Linux 4.19-rc7
[linux-2.6/btrfs-unstable.git] / drivers / iommu / dmar.c
blobd9c748b6f9e452bf521626d4c1174af8694558bf
1 /*
2 * Copyright (c) 2006, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
17 * Copyright (C) 2006-2008 Intel Corporation
18 * Author: Ashok Raj <ashok.raj@intel.com>
19 * Author: Shaohua Li <shaohua.li@intel.com>
20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
22 * This file implements early detection/parsing of Remapping Devices
23 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
24 * tables.
26 * These routines are used by both DMA-remapping and Interrupt-remapping
29 #define pr_fmt(fmt) "DMAR: " fmt
31 #include <linux/pci.h>
32 #include <linux/dmar.h>
33 #include <linux/iova.h>
34 #include <linux/intel-iommu.h>
35 #include <linux/timer.h>
36 #include <linux/irq.h>
37 #include <linux/interrupt.h>
38 #include <linux/tboot.h>
39 #include <linux/dmi.h>
40 #include <linux/slab.h>
41 #include <linux/iommu.h>
42 #include <asm/irq_remapping.h>
43 #include <asm/iommu_table.h>
45 #include "irq_remapping.h"
47 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
48 struct dmar_res_callback {
49 dmar_res_handler_t cb[ACPI_DMAR_TYPE_RESERVED];
50 void *arg[ACPI_DMAR_TYPE_RESERVED];
51 bool ignore_unhandled;
52 bool print_entry;
56 * Assumptions:
57 * 1) The hotplug framework guarentees that DMAR unit will be hot-added
58 * before IO devices managed by that unit.
59 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
60 * after IO devices managed by that unit.
61 * 3) Hotplug events are rare.
63 * Locking rules for DMA and interrupt remapping related global data structures:
64 * 1) Use dmar_global_lock in process context
65 * 2) Use RCU in interrupt context
67 DECLARE_RWSEM(dmar_global_lock);
68 LIST_HEAD(dmar_drhd_units);
70 struct acpi_table_header * __initdata dmar_tbl;
71 static int dmar_dev_scope_status = 1;
72 static unsigned long dmar_seq_ids[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED)];
74 static int alloc_iommu(struct dmar_drhd_unit *drhd);
75 static void free_iommu(struct intel_iommu *iommu);
77 extern const struct iommu_ops intel_iommu_ops;
79 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
82 * add INCLUDE_ALL at the tail, so scan the list will find it at
83 * the very end.
85 if (drhd->include_all)
86 list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
87 else
88 list_add_rcu(&drhd->list, &dmar_drhd_units);
91 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
93 struct acpi_dmar_device_scope *scope;
95 *cnt = 0;
96 while (start < end) {
97 scope = start;
98 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
99 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
100 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
101 (*cnt)++;
102 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
103 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
104 pr_warn("Unsupported device scope\n");
106 start += scope->length;
108 if (*cnt == 0)
109 return NULL;
111 return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
114 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
116 int i;
117 struct device *tmp_dev;
119 if (*devices && *cnt) {
120 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
121 put_device(tmp_dev);
122 kfree(*devices);
125 *devices = NULL;
126 *cnt = 0;
129 /* Optimize out kzalloc()/kfree() for normal cases */
130 static char dmar_pci_notify_info_buf[64];
132 static struct dmar_pci_notify_info *
133 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
135 int level = 0;
136 size_t size;
137 struct pci_dev *tmp;
138 struct dmar_pci_notify_info *info;
140 BUG_ON(dev->is_virtfn);
142 /* Only generate path[] for device addition event */
143 if (event == BUS_NOTIFY_ADD_DEVICE)
144 for (tmp = dev; tmp; tmp = tmp->bus->self)
145 level++;
147 size = sizeof(*info) + level * sizeof(struct acpi_dmar_pci_path);
148 if (size <= sizeof(dmar_pci_notify_info_buf)) {
149 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
150 } else {
151 info = kzalloc(size, GFP_KERNEL);
152 if (!info) {
153 pr_warn("Out of memory when allocating notify_info "
154 "for %s.\n", pci_name(dev));
155 if (dmar_dev_scope_status == 0)
156 dmar_dev_scope_status = -ENOMEM;
157 return NULL;
161 info->event = event;
162 info->dev = dev;
163 info->seg = pci_domain_nr(dev->bus);
164 info->level = level;
165 if (event == BUS_NOTIFY_ADD_DEVICE) {
166 for (tmp = dev; tmp; tmp = tmp->bus->self) {
167 level--;
168 info->path[level].bus = tmp->bus->number;
169 info->path[level].device = PCI_SLOT(tmp->devfn);
170 info->path[level].function = PCI_FUNC(tmp->devfn);
171 if (pci_is_root_bus(tmp->bus))
172 info->bus = tmp->bus->number;
176 return info;
179 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
181 if ((void *)info != dmar_pci_notify_info_buf)
182 kfree(info);
185 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
186 struct acpi_dmar_pci_path *path, int count)
188 int i;
190 if (info->bus != bus)
191 goto fallback;
192 if (info->level != count)
193 goto fallback;
195 for (i = 0; i < count; i++) {
196 if (path[i].device != info->path[i].device ||
197 path[i].function != info->path[i].function)
198 goto fallback;
201 return true;
203 fallback:
205 if (count != 1)
206 return false;
208 i = info->level - 1;
209 if (bus == info->path[i].bus &&
210 path[0].device == info->path[i].device &&
211 path[0].function == info->path[i].function) {
212 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
213 bus, path[0].device, path[0].function);
214 return true;
217 return false;
220 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
221 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
222 void *start, void*end, u16 segment,
223 struct dmar_dev_scope *devices,
224 int devices_cnt)
226 int i, level;
227 struct device *tmp, *dev = &info->dev->dev;
228 struct acpi_dmar_device_scope *scope;
229 struct acpi_dmar_pci_path *path;
231 if (segment != info->seg)
232 return 0;
234 for (; start < end; start += scope->length) {
235 scope = start;
236 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
237 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
238 continue;
240 path = (struct acpi_dmar_pci_path *)(scope + 1);
241 level = (scope->length - sizeof(*scope)) / sizeof(*path);
242 if (!dmar_match_pci_path(info, scope->bus, path, level))
243 continue;
246 * We expect devices with endpoint scope to have normal PCI
247 * headers, and devices with bridge scope to have bridge PCI
248 * headers. However PCI NTB devices may be listed in the
249 * DMAR table with bridge scope, even though they have a
250 * normal PCI header. NTB devices are identified by class
251 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
252 * for this special case.
