xref: /linux/drivers/iommu/intel/dmar.c (revision 3269d6fb7580e91313f40dffcff70c01cd3f0717)
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 "../iommu-pages.h"
36 #include "perf.h"
37 #include "trace.h"
38 #include "perfmon.h"
39 
40 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
41 struct dmar_res_callback {
42 	dmar_res_handler_t	cb[ACPI_DMAR_TYPE_RESERVED];
43 	void			*arg[ACPI_DMAR_TYPE_RESERVED];
44 	bool			ignore_unhandled;
45 	bool			print_entry;
46 };
47 
48 /*
49  * Assumptions:
50  * 1) The hotplug framework guarentees that DMAR unit will be hot-added
51  *    before IO devices managed by that unit.
52  * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
53  *    after IO devices managed by that unit.
54  * 3) Hotplug events are rare.
55  *
56  * Locking rules for DMA and interrupt remapping related global data structures:
57  * 1) Use dmar_global_lock in process context
58  * 2) Use RCU in interrupt context
59  */
60 DECLARE_RWSEM(dmar_global_lock);
61 LIST_HEAD(dmar_drhd_units);
62 
63 struct acpi_table_header * __initdata dmar_tbl;
64 static int dmar_dev_scope_status = 1;
65 static DEFINE_IDA(dmar_seq_ids);
66 
67 static int alloc_iommu(struct dmar_drhd_unit *drhd);
68 static void free_iommu(struct intel_iommu *iommu);
69 
70 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
71 {
72 	/*
73 	 * add INCLUDE_ALL at the tail, so scan the list will find it at
74 	 * the very end.
75 	 */
76 	if (drhd->include_all)
77 		list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
78 	else
79 		list_add_rcu(&drhd->list, &dmar_drhd_units);
80 }
81 
82 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
83 {
84 	struct acpi_dmar_device_scope *scope;
85 
86 	*cnt = 0;
87 	while (start < end) {
88 		scope = start;
89 		if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
90 		    scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
91 		    scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
92 			(*cnt)++;
93 		else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
94 			scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
95 			pr_warn("Unsupported device scope\n");
96 		}
97 		start += scope->length;
98 	}
99 	if (*cnt == 0)
100 		return NULL;
101 
102 	return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
103 }
104 
105 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
106 {
107 	int i;
108 	struct device *tmp_dev;
109 
110 	if (*devices && *cnt) {
111 		for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
112 			put_device(tmp_dev);
113 		kfree(*devices);
114 	}
115 
116 	*devices = NULL;
117 	*cnt = 0;
118 }
119 
120 /* Optimize out kzalloc()/kfree() for normal cases */
121 static char dmar_pci_notify_info_buf[64];
122 
123 static struct dmar_pci_notify_info *
124 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
125 {
126 	int level = 0;
127 	size_t size;
128 	struct pci_dev *tmp;
129 	struct dmar_pci_notify_info *info;
130 
131 	/*
132 	 * Ignore devices that have a domain number higher than what can
133 	 * be looked up in DMAR, e.g. VMD subdevices with domain 0x10000
134 	 */
135 	if (pci_domain_nr(dev->bus) > U16_MAX)
136 		return NULL;
137 
138 	/* Only generate path[] for device addition event */
139 	if (event == BUS_NOTIFY_ADD_DEVICE)
140 		for (tmp = dev; tmp; tmp = tmp->bus->self)
141 			level++;
142 
143 	size = struct_size(info, path, level);
144 	if (size <= sizeof(dmar_pci_notify_info_buf)) {
145 		info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
146 	} else {
147 		info = kzalloc(size, GFP_KERNEL);
148 		if (!info) {
149 			if (dmar_dev_scope_status == 0)
150 				dmar_dev_scope_status = -ENOMEM;
151 			return NULL;
152 		}
153 	}
154 
155 	info->event = event;
156 	info->dev = dev;
157 	info->seg = pci_domain_nr(dev->bus);
158 	info->level = level;
159 	if (event == BUS_NOTIFY_ADD_DEVICE) {
160 		for (tmp = dev; tmp; tmp = tmp->bus->self) {
161 			level--;
162 			info->path[level].bus = tmp->bus->number;
163 			info->path[level].device = PCI_SLOT(tmp->devfn);
164 			info->path[level].function = PCI_FUNC(tmp->devfn);
165 			if (pci_is_root_bus(tmp->bus))
166 				info->bus = tmp->bus->number;
167 		}
168 	}
169 
170 	return info;
171 }
172 
173 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
174 {
175 	if ((void *)info != dmar_pci_notify_info_buf)
176 		kfree(info);
177 }
178 
179 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
180 				struct acpi_dmar_pci_path *path, int count)
181 {
182 	int i;
183 
184 	if (info->bus != bus)
185 		goto fallback;
186 	if (info->level != count)
187 		goto fallback;
188 
189 	for (i = 0; i < count; i++) {
190 		if (path[i].device != info->path[i].device ||
191 		    path[i].function != info->path[i].function)
192 			goto fallback;
193 	}
194 
195 	return true;
196 
197 fallback:
198 
199 	if (count != 1)
200 		return false;
201 
202 	i = info->level - 1;
203 	if (bus              == info->path[i].bus &&
204 	    path[0].device   == info->path[i].device &&
205 	    path[0].function == info->path[i].function) {
206 		pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
207 			bus, path[0].device, path[0].function);
208 		return true;
209 	}
210 
211 	return false;
212 }
213 
214 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
215 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
216 			  void *start, void*end, u16 segment,
217 			  struct dmar_dev_scope *devices,
218 			  int devices_cnt)
219 {
220 	int i, level;
221 	struct device *tmp, *dev = &info->dev->dev;
222 	struct acpi_dmar_device_scope *scope;
223 	struct acpi_dmar_pci_path *path;
224 
225 	if (segment != info->seg)
226 		return 0;
227 
228 	for (; start < end; start += scope->length) {
229 		scope = start;
230 		if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
231 		    scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
232 			continue;
233 
234 		path = (struct acpi_dmar_pci_path *)(scope + 1);
235 		level = (scope->length - sizeof(*scope)) / sizeof(*path);
236 		if (!dmar_match_pci_path(info, scope->bus, path, level))
237 			continue;
238 
239 		/*
240 		 * We expect devices with endpoint scope to have normal PCI
241 		 * headers, and devices with bridge scope to have bridge PCI
242 		 * headers.  However PCI NTB devices may be listed in the
243 		 * DMAR table with bridge scope, even though they have a
244 		 * normal PCI header.  NTB devices are identified by class
245 		 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
246 		 * for this special case.
