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
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
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 <linux/limits.h>
43 #include <asm/irq_remapping.h>
44 #include <asm/iommu_table.h>
46 #include "irq_remapping.h"
48 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
49 struct dmar_res_callback {
50 dmar_res_handler_t cb[ACPI_DMAR_TYPE_RESERVED];
51 void *arg[ACPI_DMAR_TYPE_RESERVED];
52 bool ignore_unhandled;
58 * 1) The hotplug framework guarentees that DMAR unit will be hot-added
59 * before IO devices managed by that unit.
60 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
61 * after IO devices managed by that unit.
62 * 3) Hotplug events are rare.
64 * Locking rules for DMA and interrupt remapping related global data structures:
65 * 1) Use dmar_global_lock in process context
66 * 2) Use RCU in interrupt context
68 DECLARE_RWSEM(dmar_global_lock);
69 LIST_HEAD(dmar_drhd_units);
71 struct acpi_table_header * __initdata dmar_tbl;
72 static int dmar_dev_scope_status = 1;
73 static unsigned long dmar_seq_ids[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED)];
75 static int alloc_iommu(struct dmar_drhd_unit *drhd);
76 static void free_iommu(struct intel_iommu *iommu);
78 extern const struct iommu_ops intel_iommu_ops;
80 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
83 * add INCLUDE_ALL at the tail, so scan the list will find it at
86 if (drhd->include_all)
87 list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
89 list_add_rcu(&drhd->list, &dmar_drhd_units);
92 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
94 struct acpi_dmar_device_scope *scope;
99 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
100 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
101 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
103 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
104 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
105 pr_warn("Unsupported device scope\n");
107 start += scope->length;
112 return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
115 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
118 struct device *tmp_dev;
120 if (*devices && *cnt) {
121 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
130 /* Optimize out kzalloc()/kfree() for normal cases */
131 static char dmar_pci_notify_info_buf[64];
133 static struct dmar_pci_notify_info *
134 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
139 struct dmar_pci_notify_info *info;
141 BUG_ON(dev->is_virtfn);
144 * Ignore devices that have a domain number higher than what can
145 * be looked up in DMAR, e.g. VMD subdevices with domain 0x10000
147 if (pci_domain_nr(dev->bus) > U16_MAX)
150 /* Only generate path[] for device addition event */
151 if (event == BUS_NOTIFY_ADD_DEVICE)
152 for (tmp = dev; tmp; tmp = tmp->bus->self)
155 size = sizeof(*info) + level * sizeof(info->path[0]);
156 if (size <= sizeof(dmar_pci_notify_info_buf)) {
157 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
159 info = kzalloc(size, GFP_KERNEL);
161 pr_warn("Out of memory when allocating notify_info "
162 "for %s.\n", pci_name(dev));
163 if (dmar_dev_scope_status == 0)
164 dmar_dev_scope_status = -ENOMEM;
171 info->seg = pci_domain_nr(dev->bus);
173 if (event == BUS_NOTIFY_ADD_DEVICE) {
174 for (tmp = dev; tmp; tmp = tmp->bus->self) {
176 info->path[level].bus = tmp->bus->number;
177 info->path[level].device = PCI_SLOT(tmp->devfn);
178 info->path[level].function = PCI_FUNC(tmp->devfn);
179 if (pci_is_root_bus(tmp->bus))
180 info->bus = tmp->bus->number;
187 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
189 if ((void *)info != dmar_pci_notify_info_buf)
193 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
194 struct acpi_dmar_pci_path *path, int count)
198 if (info->bus != bus)
200 if (info->level != count)
203 for (i = 0; i < count; i++) {
204 if (path[i].device != info->path[i].device ||
205 path[i].function != info->path[i].function)
217 if (bus == info->path[i].bus &&
218 path[0].device == info->path[i].device &&
219 path[0].function == info->path[i].function) {
220 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
221 bus, path[0].device, path[0].function);
228 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
229 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
230 void *start, void*end, u16 segment,
231 struct dmar_dev_scope *devices,
235 struct device *tmp, *dev = &info->dev->dev;
236 struct acpi_dmar_device_scope *scope;
237 struct acpi_dmar_pci_path *path;
239 if (segment != info->seg)
242 for (; start < end; start += scope->length) {
244 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
245 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
248 path = (struct acpi_dmar_pci_path *)(scope + 1);
249 level = (scope->length - sizeof(*scope)) / sizeof(*path);
250 if (!dmar_match_pci_path(info, scope->bus, path, level))
254 * We expect devices with endpoint scope to have normal PCI
255 * headers, and devices with bridge scope to have bridge PCI
256 * headers. However PCI NTB devices may be listed in the
257 * DMAR table with bridge scope, even though they have a
258 * normal PCI header. NTB devices are identified by class
259 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
260 * for this special case.
