH) << shift;
1422 		memcpy(qi->desc + offset, &desc[i], 1 << shift);
1423 		qi->desc_status[(index + i) % QI_LENGTH] = QI_IN_USE;
1424 		trace_qi_submit(iommu, desc[i].qw0, desc[i].qw1,
1425 				desc[i].qw2, desc[i].qw3);
1426 	}
1427 	qi->desc_status[wait_index] = QI_IN_USE;
1428 
1429 	wait_desc.qw0 = QI_IWD_STATUS_DATA(QI_DONE) |
1430 			QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1431 	if (options & QI_OPT_WAIT_DRAIN)
1432 		wait_desc.qw0 |= QI_IWD_PRQ_DRAIN;
1433 	wait_desc.qw1 = virt_to_phys(&qi->desc_status[wait_index]);
1434 	wait_desc.qw2 = 0;
1435 	wait_desc.qw3 = 0;
1436 
1437 	offset = wait_index << shift;
1438 	memcpy(qi->desc + offset, &wait_desc, 1 << shift);
1439 
1440 	qi->free_head = (qi->free_head + count + 1) % QI_LENGTH;
1441 	qi->free_cnt -= count + 1;
1442 
1443 	/*
1444 	 * update the HW tail register indicating the presence of
1445 	 * new descriptors.
1446 	 */
1447 	writel(qi->free_head << shift, iommu->reg + DMAR_IQT_REG);
1448 
1449 	while (qi->desc_status[wait_index] != QI_DONE) {
1450 		/*
1451 		 * We will leave the interrupts disabled, to prevent interrupt
1452 		 * context to queue another cmd while a cmd is already submitted
1453 		 * and waiting for completion on this cpu. This is to avoid
1454 		 * a deadlock where the interrupt context can wait indefinitely
1455 		 * for free slots in the queue.
1456 		 */
1457 		rc = qi_check_fault(iommu, index, wait_index);
1458 		if (rc)
1459 			break;
1460 
1461 		raw_spin_unlock(&qi->q_lock);
1462 		cpu_relax();
1463 		raw_spin_lock(&qi->q_lock);
1464 	}
1465 
1466 	for (i = 0; i < count; i++)
1467 		qi->desc_status[(index + i) % QI_LENGTH] = QI_DONE;
1468 
1469 	reclaim_free_desc(qi);
1470 	raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1471 
1472 	if (rc == -EAGAIN)
1473 		goto restart;
1474 
1475 	if (iotlb_start_ktime)
1476 		dmar_latency_update(iommu, DMAR_LATENCY_INV_IOTLB,
1477 				ktime_to_ns(ktime_get()) - iotlb_start_ktime);
1478 
1479 	if (devtlb_start_ktime)
1480 		dmar_latency_update(iommu, DMAR_LATENCY_INV_DEVTLB,
1481 				ktime_to_ns(ktime_get()) - devtlb_start_ktime);
1482 
1483 	if (iec_start_ktime)
1484 		dmar_latency_update(iommu, DMAR_LATENCY_INV_IEC,
1485 				ktime_to_ns(ktime_get()) - iec_start_ktime);
1486 
1487 	return rc;
1488 }
1489 
1490 /*
1491  * Flush the global interrupt entry cache.
1492  */
1493 void qi_global_iec(struct intel_iommu *iommu)
1494 {
1495 	struct qi_desc desc;
1496 
1497 	desc.qw0 = QI_IEC_TYPE;
1498 	desc.qw1 = 0;
1499 	desc.qw2 = 0;
1500 	desc.qw3 = 0;
1501 
1502 	/* should never fail */
1503 	qi_submit_sync(iommu, &desc, 1, 0);
1504 }
1505 
1506 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1507 		      u64 type)
1508 {
1509 	struct qi_desc desc;
1510 
1511 	desc.qw0 = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1512 			| QI_CC_GRAN(type) | QI_CC_TYPE;
1513 	desc.qw1 = 0;
1514 	desc.qw2 = 0;
1515 	desc.qw3 = 0;
1516 
1517 	qi_submit_sync(iommu, &desc, 1, 0);
1518 }
1519 
1520 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1521 		    unsigned int size_order, u64 type)
1522 {
1523 	u8 dw = 0, dr = 0;
1524 
1525 	struct qi_desc desc;
1526 	int ih = 0;
1527 
1528 	if (cap_write_drain(iommu->cap))
1529 		dw = 1;
1530 
1531 	if (cap_read_drain(iommu->cap))
1532 		dr = 1;
1533 
1534 	desc.qw0 = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1535 		| QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1536 	desc.qw1 = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1537 		| QI_IOTLB_AM(size_order);
1538 	desc.qw2 = 0;
1539 	desc.qw3 = 0;
1540 
1541 	qi_submit_sync(iommu, &desc, 1, 0);
1542 }
1543 
1544 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 pfsid,
1545 			u16 qdep, u64 addr, unsigned mask)
1546 {
1547 	struct qi_desc desc;
1548 
1549 	/*
1550 	 * VT-d spec, section 4.3:
1551 	 *
1552 	 * Software is recommended to not submit any Device-TLB invalidation
1553 	 * requests while address remapping hardware is disabled.