247 		 */
248 		if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
249 		     info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
250 		    (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
251 		     (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
252 		      info->dev->class >> 16 != PCI_BASE_CLASS_BRIDGE))) {
253 			pr_warn("Device scope type does not match for %s\n",
254 				pci_name(info->dev));
255 			return -EINVAL;
256 		}
257 
258 		for_each_dev_scope(devices, devices_cnt, i, tmp)
259 			if (tmp == NULL) {
260 				devices[i].bus = info->dev->bus->number;
261 				devices[i].devfn = info->dev->devfn;
262 				rcu_assign_pointer(devices[i].dev,
263 						   get_device(dev));
264 				return 1;
265 			}
266 		if (WARN_ON(i >= devices_cnt))
267 			return -EINVAL;
268 	}
269 
270 	return 0;
271 }
272 
273 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
274 			  struct dmar_dev_scope *devices, int count)
275 {
276 	int index;
277 	struct device *tmp;
278 
279 	if (info->seg != segment)
280 		return 0;
281 
282 	for_each_active_dev_scope(devices, count, index, tmp)
283 		if (tmp == &info->dev->dev) {
284 			RCU_INIT_POINTER(devices[index].dev, NULL);
285 			synchronize_rcu();
286 			put_device(tmp);
287 			return 1;
288 		}
289 
290 	return 0;
291 }
292 
293 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
294 {
295 	int ret = 0;
296 	struct dmar_drhd_unit *dmaru;
297 	struct acpi_dmar_hardware_unit *drhd;
298 
299 	for_each_drhd_unit(dmaru) {
300 		if (dmaru->include_all)
301 			continue;
302 
303 		drhd = container_of(dmaru->hdr,
304 				    struct acpi_dmar_hardware_unit, header);
305 		ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
306 				((void *)drhd) + drhd->header.length,
307 				dmaru->segment,
308 				dmaru->devices, dmaru->devices_cnt);
309 		if (ret)
310 			break;
311 	}
312 	if (ret >= 0)
313 		ret = dmar_iommu_notify_scope_dev(info);
314 	if (ret < 0 && dmar_dev_scope_status == 0)
315 		dmar_dev_scope_status = ret;
316 
317 	if (ret >= 0)
318 		intel_irq_remap_add_device(info);
319 
320 	return ret;
321 }
322 
323 static void  dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
324 {
325 	struct dmar_drhd_unit *dmaru;
326 
327 	for_each_drhd_unit(dmaru)
328 		if (dmar_remove_dev_scope(info, dmaru->segment,
329 			dmaru->devices, dmaru->devices_cnt))
330 			break;
331 	dmar_iommu_notify_scope_dev(info);
332 }
333 
334 static inline void vf_inherit_msi_domain(struct pci_dev *pdev)
335 {
336 	struct pci_dev *physfn = pci_physfn(pdev);
337 
338 	dev_set_msi_domain(&pdev->dev, dev_get_msi_domain(&physfn->dev));
339 }
340 
341 static int dmar_pci_bus_notifier(struct notifier_block *nb,
342 				 unsigned long action, void *data)
343 {
344 	struct pci_dev *pdev = to_pci_dev(data);
345 	struct dmar_pci_notify_info *info;
346 
347 	/* Only care about add/remove events for physical functions.
348 	 * For VFs we actually do the lookup based on the corresponding
349 	 * PF in device_to_iommu() anyway. */
350 	if (pdev->is_virtfn) {
351 		/*
352 		 * Ensure that the VF device inherits the irq domain of the
353 		 * PF device. Ideally the device would inherit the domain
354 		 * from the bus, but DMAR can have multiple units per bus
355 		 * which makes this impossible. The VF 'bus' could inherit
356 		 * from the PF device, but that's yet another x86'sism to
357 		 * inflict on everybody else.
358 		 */
359 		if (action == BUS_NOTIFY_ADD_DEVICE)
360 			vf_inherit_msi_domain(pdev);
361 		return NOTIFY_DONE;
362 	}
363 
364 	if (action != BUS_NOTIFY_ADD_DEVICE &&
365 	    action != BUS_NOTIFY_REMOVED_DEVICE)
366 		return NOTIFY_DONE;
367 
368 	info = dmar_alloc_pci_notify_info(pdev, action);
369 	if (!info)
370 		return NOTIFY_DONE;
371 
372 	down_write(&dmar_global_lock);
373 	if (action == BUS_NOTIFY_ADD_DEVICE)
374 		dmar_pci_bus_add_dev(info);
375 	else if (action == BUS_NOTIFY_REMOVED_DEVICE)
376 		dmar_pci_bus_del_dev(info);
377 	up_write(&dmar_global_lock);
378 
379 	dmar_free_pci_notify_info(info);
380 
381 	return NOTIFY_OK;
382 }
383 
384 static struct notifier_block dmar_pci_bus_nb = {
385 	.notifier_call = dmar_pci_bus_notifier,
386 	.priority = 1,
387 };
388 
389 static struct dmar_drhd_unit *
390 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
391 {
392 	struct dmar_drhd_unit *dmaru;
393 
394 	list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list,
395 				dmar_rcu_check())
396 		if (dmaru->segment == drhd->segment &&
397 		    dmaru->reg_base_addr == drhd->address)
398 			return dmaru;
399 
400 	return NULL;
401 }
402 
403 /*
404  * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
405  * structure which uniquely represent one DMA remapping hardware unit
406  * present in the platform
407  */
408 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
409 {
410 	struct acpi_dmar_hardware_unit *drhd;
411 	struct dmar_drhd_unit *dmaru;
412 	int ret;
413 
414 	drhd = (struct acpi_dmar_hardware_unit *)header;
415 	dmaru = dmar_find_dmaru(drhd);
416 	if (dmaru)
417 		goto out;
418 
419 	dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
420 	if (!dmaru)
421 		return -ENOMEM;
422 
423 	/*
424 	 * If header is allocated from slab by ACPI _DSM method, we need to
425 	 * copy the content because the memory buffer will be freed on return.
426 	 */
427 	dmaru->hdr = (void *)(dmaru + 1);
428 	memcpy(dmaru->hdr, header, header->length);
429 	dmaru->reg_base_addr = drhd->address;
430 	dmaru->segment = drhd->segment;
431 	/* The size of the register set is 2 ^ N 4 KB pages. */
432 	dmaru->reg_size = 1UL << (drhd->size + 12);
433 	dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
434 	dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
435 					      ((void *)drhd) + drhd->header.length,
436 					      &dmaru->devices_cnt);
437 	if (dmaru->devices_cnt && dmaru->devices == NULL) {
438 		kfree(dmaru);
439 		return -ENOMEM;
440 	}
441 
442 	ret = alloc_iommu(dmaru);
443 	if (ret) {
444 		dmar_free_dev_scope(&dmaru->devices,
445 				    &dmaru->devices_cnt);
446 		kfree(dmaru);
447 		return ret;
448 	}
449 	dmar_register_drhd_unit(dmaru);
450 
451 out:
452 	if (arg)
453 		(*(int *)arg)++;
454 
455 	return 0;
456 }
457 
458 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
459 {
460 	if (dmaru->devices && dmaru->devices_cnt)