262 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
263 info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
264 (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
265 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
266 info->dev->class >> 8 != PCI_CLASS_BRIDGE_OTHER))) {
267 pr_warn("Device scope type does not match for %s\n",
268 pci_name(info->dev));
272 for_each_dev_scope(devices, devices_cnt, i, tmp)
274 devices[i].bus = info->dev->bus->number;
275 devices[i].devfn = info->dev->devfn;
276 rcu_assign_pointer(devices[i].dev,
280 BUG_ON(i >= devices_cnt);
286 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
287 struct dmar_dev_scope *devices, int count)
292 if (info->seg != segment)
295 for_each_active_dev_scope(devices, count, index, tmp)
296 if (tmp == &info->dev->dev) {
297 RCU_INIT_POINTER(devices[index].dev, NULL);
306 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
309 struct dmar_drhd_unit *dmaru;
310 struct acpi_dmar_hardware_unit *drhd;
312 for_each_drhd_unit(dmaru) {
313 if (dmaru->include_all)
316 drhd = container_of(dmaru->hdr,
317 struct acpi_dmar_hardware_unit, header);
318 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
319 ((void *)drhd) + drhd->header.length,
321 dmaru->devices, dmaru->devices_cnt);
326 ret = dmar_iommu_notify_scope_dev(info);
327 if (ret < 0 && dmar_dev_scope_status == 0)
328 dmar_dev_scope_status = ret;
333 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
335 struct dmar_drhd_unit *dmaru;
337 for_each_drhd_unit(dmaru)
338 if (dmar_remove_dev_scope(info, dmaru->segment,
339 dmaru->devices, dmaru->devices_cnt))
341 dmar_iommu_notify_scope_dev(info);
344 static int dmar_pci_bus_notifier(struct notifier_block *nb,
345 unsigned long action, void *data)
347 struct pci_dev *pdev = to_pci_dev(data);
348 struct dmar_pci_notify_info *info;
350 /* Only care about add/remove events for physical functions.
351 * For VFs we actually do the lookup based on the corresponding
352 * PF in device_to_iommu() anyway. */
355 if (action != BUS_NOTIFY_ADD_DEVICE &&
356 action != BUS_NOTIFY_REMOVED_DEVICE)
359 info = dmar_alloc_pci_notify_info(pdev, action);
363 down_write(&dmar_global_lock);
364 if (action == BUS_NOTIFY_ADD_DEVICE)
365 dmar_pci_bus_add_dev(info);
366 else if (action == BUS_NOTIFY_REMOVED_DEVICE)
367 dmar_pci_bus_del_dev(info);
368 up_write(&dmar_global_lock);
370 dmar_free_pci_notify_info(info);
375 static struct notifier_block dmar_pci_bus_nb = {
376 .notifier_call = dmar_pci_bus_notifier,
380 static struct dmar_drhd_unit *
381 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
383 struct dmar_drhd_unit *dmaru;
385 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
386 if (dmaru->segment == drhd->segment &&
387 dmaru->reg_base_addr == drhd->address)
394 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
395 * structure which uniquely represent one DMA remapping hardware unit
396 * present in the platform
398 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
400 struct acpi_dmar_hardware_unit *drhd;
401 struct dmar_drhd_unit *dmaru;
404 drhd = (struct acpi_dmar_hardware_unit *)header;
405 dmaru = dmar_find_dmaru(drhd);
409 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
414 * If header is allocated from slab by ACPI _DSM method, we need to
415 * copy the content because the memory buffer will be freed on return.
417 dmaru->hdr = (void *)(dmaru + 1);
418 memcpy(dmaru->hdr, header, header->length);
419 dmaru->reg_base_addr = drhd->address;
420 dmaru->segment = drhd->segment;
421 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
422 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
423 ((void *)drhd) + drhd->header.length,
424 &dmaru->devices_cnt);
425 if (dmaru->devices_cnt && dmaru->devices == NULL) {
430 ret = alloc_iommu(dmaru);
432 dmar_free_dev_scope(&dmaru->devices,
433 &dmaru->devices_cnt);
437 dmar_register_drhd_unit(dmaru);
446 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
448 if (dmaru->devices && dmaru->devices_cnt)
449 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
451 free_iommu(dmaru->iommu);
455 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
458 struct acpi_dmar_andd *andd = (void *)header;
460 /* Check for NUL termination within the designated length */
461 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
463 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
464 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
465 dmi_get_system_info(DMI_BIOS_VENDOR),
466 dmi_get_system_info(DMI_BIOS_VERSION),
467 dmi_get_system_info(DMI_PRODUCT_VERSION));
468 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
471 pr_info("ANDD device: %x name: %s\n", andd->device_number,
477 #ifdef CONFIG_ACPI_NUMA
478 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
480 struct acpi_dmar_rhsa *rhsa;
481 struct dmar_drhd_unit *drhd;
483 rhsa = (struct acpi_dmar_rhsa *)header;
484 for_each_drhd_unit(drhd) {