1554 	 */
1555 	if (!(iommu->gcmd & DMA_GCMD_TE))
1556 		return;
1557 
1558 	if (mask) {
1559 		addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1560 		desc.qw1 = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1561 	} else
1562 		desc.qw1 = QI_DEV_IOTLB_ADDR(addr);
1563 
1564 	if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1565 		qdep = 0;
1566 
1567 	desc.qw0 = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1568 		   QI_DIOTLB_TYPE | QI_DEV_IOTLB_PFSID(pfsid);
1569 	desc.qw2 = 0;
1570 	desc.qw3 = 0;
1571 
1572 	qi_submit_sync(iommu, &desc, 1, 0);
1573 }
1574 
1575 /* PASID-based IOTLB invalidation */
1576 void qi_flush_piotlb(struct intel_iommu *iommu, u16 did, u32 pasid, u64 addr,
1577 		     unsigned long npages, bool ih)
1578 {
1579 	struct qi_desc desc = {.qw2 = 0, .qw3 = 0};
1580 
1581 	/*
1582 	 * npages == -1 means a PASID-selective invalidation, otherwise,
1583 	 * a positive value for Page-selective-within-PASID invalidation.
1584 	 * 0 is not a valid input.
1585 	 */
1586 	if (WARN_ON(!npages)) {
1587 		pr_err("Invalid input npages = %ld\n", npages);
1588 		return;
1589 	}
1590 
1591 	if (npages == -1) {
1592 		desc.qw0 = QI_EIOTLB_PASID(pasid) |
1593 				QI_EIOTLB_DID(did) |
1594 				QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) |
1595 				QI_EIOTLB_TYPE;
1596 		desc.qw1 = 0;
1597 	} else {
1598 		int mask = ilog2(__roundup_pow_of_two(npages));
1599 		unsigned long align = (1ULL << (VTD_PAGE_SHIFT + mask));
1600 
1601 		if (WARN_ON_ONCE(!IS_ALIGNED(addr, align)))
1602 			addr = ALIGN_DOWN(addr, align);
1603 
1604 		desc.qw0 = QI_EIOTLB_PASID(pasid) |
1605 				QI_EIOTLB_DID(did) |
1606 				QI_EIOTLB_GRAN(QI_GRAN_PSI_PASID) |
1607 				QI_EIOTLB_TYPE;
1608 		desc.qw1 = QI_EIOTLB_ADDR(addr) |
1609 				QI_EIOTLB_IH(ih) |
1610 				QI_EIOTLB_AM(mask);
1611 	}
1612 
1613 	qi_submit_sync(iommu, &desc, 1, 0);
1614 }
1615 
1616 /* PASID-based device IOTLB Invalidate */
1617 void qi_flush_dev_iotlb_pasid(struct intel_iommu *iommu, u16 sid, u16 pfsid,
1618 			      u32 pasid,  u16 qdep, u64 addr, unsigned int size_order)
1619 {
1620 	unsigned long mask = 1UL << (VTD_PAGE_SHIFT + size_order - 1);
1621 	struct qi_desc desc = {.qw1 = 0, .qw2 = 0, .qw3 = 0};
1622 
1623 	/*
1624 	 * VT-d spec, section 4.3:
1625 	 *
1626 	 * Software is recommended to not submit any Device-TLB invalidation
1627 	 * requests while address remapping hardware is disabled.
1628 	 */
1629 	if (!(iommu->gcmd & DMA_GCMD_TE))
1630 		return;
1631 
1632 	desc.qw0 = QI_DEV_EIOTLB_PASID(pasid) | QI_DEV_EIOTLB_SID(sid) |
1633 		QI_DEV_EIOTLB_QDEP(qdep) | QI_DEIOTLB_TYPE |
1634 		QI_DEV_IOTLB_PFSID(pfsid);
1635 
1636 	/*
1637 	 * If S bit is 0, we only flush a single page. If S bit is set,
1638 	 * The least significant zero bit indicates the invalidation address
1639 	 * range. VT-d spec 6.5.2.6.
1640 	 * e.g. address bit 12[0] indicates 8KB, 13[0] indicates 16KB.
1641 	 * size order = 0 is PAGE_SIZE 4KB
1642 	 * Max Invs Pending (MIP) is set to 0 for now until we have DIT in
1643 	 * ECAP.
1644 	 */
1645 	if (!IS_ALIGNED(addr, VTD_PAGE_SIZE << size_order))
1646 		pr_warn_ratelimited("Invalidate non-aligned address %llx, order %d\n",
1647 				    addr, size_order);
1648 
1649 	/* Take page address */
1650 	desc.qw1 = QI_DEV_EIOTLB_ADDR(addr);
1651 
1652 	if (size_order) {
1653 		/*
1654 		 * Existing 0s in address below size_order may be the least
1655 		 * significant bit, we must set them to 1s to avoid having
1656 		 * smaller size than desired.