461 		dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
462 	if (dmaru->iommu)
463 		free_iommu(dmaru->iommu);
464 	kfree(dmaru);
465 }
466 
467 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
468 				      void *arg)
469 {
470 	struct acpi_dmar_andd *andd = (void *)header;
471 
472 	/* Check for NUL termination within the designated length */
473 	if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
474 		pr_warn(FW_BUG
475 			   "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
476 			   "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
477 			   dmi_get_system_info(DMI_BIOS_VENDOR),
478 			   dmi_get_system_info(DMI_BIOS_VERSION),
479 			   dmi_get_system_info(DMI_PRODUCT_VERSION));
480 		add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
481 		return -EINVAL;
482 	}
483 	pr_info("ANDD device: %x name: %s\n", andd->device_number,
484 		andd->device_name);
485 
486 	return 0;
487 }
488 
489 #ifdef CONFIG_ACPI_NUMA
490 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
491 {
492 	struct acpi_dmar_rhsa *rhsa;
493 	struct dmar_drhd_unit *drhd;
494 
495 	rhsa = (struct acpi_dmar_rhsa *)header;
496 	for_each_drhd_unit(drhd) {
497 		if (drhd->reg_base_addr == rhsa->base_address) {
498 			int node = pxm_to_node(rhsa->proximity_domain);
499 
500 			if (node != NUMA_NO_NODE && !node_online(node))
501 				node = NUMA_NO_NODE;
502 			drhd->iommu->node = node;
503 			return 0;
504 		}
505 	}
506 	pr_warn(FW_BUG
507 		"Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
508 		"BIOS vendor: %s; Ver: %s; Product Version: %s\n",
509 		rhsa->base_address,
510 		dmi_get_system_info(DMI_BIOS_VENDOR),
511 		dmi_get_system_info(DMI_BIOS_VERSION),
512 		dmi_get_system_info(DMI_PRODUCT_VERSION));
513 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
514 
515 	return 0;
516 }
517 #else
518 #define	dmar_parse_one_rhsa		dmar_res_noop
519 #endif
520 
521 static void
522 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
523 {
524 	struct acpi_dmar_hardware_unit *drhd;
525 	struct acpi_dmar_reserved_memory *rmrr;
526 	struct acpi_dmar_atsr *atsr;
527 	struct acpi_dmar_rhsa *rhsa;
528 	struct acpi_dmar_satc *satc;
529 
530 	switch (header->type) {
531 	case ACPI_DMAR_TYPE_HARDWARE_UNIT:
532 		drhd = container_of(header, struct acpi_dmar_hardware_unit,
533 				    header);
534 		pr_info("DRHD base: %#016Lx flags: %#x\n",
535 			(unsigned long long)drhd->address, drhd->flags);
536 		break;
537 	case ACPI_DMAR_TYPE_RESERVED_MEMORY:
538 		rmrr = container_of(header, struct acpi_dmar_reserved_memory,
539 				    header);
540 		pr_info("RMRR base: %#016Lx end: %#016Lx\n",
541 			(unsigned long long)rmrr->base_address,
542 			(unsigned long long)rmrr->end_address);
543 		break;
544 	case ACPI_DMAR_TYPE_ROOT_ATS:
545 		atsr = container_of(header, struct acpi_dmar_atsr, header);
546 		pr_info("ATSR flags: %#x\n", atsr->flags);
547 		break;
548 	case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
549 		rhsa = container_of(header, struct acpi_dmar_rhsa, header);
550 		pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
551 		       (unsigned long long)rhsa->base_address,
552 		       rhsa->proximity_domain);
553 		break;
554 	case ACPI_DMAR_TYPE_NAMESPACE:
555 		/* We don't print this here because we need to sanity-check
556 		   it first. So print it in dmar_parse_one_andd() instead. */
557 		break;
558 	case ACPI_DMAR_TYPE_SATC:
559 		satc = container_of(header, struct acpi_dmar_satc, header);
560 		pr_info("SATC flags: 0x%x\n", satc->flags);
561 		break;
562 	}
563 }
564 
565 /**
566  * dmar_table_detect - checks to see if the platform supports DMAR devices
567  */
568 static int __init dmar_table_detect(void)
569 {
570 	acpi_status status = AE_OK;
571 
572 	/* if we could find DMAR table, then there are DMAR devices */
573 	status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
574 
575 	if (ACPI_SUCCESS(status) && !dmar_tbl) {
576 		pr_warn("Unable to map DMAR\n");
577 		status = AE_NOT_FOUND;
578 	}
579 
580 	return ACPI_SUCCESS(status) ? 0 : -ENOENT;
581 }
582 
583 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
584 				       size_t len, struct dmar_res_callback *cb)
585 {
586 	struct acpi_dmar_header *iter, *next;
587 	struct acpi_dmar_header *end = ((void *)start) + len;
588 
589 	for (iter = start; iter < end; iter = next) {
590 		next = (void *)iter + iter->length;
591 		if (iter->length == 0) {
592 			/* Avoid looping forever on bad ACPI tables */
593 			pr_debug(FW_BUG "Invalid 0-length structure\n");
594 			break;
595 		} else if (next > end) {
596 			/* Avoid passing table end */
597 			pr_warn(FW_BUG "Record passes table end\n");
598 			return -EINVAL;
599 		}
600 
601 		if (cb->print_entry)
602 			dmar_table_print_dmar_entry(iter);
603 
604 		if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
605 			/* continue for forward compatibility */
606 			pr_debug("Unknown DMAR structure type %d\n",
607 				 iter->type);
608 		} else if (cb->cb[iter->type]) {
609 			int ret;
610 
611 			ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
612 			if (ret)
613 				return ret;
614 		} else if (!cb->ignore_unhandled) {
615 			pr_warn("No handler for DMAR structure type %d\n",
616 				iter->type);
617 			return -EINVAL;
618 		}
619 	}
620 
621 	return 0;
622 }
623 
624 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
625 				       struct dmar_res_callback *cb)
626 {
627 	return dmar_walk_remapping_entries((void *)(dmar + 1),
628 			dmar->header.length - sizeof(*dmar), cb);
629 }
630 
631 /**
632  * parse_dmar_table - parses the DMA reporting table
633  */
634 static int __init
635 parse_dmar_table(void)
636 {
637 	struct acpi_table_dmar *dmar;
638 	int drhd_count = 0;
639 	int ret;
640 	struct dmar_res_callback cb = {
641 		.print_entry = true,
642 		.ignore_unhandled = true,
643 		.arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
644 		.cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
645 		.cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
646 		.cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
647 		.cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
648 		.cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
649 		.cb[ACPI_DMAR_TYPE_SATC] = &dmar_parse_one_satc,
650 	};
651 
652 	/*
653 	 * Do it again, earlier dmar_tbl mapping could be mapped with
654 	 * fixed map.