485 if (drhd->reg_base_addr == rhsa->base_address) {
486 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
488 if (!node_online(node))
490 drhd->iommu->node = node;
495 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
496 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
498 dmi_get_system_info(DMI_BIOS_VENDOR),
499 dmi_get_system_info(DMI_BIOS_VERSION),
500 dmi_get_system_info(DMI_PRODUCT_VERSION));
501 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
506 #define dmar_parse_one_rhsa dmar_res_noop
510 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
512 struct acpi_dmar_hardware_unit *drhd;
513 struct acpi_dmar_reserved_memory *rmrr;
514 struct acpi_dmar_atsr *atsr;
515 struct acpi_dmar_rhsa *rhsa;
517 switch (header->type) {
518 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
519 drhd = container_of(header, struct acpi_dmar_hardware_unit,
521 pr_info("DRHD base: %#016Lx flags: %#x\n",
522 (unsigned long long)drhd->address, drhd->flags);
524 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
525 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
527 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
528 (unsigned long long)rmrr->base_address,
529 (unsigned long long)rmrr->end_address);
531 case ACPI_DMAR_TYPE_ROOT_ATS:
532 atsr = container_of(header, struct acpi_dmar_atsr, header);
533 pr_info("ATSR flags: %#x\n", atsr->flags);
535 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
536 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
537 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
538 (unsigned long long)rhsa->base_address,
539 rhsa->proximity_domain);
541 case ACPI_DMAR_TYPE_NAMESPACE:
542 /* We don't print this here because we need to sanity-check
543 it first. So print it in dmar_parse_one_andd() instead. */
549 * dmar_table_detect - checks to see if the platform supports DMAR devices
551 static int __init dmar_table_detect(void)
553 acpi_status status = AE_OK;
555 /* if we could find DMAR table, then there are DMAR devices */
556 status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
558 if (ACPI_SUCCESS(status) && !dmar_tbl) {
559 pr_warn("Unable to map DMAR\n");
560 status = AE_NOT_FOUND;
563 return ACPI_SUCCESS(status) ? 0 : -ENOENT;
566 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
567 size_t len, struct dmar_res_callback *cb)
569 struct acpi_dmar_header *iter, *next;
570 struct acpi_dmar_header *end = ((void *)start) + len;
572 for (iter = start; iter < end; iter = next) {
573 next = (void *)iter + iter->length;
574 if (iter->length == 0) {
575 /* Avoid looping forever on bad ACPI tables */
576 pr_debug(FW_BUG "Invalid 0-length structure\n");
578 } else if (next > end) {
579 /* Avoid passing table end */
580 pr_warn(FW_BUG "Record passes table end\n");
585 dmar_table_print_dmar_entry(iter);
587 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
588 /* continue for forward compatibility */
589 pr_debug("Unknown DMAR structure type %d\n",
591 } else if (cb->cb[iter->type]) {
594 ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
597 } else if (!cb->ignore_unhandled) {
598 pr_warn("No handler for DMAR structure type %d\n",
607 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
608 struct dmar_res_callback *cb)
610 return dmar_walk_remapping_entries((void *)(dmar + 1),
611 dmar->header.length - sizeof(*dmar), cb);
615 * parse_dmar_table - parses the DMA reporting table
618 parse_dmar_table(void)
620 struct acpi_table_dmar *dmar;
623 struct dmar_res_callback cb = {
625 .ignore_unhandled = true,
626 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
627 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
628 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
629 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
630 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
631 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
635 * Do it again, earlier dmar_tbl mapping could be mapped with
641 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
642 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
644 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
646 dmar = (struct acpi_table_dmar *)dmar_tbl;
650 if (dmar->width < PAGE_SHIFT - 1) {
651 pr_warn("Invalid DMAR haw\n");
655 pr_info("Host address width %d\n", dmar->width + 1);
656 ret = dmar_walk_dmar_table(dmar, &cb);
657 if (ret == 0 && drhd_count == 0)
658 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
663 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
664 int cnt, struct pci_dev *dev)
670 for_each_active_dev_scope(devices, cnt, index, tmp)
671 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
674 /* Check our parent */
675 dev = dev->bus->self;
681 struct dmar_drhd_unit *
682 dmar_find_matched_drhd_unit(struct pci_dev *dev)
684 struct dmar_drhd_unit *dmaru;
685 struct acpi_dmar_hardware_unit *drhd;
687 dev = pci_physfn(dev);
690 for_each_drhd_unit(dmaru) {
691 drhd = container_of(dmaru->hdr,