1657 		 */
1658 		desc.qw1 |= GENMASK_ULL(size_order + VTD_PAGE_SHIFT - 1,
1659 					VTD_PAGE_SHIFT);
1660 		/* Clear size_order bit to indicate size */
1661 		desc.qw1 &= ~mask;
1662 		/* Set the S bit to indicate flushing more than 1 page */
1663 		desc.qw1 |= QI_DEV_EIOTLB_SIZE;
1664 	}
1665 
1666 	qi_submit_sync(iommu, &desc, 1, 0);
1667 }
1668 
1669 void qi_flush_pasid_cache(struct intel_iommu *iommu, u16 did,
1670 			  u64 granu, u32 pasid)
1671 {
1672 	struct qi_desc desc = {.qw1 = 0, .qw2 = 0, .qw3 = 0};
1673 
1674 	desc.qw0 = QI_PC_PASID(pasid) | QI_PC_DID(did) |
1675 			QI_PC_GRAN(granu) | QI_PC_TYPE;
1676 	qi_submit_sync(iommu, &desc, 1, 0);
1677 }
1678 
1679 /*
1680  * Disable Queued Invalidation interface.
1681  */
1682 void dmar_disable_qi(struct intel_iommu *iommu)
1683 {
1684 	unsigned long flags;
1685 	u32 sts;
1686 	cycles_t start_time = get_cycles();
1687 
1688 	if (!ecap_qis(iommu->ecap))
1689 		return;
1690 
1691 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
1692 
1693 	sts =  readl(iommu->reg + DMAR_GSTS_REG);
1694 	if (!(sts & DMA_GSTS_QIES))
1695 		goto end;
1696 
1697 	/*
1698 	 * Give a chance to HW to complete the pending invalidation requests.
1699 	 */
1700 	while ((readl(iommu->reg + DMAR_IQT_REG) !=
1701 		readl(iommu->reg + DMAR_IQH_REG)) &&
1702 		(DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1703 		cpu_relax();
1704 
1705 	iommu->gcmd &= ~DMA_GCMD_QIE;
1706 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1707 
1708 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1709 		      !(sts & DMA_GSTS_QIES), sts);
1710 end:
1711 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1712 }
1713 
1714 /*
1715  * Enable queued invalidation.
1716  */
1717 static void __dmar_enable_qi(struct intel_iommu *iommu)
1718 {
1719 	u32 sts;
1720 	unsigned long flags;
1721 	struct q_inval *qi = iommu->qi;
1722 	u64 val = virt_to_phys(qi->desc);
1723 
1724 	qi->free_head = qi->free_tail = 0;
1725 	qi->free_cnt = QI_LENGTH;
1726 
1727 	/*
1728 	 * Set DW=1 and QS=1 in IQA_REG when Scalable Mode capability
1729 	 * is present.
1730 	 */
1731 	if (ecap_smts(iommu->ecap))
1732 		val |= BIT_ULL(11) | BIT_ULL(0);
1733 
1734 	raw_spin_lock_irqsave(&iommu->register_lock, flags);
1735 
1736 	/* write zero to the tail reg */
1737 	writel(0, iommu->reg + DMAR_IQT_REG);
1738 
1739 	dmar_writeq(iommu->reg + DMAR_IQA_REG, val);
1740 
1741 	iommu->gcmd |= DMA_GCMD_QIE;
1742 	writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1743 
1744 	/* Make sure hardware complete it */
1745 	IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1746 
1747 	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1748 }
1749 
1750 /*
1751  * Enable Queued Invalidation interface. This is a must to support
1752  * interrupt-remapping. Also used by DMA-remapping, which replaces
1753  * register based IOTLB invalidation.
1754  */
1755 int dmar_enable_qi(struct intel_iommu *iommu)
1756 {
1757 	struct q_inval *qi;
1758 	struct page *desc_page;
1759 
1760 	if (!ecap_qis(iommu->ecap))
1761 		return -ENOENT;
1762 
1763 	/*
1764 	 * queued invalidation is already setup and enabled.
1765 	 */
1766 	if (iommu->qi)
1767 		return 0;
1768 
1769 	iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1770 	if (!iommu->qi)
1771 		return -ENOMEM;
1772 
1773 	qi = iommu->qi;
1774 
1775 	/*
1776 	 * Need two pages to accommodate 256 descriptors of 256 bits each
1777 	 * if the remapping hardware supports scalable mode translation.