254 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
255 info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
256 (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
257 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
258 info->dev->class >> 8 != PCI_CLASS_BRIDGE_OTHER))) {
259 pr_warn("Device scope type does not match for %s\n",
260 pci_name(info->dev));
261 return -EINVAL;
264 for_each_dev_scope(devices, devices_cnt, i, tmp)
265 if (tmp == NULL) {
266 devices[i].bus = info->dev->bus->number;
267 devices[i].devfn = info->dev->devfn;
268 rcu_assign_pointer(devices[i].dev,
269 get_device(dev));
270 return 1;
272 BUG_ON(i >= devices_cnt);
275 return 0;
278 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
279 struct dmar_dev_scope *devices, int count)
281 int index;
282 struct device *tmp;
284 if (info->seg != segment)
285 return 0;
287 for_each_active_dev_scope(devices, count, index, tmp)
288 if (tmp == &info->dev->dev) {
289 RCU_INIT_POINTER(devices[index].dev, NULL);
290 synchronize_rcu();
291 put_device(tmp);
292 return 1;
295 return 0;
298 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
300 int ret = 0;
301 struct dmar_drhd_unit *dmaru;
302 struct acpi_dmar_hardware_unit *drhd;
304 for_each_drhd_unit(dmaru) {
305 if (dmaru->include_all)
306 continue;
308 drhd = container_of(dmaru->hdr,
309 struct acpi_dmar_hardware_unit, header);
310 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
311 ((void *)drhd) + drhd->header.length,
312 dmaru->segment,
313 dmaru->devices, dmaru->devices_cnt);
314 if (ret)
315 break;
317 if (ret >= 0)
318 ret = dmar_iommu_notify_scope_dev(info);
319 if (ret < 0 && dmar_dev_scope_status == 0)
320 dmar_dev_scope_status = ret;
322 return ret;
325 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
327 struct dmar_drhd_unit *dmaru;
329 for_each_drhd_unit(dmaru)
330 if (dmar_remove_dev_scope(info, dmaru->segment,
331 dmaru->devices, dmaru->devices_cnt))
332 break;
333 dmar_iommu_notify_scope_dev(info);
336 static int dmar_pci_bus_notifier(struct notifier_block *nb,
337 unsigned long action, void *data)
339 struct pci_dev *pdev = to_pci_dev(data);
340 struct dmar_pci_notify_info *info;
342 /* Only care about add/remove events for physical functions.
343 * For VFs we actually do the lookup based on the corresponding
344 * PF in device_to_iommu() anyway. */
345 if (pdev->is_virtfn)
346 return NOTIFY_DONE;
347 if (action != BUS_NOTIFY_ADD_DEVICE &&
348 action != BUS_NOTIFY_REMOVED_DEVICE)
349 return NOTIFY_DONE;
351 info = dmar_alloc_pci_notify_info(pdev, action);
352 if (!info)
353 return NOTIFY_DONE;
355 down_write(&dmar_global_lock);
356 if (action == BUS_NOTIFY_ADD_DEVICE)
357 dmar_pci_bus_add_dev(info);
358 else if (action == BUS_NOTIFY_REMOVED_DEVICE)
359 dmar_pci_bus_del_dev(info);
360 up_write(&dmar_global_lock);
362 dmar_free_pci_notify_info(info);
364 return NOTIFY_OK;
367 static struct notifier_block dmar_pci_bus_nb = {
368 .notifier_call = dmar_pci_bus_notifier,
369 .priority = INT_MIN,
372 static struct dmar_drhd_unit *
373 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
375 struct dmar_drhd_unit *dmaru;
377 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
378 if (dmaru->segment == drhd->segment &&
379 dmaru->reg_base_addr == drhd->address)
380 return dmaru;
382 return NULL;
386 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
387 * structure which uniquely represent one DMA remapping hardware unit
388 * present in the platform
390 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
392 struct acpi_dmar_hardware_unit *drhd;
393 struct dmar_drhd_unit *dmaru;
394 int ret;
396 drhd = (struct acpi_dmar_hardware_unit *)header;
397 dmaru = dmar_find_dmaru(drhd);
398 if (dmaru)
399 goto out;
401 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
402 if (!dmaru)
403 return -ENOMEM;
406 * If header is allocated from slab by ACPI _DSM method, we need to
407 * copy the content because the memory buffer will be freed on return.
409 dmaru->hdr = (void *)(dmaru + 1);
410 memcpy(dmaru->hdr, header, header->length);
411 dmaru->reg_base_addr = drhd->address;
412 dmaru->segment = drhd->segment;
413 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
414 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
415 ((void *)drhd) + drhd->header.length,
416 &dmaru->devices_cnt);
417 if (dmaru->devices_cnt && dmaru->devices == NULL) {
418 kfree(dmaru);
419 return -ENOMEM;
422 ret = alloc_iommu(dmaru);
423 if (ret) {
424 dmar_free_dev_scope(&dmaru->devices,
425 &dmaru->devices_cnt);
426 kfree(dmaru);
427 return ret;
429 dmar_register_drhd_unit(dmaru);
431 out:
432 if (arg)
433 (*(int *)arg)++;
435 return 0;
438 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
440 if (dmaru->devices && dmaru->devices_cnt)
441 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
442 if (dmaru->iommu)
443 free_iommu(dmaru->iommu);
444 kfree(dmaru);
447 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
448 void *arg)
450 struct acpi_dmar_andd *andd = (void *)header;
452 /* Check for NUL termination within the designated length */
453 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
454 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND,
455 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
456 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
457 dmi_get_system_info(DMI_BIOS_VENDOR),
458 dmi_get_system_info(DMI_BIOS_VERSION),
459 dmi_get_system_info(DMI_PRODUCT_VERSION));
460 return -EINVAL;
462 pr_info("ANDD device: %x name: %s\n", andd->device_number,
463 andd->device_name);
465 return 0;
468 #ifdef CONFIG_ACPI_NUMA
469 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
471 struct acpi_dmar_rhsa *rhsa;
472 struct dmar_drhd_unit *drhd;
474 rhsa = (struct acpi_dmar_rhsa *)header;
475 for_each_drhd_unit(drhd) {