692 struct acpi_dmar_hardware_unit,
695 if (dmaru->include_all &&
696 drhd->segment == pci_domain_nr(dev->bus))
699 if (dmar_pci_device_match(dmaru->devices,
700 dmaru->devices_cnt, dev))
710 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
711 struct acpi_device *adev)
713 struct dmar_drhd_unit *dmaru;
714 struct acpi_dmar_hardware_unit *drhd;
715 struct acpi_dmar_device_scope *scope;
718 struct acpi_dmar_pci_path *path;
720 for_each_drhd_unit(dmaru) {
721 drhd = container_of(dmaru->hdr,
722 struct acpi_dmar_hardware_unit,
725 for (scope = (void *)(drhd + 1);
726 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
727 scope = ((void *)scope) + scope->length) {
728 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
730 if (scope->enumeration_id != device_number)
733 path = (void *)(scope + 1);
734 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
735 dev_name(&adev->dev), dmaru->reg_base_addr,
736 scope->bus, path->device, path->function);
737 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
739 dmaru->devices[i].bus = scope->bus;
740 dmaru->devices[i].devfn = PCI_DEVFN(path->device,
742 rcu_assign_pointer(dmaru->devices[i].dev,
743 get_device(&adev->dev));
746 BUG_ON(i >= dmaru->devices_cnt);
749 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
750 device_number, dev_name(&adev->dev));
753 static int __init dmar_acpi_dev_scope_init(void)
755 struct acpi_dmar_andd *andd;
757 if (dmar_tbl == NULL)
760 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
761 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
762 andd = ((void *)andd) + andd->header.length) {
763 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
765 struct acpi_device *adev;
767 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
770 pr_err("Failed to find handle for ACPI object %s\n",
774 if (acpi_bus_get_device(h, &adev)) {
775 pr_err("Failed to get device for ACPI object %s\n",
779 dmar_acpi_insert_dev_scope(andd->device_number, adev);
785 int __init dmar_dev_scope_init(void)
787 struct pci_dev *dev = NULL;
788 struct dmar_pci_notify_info *info;
790 if (dmar_dev_scope_status != 1)
791 return dmar_dev_scope_status;
793 if (list_empty(&dmar_drhd_units)) {
794 dmar_dev_scope_status = -ENODEV;
796 dmar_dev_scope_status = 0;
798 dmar_acpi_dev_scope_init();
800 for_each_pci_dev(dev) {
804 info = dmar_alloc_pci_notify_info(dev,
805 BUS_NOTIFY_ADD_DEVICE);
807 return dmar_dev_scope_status;
809 dmar_pci_bus_add_dev(info);
810 dmar_free_pci_notify_info(info);
815 return dmar_dev_scope_status;
818 void __init dmar_register_bus_notifier(void)
820 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
824 int __init dmar_table_init(void)
826 static int dmar_table_initialized;
829 if (dmar_table_initialized == 0) {
830 ret = parse_dmar_table();
833 pr_info("Parse DMAR table failure.\n");
834 } else if (list_empty(&dmar_drhd_units)) {
835 pr_info("No DMAR devices found\n");
840 dmar_table_initialized = ret;
842 dmar_table_initialized = 1;
845 return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
848 static void warn_invalid_dmar(u64 addr, const char *message)
851 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
852 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
854 dmi_get_system_info(DMI_BIOS_VENDOR),
855 dmi_get_system_info(DMI_BIOS_VERSION),
856 dmi_get_system_info(DMI_PRODUCT_VERSION));
857 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
861 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
863 struct acpi_dmar_hardware_unit *drhd;
867 drhd = (void *)entry;
868 if (!drhd->address) {
869 warn_invalid_dmar(0, "");
874 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
876 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
878 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
882 cap = dmar_readq(addr + DMAR_CAP_REG);
883 ecap = dmar_readq(addr + DMAR_ECAP_REG);
888 early_iounmap(addr, VTD_PAGE_SIZE);
890 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
891 warn_invalid_dmar(drhd->address, " returns all ones");
898 int __init detect_intel_iommu(void)
901 struct dmar_res_callback validate_drhd_cb = {
902 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
903 .ignore_unhandled = true,
906 down_write(&dmar_global_lock);
907 ret = dmar_table_detect();
909 ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
911 if (!ret && !no_iommu && !iommu_detected && !dmar_disabled) {
913 /* Make sure ACS will be enabled */
919 x86_init.iommu.iommu_init = intel_iommu_init;
923 acpi_put_table(dmar_tbl);
926 up_write(&dmar_global_lock);
928 return ret ? ret : 1;
931 static void unmap_iommu(struct intel_iommu *iommu)
934 release_mem_region(iommu->reg_phys, iommu->reg_size);
938 * map_iommu: map the iommu's registers
939 * @iommu: the iommu to map
940 * @phys_addr: the physical address of the base resgister
942 * Memory map the iommu's registers. Start w/ a single page, and
943 * possibly expand if that turns out to be insufficent.