655 	 */
656 	dmar_table_detect();
657 
658 	/*
659 	 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
660 	 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
661 	 */
662 	dmar_tbl = tboot_get_dmar_table(dmar_tbl);
663 
664 	dmar = (struct acpi_table_dmar *)dmar_tbl;
665 	if (!dmar)
666 		return -ENODEV;
667 
668 	if (dmar->width < PAGE_SHIFT - 1) {
669 		pr_warn("Invalid DMAR haw\n");
670 		return -EINVAL;
671 	}
672 
673 	pr_info("Host address width %d\n", dmar->width + 1);
674 	ret = dmar_walk_dmar_table(dmar, &cb);
675 	if (ret == 0 && drhd_count == 0)
676 		pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
677 
678 	return ret;
679 }
680 
681 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
682 				 int cnt, struct pci_dev *dev)
683 {
684 	int index;
685 	struct device *tmp;
686 
687 	while (dev) {
688 		for_each_active_dev_scope(devices, cnt, index, tmp)
689 			if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
690 				return 1;
691 
692 		/* Check our parent */
693 		dev = dev->bus->self;
694 	}
695 
696 	return 0;
697 }
698 
699 struct dmar_drhd_unit *
700 dmar_find_matched_drhd_unit(struct pci_dev *dev)
701 {
702 	struct dmar_drhd_unit *dmaru;
703 	struct acpi_dmar_hardware_unit *drhd;
704 
705 	dev = pci_physfn(dev);
706 
707 	rcu_read_lock();
708 	for_each_drhd_unit(dmaru) {
709 		drhd = container_of(dmaru->hdr,
710 				    struct acpi_dmar_hardware_unit,
711 				    header);
712 
713 		if (dmaru->include_all &&
714 		    drhd->segment == pci_domain_nr(dev->bus))
715 			goto out;
716 
717 		if (dmar_pci_device_match(dmaru->devices,
718 					  dmaru->devices_cnt, dev))
719 			goto out;
720 	}
721 	dmaru = NULL;
722 out:
723 	rcu_read_unlock();
724 
725 	return dmaru;
726 }
727 
728 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
729 					      struct acpi_device *adev)
730 {
731 	struct dmar_drhd_unit *dmaru;
732 	struct acpi_dmar_hardware_unit *drhd;
733 	struct acpi_dmar_device_scope *scope;
734 	struct device *tmp;
735 	int i;
736 	struct acpi_dmar_pci_path *path;
737 
738 	for_each_drhd_unit(dmaru) {
739 		drhd = container_of(dmaru->hdr,
740 				    struct acpi_dmar_hardware_unit,
741 				    header);
742 
743 		for (scope = (void *)(drhd + 1);
744 		     (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
745 		     scope = ((void *)scope) + scope->length) {
746 			if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
747 				continue;
748 			if (scope->enumeration_id != device_number)
749 				continue;
750 
751 			path = (void *)(scope + 1);
752 			pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
753 				dev_name(&adev->dev), dmaru->reg_base_addr,
754 				scope->bus, path->device, path->function);
755 			for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
756 				if (tmp == NULL) {
757 					dmaru->devices[i].bus = scope->bus;
758 					dmaru->devices[i].devfn = PCI_DEVFN(path->device,
759 									    path->function);
760 					rcu_assign_pointer(dmaru->devices[i].dev,
761 							   get_device(&adev->dev));
762 					return;
763 				}
764 			BUG_ON(i >= dmaru->devices_cnt);
765 		}
766 	}
767 	pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
768 		device_number, dev_name(&adev->dev));
769 }
770 
771 static int __init dmar_acpi_dev_scope_init(void)
772 {
773 	struct acpi_dmar_andd *andd;
774 
775 	if (dmar_tbl == NULL)
776 		return -ENODEV;
777 
778 	for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
779 	     ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
780 	     andd = ((void *)andd) + andd->header.length) {
781 		if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
782 			acpi_handle h;
783 			struct acpi_device *adev;
784 
785 			if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
786 							  andd->device_name,
787 							  &h))) {
788 				pr_err("Failed to find handle for ACPI object %s\n",
789 				       andd->device_name);
790 				continue;
791 			}
792 			adev = acpi_fetch_acpi_dev(h);
793 			if (!adev) {
794 				pr_err("Failed to get device for ACPI object %s\n",
795 				       andd->device_name);
796 				continue;
797 			}
798 			dmar_acpi_insert_dev_scope(andd->device_number, adev);
799 		}
800 	}
801 	return 0;
802 }
803 
804 int __init dmar_dev_scope_init(void)
805 {
806 	struct pci_dev *dev = NULL;
807 	struct dmar_pci_notify_info *info;
808 
809 	if (dmar_dev_scope_status != 1)
810 		return dmar_dev_scope_status;
811 
812 	if (list_empty(&dmar_drhd_units)) {
813 		dmar_dev_scope_status = -ENODEV;
814 	} else {
815 		dmar_dev_scope_status = 0;
816 
817 		dmar_acpi_dev_scope_init();
818 
819 		for_each_pci_dev(dev) {
820 			if (dev->is_virtfn)
821 				continue;
822 
823 			info = dmar_alloc_pci_notify_info(dev,
824 					BUS_NOTIFY_ADD_DEVICE);
825 			if (!info) {
826 				pci_dev_put(dev);
827 				return dmar_dev_scope_status;
828 			} else {
829 				dmar_pci_bus_add_dev(info);
830 				dmar_free_pci_notify_info(info);
831 			}
832 		}
833 	}
834 
835 	return dmar_dev_scope_status;
836 }
837 
838 void __init dmar_register_bus_notifier(void)
839 {
840 	bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
841 }
842 
843 
844 int __init dmar_table_init(void)
845 {
846 	static int dmar_table_initialized;
847 	int ret;
848 
849 	if (dmar_table_initialized == 0) {
850 		ret = parse_dmar_table();
851 		if (ret < 0) {
852 			if (ret != -ENODEV)
853 				pr_info("Parse DMAR table failure.\n");
854 		} else  if (list_empty(&dmar_drhd_units)) {
855 			pr_info("No DMAR devices found\n");
856 			ret = -ENODEV;
857 		}
858 
859 		if (ret < 0)
860 			dmar_table_initialized = ret;
861 		else
862 			dmar_table_initialized = 1;
863 	}
864 
865 	return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
866 }
867 
868 static void warn_invalid_dmar(u64 addr, const char *message)
869 {
870 	pr_warn_once(FW_BUG
871 		"Your BIOS is broken; DMAR reported at address %llx%s!\n"
872 		"BIOS vendor: %s; Ver: %s; Product Version: %s\n",
873 		addr, message,
874 		dmi_get_system_info(DMI_BIOS_VENDOR),
875 		dmi_get_system_info(DMI_BIOS_VERSION),
876 		dmi_get_system_info(DMI_PRODUCT_VERSION));
877 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
878 }
879 
880 static int __ref
881 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
882 {
883 	struct acpi_dmar_hardware_unit *drhd;
884 	void __iomem *addr;
885 	u64 cap, ecap;
886 
887 	drhd = (void *)entry;
888 	if (!drhd->address) {
889 		warn_invalid_dmar(0, "");
890 		return -EINVAL;
891 	}
892 
893 	if (arg)
894 		addr = ioremap(drhd->address, VTD_PAGE_SIZE);
895 	else
896 		addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
897 	if (!addr) {
898 		pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
899 		return -EINVAL;
900 	}
901 
902 	cap = dmar_readq(addr + DMAR_CAP_REG);
903 	ecap = dmar_readq(addr + DMAR_ECAP_REG);
904 
905 	if (arg)
906 		iounmap(addr);
907 	else
908 		early_iounmap(addr, VTD_PAGE_SIZE);
909 
910 	if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
911 		warn_invalid_dmar(drhd->address, " returns all ones");
912 		return -EINVAL;
913 	}
914 
915 	return 0;
916 }
917 
918 void __init detect_intel_iommu(void)
919 {
920 	int ret;
921 	struct dmar_res_callback validate_drhd_cb = {
922 		.cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
923 		.ignore_unhandled = true,
924 	};
925 
926 	down_write(&dmar_global_lock);
927 	ret = dmar_table_detect();
928 	if (!ret)
929 		ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
930 					   &validate_drhd_cb);
931 	if (!ret && !no_iommu && !iommu_detected &&
932 	    (!dmar_disabled || dmar_platform_optin())) {
933 		iommu_detected = 1;
934 		/* Make sure ACS will be enabled */
935 		pci_request_acs();
936 	}
937 
938 #ifdef CONFIG_X86
939 	if (!ret) {
940 		x86_init.iommu.iommu_init = intel_iommu_init;
941 		x86_platform.iommu_shutdown = intel_iommu_shutdown;
942 	}
943 
944 #endif
945 
946 	if (dmar_tbl) {
947 		acpi_put_table(dmar_tbl);
948 		dmar_tbl = NULL;
949 	}
950 	up_write(&dmar_global_lock);
951 }
952 
953 static void unmap_iommu(struct intel_iommu *iommu)
954 {
955 	iounmap(iommu->reg);
956 	release_mem_region(iommu->reg_phys, iommu->reg_size);
957 }
958 
959 /**
960  * map_iommu: map the iommu's registers
961  * @iommu: the iommu to map
962  * @drhd: DMA remapping hardware definition structure
963  *
964  * Memory map the iommu's registers.  Start w/ a single page, and
965  * possibly expand if that turns out to be insufficent.