476 if (drhd->reg_base_addr == rhsa->base_address) {
477 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
479 if (!node_online(node))
480 node = -1;
481 drhd->iommu->node = node;
482 return 0;
485 WARN_TAINT(
486 1, TAINT_FIRMWARE_WORKAROUND,
487 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
488 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
489 drhd->reg_base_addr,
490 dmi_get_system_info(DMI_BIOS_VENDOR),
491 dmi_get_system_info(DMI_BIOS_VERSION),
492 dmi_get_system_info(DMI_PRODUCT_VERSION));
494 return 0;
496 #else
497 #define dmar_parse_one_rhsa dmar_res_noop
498 #endif
500 static void
501 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
503 struct acpi_dmar_hardware_unit *drhd;
504 struct acpi_dmar_reserved_memory *rmrr;
505 struct acpi_dmar_atsr *atsr;
506 struct acpi_dmar_rhsa *rhsa;
508 switch (header->type) {
509 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
510 drhd = container_of(header, struct acpi_dmar_hardware_unit,
511 header);
512 pr_info("DRHD base: %#016Lx flags: %#x\n",
513 (unsigned long long)drhd->address, drhd->flags);
514 break;
515 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
516 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
517 header);
518 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
519 (unsigned long long)rmrr->base_address,
520 (unsigned long long)rmrr->end_address);
521 break;
522 case ACPI_DMAR_TYPE_ROOT_ATS:
523 atsr = container_of(header, struct acpi_dmar_atsr, header);
524 pr_info("ATSR flags: %#x\n", atsr->flags);
525 break;
526 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
527 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
528 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
529 (unsigned long long)rhsa->base_address,
530 rhsa->proximity_domain);
531 break;
532 case ACPI_DMAR_TYPE_NAMESPACE:
533 /* We don't print this here because we need to sanity-check
534 it first. So print it in dmar_parse_one_andd() instead. */
535 break;
540 * dmar_table_detect - checks to see if the platform supports DMAR devices
542 static int __init dmar_table_detect(void)
544 acpi_status status = AE_OK;
546 /* if we could find DMAR table, then there are DMAR devices */
547 status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
549 if (ACPI_SUCCESS(status) && !dmar_tbl) {
550 pr_warn("Unable to map DMAR\n");
551 status = AE_NOT_FOUND;
554 return ACPI_SUCCESS(status) ? 0 : -ENOENT;
557 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
558 size_t len, struct dmar_res_callback *cb)
560 struct acpi_dmar_header *iter, *next;
561 struct acpi_dmar_header *end = ((void *)start) + len;
563 for (iter = start; iter < end; iter = next) {
564 next = (void *)iter + iter->length;
565 if (iter->length == 0) {
566 /* Avoid looping forever on bad ACPI tables */
567 pr_debug(FW_BUG "Invalid 0-length structure\n");
568 break;
569 } else if (next > end) {
570 /* Avoid passing table end */
571 pr_warn(FW_BUG "Record passes table end\n");
572 return -EINVAL;
575 if (cb->print_entry)
576 dmar_table_print_dmar_entry(iter);
578 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
579 /* continue for forward compatibility */
580 pr_debug("Unknown DMAR structure type %d\n",
581 iter->type);
582 } else if (cb->cb[iter->type]) {
583 int ret;
585 ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
586 if (ret)
587 return ret;
588 } else if (!cb->ignore_unhandled) {
589 pr_warn("No handler for DMAR structure type %d\n",
590 iter->type);
591 return -EINVAL;
595 return 0;
598 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
599 struct dmar_res_callback *cb)
601 return dmar_walk_remapping_entries((void *)(dmar + 1),
602 dmar->header.length - sizeof(*dmar), cb);
606 * parse_dmar_table - parses the DMA reporting table
608 static int __init
609 parse_dmar_table(void)
611 struct acpi_table_dmar *dmar;
612 int drhd_count = 0;
613 int ret;
614 struct dmar_res_callback cb = {
615 .print_entry = true,
616 .ignore_unhandled = true,
617 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
618 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
619 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
620 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
621 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
622 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
626 * Do it again, earlier dmar_tbl mapping could be mapped with
627 * fixed map.
629 dmar_table_detect();
632 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
633 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
635 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
637 dmar = (struct acpi_table_dmar *)dmar_tbl;
638 if (!dmar)
639 return -ENODEV;
641 if (dmar->width < PAGE_SHIFT - 1) {
642 pr_warn("Invalid DMAR haw\n");
643 return -EINVAL;
646 pr_info("Host address width %d\n", dmar->width + 1);
647 ret = dmar_walk_dmar_table(dmar, &cb);
648 if (ret == 0 && drhd_count == 0)
649 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
651 return ret;
654 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
655 int cnt, struct pci_dev *dev)
657 int index;
658 struct device *tmp;
660 while (dev) {
661 for_each_active_dev_scope(devices, cnt, index, tmp)
662 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
663 return 1;
665 /* Check our parent */
666 dev = dev->bus->self;
669 return 0;
672 struct dmar_drhd_unit *
673 dmar_find_matched_drhd_unit(struct pci_dev *dev)
675 struct dmar_drhd_unit *dmaru;
676 struct acpi_dmar_hardware_unit *drhd;
678 dev = pci_physfn(dev);
680 rcu_read_lock();
681 for_each_drhd_unit(dmaru) {
682 drhd = container_of(dmaru->hdr,
683 struct acpi_dmar_hardware_unit,
684 header);
686 if (dmaru->include_all &&
687 drhd->segment == pci_domain_nr(dev->bus))
688 goto out;
690 if (dmar_pci_device_match(dmaru->devices,
