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