966  */
967 static int map_iommu(struct intel_iommu *iommu, struct dmar_drhd_unit *drhd)
968 {
969 	u64 phys_addr = drhd->reg_base_addr;
970 	int map_size, err=0;
971 
972 	iommu->reg_phys = phys_addr;
973 	iommu->reg_size = drhd->reg_size;
974 
975 	if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
976 		pr_err("Can't reserve memory\n");
977 		err = -EBUSY;
978 		goto out;
979 	}
980 
981 	iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
982 	if (!iommu->reg) {
983 		pr_err("Can't map the region\n");
984 		err = -ENOMEM;
985 		goto release;
986 	}
987 
988 	iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
989 	iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
990 
991 	if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
992 		err = -EINVAL;
993 		warn_invalid_dmar(phys_addr, " returns all ones");
994 		goto unmap;
995 	}
996 
997 	/* the registers might be more than one page */
998 	map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
999 			 cap_max_fault_reg_offset(iommu->cap));
1000 	map_size = VTD_PAGE_ALIGN(map_size);
1001 	if (map_size > iommu->reg_size) {
1002 		iounmap(iommu->reg);
1003 		release_mem_region(iommu->reg_phys, iommu->reg_size);
1004 		iommu->reg_size = map_size;
1005 		if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
1006 					iommu->name)) {
1007 			pr_err("Can't reserve memory\n");
1008 			err = -EBUSY;
1009 			goto out;
1010 		}
1011 		iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
1012 		if (!iommu->reg) {
1013 			pr_err("Can't map the region\n");
1014 			err = -ENOMEM;
1015 			goto release;
1016 		}
1017 	}
1018 
1019 	if (cap_ecmds(iommu->cap)) {
1020 		int i;
1021 
1022 		for (i = 0; i < DMA_MAX_NUM_ECMDCAP; i++) {
1023 			iommu->ecmdcap[i] = dmar_readq(iommu->reg + DMAR_ECCAP_REG +
1024 						       i * DMA_ECMD_REG_STEP);
1025 		}
1026 	}
1027 
1028 	err = 0;
1029 	goto out;
1030 
1031 unmap:
1032 	iounmap(iommu->reg);
1033 release:
1034 	release_mem_region(iommu->reg_phys, iommu->reg_size);
1035 out:
1036 	return err;
1037 }
1038 
1039 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1040 {
1041 	struct intel_iommu *iommu;
1042 	u32 ver, sts;
1043 	int agaw = -1;
1044 	int msagaw = -1;
1045 	int err;
1046 
1047 	if (!drhd->reg_base_addr) {
1048 		warn_invalid_dmar(0, "");
1049 		return -EINVAL;
1050 	}
1051 
1052 	iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1053 	if (!iommu)
1054 		return -ENOMEM;
1055 
1056 	iommu->seq_id = ida_alloc_range(&dmar_seq_ids, 0,
1057 					DMAR_UNITS_SUPPORTED - 1, GFP_KERNEL);
1058 	if (iommu->seq_id < 0) {
1059 		pr_err("Failed to allocate seq_id\n");
1060 		err = iommu->seq_id;
1061 		goto error;
1062 	}
1063 	sprintf(iommu->name, "dmar%d", iommu->seq_id);
1064 
1065 	err = map_iommu(iommu, drhd);
1066 	if (err) {
1067 		pr_err("Failed to map %s\n", iommu->name);
1068 		goto error_free_seq_id;
1069 	}
1070 
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 		iommu_free_page(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 	void *desc;
1759 	int order;
1760 
1761 	if (!ecap_qis(iommu->ecap))
1762 		return -ENOENT;
1763 
1764 	/*
1765 	 * queued invalidation is already setup and enabled.
1766 	 */
1767 	if (iommu->qi)
1768 		return 0;
1769 
1770 	iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1771 	if (!iommu->qi)
1772 		return -ENOMEM;
1773 
1774 	qi = iommu->qi;
1775 
1776 	/*
1777 	 * Need two pages to accommodate 256 descriptors of 256 bits each
1778 	 * if the remapping hardware supports scalable mode translation.
1779 	 */
1780 	order = ecap_smts(iommu->ecap) ? 1 : 0;
1781 	desc = iommu_alloc_pages_node(iommu->node, GFP_ATOMIC, order);
1782 	if (!desc) {
1783 		kfree(qi);
1784 		iommu->qi = NULL;
1785 		return -ENOMEM;
1786 	}
1787 
1788 	qi->desc = desc;
1789 
1790 	qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC);
1791 	if (!qi->desc_status) {
1792 		iommu_free_page(qi->desc);
1793 		kfree(qi);
1794 		iommu->qi = NULL;
1795 		return -ENOMEM;
1796 	}
1797 
1798 	raw_spin_lock_init(&qi->q_lock);
1799 
1800 	__dmar_enable_qi(iommu);
1801 
1802 	return 0;
1803 }
1804 
1805 /* iommu interrupt handling. Most stuff are MSI-like. */
1806 
1807 enum faulttype {
1808 	DMA_REMAP,
1809 	INTR_REMAP,
1810 	UNKNOWN,
1811 };
1812 
1813 static const char *dma_remap_fault_reasons[] =
1814 {
1815 	"Software",
1816 	"Present bit in root entry is clear",
1817 	"Present bit in context entry is clear",
1818 	"Invalid context entry",
1819 	"Access beyond MGAW",
1820 	"PTE Write access is not set",
1821 	"PTE Read access is not set",
1822 	"Next page table ptr is invalid",
1823 	"Root table address invalid",
1824 	"Context table ptr is invalid",
1825 	"non-zero reserved fields in RTP",
1826 	"non-zero reserved fields in CTP",
1827 	"non-zero reserved fields in PTE",
1828 	"PCE for translation request specifies blocking",
1829 };
1830 
1831 static const char * const dma_remap_sm_fault_reasons[] = {
1832 	"SM: Invalid Root Table Address",
1833 	"SM: TTM 0 for request with PASID",
1834 	"SM: TTM 0 for page group request",
1835 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x33-0x37 */
1836 	"SM: Error attempting to access Root Entry",
1837 	"SM: Present bit in Root Entry is clear",
1838 	"SM: Non-zero reserved field set in Root Entry",
1839 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x3B-0x3F */
1840 	"SM: Error attempting to access Context Entry",
1841 	"SM: Present bit in Context Entry is clear",
1842 	"SM: Non-zero reserved field set in the Context Entry",
1843 	"SM: Invalid Context Entry",
1844 	"SM: DTE field in Context Entry is clear",
1845 	"SM: PASID Enable field in Context Entry is clear",
1846 	"SM: PASID is larger than the max in Context Entry",
1847 	"SM: PRE field in Context-Entry is clear",
1848 	"SM: RID_PASID field error in Context-Entry",
1849 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x49-0x4F */
1850 	"SM: Error attempting to access the PASID Directory Entry",
1851 	"SM: Present bit in Directory Entry is clear",
1852 	"SM: Non-zero reserved field set in PASID Directory Entry",
1853 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x53-0x57 */
1854 	"SM: Error attempting to access PASID Table Entry",
1855 	"SM: Present bit in PASID Table Entry is clear",
1856 	"SM: Non-zero reserved field set in PASID Table Entry",
1857 	"SM: Invalid Scalable-Mode PASID Table Entry",
1858 	"SM: ERE field is clear in PASID Table Entry",