691 dmaru->devices_cnt, dev))
692 goto out;
694 dmaru = NULL;
695 out:
696 rcu_read_unlock();
698 return dmaru;
701 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
702 struct acpi_device *adev)
704 struct dmar_drhd_unit *dmaru;
705 struct acpi_dmar_hardware_unit *drhd;
706 struct acpi_dmar_device_scope *scope;
707 struct device *tmp;
708 int i;
709 struct acpi_dmar_pci_path *path;
711 for_each_drhd_unit(dmaru) {
712 drhd = container_of(dmaru->hdr,
713 struct acpi_dmar_hardware_unit,
714 header);
716 for (scope = (void *)(drhd + 1);
717 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
718 scope = ((void *)scope) + scope->length) {
719 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
720 continue;
721 if (scope->enumeration_id != device_number)
722 continue;
724 path = (void *)(scope + 1);
725 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
726 dev_name(&adev->dev), dmaru->reg_base_addr,
727 scope->bus, path->device, path->function);
728 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
729 if (tmp == NULL) {
730 dmaru->devices[i].bus = scope->bus;
731 dmaru->devices[i].devfn = PCI_DEVFN(path->device,
732 path->function);
733 rcu_assign_pointer(dmaru->devices[i].dev,
734 get_device(&adev->dev));
735 return;
737 BUG_ON(i >= dmaru->devices_cnt);
740 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
741 device_number, dev_name(&adev->dev));
744 static int __init dmar_acpi_dev_scope_init(void)
746 struct acpi_dmar_andd *andd;
748 if (dmar_tbl == NULL)
749 return -ENODEV;
751 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
752 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
753 andd = ((void *)andd) + andd->header.length) {
754 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
755 acpi_handle h;
756 struct acpi_device *adev;
758 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
759 andd->device_name,
760 &h))) {
761 pr_err("Failed to find handle for ACPI object %s\n",
762 andd->device_name);
763 continue;
765 if (acpi_bus_get_device(h, &adev)) {
766 pr_err("Failed to get device for ACPI object %s\n",
767 andd->device_name);
768 continue;
770 dmar_acpi_insert_dev_scope(andd->device_number, adev);
773 return 0;
776 int __init dmar_dev_scope_init(void)
778 struct pci_dev *dev = NULL;
779 struct dmar_pci_notify_info *info;
781 if (dmar_dev_scope_status != 1)
782 return dmar_dev_scope_status;
784 if (list_empty(&dmar_drhd_units)) {
785 dmar_dev_scope_status = -ENODEV;
786 } else {
787 dmar_dev_scope_status = 0;
789 dmar_acpi_dev_scope_init();
791 for_each_pci_dev(dev) {
792 if (dev->is_virtfn)
793 continue;
795 info = dmar_alloc_pci_notify_info(dev,
796 BUS_NOTIFY_ADD_DEVICE);
797 if (!info) {
798 return dmar_dev_scope_status;
799 } else {
800 dmar_pci_bus_add_dev(info);
801 dmar_free_pci_notify_info(info);
806 return dmar_dev_scope_status;
809 void __init dmar_register_bus_notifier(void)
811 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
815 int __init dmar_table_init(void)
817 static int dmar_table_initialized;
818 int ret;
820 if (dmar_table_initialized == 0) {
821 ret = parse_dmar_table();
822 if (ret < 0) {
823 if (ret != -ENODEV)
824 pr_info("Parse DMAR table failure.\n");
825 } else if (list_empty(&dmar_drhd_units)) {
826 pr_info("No DMAR devices found\n");
827 ret = -ENODEV;
830 if (ret < 0)
831 dmar_table_initialized = ret;
832 else
833 dmar_table_initialized = 1;
836 return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
839 static void warn_invalid_dmar(u64 addr, const char *message)
841 WARN_TAINT_ONCE(
842 1, TAINT_FIRMWARE_WORKAROUND,
843 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
844 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
845 addr, message,
846 dmi_get_system_info(DMI_BIOS_VENDOR),
847 dmi_get_system_info(DMI_BIOS_VERSION),
848 dmi_get_system_info(DMI_PRODUCT_VERSION));
851 static int __ref
852 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
854 struct acpi_dmar_hardware_unit *drhd;
855 void __iomem *addr;
856 u64 cap, ecap;
858 drhd = (void *)entry;
859 if (!drhd->address) {
860 warn_invalid_dmar(0, "");
861 return -EINVAL;
864 if (arg)
865 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
866 else
867 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
868 if (!addr) {
869 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
870 return -EINVAL;
873 cap = dmar_readq(addr + DMAR_CAP_REG);
874 ecap = dmar_readq(addr + DMAR_ECAP_REG);
876 if (arg)
877 iounmap(addr);
878 else
879 early_iounmap(addr, VTD_PAGE_SIZE);
881 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
882 warn_invalid_dmar(drhd->address, " returns all ones");
883 return -EINVAL;
886 return 0;
889 int __init detect_intel_iommu(void)
891 int ret;
892 struct dmar_res_callback validate_drhd_cb = {
893 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
894 .ignore_unhandled = true,
897 down_write(&dmar_global_lock);
898 ret = dmar_table_detect();
899 if (!ret)
900 ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
901 &validate_drhd_cb);
902 if (!ret && !no_iommu && !iommu_detected && !dmar_disabled) {
903 iommu_detected = 1;
904 /* Make sure ACS will be enabled */
905 pci_request_acs();
908 #ifdef CONFIG_X86
909 if (!ret)
910 x86_init.iommu.iommu_init = intel_iommu_init;
911 #endif
913 if (dmar_tbl) {
914 acpi_put_table(dmar_tbl);
915 dmar_tbl = NULL;
917 up_write(&dmar_global_lock);
919 return ret ? ret : 1;
922 static void unmap_iommu(struct intel_iommu *iommu)
924 iounmap(iommu->reg);
925 release_mem_region(iommu->reg_phys, iommu->reg_size);
929 * map_iommu: map the iommu's registers
930 * @iommu: the iommu to map
931 * @phys_addr: the physical address of the base resgister
933 * Memory map the iommu's registers. Start w/ a single page, and
934 * possibly expand if that turns out to be insufficent.