1778 	 */
1779 	desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
1780 				     !!ecap_smts(iommu->ecap));
1781 	if (!desc_page) {
1782 		kfree(qi);
1783 		iommu->qi = NULL;
1784 		return -ENOMEM;
1785 	}
1786 
1787 	qi->desc = page_address(desc_page);
1788 
1789 	qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC);
1790 	if (!qi->desc_status) {
1791 		free_page((unsigned long) qi->desc);
1792 		kfree(qi);
1793 		iommu->qi = NULL;
1794 		return -ENOMEM;
1795 	}
1796 
1797 	raw_spin_lock_init(&qi->q_lock);
1798 
1799 	__dmar_enable_qi(iommu);
1800 
1801 	return 0;
1802 }
1803 
1804 /* iommu interrupt handling. Most stuff are MSI-like. */
1805 
1806 enum faulttype {
1807 	DMA_REMAP,
1808 	INTR_REMAP,
1809 	UNKNOWN,
1810 };
1811 
1812 static const char *dma_remap_fault_reasons[] =
1813 {
1814 	"Software",
1815 	"Present bit in root entry is clear",
1816 	"Present bit in context entry is clear",
1817 	"Invalid context entry",
1818 	"Access beyond MGAW",
1819 	"PTE Write access is not set",
1820 	"PTE Read access is not set",
1821 	"Next page table ptr is invalid",
1822 	"Root table address invalid",
1823 	"Context table ptr is invalid",
1824 	"non-zero reserved fields in RTP",
1825 	"non-zero reserved fields in CTP",
1826 	"non-zero reserved fields in PTE",
1827 	"PCE for translation request specifies blocking",
1828 };
1829 
1830 static const char * const dma_remap_sm_fault_reasons[] = {
1831 	"SM: Invalid Root Table Address",
1832 	"SM: TTM 0 for request with PASID",
1833 	"SM: TTM 0 for page group request",
1834 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x33-0x37 */
1835 	"SM: Error attempting to access Root Entry",
1836 	"SM: Present bit in Root Entry is clear",
1837 	"SM: Non-zero reserved field set in Root Entry",
1838 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x3B-0x3F */
1839 	"SM: Error attempting to access Context Entry",
1840 	"SM: Present bit in Context Entry is clear",
1841 	"SM: Non-zero reserved field set in the Context Entry",
1842 	"SM: Invalid Context Entry",
1843 	"SM: DTE field in Context Entry is clear",
1844 	"SM: PASID Enable field in Context Entry is clear",
1845 	"SM: PASID is larger than the max in Context Entry",
1846 	"SM: PRE field in Context-Entry is clear",
1847 	"SM: RID_PASID field error in Context-Entry",
1848 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x49-0x4F */
1849 	"SM: Error attempting to access the PASID Directory Entry",
1850 	"SM: Present bit in Directory Entry is clear",
1851 	"SM: Non-zero reserved field set in PASID Directory Entry",
1852 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x53-0x57 */
1853 	"SM: Error attempting to access PASID Table Entry",
1854 	"SM: Present bit in PASID Table Entry is clear",
1855 	"SM: Non-zero reserved field set in PASID Table Entry",
1856 	"SM: Invalid Scalable-Mode PASID Table Entry",
1857 	"SM: ERE field is clear in PASID Table Entry",
1858 	"SM: SRE field is clear in PASID Table Entry",
1859 	"Unknown", "Unknown",/* 0x5E-0x5F */
1860 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x60-0x67 */
1861 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x68-0x6F */
1862 	"SM: Error attempting to access first-level paging entry",
1863 	"SM: Present bit in first-level paging entry is clear",
1864 	"SM: Non-zero reserved field set in first-level paging entry",
1865 	"SM: Error attempting to access FL-PML4 entry",
1866 	"SM: First-level entry address beyond MGAW in Nested translation",
1867 	"SM: Read permission error in FL-PML4 entry in Nested translation",
1868 	"SM: Read permission error in first-level paging entry in Nested translation",
1869 	"SM: Write permission error in first-level paging entry in Nested translation",
1870 	"SM: Error attempting to access second-level paging entry",
1871 	"SM: Read/Write permission error in second-level paging entry",
1872 	"SM: Non-zero reserved field set in second-level paging entry",
1873 	"SM: Invalid second-level page table pointer",
1874 	"SM: A/D bit update needed in second-level entry when set up in no snoop",
1875 	"Unknown", "Unknown", "Unknown", /* 0x7D-0x7F */
1876 	"SM: Address in first-level translation is not canonical",
1877 	"SM: U/S set 0 for first-level translation with user privilege",
1878 	"SM: No execute permission for request with PASID and ER=1",
1879 	"SM: Address beyond the DMA hardware max",
1880 	"SM: Second-level entry address beyond the max",
1881 	"SM: No write permission for Write/AtomicOp request",
1882 	"SM: No read permission for Read/AtomicOp request",