1859 	"SM: SRE field is clear in PASID Table Entry",
1860 	"Unknown", "Unknown",/* 0x5E-0x5F */
1861 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x60-0x67 */
1862 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x68-0x6F */
1863 	"SM: Error attempting to access first-level paging entry",
1864 	"SM: Present bit in first-level paging entry is clear",
1865 	"SM: Non-zero reserved field set in first-level paging entry",
1866 	"SM: Error attempting to access FL-PML4 entry",
1867 	"SM: First-level entry address beyond MGAW in Nested translation",
1868 	"SM: Read permission error in FL-PML4 entry in Nested translation",
1869 	"SM: Read permission error in first-level paging entry in Nested translation",
1870 	"SM: Write permission error in first-level paging entry in Nested translation",
1871 	"SM: Error attempting to access second-level paging entry",
1872 	"SM: Read/Write permission error in second-level paging entry",
1873 	"SM: Non-zero reserved field set in second-level paging entry",
1874 	"SM: Invalid second-level page table pointer",
1875 	"SM: A/D bit update needed in second-level entry when set up in no snoop",
1876 	"Unknown", "Unknown", "Unknown", /* 0x7D-0x7F */
1877 	"SM: Address in first-level translation is not canonical",
1878 	"SM: U/S set 0 for first-level translation with user privilege",
1879 	"SM: No execute permission for request with PASID and ER=1",
1880 	"SM: Address beyond the DMA hardware max",
1881 	"SM: Second-level entry address beyond the max",
1882 	"SM: No write permission for Write/AtomicOp request",
1883 	"SM: No read permission for Read/AtomicOp request",
1884 	"SM: Invalid address-interrupt address",
1885 	"Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x88-0x8F */
1886 	"SM: A/D bit update needed in first-level entry when set up in no snoop",
1887 };
1888 
1889 static const char *irq_remap_fault_reasons[] =
1890 {
1891 	"Detected reserved fields in the decoded interrupt-remapped request",
1892 	"Interrupt index exceeded the interrupt-remapping table size",
1893 	"Present field in the IRTE entry is clear",
1894 	"Error accessing interrupt-remapping table pointed by IRTA_REG",
1895 	"Detected reserved fields in the IRTE entry",
1896 	"Blocked a compatibility format interrupt request",
1897 	"Blocked an interrupt request due to source-id verification failure",
1898 };
1899 
1900 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1901 {
1902 	if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1903 					ARRAY_SIZE(irq_remap_fault_reasons))) {
1904 		*fault_type = INTR_REMAP;
1905 		return irq_remap_fault_reasons[fault_reason - 0x20];
1906 	} else if (fault_reason >= 0x30 && (fault_reason - 0x30 <
1907 			ARRAY_SIZE(dma_remap_sm_fault_reasons))) {
1908 		*fault_type = DMA_REMAP;
1909 		return dma_remap_sm_fault_reasons[fault_reason - 0x30];
1910 	} else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1911 		*fault_type = DMA_REMAP;
1912 		return dma_remap_fault_reasons[fault_reason];
1913 	} else {
1914 		*fault_type = UNKNOWN;
1915 		return "Unknown";
1916 	}
1917 }
1918 
1919 
1920 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1921 {
1922 	if (iommu->irq == irq)
1923 		return DMAR_FECTL_REG;
1924 	else if (iommu->pr_irq == irq)
1925 		return DMAR_PECTL_REG;
1926 	else if (iommu->perf_irq == irq)
1927 		return DMAR_PERFINTRCTL_REG;
1928 	else
1929 		BUG();
1930 }
1931 
1932 void dmar_msi_unmask(struct irq_data *data)
1933 {
1934 	struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1935 	int reg = dmar_msi_reg(iommu, data->irq);
1936 	unsigned long flag;
1937 
1938 	/* unmask it */
1939 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1940 	writel(0, iommu->reg + reg);
1941 	/* Read a reg to force flush the post write */
1942 	readl(iommu->reg + reg);
1943 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1944 }
1945 
1946 void dmar_msi_mask(struct irq_data *data)
1947 {
1948 	struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1949 	int reg = dmar_msi_reg(iommu, data->irq);
1950 	unsigned long flag;
1951 
1952 	/* mask it */
1953 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1954 	writel(DMA_FECTL_IM, iommu->reg + reg);
1955 	/* Read a reg to force flush the post write */
1956 	readl(iommu->reg + reg);
1957 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1958 }
1959 
1960 void dmar_msi_write(int irq, struct msi_msg *msg)
1961 {
1962 	struct intel_iommu *iommu = irq_get_handler_data(irq);
1963 	int reg = dmar_msi_reg(iommu, irq);
1964 	unsigned long flag;
1965 
1966 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1967 	writel(msg->data, iommu->reg + reg + 4);
1968 	writel(msg->address_lo, iommu->reg + reg + 8);
1969 	writel(msg->address_hi, iommu->reg + reg + 12);
1970 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1971 }
1972 
1973 void dmar_msi_read(int irq, struct msi_msg *msg)
1974 {
1975 	struct intel_iommu *iommu = irq_get_handler_data(irq);
1976 	int reg = dmar_msi_reg(iommu, irq);
1977 	unsigned long flag;
1978 
1979 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
1980 	msg->data = readl(iommu->reg + reg + 4);
1981 	msg->address_lo = readl(iommu->reg + reg + 8);
1982 	msg->address_hi = readl(iommu->reg + reg + 12);
1983 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1984 }
1985 
1986 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1987 		u8 fault_reason, u32 pasid, u16 source_id,
1988 		unsigned long long addr)
1989 {
1990 	const char *reason;
1991 	int fault_type;
1992 
1993 	reason = dmar_get_fault_reason(fault_reason, &fault_type);
1994 
1995 	if (fault_type == INTR_REMAP) {
1996 		pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index 0x%llx [fault reason 0x%02x] %s\n",
1997 		       source_id >> 8, PCI_SLOT(source_id & 0xFF),
1998 		       PCI_FUNC(source_id & 0xFF), addr >> 48,
1999 		       fault_reason, reason);
2000 
2001 		return 0;
2002 	}
2003 
2004 	if (pasid == IOMMU_PASID_INVALID)
2005 		pr_err("[%s NO_PASID] Request device [%02x:%02x.%d] fault addr 0x%llx [fault reason 0x%02x] %s\n",
2006 		       type ? "DMA Read" : "DMA Write",
2007 		       source_id >> 8, PCI_SLOT(source_id & 0xFF),
2008 		       PCI_FUNC(source_id & 0xFF), addr,
2009 		       fault_reason, reason);
2010 	else
2011 		pr_err("[%s PASID 0x%x] Request device [%02x:%02x.%d] fault addr 0x%llx [fault reason 0x%02x] %s\n",
2012 		       type ? "DMA Read" : "DMA Write", pasid,