945 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
949 iommu->reg_phys = phys_addr;
950 iommu->reg_size = VTD_PAGE_SIZE;
952 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
953 pr_err("Can't reserve memory\n");
958 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
960 pr_err("Can't map the region\n");
965 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
966 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
968 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
970 warn_invalid_dmar(phys_addr, " returns all ones");
974 /* the registers might be more than one page */
975 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
976 cap_max_fault_reg_offset(iommu->cap));
977 map_size = VTD_PAGE_ALIGN(map_size);
978 if (map_size > iommu->reg_size) {
980 release_mem_region(iommu->reg_phys, iommu->reg_size);
981 iommu->reg_size = map_size;
982 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
984 pr_err("Can't reserve memory\n");
988 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
990 pr_err("Can't map the region\n");
1001 release_mem_region(iommu->reg_phys, iommu->reg_size);
1006 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
1008 iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
1009 DMAR_UNITS_SUPPORTED);
1010 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
1013 set_bit(iommu->seq_id, dmar_seq_ids);
1014 sprintf(iommu->name, "dmar%d", iommu->seq_id);
1017 return iommu->seq_id;
1020 static void dmar_free_seq_id(struct intel_iommu *iommu)
1022 if (iommu->seq_id >= 0) {
1023 clear_bit(iommu->seq_id, dmar_seq_ids);
1028 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1030 struct intel_iommu *iommu;
1036 if (!drhd->reg_base_addr) {
1037 warn_invalid_dmar(0, "");
1041 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1045 if (dmar_alloc_seq_id(iommu) < 0) {
1046 pr_err("Failed to allocate seq_id\n");
1051 err = map_iommu(iommu, drhd->reg_base_addr);
1053 pr_err("Failed to map %s\n", iommu->name);
1054 goto error_free_seq_id;
1058 agaw = iommu_calculate_agaw(iommu);
1060 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1064 msagaw = iommu_calculate_max_sagaw(iommu);
1066 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1071 iommu->msagaw = msagaw;
1072 iommu->segment = drhd->segment;
1076 ver = readl(iommu->reg + DMAR_VER_REG);
1077 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1079 (unsigned long long)drhd->reg_base_addr,
1080 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1081 (unsigned long long)iommu->cap,
1082 (unsigned long long)iommu->ecap);
1084 /* Reflect status in gcmd */
1085 sts = readl(iommu->reg + DMAR_GSTS_REG);
1086 if (sts & DMA_GSTS_IRES)
1087 iommu->gcmd |= DMA_GCMD_IRE;
1088 if (sts & DMA_GSTS_TES)
1089 iommu->gcmd |= DMA_GCMD_TE;
1090 if (sts & DMA_GSTS_QIES)
1091 iommu->gcmd |= DMA_GCMD_QIE;
1093 raw_spin_lock_init(&iommu->register_lock);
1095 if (intel_iommu_enabled) {
1096 err = iommu_device_sysfs_add(&iommu->iommu, NULL,
1102 iommu_device_set_ops(&iommu->iommu, &intel_iommu_ops);
1104 err = iommu_device_register(&iommu->iommu);
1109 drhd->iommu = iommu;
1116 dmar_free_seq_id(iommu);
1122 static void free_iommu(struct intel_iommu *iommu)
1124 if (intel_iommu_enabled) {
1125 iommu_device_unregister(&iommu->iommu);
1126 iommu_device_sysfs_remove(&iommu->iommu);
1130 if (iommu->pr_irq) {
1131 free_irq(iommu->pr_irq, iommu);
1132 dmar_free_hwirq(iommu->pr_irq);
1135 free_irq(iommu->irq, iommu);
1136 dmar_free_hwirq(iommu->irq);
1141 free_page((unsigned long)iommu->qi->desc);
1142 kfree(iommu->qi->desc_status);
1149 dmar_free_seq_id(iommu);
1154 * Reclaim all the submitted descriptors which have completed its work.
1156 static inline void reclaim_free_desc(struct q_inval *qi)
1158 while (qi->desc_status[qi->free_tail] == QI_DONE ||
1159 qi->desc_status[qi->free_tail] == QI_ABORT) {
1160 qi->desc_status[qi->free_tail] = QI_FREE;
1161 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1166 static int qi_check_fault(struct intel_iommu *iommu, int index)
1170 struct q_inval *qi = iommu->qi;
1171 int wait_index = (index + 1) % QI_LENGTH;
1173 if (qi->desc_status[wait_index] == QI_ABORT)
1176 fault = readl(iommu->reg + DMAR_FSTS_REG);
1179 * If IQE happens, the head points to the descriptor associated
1180 * with the error. No new descriptors are fetched until the IQE
1183 if (fault & DMA_FSTS_IQE) {
1184 head = readl(iommu->reg + DMAR_IQH_REG);
1185 if ((head >> DMAR_IQ_SHIFT) == index) {
1186 pr_err("VT-d detected invalid descriptor: "
1187 "low=%llx, high=%llx\n",
1188 (unsigned long long)qi->desc[index].low,
1189 (unsigned long long)qi->desc[index].high);
1190 memcpy(&qi->desc[index], &qi->desc[wait_index],
1191 sizeof(struct qi_desc));
1192 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1198 * If ITE happens, all pending wait_desc commands are aborted.
1199 * No new descriptors are fetched until the ITE is cleared.
1201 if (fault & DMA_FSTS_ITE) {
1202 head = readl(iommu->reg + DMAR_IQH_REG);
1203 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1205 tail = readl(iommu->reg + DMAR_IQT_REG);
1206 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1208 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1211 if (qi->desc_status[head] == QI_IN_USE)
1212 qi->desc_status[head] = QI_ABORT;
1213 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1214 } while (head != tail);
1216 if (qi->desc_status[wait_index] == QI_ABORT)
1220 if (fault & DMA_FSTS_ICE)
1221 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1227 * Submit the queued invalidation descriptor to the remapping
1228 * hardware unit and wait for its completion.
1230 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1233 struct q_inval *qi = iommu->qi;
1234 struct qi_desc *hw, wait_desc;
1235 int wait_index, index;
1236 unsigned long flags;
1246 raw_spin_lock_irqsave(&qi->q_lock, flags);
1247 while (qi->free_cnt < 3) {
1248 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1250 raw_spin_lock_irqsave(&qi->q_lock, flags);
1253 index = qi->free_head;
1254 wait_index = (index + 1) % QI_LENGTH;
1256 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1260 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
1261 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1262 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
1264 hw[wait_index] = wait_desc;
1266 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1270 * update the HW tail register indicating the presence of
1273 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
1275 while (qi->desc_status[wait_index] != QI_DONE) {
1277 * We will leave the interrupts disabled, to prevent interrupt
1278 * context to queue another cmd while a cmd is already submitted
1279 * and waiting for completion on this cpu. This is to avoid
1280 * a deadlock where the interrupt context can wait indefinitely
1281 * for free slots in the queue.