936 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
938 int map_size, err=0;
940 iommu->reg_phys = phys_addr;
941 iommu->reg_size = VTD_PAGE_SIZE;
943 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
944 pr_err("Can't reserve memory\n");
945 err = -EBUSY;
946 goto out;
949 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
950 if (!iommu->reg) {
951 pr_err("Can't map the region\n");
952 err = -ENOMEM;
953 goto release;
956 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
957 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
959 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
960 err = -EINVAL;
961 warn_invalid_dmar(phys_addr, " returns all ones");
962 goto unmap;
965 /* the registers might be more than one page */
966 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
967 cap_max_fault_reg_offset(iommu->cap));
968 map_size = VTD_PAGE_ALIGN(map_size);
969 if (map_size > iommu->reg_size) {
970 iounmap(iommu->reg);
971 release_mem_region(iommu->reg_phys, iommu->reg_size);
972 iommu->reg_size = map_size;
973 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
974 iommu->name)) {
975 pr_err("Can't reserve memory\n");
976 err = -EBUSY;
977 goto out;
979 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
980 if (!iommu->reg) {
981 pr_err("Can't map the region\n");
982 err = -ENOMEM;
983 goto release;
986 err = 0;
987 goto out;
989 unmap:
990 iounmap(iommu->reg);
991 release:
992 release_mem_region(iommu->reg_phys, iommu->reg_size);
993 out:
994 return err;
997 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
999 iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
1000 DMAR_UNITS_SUPPORTED);
1001 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
1002 iommu->seq_id = -1;
1003 } else {
1004 set_bit(iommu->seq_id, dmar_seq_ids);
1005 sprintf(iommu->name, "dmar%d", iommu->seq_id);
1008 return iommu->seq_id;
1011 static void dmar_free_seq_id(struct intel_iommu *iommu)
1013 if (iommu->seq_id >= 0) {
1014 clear_bit(iommu->seq_id, dmar_seq_ids);
1015 iommu->seq_id = -1;
1019 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1021 struct intel_iommu *iommu;
1022 u32 ver, sts;
1023 int agaw = 0;
1024 int msagaw = 0;
1025 int err;
1027 if (!drhd->reg_base_addr) {
1028 warn_invalid_dmar(0, "");
1029 return -EINVAL;
1032 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1033 if (!iommu)
1034 return -ENOMEM;
1036 if (dmar_alloc_seq_id(iommu) < 0) {
1037 pr_err("Failed to allocate seq_id\n");
1038 err = -ENOSPC;
1039 goto error;
1042 err = map_iommu(iommu, drhd->reg_base_addr);
1043 if (err) {
1044 pr_err("Failed to map %s\n", iommu->name);
1045 goto error_free_seq_id;
1048 err = -EINVAL;
1049 agaw = iommu_calculate_agaw(iommu);
1050 if (agaw < 0) {
1051 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1052 iommu->seq_id);
1053 goto err_unmap;
1055 msagaw = iommu_calculate_max_sagaw(iommu);
1056 if (msagaw < 0) {
1057 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1058 iommu->seq_id);
1059 goto err_unmap;
1061 iommu->agaw = agaw;
1062 iommu->msagaw = msagaw;
1063 iommu->segment = drhd->segment;
1065 iommu->node = -1;
1067 ver = readl(iommu->reg + DMAR_VER_REG);
1068 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1069 iommu->name,
1070 (unsigned long long)drhd->reg_base_addr,
1071 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1072 (unsigned long long)iommu->cap,
1073 (unsigned long long)iommu->ecap);
1075 /* Reflect status in gcmd */
1076 sts = readl(iommu->reg + DMAR_GSTS_REG);
1077 if (sts & DMA_GSTS_IRES)
1078 iommu->gcmd |= DMA_GCMD_IRE;
1079 if (sts & DMA_GSTS_TES)
1080 iommu->gcmd |= DMA_GCMD_TE;
1081 if (sts & DMA_GSTS_QIES)
1082 iommu->gcmd |= DMA_GCMD_QIE;
1084 raw_spin_lock_init(&iommu->register_lock);
1086 if (intel_iommu_enabled) {
1087 err = iommu_device_sysfs_add(&iommu->iommu, NULL,
1088 intel_iommu_groups,
1089 "%s", iommu->name);
1090 if (err)
1091 goto err_unmap;
1093 iommu_device_set_ops(&iommu->iommu, &intel_iommu_ops);
1095 err = iommu_device_register(&iommu->iommu);
1096 if (err)
1097 goto err_unmap;
1100 drhd->iommu = iommu;
1102 return 0;
1104 err_unmap:
1105 unmap_iommu(iommu);
1106 error_free_seq_id:
1107 dmar_free_seq_id(iommu);
1108 error:
1109 kfree(iommu);
1110 return err;
1113 static void free_iommu(struct intel_iommu *iommu)
1115 if (intel_iommu_enabled) {
1116 iommu_device_unregister(&iommu->iommu);
1117 iommu_device_sysfs_remove(&iommu->iommu);
1120 if (iommu->irq) {
1121 if (iommu->pr_irq) {
1122 free_irq(iommu->pr_irq, iommu);
1123 dmar_free_hwirq(iommu->pr_irq);
1124 iommu->pr_irq = 0;
1126 free_irq(iommu->irq, iommu);
1127 dmar_free_hwirq(iommu->irq);
1128 iommu->irq = 0;
1131 if (iommu->qi) {
1132 free_page((unsigned long)iommu->qi->desc);
1133 kfree(iommu->qi->desc_status);
1134 kfree(iommu->qi);
1137 if (iommu->reg)
1138 unmap_iommu(iommu);
1140 dmar_free_seq_id(iommu);
1141 kfree(iommu);
1145 * Reclaim all the submitted descriptors which have completed its work.
1147 static inline void reclaim_free_desc(struct q_inval *qi)
1149 while (qi->desc_status[qi->free_tail] == QI_DONE ||
1150 qi->desc_status[qi->free_tail] == QI_ABORT) {
1151 qi->desc_status[qi->free_tail] = QI_FREE;
1152 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1153 qi->free_cnt++;
1157 static int qi_check_fault(struct intel_iommu *iommu, int index)
1159 u32 fault;
1160 int head, tail;
1161 struct q_inval *qi = iommu->qi;
1162 int wait_index = (index + 1) % QI_LENGTH;
1164 if (qi->desc_status[wait_index] == QI_ABORT)
1165 return -EAGAIN;
1167 fault = readl(iommu->reg + DMAR_FSTS_REG);
1170 * If IQE happens, the head points to the descriptor associated
1171 * with the error. No new descriptors are fetched until the IQE
1172 * is cleared.
1174 if (fault & DMA_FSTS_IQE) {
1175 head = readl(iommu->reg + DMAR_IQH_REG);
1176 if ((head >> DMAR_IQ_SHIFT) == index) {
1177 pr_err("VT-d detected invalid descriptor: "
1178 "low=%llx, high=%llx\n",
1179 (unsigned long long)qi->desc[index].low,
1180 (unsigned long long)qi->desc[index].high);
1181 memcpy(&qi->desc[index], &qi->desc[wait_index],
1182 sizeof(struct qi_desc));
1183 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1184 return -EINVAL;
1189 * If ITE happens, all pending wait_desc commands are aborted.
1190 * No new descriptors are fetched until the ITE is cleared.
1192 if (fault & DMA_FSTS_ITE) {
1193 head = readl(iommu->reg + DMAR_IQH_REG);
1194 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1195 head |= 1;
1196 tail = readl(iommu->reg + DMAR_IQT_REG);
1197 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1199 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1201 do {
1202 if (qi->desc_status[head] == QI_IN_USE)
1203 qi->desc_status[head] = QI_ABORT;
1204 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1205 } while (head != tail);
1207 if (qi->desc_status[wait_index] == QI_ABORT)
1208 return -EAGAIN;
1211 if (fault & DMA_FSTS_ICE)
1212 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1214 return 0;
1218 * Submit the queued invalidation descriptor to the remapping
1219 * hardware unit and wait for its completion.
1221 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1223 int rc;
1224 struct q_inval *qi = iommu->qi;
1225 struct qi_desc *hw, wait_desc;
1226 int wait_index, index;
1227 unsigned long flags;
1229 if (!qi)
1230 return 0;
1232 hw = qi->desc;
1234 restart:
1235 rc = 0;
1237 raw_spin_lock_irqsave(&qi->q_lock, flags);
1238 while (qi->free_cnt < 3) {
1239 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1240 cpu_relax();
1241 raw_spin_lock_irqsave(&qi->q_lock, flags);
1244 index = qi->free_head;
1245 wait_index = (index + 1) % QI_LENGTH;
1247 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1249 hw[index] = *desc;
1251 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
1252 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1253 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
1255 hw[wait_index] = wait_desc;
1257 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1258 qi->free_cnt -= 2;
1261 * update the HW tail register indicating the presence of
1262 * new descriptors.