2013 		       source_id >> 8, PCI_SLOT(source_id & 0xFF),
2014 		       PCI_FUNC(source_id & 0xFF), addr,
2015 		       fault_reason, reason);
2016 
2017 	dmar_fault_dump_ptes(iommu, source_id, addr, pasid);
2018 
2019 	return 0;
2020 }
2021 
2022 #define PRIMARY_FAULT_REG_LEN (16)
2023 irqreturn_t dmar_fault(int irq, void *dev_id)
2024 {
2025 	struct intel_iommu *iommu = dev_id;
2026 	int reg, fault_index;
2027 	u32 fault_status;
2028 	unsigned long flag;
2029 	static DEFINE_RATELIMIT_STATE(rs,
2030 				      DEFAULT_RATELIMIT_INTERVAL,
2031 				      DEFAULT_RATELIMIT_BURST);
2032 
2033 	raw_spin_lock_irqsave(&iommu->register_lock, flag);
2034 	fault_status = readl(iommu->reg + DMAR_FSTS_REG);
2035 	if (fault_status && __ratelimit(&rs))
2036 		pr_err("DRHD: handling fault status reg %x\n", fault_status);
2037 
2038 	/* TBD: ignore advanced fault log currently */
2039 	if (!(fault_status & DMA_FSTS_PPF))
2040 		goto unlock_exit;
2041 
2042 	fault_index = dma_fsts_fault_record_index(fault_status);
2043 	reg = cap_fault_reg_offset(iommu->cap);
2044 	while (1) {
2045 		/* Disable printing, simply clear the fault when ratelimited */
2046 		bool ratelimited = !__ratelimit(&rs);
2047 		u8 fault_reason;
2048 		u16 source_id;
2049 		u64 guest_addr;
2050 		u32 pasid;
2051 		int type;
2052 		u32 data;
2053 		bool pasid_present;
2054 
2055 		/* highest 32 bits */
2056 		data = readl(iommu->reg + reg +
2057 				fault_index * PRIMARY_FAULT_REG_LEN + 12);
2058 		if (!(data & DMA_FRCD_F))
2059 			break;
2060 
2061 		if (!ratelimited) {
2062 			fault_reason = dma_frcd_fault_reason(data);
2063 			type = dma_frcd_type(data);
2064 
2065 			pasid = dma_frcd_pasid_value(data);
2066 			data = readl(iommu->reg + reg +
2067 				     fault_index * PRIMARY_FAULT_REG_LEN + 8);
2068 			source_id = dma_frcd_source_id(data);
2069 
2070 			pasid_present = dma_frcd_pasid_present(data);
2071 			guest_addr = dmar_readq(iommu->reg + reg +
2072 					fault_index * PRIMARY_FAULT_REG_LEN);
2073 			guest_addr = dma_frcd_page_addr(guest_addr);
2074 		}
2075 
2076 		/* clear the fault */
2077 		writel(DMA_FRCD_F, iommu->reg + reg +
2078 			fault_index * PRIMARY_FAULT_REG_LEN + 12);
2079 
2080 		raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
2081 
2082 		if (!ratelimited)
2083 			/* Using pasid -1 if pasid is not present */
2084 			dmar_fault_do_one(iommu, type, fault_reason,
2085 					  pasid_present ? pasid : IOMMU_PASID_INVALID,
2086 					  source_id, guest_addr);
2087 
2088 		fault_index++;
2089 		if (fault_index >= cap_num_fault_regs(iommu->cap))
2090 			fault_index = 0;
2091 		raw_spin_lock_irqsave(&iommu->register_lock, flag);
2092 	}
2093 
2094 	writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO,
2095 	       iommu->reg + DMAR_FSTS_REG);
2096 
2097 unlock_exit:
2098 	raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
2099 	return IRQ_HANDLED;
2100 }
2101 
2102 int dmar_set_interrupt(struct intel_iommu *iommu)
2103 {
2104 	int irq, ret;
2105 
2106 	/*
2107 	 * Check if the fault interrupt is already initialized.
2108 	 */
2109 	if (iommu->irq)
2110 		return 0;
2111 
2112 	irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
2113 	if (irq > 0) {
2114 		iommu->irq = irq;
2115 	} else {
2116 		pr_err("No free IRQ vectors\n");
2117 		return -EINVAL;
2118 	}
2119 
2120 	ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
2121 	if (ret)
2122 		pr_err("Can't request irq\n");
2123 	return ret;
2124 }
2125 
2126 int enable_drhd_fault_handling(unsigned int cpu)
2127 {
2128 	struct dmar_drhd_unit *drhd;
2129 	struct intel_iommu *iommu;
2130 
2131 	/*
2132 	 * Enable fault control interrupt.
2133 	 */
2134 	for_each_iommu(iommu, drhd) {
2135 		u32 fault_status;
2136 		int ret;
2137 
2138 		if (iommu->irq || iommu->node != cpu_to_node(cpu))
2139 			continue;
2140 
2141 		ret = dmar_set_interrupt(iommu);
2142 
2143 		if (ret) {
2144 			pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
2145 			       (unsigned long long)drhd->reg_base_addr, ret);
2146 			return -1;
2147 		}
2148 
2149 		/*
2150 		 * Clear any previous faults.
2151 		 */
2152 		dmar_fault(iommu->irq, iommu);
2153 		fault_status = readl(iommu->reg + DMAR_FSTS_REG);
2154 		writel(fault_status, iommu->reg + DMAR_FSTS_REG);
2155 	}
2156 
2157 	return 0;
2158 }
2159 
2160 /*
2161  * Re-enable Queued Invalidation interface.
2162  */
2163 int dmar_reenable_qi(struct intel_iommu *iommu)
2164 {
2165 	if (!ecap_qis(iommu->ecap))
2166 		return -ENOENT;
2167 
2168 	if (!iommu->qi)
2169 		return -ENOENT;
2170 
2171 	/*
2172 	 * First disable queued invalidation.
2173 	 */
2174 	dmar_disable_qi(iommu);
2175 	/*
2176 	 * Then enable queued invalidation again. Since there is no pending
2177 	 * invalidation requests now, it's safe to re-enable queued
2178 	 * invalidation.
2179 	 */
2180 	__dmar_enable_qi(iommu);
2181 
2182 	return 0;
2183 }
2184 
2185 /*
2186  * Check interrupt remapping support in DMAR table description.
2187  */
2188 int __init dmar_ir_support(void)
2189 {
2190 	struct acpi_table_dmar *dmar;
2191 	dmar = (struct acpi_table_dmar *)dmar_tbl;
2192 	if (!dmar)
2193 		return 0;
2194 	return dmar->flags & 0x1;
2195 }
2196 
2197 /* Check whether DMAR units are in use */
2198 static inline bool dmar_in_use(void)
2199 {
2200 	return irq_remapping_enabled || intel_iommu_enabled;
2201 }
2202 
2203 static int __init dmar_free_unused_resources(void)
2204 {
2205 	struct dmar_drhd_unit *dmaru, *dmaru_n;
2206 
2207 	if (dmar_in_use())
2208 		return 0;
2209 
2210 	if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
2211 		bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
2212 
2213 	down_write(&dmar_global_lock);
2214 	list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
2215 		list_del(&dmaru->list);
2216 		dmar_free_drhd(dmaru);
2217 	}
2218 	up_write(&dmar_global_lock);
2219 
2220 	return 0;
2221 }
2222 
2223 late_initcall(dmar_free_unused_resources);
2224 
2225 /*
2226  * DMAR Hotplug Support
2227  * For more details, please refer to Intel(R) Virtualization Technology
2228  * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
2229  * "Remapping Hardware Unit Hot Plug".
2230  */
2231 static guid_t dmar_hp_guid =
2232 	GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
2233 		  0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
2234 
2235 /*
2236  * Currently there's only one revision and BIOS will not check the revision id,
2237  * so use 0 for safety.