1283 rc = qi_check_fault(iommu, index);
1287 raw_spin_unlock(&qi->q_lock);
1289 raw_spin_lock(&qi->q_lock);
1292 qi->desc_status[index] = QI_DONE;
1294 reclaim_free_desc(qi);
1295 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1304 * Flush the global interrupt entry cache.
1306 void qi_global_iec(struct intel_iommu *iommu)
1308 struct qi_desc desc;
1310 desc.low = QI_IEC_TYPE;
1313 /* should never fail */
1314 qi_submit_sync(&desc, iommu);
1317 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1320 struct qi_desc desc;
1322 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1323 | QI_CC_GRAN(type) | QI_CC_TYPE;
1326 qi_submit_sync(&desc, iommu);
1329 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1330 unsigned int size_order, u64 type)
1334 struct qi_desc desc;
1337 if (cap_write_drain(iommu->cap))
1340 if (cap_read_drain(iommu->cap))
1343 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1344 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1345 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1346 | QI_IOTLB_AM(size_order);
1348 qi_submit_sync(&desc, iommu);
1351 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 pfsid,
1352 u16 qdep, u64 addr, unsigned mask)
1354 struct qi_desc desc;
1357 addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1358 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1360 desc.high = QI_DEV_IOTLB_ADDR(addr);
1362 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1365 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1366 QI_DIOTLB_TYPE | QI_DEV_IOTLB_PFSID(pfsid);
1368 qi_submit_sync(&desc, iommu);
1372 * Disable Queued Invalidation interface.
1374 void dmar_disable_qi(struct intel_iommu *iommu)
1376 unsigned long flags;
1378 cycles_t start_time = get_cycles();
1380 if (!ecap_qis(iommu->ecap))
1383 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1385 sts = readl(iommu->reg + DMAR_GSTS_REG);
1386 if (!(sts & DMA_GSTS_QIES))
1390 * Give a chance to HW to complete the pending invalidation requests.
1392 while ((readl(iommu->reg + DMAR_IQT_REG) !=
1393 readl(iommu->reg + DMAR_IQH_REG)) &&
1394 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1397 iommu->gcmd &= ~DMA_GCMD_QIE;
1398 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1400 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1401 !(sts & DMA_GSTS_QIES), sts);
1403 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1407 * Enable queued invalidation.
1409 static void __dmar_enable_qi(struct intel_iommu *iommu)
1412 unsigned long flags;
1413 struct q_inval *qi = iommu->qi;
1415 qi->free_head = qi->free_tail = 0;
1416 qi->free_cnt = QI_LENGTH;
1418 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1420 /* write zero to the tail reg */
1421 writel(0, iommu->reg + DMAR_IQT_REG);
1423 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1425 iommu->gcmd |= DMA_GCMD_QIE;
1426 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1428 /* Make sure hardware complete it */
1429 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1431 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1435 * Enable Queued Invalidation interface. This is a must to support
1436 * interrupt-remapping. Also used by DMA-remapping, which replaces
1437 * register based IOTLB invalidation.
1439 int dmar_enable_qi(struct intel_iommu *iommu)
1442 struct page *desc_page;
1444 if (!ecap_qis(iommu->ecap))
1448 * queued invalidation is already setup and enabled.