1883 	"SM: Invalid address-interrupt address",
1884 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x88-0x8F */
1885 	"SM: A/D bit update needed in first-level entry when set up in no snoop",
1886 };
1887 
1888 static const char *irq_remap_fault_reasons[] =
1889 {
1890 	"Detected reserved fields in the decoded interrupt-remapped request",
1891 	"Interrupt index exceeded the interrupt-remapping table size",
1892 	"Present field in the IRTE entry is clear",
1893 	"Error accessing interrupt-remapping table pointed by IRTA_REG",
1894 	"Detected reserved fields in the IRTE entry",
1895 	"Blocked a compatibility format interrupt request",
1896 	"Blocked an interrupt request due to source-id verification failure",
1897 };
1898 
1899 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1900 {
1901 	if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1902 					ARRAY_SIZE(irq_remap_fault_reasons))) {
1903 		*fault_type = INTR_REMAP;
1904 		return irq_remap_fault_reasons[fault_reason - 0x20];
1905 	} else if (fault_reason >= 0x30 && (fault_reason - 0x30 <
1906 			ARRAY_SIZE(dma_remap_sm_fault_reasons))) {
1907 		*fault_type = DMA_REMAP;
1908 		return dma_remap_sm_fault_reasons[fault_reason - 0x30];
1909 	} else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1910 		*fault_type = DMA_REMAP;
1911 		return dma_remap_fault_reasons[fault_reason];
1912 	} else {
1913 		*fault_type = UNKNOWN;
1914 		return "Unknown";
1915 	}
1916 }
1917 
1918 
1919 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1920 {
1921 	if (iommu->irq == irq)
1922 		return DMAR_FECTL_REG;
1923 	else if (iommu->pr_irq == irq)
1924 		return DMAR_PECTL_REG;
1925 	else if (iommu->perf_irq == irq)
1926 		return DMAR_PERFINTRCTL_REG;
1927 	else
1928 		BUG();
1929 }
1930 
1931 void dmar_msi_unmask(struct irq_data *data)
1932 {
1933 	struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1934 	int reg = dmar_msi_reg(iommu, data->irq);
1935 	unsigned long flag;
1936 
1937 	/* unmask it */
1938 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1939 	writel(0, iommu->reg + reg);
1940 	/* Read a reg to force flush the post write */
1941 	readl(iommu->reg + reg);
1942 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1943 }
1944 
1945 void dmar_msi_mask(struct irq_data *data)
1946 {
1947 	struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1948 	int reg = dmar_msi_reg(iommu, data->irq);
1949 	unsigned long flag;
1950 
1951 	/* mask it */
1952 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1953 	writel(DMA_FECTL_IM, iommu->reg + reg);
1954 	/* Read a reg to force flush the post write */
1955 	readl(iommu->reg + reg);
1956 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1957 }
1958 
1959 void dmar_msi_write(int irq, struct msi_msg *msg)
1960 {
1961 	struct intel_iommu *iommu = irq_get_handler_data(irq);
1962 	int reg = dmar_msi_reg(iommu, irq);
1963 	unsigned long flag;
1964 
1965 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1966 	writel(msg->data, iommu->reg + reg + 4);
1967 	writel(msg->address_lo, iommu->reg + reg + 8);
1968 	writel(msg->address_hi, iommu->reg + reg + 12);
1969 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1970 }
1971 
1972 void dmar_msi_read(int irq, struct msi_msg *msg)
1973 {
1974 	struct intel_iommu *iommu = irq_get_handler_data(irq);
1975 	int reg = dmar_msi_reg(iommu, irq);
1976 	unsigned long flag;
1977 
1978 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1979 	msg->data = readl(iommu->reg + reg + 4);
1980 	msg->address_lo = readl(iommu->reg + reg + 8);
1981 	msg->address_hi = readl(iommu->reg + reg + 12);
1982 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1983 }
1984 
1985 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1986 		u8 fault_reason, u32 pasid, u16 source_id,
1987 		unsigned long long addr)
1988 {
1989 	const char *reason;
1990 	int fault_type;
1991 
1992 	reason = dmar_get_fault_reason(fault_reason, &fault_type);
1993 
1994 	if (fault_type == INTR_REMAP) {
1995 		pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index 0x%llx [fault reason 0x%02x] %s\n",
1996 		       source_id >> 8, PCI_SLOT(source_id & 0xFF),
1997 		       PCI_FUNC(source_id & 0xFF), addr >> 48,
1998 		       fault_reason, reason);
1999 
2000 		return 0;
2001 	}
2002 
2003 	if (pasid == IOMMU_PASID_INVALID)
2004 		pr_err("[%s NO_PASID] Request device [%02x:%02x.%d] fault addr 0x%llx [fault reason 0x%02x] %s\n",
2005 		       type ? "DMA Read" : "DMA Write",
2006 		       source_id >> 8, PCI_SLOT(source_id & 0xFF),