1264 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
1266 while (qi->desc_status[wait_index] != QI_DONE) {
1268 * We will leave the interrupts disabled, to prevent interrupt
1269 * context to queue another cmd while a cmd is already submitted
1270 * and waiting for completion on this cpu. This is to avoid
1271 * a deadlock where the interrupt context can wait indefinitely
1272 * for free slots in the queue.
1274 rc = qi_check_fault(iommu, index);
1275 if (rc)
1276 break;
1278 raw_spin_unlock(&qi->q_lock);
1279 cpu_relax();
1280 raw_spin_lock(&qi->q_lock);
1283 qi->desc_status[index] = QI_DONE;
1285 reclaim_free_desc(qi);
1286 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1288 if (rc == -EAGAIN)
1289 goto restart;
1291 return rc;
1295 * Flush the global interrupt entry cache.
1297 void qi_global_iec(struct intel_iommu *iommu)
1299 struct qi_desc desc;
1301 desc.low = QI_IEC_TYPE;
1302 desc.high = 0;
1304 /* should never fail */
1305 qi_submit_sync(&desc, iommu);
1308 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1309 u64 type)
1311 struct qi_desc desc;
1313 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1314 | QI_CC_GRAN(type) | QI_CC_TYPE;
1315 desc.high = 0;
1317 qi_submit_sync(&desc, iommu);
1320 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1321 unsigned int size_order, u64 type)
1323 u8 dw = 0, dr = 0;
1325 struct qi_desc desc;
1326 int ih = 0;
1328 if (cap_write_drain(iommu->cap))
1329 dw = 1;
1331 if (cap_read_drain(iommu->cap))
1332 dr = 1;
1334 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1335 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1336 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1337 | QI_IOTLB_AM(size_order);
1339 qi_submit_sync(&desc, iommu);
1342 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 pfsid,
1343 u16 qdep, u64 addr, unsigned mask)
1345 struct qi_desc desc;
1347 if (mask) {
1348 WARN_ON_ONCE(addr & ((1ULL << (VTD_PAGE_SHIFT + mask)) - 1));
1349 addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1350 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1351 } else
1352 desc.high = QI_DEV_IOTLB_ADDR(addr);
1354 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1355 qdep = 0;
1357 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1358 QI_DIOTLB_TYPE | QI_DEV_IOTLB_PFSID(pfsid);
1360 qi_submit_sync(&desc, iommu);
1364 * Disable Queued Invalidation interface.
1366 void dmar_disable_qi(struct intel_iommu *iommu)
1368 unsigned long flags;
1369 u32 sts;
1370 cycles_t start_time = get_cycles();
1372 if (!ecap_qis(iommu->ecap))
1373 return;
1375 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1377 sts = readl(iommu->reg + DMAR_GSTS_REG);
1378 if (!(sts & DMA_GSTS_QIES))
1379 goto end;
1382 * Give a chance to HW to complete the pending invalidation requests.
1384 while ((readl(iommu->reg + DMAR_IQT_REG) !=
1385 readl(iommu->reg + DMAR_IQH_REG)) &&
1386 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1387 cpu_relax();
1389 iommu->gcmd &= ~DMA_GCMD_QIE;
1390 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1392 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1393 !(sts & DMA_GSTS_QIES), sts);
1394 end:
1395 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1399 * Enable queued invalidation.
1401 static void __dmar_enable_qi(struct intel_iommu *iommu)
1403 u32 sts;
1404 unsigned long flags;
1405 struct q_inval *qi = iommu->qi;
1407 qi->free_head = qi->free_tail = 0;
1408 qi->free_cnt = QI_LENGTH;
1410 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1412 /* write zero to the tail reg */
1413 writel(0, iommu->reg + DMAR_IQT_REG);
1415 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1417 iommu->gcmd |= DMA_GCMD_QIE;
1418 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1420 /* Make sure hardware complete it */
1421 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1423 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1427 * Enable Queued Invalidation interface. This is a must to support
1428 * interrupt-remapping. Also used by DMA-remapping, which replaces
1429 * register based IOTLB invalidation.
1431 int dmar_enable_qi(struct intel_iommu *iommu)
1433 struct q_inval *qi;
1434 struct page *desc_page;
1436 if (!ecap_qis(iommu->ecap))
1437 return -ENOENT;
1440 * queued invalidation is already setup and enabled.