1453 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1460 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1467 qi->desc = page_address(desc_page);
1469 qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC);
1470 if (!qi->desc_status) {
1471 free_page((unsigned long) qi->desc);
1477 raw_spin_lock_init(&qi->q_lock);
1479 __dmar_enable_qi(iommu);
1484 /* iommu interrupt handling. Most stuff are MSI-like. */
1492 static const char *dma_remap_fault_reasons[] =
1495 "Present bit in root entry is clear",
1496 "Present bit in context entry is clear",
1497 "Invalid context entry",
1498 "Access beyond MGAW",
1499 "PTE Write access is not set",
1500 "PTE Read access is not set",
1501 "Next page table ptr is invalid",
1502 "Root table address invalid",
1503 "Context table ptr is invalid",
1504 "non-zero reserved fields in RTP",
1505 "non-zero reserved fields in CTP",
1506 "non-zero reserved fields in PTE",
1507 "PCE for translation request specifies blocking",
1510 static const char *irq_remap_fault_reasons[] =
1512 "Detected reserved fields in the decoded interrupt-remapped request",
1513 "Interrupt index exceeded the interrupt-remapping table size",
1514 "Present field in the IRTE entry is clear",
1515 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1516 "Detected reserved fields in the IRTE entry",
1517 "Blocked a compatibility format interrupt request",
1518 "Blocked an interrupt request due to source-id verification failure",
1521 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1523 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1524 ARRAY_SIZE(irq_remap_fault_reasons))) {
1525 *fault_type = INTR_REMAP;
1526 return irq_remap_fault_reasons[fault_reason - 0x20];
1527 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1528 *fault_type = DMA_REMAP;
1529 return dma_remap_fault_reasons[fault_reason];
1531 *fault_type = UNKNOWN;
1537 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1539 if (iommu->irq == irq)
1540 return DMAR_FECTL_REG;
1541 else if (iommu->pr_irq == irq)
1542 return DMAR_PECTL_REG;
1547 void dmar_msi_unmask(struct irq_data *data)
1549 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1550 int reg = dmar_msi_reg(iommu, data->irq);
1554 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1555 writel(0, iommu->reg + reg);
1556 /* Read a reg to force flush the post write */
1557 readl(iommu->reg + reg);
1558 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1561 void dmar_msi_mask(struct irq_data *data)
1563 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1564 int reg = dmar_msi_reg(iommu, data->irq);
1568 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1569 writel(DMA_FECTL_IM, iommu->reg + reg);
1570 /* Read a reg to force flush the post write */
1571 readl(iommu->reg + reg);
1572 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1575 void dmar_msi_write(int irq, struct msi_msg *msg)
1577 struct intel_iommu *iommu = irq_get_handler_data(irq);
1578 int reg = dmar_msi_reg(iommu, irq);
1581 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1582 writel(msg->data, iommu->reg + reg + 4);
1583 writel(msg->address_lo, iommu->reg + reg + 8);
1584 writel(msg->address_hi, iommu->reg + reg + 12);
1585 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1588 void dmar_msi_read(int irq, struct msi_msg *msg)
1590 struct intel_iommu *iommu = irq_get_handler_data(irq);
1591 int reg = dmar_msi_reg(iommu, irq);
1594 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1595 msg->data = readl(iommu->reg + reg + 4);
1596 msg->address_lo = readl(iommu->reg + reg + 8);
1597 msg->address_hi = readl(iommu->reg + reg + 12);
1598 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1601 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1602 u8 fault_reason, u16 source_id, unsigned long long addr)
1607 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1609 if (fault_type == INTR_REMAP)
1610 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1611 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1612 PCI_FUNC(source_id & 0xFF), addr >> 48,
1613 fault_reason, reason);
1615 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1616 type ? "DMA Read" : "DMA Write",
1617 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1618 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1622 #define PRIMARY_FAULT_REG_LEN (16)
1623 irqreturn_t dmar_fault(int irq, void *dev_id)
1625 struct intel_iommu *iommu = dev_id;
1626 int reg, fault_index;
1629 static DEFINE_RATELIMIT_STATE(rs,
1630 DEFAULT_RATELIMIT_INTERVAL,
1631 DEFAULT_RATELIMIT_BURST);
1633 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1634 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1635 if (fault_status && __ratelimit(&rs))
1636 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1638 /* TBD: ignore advanced fault log currently */
1639 if (!(fault_status & DMA_FSTS_PPF))
1642 fault_index = dma_fsts_fault_record_index(fault_status);
1643 reg = cap_fault_reg_offset(iommu->cap);
1645 /* Disable printing, simply clear the fault when ratelimited */
1646 bool ratelimited = !__ratelimit(&rs);
1653 /* highest 32 bits */
1654 data = readl(iommu->reg + reg +
1655 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1656 if (!(data & DMA_FRCD_F))
1660 fault_reason = dma_frcd_fault_reason(data);
1661 type = dma_frcd_type(data);
1663 data = readl(iommu->reg + reg +
1664 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1665 source_id = dma_frcd_source_id(data);
1667 guest_addr = dmar_readq(iommu->reg + reg +
1668 fault_index * PRIMARY_FAULT_REG_LEN);
1669 guest_addr = dma_frcd_page_addr(guest_addr);
1672 /* clear the fault */
1673 writel(DMA_FRCD_F, iommu->reg + reg +
1674 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1676 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1679 dmar_fault_do_one(iommu, type, fault_reason,
1680 source_id, guest_addr);
1683 if (fault_index >= cap_num_fault_regs(iommu->cap))
1685 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1688 writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO,
1689 iommu->reg + DMAR_FSTS_REG);
1692 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1696 int dmar_set_interrupt(struct intel_iommu *iommu)
1701 * Check if the fault interrupt is already initialized.
1706 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1710 pr_err("No free IRQ vectors\n");
1714 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1716 pr_err("Can't request irq\n");
1720 int __init enable_drhd_fault_handling(void)
1722 struct dmar_drhd_unit *drhd;
1723 struct intel_iommu *iommu;
1726 * Enable fault control interrupt.
1728 for_each_iommu(iommu, drhd) {
1730 int ret = dmar_set_interrupt(iommu);
1733 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1734 (unsigned long long)drhd->reg_base_addr, ret);
1739 * Clear any previous faults.
1741 dmar_fault(iommu->irq, iommu);
1742 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1743 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1750 * Re-enable Queued Invalidation interface.
1752 int dmar_reenable_qi(struct intel_iommu *iommu)
1754 if (!ecap_qis(iommu->ecap))
1761 * First disable queued invalidation.