2007 		       PCI_FUNC(source_id & 0xFF), addr,
2008 		       fault_reason, reason);
2009 	else
2010 		pr_err("[%s PASID 0x%x] Request device [%02x:%02x.%d] fault addr 0x%llx [fault reason 0x%02x] %s\n",
2011 		       type ? "DMA Read" : "DMA Write", pasid,
2012 		       source_id >> 8, PCI_SLOT(source_id & 0xFF),
2013 		       PCI_FUNC(source_id & 0xFF), addr,
2014 		       fault_reason, reason);
2015 
2016 	dmar_fault_dump_ptes(iommu, source_id, addr, pasid);
2017 
2018 	return 0;
2019 }
2020 
2021 #define PRIMARY_FAULT_REG_LEN (16)
2022 irqreturn_t dmar_fault(int irq, void *dev_id)
2023 {
2024 	struct intel_iommu *iommu = dev_id;
2025 	int reg, fault_index;
2026 	u32 fault_status;
2027 	unsigned long flag;
2028 	static DEFINE_RATELIMIT_STATE(rs,
2029 				      DEFAULT_RATELIMIT_INTERVAL,
2030 				      DEFAULT_RATELIMIT_BURST);
2031 
2032 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
2033 	fault_status = readl(iommu->reg + DMAR_FSTS_REG);
2034 	if (fault_status && __ratelimit(&rs))
2035 		pr_err("DRHD: handling fault status reg %x\n", fault_status);
2036 
2037 	/* TBD: ignore advanced fault log currently */
2038 	if (!(fault_status & DMA_FSTS_PPF))
2039 		goto unlock_exit;
2040 
2041 	fault_index = dma_fsts_fault_record_index(fault_status);
2042 	reg = cap_fault_reg_offset(iommu->cap);
2043 	while (1) {
2044 		/* Disable printing, simply clear the fault when ratelimited */
2045 		bool ratelimited = !__ratelimit(&rs);
2046 		u8 fault_reason;
2047 		u16 source_id;
2048 		u64 guest_addr;
2049 		u32 pasid;
2050 		int type;
2051 		u32 data;
2052 		bool pasid_present;
2053 
2054 		/* highest 32 bits */
2055 		data = readl(iommu->reg + reg +
2056 				fault_index * PRIMARY_FAULT_REG_LEN + 12);
2057 		if (!(data & DMA_FRCD_F))
2058 			break;
2059 
2060 		if (!ratelimited) {
2061 			fault_reason = dma_frcd_fault_reason(data);
2062 			type = dma_frcd_type(data);
2063 
2064 			pasid = dma_frcd_pasid_value(data);
2065 			data = readl(iommu->reg + reg +
2066 				     fault_index * PRIMARY_FAULT_REG_LEN + 8);
2067 			source_id = dma_frcd_source_id(data);
2068 
2069 			pasid_present = dma_frcd_pasid_present(data);
2070 			guest_addr = dmar_readq(iommu->reg + reg +
2071 					fault_index * PRIMARY_FAULT_REG_LEN);
2072 			guest_addr = dma_frcd_page_addr(guest_addr);
2073 		}
2074 
2075 		/* clear the fault */
2076 		writel(DMA_FRCD_F, iommu->reg + reg +
2077 			fault_index * PRIMARY_FAULT_REG_LEN + 12);
2078 
2079 		raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
2080 
2081 		if (!ratelimited)
2082 			/* Using pasid -1 if pasid is not present */
2083 			dmar_fault_do_one(iommu, type, fault_reason,
2084 					  pasid_present ? pasid : IOMMU_PASID_INVALID,
2085 					  source_id, guest_addr);
2086 
2087 		fault_index++;
2088 		if (fault_index >= cap_num_fault_regs(iommu->cap))
2089 			fault_index = 0;
2090 		raw_spin_lock_irqsave(&iommu->register_lock, flag);
2091 	}
2092 
2093 	writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO,
2094 	       iommu->reg + DMAR_FSTS_REG);
2095 
2096 unlock_exit:
2097 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
2098 	return IRQ_HANDLED;
2099 }
2100 
2101 int dmar_set_interrupt(struct intel_iommu *iommu)
2102 {
2103 	int irq, ret;
2104 
2105 	/*
2106 	 * Check if the fault interrupt is already initialized.
2107 	 */
2108 	if (iommu->irq)
2109 		return 0;
2110 
2111 	irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
2112 	if (irq > 0) {
2113 		iommu->irq = irq;
2114 	} else {
2115 		pr_err("No free IRQ vectors\n");
2116 		return -EINVAL;
2117 	}
2118 
2119 	ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
2120 	if (ret)
2121 		pr_err("Can't request irq\n");
2122 	return ret;
2123 }
2124 
2125 int __init enable_drhd_fault_handling(void)
2126 {
2127 	struct dmar_drhd_unit *drhd;
2128 	struct intel_iommu *iommu;
2129 
2130 	/*
2131 	 * Enable fault control interrupt.
2132 	 */
2133 	for_each_iommu(iommu, drhd) {
2134 		u32 fault_status;
2135 		int ret = dmar_set_interrupt(iommu);
2136 
2137 		if (ret) {
2138 			pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
2139 			       (unsigned long long)drhd->reg_base_addr, ret);
2140 			return -1;
2141 		}
2142 
2143 		/*
2144 		 * Clear any previous faults.
2145 		 */
2146 		dmar_fault(iommu->irq, iommu);
2147 		fault_status = readl(iommu->reg + DMAR_FSTS_REG);
2148 		writel(fault_status, iommu->reg + DMAR_FSTS_REG);
2149 	}
2150 
2151 	return 0;
2152 }
2153 
2154 /*
2155  * Re-enable Queued Invalidation interface.