2238  */
2239 #define	DMAR_DSM_REV_ID			0
2240 #define	DMAR_DSM_FUNC_DRHD		1
2241 #define	DMAR_DSM_FUNC_ATSR		2
2242 #define	DMAR_DSM_FUNC_RHSA		3
2243 #define	DMAR_DSM_FUNC_SATC		4
2244 
2245 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
2246 {
2247 	return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func);
2248 }
2249 
2250 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
2251 				  dmar_res_handler_t handler, void *arg)
2252 {
2253 	int ret = -ENODEV;
2254 	union acpi_object *obj;
2255 	struct acpi_dmar_header *start;
2256 	struct dmar_res_callback callback;
2257 	static int res_type[] = {
2258 		[DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
2259 		[DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
2260 		[DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
2261 		[DMAR_DSM_FUNC_SATC] = ACPI_DMAR_TYPE_SATC,
2262 	};
2263 
2264 	if (!dmar_detect_dsm(handle, func))
2265 		return 0;
2266 
2267 	obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID,
2268 				      func, NULL, ACPI_TYPE_BUFFER);
2269 	if (!obj)
2270 		return -ENODEV;
2271 
2272 	memset(&callback, 0, sizeof(callback));
2273 	callback.cb[res_type[func]] = handler;
2274 	callback.arg[res_type[func]] = arg;
2275 	start = (struct acpi_dmar_header *)obj->buffer.pointer;
2276 	ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
2277 
2278 	ACPI_FREE(obj);
2279 
2280 	return ret;
2281 }
2282 
2283 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
2284 {
2285 	int ret;
2286 	struct dmar_drhd_unit *dmaru;
2287 
2288 	dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2289 	if (!dmaru)
2290 		return -ENODEV;
2291 
2292 	ret = dmar_ir_hotplug(dmaru, true);
2293 	if (ret == 0)
2294 		ret = dmar_iommu_hotplug(dmaru, true);
2295 
2296 	return ret;
2297 }
2298 
2299 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
2300 {
2301 	int i, ret;
2302 	struct device *dev;
2303 	struct dmar_drhd_unit *dmaru;
2304 
2305 	dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2306 	if (!dmaru)
2307 		return 0;
2308 
2309 	/*
2310 	 * All PCI devices managed by this unit should have been destroyed.
2311 	 */
2312 	if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
2313 		for_each_active_dev_scope(dmaru->devices,
2314 					  dmaru->devices_cnt, i, dev)
2315 			return -EBUSY;
2316 	}
2317 
2318 	ret = dmar_ir_hotplug(dmaru, false);
2319 	if (ret == 0)
2320 		ret = dmar_iommu_hotplug(dmaru, false);
2321 
2322 	return ret;
2323 }
2324 
2325 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
2326 {
2327 	struct dmar_drhd_unit *dmaru;
2328 
2329 	dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2330 	if (dmaru) {
2331 		list_del_rcu(&dmaru->list);
2332 		synchronize_rcu();
2333 		dmar_free_drhd(dmaru);
2334 	}
2335 
2336 	return 0;
2337 }
2338 
2339 static int dmar_hotplug_insert(acpi_handle handle)
2340 {
2341 	int ret;
2342 	int drhd_count = 0;
2343 
2344 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2345 				     &dmar_validate_one_drhd, (void *)1);
2346 	if (ret)
2347 		goto out;
2348 
2349 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2350 				     &dmar_parse_one_drhd, (void *)&drhd_count);
2351 	if (ret == 0 && drhd_count == 0) {
2352 		pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
2353 		goto out;
2354 	} else if (ret) {
2355 		goto release_drhd;
2356 	}
2357 
2358 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
2359 				     &dmar_parse_one_rhsa, NULL);
2360 	if (ret)
2361 		goto release_drhd;
2362 
2363 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2364 				     &dmar_parse_one_atsr, NULL);
2365 	if (ret)
2366 		goto release_atsr;
2367 
2368 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2369 				     &dmar_hp_add_drhd, NULL);
2370 	if (!ret)
2371 		return 0;
2372 
2373 	dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2374 			       &dmar_hp_remove_drhd, NULL);
2375 release_atsr:
2376 	dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2377 			       &dmar_release_one_atsr, NULL);
2378 release_drhd:
2379 	dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2380 			       &dmar_hp_release_drhd, NULL);
2381 out:
2382 	return ret;
2383 }
2384 
2385 static int dmar_hotplug_remove(acpi_handle handle)
2386 {
2387 	int ret;
2388 
2389 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2390 				     &dmar_check_one_atsr, NULL);
2391 	if (ret)
2392 		return ret;
2393 
2394 	ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2395 				     &dmar_hp_remove_drhd, NULL);
2396 	if (ret == 0) {
2397 		WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2398 					       &dmar_release_one_atsr, NULL));
2399 		WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2400 					       &dmar_hp_release_drhd, NULL));
2401 	} else {
2402 		dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2403 				       &dmar_hp_add_drhd, NULL);
2404 	}
2405 
2406 	return ret;
2407 }
2408 
2409 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
2410 				       void *context, void **retval)
2411 {
2412 	acpi_handle *phdl = retval;
2413 
2414 	if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2415 		*phdl = handle;
2416 		return AE_CTRL_TERMINATE;
2417 	}
2418 
2419 	return AE_OK;
2420 }
2421 
2422 static int dmar_device_hotplug(acpi_handle handle, bool insert)
2423 {
2424 	int ret;
2425 	acpi_handle tmp = NULL;
2426 	acpi_status status;
2427 
2428 	if (!dmar_in_use())
2429 		return 0;
2430 
2431 	if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2432 		tmp = handle;
2433 	} else {
2434 		status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2435 					     ACPI_UINT32_MAX,
2436 					     dmar_get_dsm_handle,
2437 					     NULL, NULL, &tmp);
2438 		if (ACPI_FAILURE(status)) {
2439 			pr_warn("Failed to locate _DSM method.\n");
2440 			return -ENXIO;
2441 		}
2442 	}
2443 	if (tmp == NULL)
2444 		return 0;
2445 
2446 	down_write(&dmar_global_lock);
2447 	if (insert)
2448 		ret = dmar_hotplug_insert(tmp);
2449 	else
2450 		ret = dmar_hotplug_remove(tmp);
2451 	up_write(&dmar_global_lock);
2452 
2453 	return ret;
2454 }
2455 
2456 int dmar_device_add(acpi_handle handle)
2457 {
2458 	return dmar_device_hotplug(handle, true);
2459 }
2460 
2461 int dmar_device_remove(acpi_handle handle)
2462 {
2463 	return dmar_device_hotplug(handle, false);
2464 }
2465 
2466 /*
2467  * dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table
2468  *
2469  * Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in
2470  * the ACPI DMAR table. This means that the platform boot firmware has made
2471  * sure no device can issue DMA outside of RMRR regions.
2472  */
2473 bool dmar_platform_optin(void)
2474 {
2475 	struct acpi_table_dmar *dmar;
2476 	acpi_status status;
2477 	bool ret;
2478 
2479 	status = acpi_get_table(ACPI_SIG_DMAR, 0,
2480 				(struct acpi_table_header **)&dmar);
2481 	if (ACPI_FAILURE(status))
2482 		return false;
2483 
2484 	ret = !!(dmar->flags & DMAR_PLATFORM_OPT_IN);
2485 	acpi_put_table((struct acpi_table_header *)dmar);
2486 
2487 	return ret;
2488 }
2489 EXPORT_SYMBOL_GPL(dmar_platform_optin);
2490