1442 if (iommu->qi)
1443 return 0;
1445 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1446 if (!iommu->qi)
1447 return -ENOMEM;
1449 qi = iommu->qi;
1452 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1453 if (!desc_page) {
1454 kfree(qi);
1455 iommu->qi = NULL;
1456 return -ENOMEM;
1459 qi->desc = page_address(desc_page);
1461 qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC);
1462 if (!qi->desc_status) {
1463 free_page((unsigned long) qi->desc);
1464 kfree(qi);
1465 iommu->qi = NULL;
1466 return -ENOMEM;
1469 raw_spin_lock_init(&qi->q_lock);
1471 __dmar_enable_qi(iommu);
1473 return 0;
1476 /* iommu interrupt handling. Most stuff are MSI-like. */
1478 enum faulttype {
1479 DMA_REMAP,
1480 INTR_REMAP,
1481 UNKNOWN,
1484 static const char *dma_remap_fault_reasons[] =
1486 "Software",
1487 "Present bit in root entry is clear",
1488 "Present bit in context entry is clear",
1489 "Invalid context entry",
1490 "Access beyond MGAW",
1491 "PTE Write access is not set",
1492 "PTE Read access is not set",
1493 "Next page table ptr is invalid",
1494 "Root table address invalid",
1495 "Context table ptr is invalid",
1496 "non-zero reserved fields in RTP",
1497 "non-zero reserved fields in CTP",
1498 "non-zero reserved fields in PTE",
1499 "PCE for translation request specifies blocking",
1502 static const char *irq_remap_fault_reasons[] =
1504 "Detected reserved fields in the decoded interrupt-remapped request",
1505 "Interrupt index exceeded the interrupt-remapping table size",
1506 "Present field in the IRTE entry is clear",
1507 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1508 "Detected reserved fields in the IRTE entry",
1509 "Blocked a compatibility format interrupt request",
1510 "Blocked an interrupt request due to source-id verification failure",
1513 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1515 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1516 ARRAY_SIZE(irq_remap_fault_reasons))) {
1517 *fault_type = INTR_REMAP;
1518 return irq_remap_fault_reasons[fault_reason - 0x20];
1519 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1520 *fault_type = DMA_REMAP;
1521 return dma_remap_fault_reasons[fault_reason];
1522 } else {
1523 *fault_type = UNKNOWN;
1524 return "Unknown";
1529 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1531 if (iommu->irq == irq)
1532 return DMAR_FECTL_REG;
1533 else if (iommu->pr_irq == irq)
1534 return DMAR_PECTL_REG;
1535 else
1536 BUG();
1539 void dmar_msi_unmask(struct irq_data *data)
1541 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1542 int reg = dmar_msi_reg(iommu, data->irq);
1543 unsigned long flag;
1545 /* unmask it */
1546 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1547 writel(0, iommu->reg + reg);
1548 /* Read a reg to force flush the post write */
1549 readl(iommu->reg + reg);
1550 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1553 void dmar_msi_mask(struct irq_data *data)
1555 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1556 int reg = dmar_msi_reg(iommu, data->irq);
1557 unsigned long flag;
1559 /* mask it */
1560 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1561 writel(DMA_FECTL_IM, iommu->reg + reg);
1562 /* Read a reg to force flush the post write */
1563 readl(iommu->reg + reg);
1564 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1567 void dmar_msi_write(int irq, struct msi_msg *msg)
1569 struct intel_iommu *iommu = irq_get_handler_data(irq);
1570 int reg = dmar_msi_reg(iommu, irq);
1571 unsigned long flag;
1573 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1574 writel(msg->data, iommu->reg + reg + 4);
1575 writel(msg->address_lo, iommu->reg + reg + 8);
1576 writel(msg->address_hi, iommu->reg + reg + 12);
1577 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1580 void dmar_msi_read(int irq, struct msi_msg *msg)
1582 struct intel_iommu *iommu = irq_get_handler_data(irq);
1583 int reg = dmar_msi_reg(iommu, irq);
1584 unsigned long flag;
1586 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1587 msg->data = readl(iommu->reg + reg + 4);
1588 msg->address_lo = readl(iommu->reg + reg + 8);
1589 msg->address_hi = readl(iommu->reg + reg + 12);
1590 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1593 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1594 u8 fault_reason, u16 source_id, unsigned long long addr)
1596 const char *reason;
1597 int fault_type;
1599 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1601 if (fault_type == INTR_REMAP)
1602 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1603 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1604 PCI_FUNC(source_id & 0xFF), addr >> 48,
1605 fault_reason, reason);
1606 else
1607 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1608 type ? "DMA Read" : "DMA Write",
1609 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1610 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1611 return 0;
1614 #define PRIMARY_FAULT_REG_LEN (16)
1615 irqreturn_t dmar_fault(int irq, void *dev_id)
1617 struct intel_iommu *iommu = dev_id;
1618 int reg, fault_index;
1619 u32 fault_status;
1620 unsigned long flag;
1621 static DEFINE_RATELIMIT_STATE(rs,
1622 DEFAULT_RATELIMIT_INTERVAL,
1623 DEFAULT_RATELIMIT_BURST);
1625 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1626 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1627 if (fault_status && __ratelimit(&rs))
1628 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1630 /* TBD: ignore advanced fault log currently */
1631 if (!(fault_status & DMA_FSTS_PPF))
1632 goto unlock_exit;
1634 fault_index = dma_fsts_fault_record_index(fault_status);
1635 reg = cap_fault_reg_offset(iommu->cap);
1636 while (1) {
1637 /* Disable printing, simply clear the fault when ratelimited */
1638 bool ratelimited = !__ratelimit(&rs);
1639 u8 fault_reason;
1640 u16 source_id;
1641 u64 guest_addr;
1642 int type;
1643 u32 data;
1645 /* highest 32 bits */
1646 data = readl(iommu->reg + reg +
1647 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1648 if (!(data & DMA_FRCD_F))
1649 break;
1651 if (!ratelimited) {
1652 fault_reason = dma_frcd_fault_reason(data);
1653 type = dma_frcd_type(data);
1655 data = readl(iommu->reg + reg +
1656 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1657 source_id = dma_frcd_source_id(data);
1659 guest_addr = dmar_readq(iommu->reg + reg +
1660 fault_index * PRIMARY_FAULT_REG_LEN);
1661 guest_addr = dma_frcd_page_addr(guest_addr);
1664 /* clear the fault */
1665 writel(DMA_FRCD_F, iommu->reg + reg +
1666 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1668 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1670 if (!ratelimited)
1671 dmar_fault_do_one(iommu, type, fault_reason,
1672 source_id, guest_addr);
1674 fault_index++;
1675 if (fault_index >= cap_num_fault_regs(iommu->cap))
1676 fault_index = 0;
1677 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1680 writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO,
1681 iommu->reg + DMAR_FSTS_REG);
1683 unlock_exit:
1684 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1685 return IRQ_HANDLED;
1688 int dmar_set_interrupt(struct intel_iommu *iommu)
1690 int irq, ret;
1693 * Check if the fault interrupt is already initialized.
1695 if (iommu->irq)
1696 return 0;
1698 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1699 if (irq > 0) {
1700 iommu->irq = irq;
1701 } else {
1702 pr_err("No free IRQ vectors\n");
1703 return -EINVAL;
1706 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1707 if (ret)
1708 pr_err("Can't request irq\n");
1709 return ret;
1712 int __init enable_drhd_fault_handling(void)
1714 struct dmar_drhd_unit *drhd;
1715 struct intel_iommu *iommu;
1718 * Enable fault control interrupt.
1720 for_each_iommu(iommu, drhd) {
1721 u32 fault_status;
1722 int ret = dmar_set_interrupt(iommu);
1724 if (ret) {
1725 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1726 (unsigned long long)drhd->reg_base_addr, ret);
1727 return -1;
1731 * Clear any previous faults.
1733 dmar_fault(iommu->irq, iommu);
1734 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1735 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1738 return 0;
1742 * Re-enable Queued Invalidation interface.
1744 int dmar_reenable_qi(struct intel_iommu *iommu)
1746 if (!ecap_qis(iommu->ecap))
1747 return -ENOENT;
1749 if (!iommu->qi)
1750 return -ENOENT;
1753 * First disable queued invalidation.
1755 dmar_disable_qi(iommu);
1757 * Then enable queued invalidation again. Since there is no pending
1758 * invalidation requests now, it's safe to re-enable queued
1759 * invalidation.