2156  */
2157 int dmar_reenable_qi(struct intel_iommu *iommu)
2158 {
2159 	if (!ecap_qis(iommu->ecap))
2160 		return -ENOENT;
2161 
2162 	if (!iommu->qi)
2163 		return -ENOENT;
2164 
2165 	/*
2166 	 * First disable queued invalidation.
2167 	 */
2168 	dmar_disable_qi(iommu);
2169 	/*
2170 	 * Then enable queued invalidation again. Since there is no pending
2171 	 * invalidation requests now, it's safe to re-enable queued
2172 	 * invalidation.
2173 	 */
2174 	__dmar_enable_qi(iommu);
2175 
2176 	return 0;
2177 }
2178 
2179 /*
2180  * Check interrupt remapping support in DMAR table description.
2181  */
2182 int __init dmar_ir_support(void)
2183 {
2184 	struct acpi_table_dmar *dmar;
2185 	dmar = (struct acpi_table_dmar *)dmar_tbl;
2186 	if (!dmar)
2187 		return 0;
2188 	return dmar->flags & 0x1;
2189 }
2190 
2191 /* Check whether DMAR units are in use */
2192 static inline bool dmar_in_use(void)
2193 {
2194 	return irq_remapping_enabled || intel_iommu_enabled;
2195 }
2196 
2197 static int __init dmar_free_unused_resources(void)
2198 {
2199 	struct dmar_drhd_unit *dmaru, *dmaru_n;
2200 
2201 	if (dmar_in_use())
2202 		return 0;
2203 
2204 	if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
2205 		bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
2206 
2207 	down_write(&dmar_global_lock);
2208 	list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
2209 		list_del(&dmaru->list);
2210 		dmar_free_drhd(dmaru);
2211 	}
2212 	up_write(&dmar_global_lock);
2213 
2214 	return 0;
2215 }
2216 
2217 late_initcall(dmar_free_unused_resources);
2218 
2219 /*
2220  * DMAR Hotplug Support
2221  * For more details, please refer to Intel(R) Virtualization Technology
2222  * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
2223  * "Remapping Hardware Unit Hot Plug".
2224  */
2225 static guid_t dmar_hp_guid =
2226 	GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
2227 		  0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
2228 
2229 /*
2230  * Currently there's only one revision and BIOS will not check the revision id,
2231  * so use 0 for safety.
2232  */
2233 #define	DMAR_DSM_REV_ID			0
2234 #define	DMAR_DSM_FUNC_DRHD		1
2235 #define	DMAR_DSM_FUNC_ATSR		2
2236 #define	DMAR_DSM_FUNC_RHSA		3
2237 #define	DMAR_DSM_FUNC_SATC		4
2238 
2239 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
2240 {
2241 	return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func);
2242 }
2243 
2244 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
2245 				  dmar_res_handler_t handler, void *arg)
2246 {
2247 	int ret = -ENODEV;
2248 	union acpi_object *obj;
2249 	struct acpi_dmar_header *start;
2250 	struct dmar_res_callback callback;
2251 	static int res_type[] = {
2252 		[DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
2253 		[DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
2254 		[DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
2255 		[DMAR_DSM_FUNC_SATC] = ACPI_DMAR_TYPE_SATC,
2256 	};
2257 
2258 	if (!dmar_detect_dsm(handle, func))
2259 		return 0;
2260 
2261 	obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID,
2262 				      func, NULL, ACPI_TYPE_BUFFER);
2263 	if (!obj)
2264 		return -ENODEV;
2265 
2266 	memset(&callback, 0, sizeof(callback));
2267 	callback.cb[res_type[func]] = handler;
2268 	callback.arg[res_type[func]] = arg;
2269 	start = (struct acpi_dmar_header *)obj->buffer.pointer;
2270 	ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
2271 
2272 	ACPI_FREE(obj);
2273 
2274 	return ret;
2275 }
2276 
2277 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
2278 {
2279 	int ret;
2280 	struct dmar_drhd_unit *dmaru;
2281 
2282 	dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2283 	if (!dmaru)
2284 		return -ENODEV;
2285 
2286 	ret = dmar_ir_hotplug(dmaru, true);
2287 	if (ret == 0)
2288 		ret = dmar_iommu_hotplug(dmaru, true);
2289 
2290 	return ret;
2291 }
2292 
2293 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
2294 {
2295 	int i, ret;
2296 	struct device *dev;
2297 	struct dmar_drhd_unit *dmaru;
2298 
2299 	dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2300 	if (!dmaru)
2301 		return 0;
2302 
2303 	/*
2304 	 * All PCI devices managed by this unit should have been destroyed.