1763 dmar_disable_qi(iommu);
1765 * Then enable queued invalidation again. Since there is no pending
1766 * invalidation requests now, it's safe to re-enable queued
1769 __dmar_enable_qi(iommu);
1775 * Check interrupt remapping support in DMAR table description.
1777 int __init dmar_ir_support(void)
1779 struct acpi_table_dmar *dmar;
1780 dmar = (struct acpi_table_dmar *)dmar_tbl;
1783 return dmar->flags & 0x1;
1786 /* Check whether DMAR units are in use */
1787 static inline bool dmar_in_use(void)
1789 return irq_remapping_enabled || intel_iommu_enabled;
1792 static int __init dmar_free_unused_resources(void)
1794 struct dmar_drhd_unit *dmaru, *dmaru_n;
1799 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1800 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1802 down_write(&dmar_global_lock);
1803 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1804 list_del(&dmaru->list);
1805 dmar_free_drhd(dmaru);
1807 up_write(&dmar_global_lock);
1812 late_initcall(dmar_free_unused_resources);
1813 IOMMU_INIT_POST(detect_intel_iommu);
1816 * DMAR Hotplug Support
1817 * For more details, please refer to Intel(R) Virtualization Technology
1818 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1819 * "Remapping Hardware Unit Hot Plug".
1821 static guid_t dmar_hp_guid =
1822 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
1823 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
1826 * Currently there's only one revision and BIOS will not check the revision id,
1827 * so use 0 for safety.
1829 #define DMAR_DSM_REV_ID 0
1830 #define DMAR_DSM_FUNC_DRHD 1
1831 #define DMAR_DSM_FUNC_ATSR 2
1832 #define DMAR_DSM_FUNC_RHSA 3
1834 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1836 return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func);
1839 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1840 dmar_res_handler_t handler, void *arg)
1843 union acpi_object *obj;
1844 struct acpi_dmar_header *start;
1845 struct dmar_res_callback callback;
1846 static int res_type[] = {
1847 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1848 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1849 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1852 if (!dmar_detect_dsm(handle, func))
1855 obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID,
1856 func, NULL, ACPI_TYPE_BUFFER);
1860 memset(&callback, 0, sizeof(callback));
1861 callback.cb[res_type[func]] = handler;
1862 callback.arg[res_type[func]] = arg;
1863 start = (struct acpi_dmar_header *)obj->buffer.pointer;
1864 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1871 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1874 struct dmar_drhd_unit *dmaru;
1876 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1880 ret = dmar_ir_hotplug(dmaru, true);
1882 ret = dmar_iommu_hotplug(dmaru, true);
1887 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
1891 struct dmar_drhd_unit *dmaru;
1893 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1898 * All PCI devices managed by this unit should have been destroyed.
1900 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
1901 for_each_active_dev_scope(dmaru->devices,
1902 dmaru->devices_cnt, i, dev)
1906 ret = dmar_ir_hotplug(dmaru, false);
1908 ret = dmar_iommu_hotplug(dmaru, false);
1913 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
1915 struct dmar_drhd_unit *dmaru;
1917 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1919 list_del_rcu(&dmaru->list);
1921 dmar_free_drhd(dmaru);
1927 static int dmar_hotplug_insert(acpi_handle handle)
1932 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1933 &dmar_validate_one_drhd, (void *)1);
1937 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1938 &dmar_parse_one_drhd, (void *)&drhd_count);
1939 if (ret == 0 && drhd_count == 0) {
1940 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
1946 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
1947 &dmar_parse_one_rhsa, NULL);
1951 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1952 &dmar_parse_one_atsr, NULL);
1956 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1957 &dmar_hp_add_drhd, NULL);
1961 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1962 &dmar_hp_remove_drhd, NULL);
1964 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1965 &dmar_release_one_atsr, NULL);
1967 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1968 &dmar_hp_release_drhd, NULL);
1973 static int dmar_hotplug_remove(acpi_handle handle)
1977 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1978 &dmar_check_one_atsr, NULL);
1982 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1983 &dmar_hp_remove_drhd, NULL);
1985 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1986 &dmar_release_one_atsr, NULL));
1987 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1988 &dmar_hp_release_drhd, NULL));
1990 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1991 &dmar_hp_add_drhd, NULL);
1997 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
1998 void *context, void **retval)
2000 acpi_handle *phdl = retval;
2002 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2004 return AE_CTRL_TERMINATE;
2010 static int dmar_device_hotplug(acpi_handle handle, bool insert)
2013 acpi_handle tmp = NULL;
2019 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2022 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2024 dmar_get_dsm_handle,
2026 if (ACPI_FAILURE(status)) {
2027 pr_warn("Failed to locate _DSM method.\n");
2034 down_write(&dmar_global_lock);
2036 ret = dmar_hotplug_insert(tmp);
2038 ret = dmar_hotplug_remove(tmp);
2039 up_write(&dmar_global_lock);
2044 int dmar_device_add(acpi_handle handle)
2046 return dmar_device_hotplug(handle, true);
2049 int dmar_device_remove(acpi_handle handle)
2051 return dmar_device_hotplug(handle, false);