1761 __dmar_enable_qi(iommu);
1763 return 0;
1767 * Check interrupt remapping support in DMAR table description.
1769 int __init dmar_ir_support(void)
1771 struct acpi_table_dmar *dmar;
1772 dmar = (struct acpi_table_dmar *)dmar_tbl;
1773 if (!dmar)
1774 return 0;
1775 return dmar->flags & 0x1;
1778 /* Check whether DMAR units are in use */
1779 static inline bool dmar_in_use(void)
1781 return irq_remapping_enabled || intel_iommu_enabled;
1784 static int __init dmar_free_unused_resources(void)
1786 struct dmar_drhd_unit *dmaru, *dmaru_n;
1788 if (dmar_in_use())
1789 return 0;
1791 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1792 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1794 down_write(&dmar_global_lock);
1795 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1796 list_del(&dmaru->list);
1797 dmar_free_drhd(dmaru);
1799 up_write(&dmar_global_lock);
1801 return 0;
1804 late_initcall(dmar_free_unused_resources);
1805 IOMMU_INIT_POST(detect_intel_iommu);
1808 * DMAR Hotplug Support
1809 * For more details, please refer to Intel(R) Virtualization Technology
1810 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1811 * "Remapping Hardware Unit Hot Plug".
1813 static guid_t dmar_hp_guid =
1814 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
1815 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
1818 * Currently there's only one revision and BIOS will not check the revision id,
1819 * so use 0 for safety.
1821 #define DMAR_DSM_REV_ID 0
1822 #define DMAR_DSM_FUNC_DRHD 1
1823 #define DMAR_DSM_FUNC_ATSR 2
1824 #define DMAR_DSM_FUNC_RHSA 3
1826 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1828 return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func);
1831 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1832 dmar_res_handler_t handler, void *arg)
1834 int ret = -ENODEV;
1835 union acpi_object *obj;
1836 struct acpi_dmar_header *start;
1837 struct dmar_res_callback callback;
1838 static int res_type[] = {
1839 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1840 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1841 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1844 if (!dmar_detect_dsm(handle, func))
1845 return 0;
1847 obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID,
1848 func, NULL, ACPI_TYPE_BUFFER);
1849 if (!obj)
1850 return -ENODEV;
1852 memset(&callback, 0, sizeof(callback));
1853 callback.cb[res_type[func]] = handler;
1854 callback.arg[res_type[func]] = arg;
1855 start = (struct acpi_dmar_header *)obj->buffer.pointer;
1856 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1858 ACPI_FREE(obj);
1860 return ret;
1863 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1865 int ret;
1866 struct dmar_drhd_unit *dmaru;
1868 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1869 if (!dmaru)
1870 return -ENODEV;
1872 ret = dmar_ir_hotplug(dmaru, true);
1873 if (ret == 0)
1874 ret = dmar_iommu_hotplug(dmaru, true);
1876 return ret;
1879 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
1881 int i, ret;
1882 struct device *dev;
1883 struct dmar_drhd_unit *dmaru;
1885 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1886 if (!dmaru)
1887 return 0;
1890 * All PCI devices managed by this unit should have been destroyed.
1892 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
1893 for_each_active_dev_scope(dmaru->devices,
1894 dmaru->devices_cnt, i, dev)
1895 return -EBUSY;
1898 ret = dmar_ir_hotplug(dmaru, false);
1899 if (ret == 0)
1900 ret = dmar_iommu_hotplug(dmaru, false);
1902 return ret;
1905 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
1907 struct dmar_drhd_unit *dmaru;
1909 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1910 if (dmaru) {
1911 list_del_rcu(&dmaru->list);
1912 synchronize_rcu();
1913 dmar_free_drhd(dmaru);
1916 return 0;
1919 static int dmar_hotplug_insert(acpi_handle handle)
1921 int ret;
1922 int drhd_count = 0;
1924 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1925 &dmar_validate_one_drhd, (void *)1);
1926 if (ret)
1927 goto out;
1929 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1930 &dmar_parse_one_drhd, (void *)&drhd_count);
1931 if (ret == 0 && drhd_count == 0) {
1932 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
1933 goto out;
1934 } else if (ret) {
1935 goto release_drhd;
1938 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
1939 &dmar_parse_one_rhsa, NULL);
1940 if (ret)
1941 goto release_drhd;
1943 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1944 &dmar_parse_one_atsr, NULL);
1945 if (ret)
1946 goto release_atsr;
1948 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1949 &dmar_hp_add_drhd, NULL);
1950 if (!ret)
1951 return 0;
1953 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1954 &dmar_hp_remove_drhd, NULL);
1955 release_atsr:
1956 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1957 &dmar_release_one_atsr, NULL);
1958 release_drhd:
1959 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1960 &dmar_hp_release_drhd, NULL);
1961 out:
1962 return ret;
1965 static int dmar_hotplug_remove(acpi_handle handle)
1967 int ret;
1969 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1970 &dmar_check_one_atsr, NULL);
1971 if (ret)
1972 return ret;
1974 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1975 &dmar_hp_remove_drhd, NULL);
1976 if (ret == 0) {
1977 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1978 &dmar_release_one_atsr, NULL));
1979 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1980 &dmar_hp_release_drhd, NULL));
1981 } else {
1982 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1983 &dmar_hp_add_drhd, NULL);
1986 return ret;
1989 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
1990 void *context, void **retval)
1992 acpi_handle *phdl = retval;
1994 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
1995 *phdl = handle;
1996 return AE_CTRL_TERMINATE;
1999 return AE_OK;
2002 static int dmar_device_hotplug(acpi_handle handle, bool insert)
2004 int ret;
2005 acpi_handle tmp = NULL;
2006 acpi_status status;
2008 if (!dmar_in_use())
2009 return 0;
2011 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2012 tmp = handle;
2013 } else {
2014 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2015 ACPI_UINT32_MAX,
2016 dmar_get_dsm_handle,
2017 NULL, NULL, &tmp);
2018 if (ACPI_FAILURE(status)) {
2019 pr_warn("Failed to locate _DSM method.\n");
2020 return -ENXIO;
2023 if (tmp == NULL)
2024 return 0;
2026 down_write(&dmar_global_lock);
2027 if (insert)
2028 ret = dmar_hotplug_insert(tmp);
2029 else
2030 ret = dmar_hotplug_remove(tmp);
2031 up_write(&dmar_global_lock);
2033 return ret;
2036 int dmar_device_add(acpi_handle handle)
2038 return dmar_device_hotplug(handle, true);
2041 int dmar_device_remove(acpi_handle handle)
2043 return dmar_device_hotplug(handle, false);