2305 	 */
2306 	if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
2307 		for_each_active_dev_scope(dmaru->devices,
2308 					  dmaru->devices_cnt, i, dev)
2309 			return -EBUSY;
2310 	}
2311 
2312 	ret = dmar_ir_hotplug(dmaru, false);
2313 	if (ret == 0)
2314 		ret = dmar_iommu_hotplug(dmaru, false);
2315 
2316 	return ret;
2317 }
2318 
2319 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
2320 {
2321 	struct dmar_drhd_unit *dmaru;
2322 
2323 	dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2324 	if (dmaru) {
2325 		list_del_rcu(&dmaru->list);
2326 		synchronize_rcu();
2327 		dmar_free_drhd(dmaru);
2328 	}
2329 
2330 	return 0;
2331 }
2332 
2333 static int dmar_hotplug_insert(acpi_handle handle)
2334 {
2335 	int ret;
2336 	int drhd_count = 0;
2337 
2338 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2339 				     &dmar_validate_one_drhd, (void *)1);
2340 	if (ret)
2341 		goto out;
2342 
2343 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2344 				     &dmar_parse_one_drhd, (void *)&drhd_count);
2345 	if (ret == 0 && drhd_count == 0) {
2346 		pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
2347 		goto out;
2348 	} else if (ret) {
2349 		goto release_drhd;
2350 	}
2351 
2352 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
2353 				     &dmar_parse_one_rhsa, NULL);
2354 	if (ret)
2355 		goto release_drhd;
2356 
2357 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2358 				     &dmar_parse_one_atsr, NULL);
2359 	if (ret)
2360 		goto release_atsr;
2361 
2362 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2363 				     &dmar_hp_add_drhd, NULL);
2364 	if (!ret)
2365 		return 0;
2366 
2367 	dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2368 			       &dmar_hp_remove_drhd, NULL);
2369 release_atsr:
2370 	dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2371 			       &dmar_release_one_atsr, NULL);
2372 release_drhd:
2373 	dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2374 			       &dmar_hp_release_drhd, NULL);
2375 out:
2376 	return ret;
2377 }
2378 
2379 static int dmar_hotplug_remove(acpi_handle handle)
2380 {
2381 	int ret;
2382 
2383 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2384 				     &dmar_check_one_atsr, NULL);
2385 	if (ret)
2386 		return ret;
2387 
2388 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2389 				     &dmar_hp_remove_drhd, NULL);
2390 	if (ret == 0) {
2391 		WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2392 					       &dmar_release_one_atsr, NULL));
2393 		WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2394 					       &dmar_hp_release_drhd, NULL));
2395 	} else {
2396 		dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2397 				       &dmar_hp_add_drhd, NULL);
2398 	}
2399 
2400 	return ret;
2401 }
2402 
2403 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
2404 				       void *context, void **retval)
2405 {
2406 	acpi_handle *phdl = retval;
2407 
2408 	if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2409 		*phdl = handle;
2410 		return AE_CTRL_TERMINATE;
2411 	}
2412 
2413 	return AE_OK;
2414 }
2415 
2416 static int dmar_device_hotplug(acpi_handle handle, bool insert)
2417 {
2418 	int ret;
2419 	acpi_handle tmp = NULL;
2420 	acpi_status status;
2421 
2422 	if (!dmar_in_use())
2423 		return 0;
2424 
2425 	if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2426 		tmp = handle;
2427 	} else {
2428 		status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2429 					     ACPI_UINT32_MAX,
2430 					     dmar_get_dsm_handle,
2431 					     NULL, NULL, &tmp);
2432 		if (ACPI_FAILURE(status)) {
2433 			pr_warn("Failed to locate _DSM method.\n");
2434 			return -ENXIO;
2435 		}
2436 	}
2437 	if (tmp == NULL)
2438 		return 0;
2439 
2440 	down_write(&dmar_global_lock);
2441 	if (insert)
2442 		ret = dmar_hotplug_insert(tmp);
2443 	else
2444 		ret = dmar_hotplug_remove(tmp);
2445 	up_write(&dmar_global_lock);
2446 
2447 	return ret;
2448 }
2449 
2450 int dmar_device_add(acpi_handle handle)
2451 {
2452 	return dmar_device_hotplug(handle, true);
2453 }
2454 
2455 int dmar_device_remove(acpi_handle handle)
2456 {
2457 	return dmar_device_hotplug(handle, false);
2458 }
2459 
2460 /*
2461  * dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table
2462  *
2463  * Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in
2464  * the ACPI DMAR table. This means that the platform boot firmware has made
2465  * sure no device can issue DMA outside of RMRR regions.
2466  */
2467 bool dmar_platform_optin(void)
2468 {
2469 	struct acpi_table_dmar *dmar;
2470 	acpi_status status;
2471 	bool ret;
2472 
2473 	status = acpi_get_table(ACPI_SIG_DMAR, 0,
2474 				(struct acpi_table_header **)&dmar);
2475 	if (ACPI_FAILURE(status))
2476 		return false;
2477 
2478 	ret = !!(dmar->flags & DMAR_PLATFORM_OPT_IN);
2479 	acpi_put_table((struct acpi_table_header *)dmar);
2480 
2481 	return ret;
2482 }
2483 EXPORT_SYMBOL_GPL(dmar_platform_optin);
2484