mm, debug: replace dump_flags() with the new printk formats
[linux/fpc-iii.git] / drivers / iommu / dmar.c
blob8ffd7568fc919e32b97354b2777db21be263d37f
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 acpi_size dmar_tbl_size;
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 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
81 * add INCLUDE_ALL at the tail, so scan the list will find it at
82 * the very end.
84 if (drhd->include_all)
85 list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
86 else
87 list_add_rcu(&drhd->list, &dmar_drhd_units);
90 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
92 struct acpi_dmar_device_scope *scope;
94 *cnt = 0;
95 while (start < end) {
96 scope = start;
97 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
98 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
99 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
100 (*cnt)++;
101 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
102 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
103 pr_warn("Unsupported device scope\n");
105 start += scope->length;
107 if (*cnt == 0)
108 return NULL;
110 return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
113 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
115 int i;
116 struct device *tmp_dev;
118 if (*devices && *cnt) {
119 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
120 put_device(tmp_dev);
121 kfree(*devices);
124 *devices = NULL;
125 *cnt = 0;
128 /* Optimize out kzalloc()/kfree() for normal cases */
129 static char dmar_pci_notify_info_buf[64];
131 static struct dmar_pci_notify_info *
132 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
134 int level = 0;
135 size_t size;
136 struct pci_dev *tmp;
137 struct dmar_pci_notify_info *info;
139 BUG_ON(dev->is_virtfn);
141 /* Only generate path[] for device addition event */
142 if (event == BUS_NOTIFY_ADD_DEVICE)
143 for (tmp = dev; tmp; tmp = tmp->bus->self)
144 level++;
146 size = sizeof(*info) + level * sizeof(struct acpi_dmar_pci_path);
147 if (size <= sizeof(dmar_pci_notify_info_buf)) {
148 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
149 } else {
150 info = kzalloc(size, GFP_KERNEL);
151 if (!info) {
152 pr_warn("Out of memory when allocating notify_info "
153 "for %s.\n", pci_name(dev));
154 if (dmar_dev_scope_status == 0)
155 dmar_dev_scope_status = -ENOMEM;
156 return NULL;
160 info->event = event;
161 info->dev = dev;
162 info->seg = pci_domain_nr(dev->bus);
163 info->level = level;
164 if (event == BUS_NOTIFY_ADD_DEVICE) {
165 for (tmp = dev; tmp; tmp = tmp->bus->self) {
166 level--;
167 info->path[level].bus = tmp->bus->number;
168 info->path[level].device = PCI_SLOT(tmp->devfn);
169 info->path[level].function = PCI_FUNC(tmp->devfn);
170 if (pci_is_root_bus(tmp->bus))
171 info->bus = tmp->bus->number;
175 return info;
178 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
180 if ((void *)info != dmar_pci_notify_info_buf)
181 kfree(info);
184 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
185 struct acpi_dmar_pci_path *path, int count)
187 int i;
189 if (info->bus != bus)
190 goto fallback;
191 if (info->level != count)
192 goto fallback;
194 for (i = 0; i < count; i++) {
195 if (path[i].device != info->path[i].device ||
196 path[i].function != info->path[i].function)
197 goto fallback;
200 return true;
202 fallback:
204 if (count != 1)
205 return false;
207 i = info->level - 1;
208 if (bus == info->path[i].bus &&
209 path[0].device == info->path[i].device &&
210 path[0].function == info->path[i].function) {
211 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
212 bus, path[0].device, path[0].function);
213 return true;
216 return false;
219 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
220 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
221 void *start, void*end, u16 segment,
222 struct dmar_dev_scope *devices,
223 int devices_cnt)
225 int i, level;
226 struct device *tmp, *dev = &info->dev->dev;
227 struct acpi_dmar_device_scope *scope;
228 struct acpi_dmar_pci_path *path;
230 if (segment != info->seg)
231 return 0;
233 for (; start < end; start += scope->length) {
234 scope = start;
235 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
236 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
237 continue;
239 path = (struct acpi_dmar_pci_path *)(scope + 1);
240 level = (scope->length - sizeof(*scope)) / sizeof(*path);
241 if (!dmar_match_pci_path(info, scope->bus, path, level))
242 continue;
244 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT) ^
245 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL)) {
246 pr_warn("Device scope type does not match for %s\n",
247 pci_name(info->dev));
248 return -EINVAL;
251 for_each_dev_scope(devices, devices_cnt, i, tmp)
252 if (tmp == NULL) {
253 devices[i].bus = info->dev->bus->number;
254 devices[i].devfn = info->dev->devfn;
255 rcu_assign_pointer(devices[i].dev,
256 get_device(dev));
257 return 1;
259 BUG_ON(i >= devices_cnt);
262 return 0;
265 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
266 struct dmar_dev_scope *devices, int count)
268 int index;
269 struct device *tmp;
271 if (info->seg != segment)
272 return 0;
274 for_each_active_dev_scope(devices, count, index, tmp)
275 if (tmp == &info->dev->dev) {
276 RCU_INIT_POINTER(devices[index].dev, NULL);
277 synchronize_rcu();
278 put_device(tmp);
279 return 1;
282 return 0;
285 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
287 int ret = 0;
288 struct dmar_drhd_unit *dmaru;
289 struct acpi_dmar_hardware_unit *drhd;
291 for_each_drhd_unit(dmaru) {
292 if (dmaru->include_all)
293 continue;
295 drhd = container_of(dmaru->hdr,
296 struct acpi_dmar_hardware_unit, header);
297 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
298 ((void *)drhd) + drhd->header.length,
299 dmaru->segment,
300 dmaru->devices, dmaru->devices_cnt);
301 if (ret != 0)
302 break;
304 if (ret >= 0)
305 ret = dmar_iommu_notify_scope_dev(info);
306 if (ret < 0 && dmar_dev_scope_status == 0)
307 dmar_dev_scope_status = ret;
309 return ret;
312 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
314 struct dmar_drhd_unit *dmaru;
316 for_each_drhd_unit(dmaru)
317 if (dmar_remove_dev_scope(info, dmaru->segment,
318 dmaru->devices, dmaru->devices_cnt))
319 break;
320 dmar_iommu_notify_scope_dev(info);
323 static int dmar_pci_bus_notifier(struct notifier_block *nb,
324 unsigned long action, void *data)
326 struct pci_dev *pdev = to_pci_dev(data);
327 struct dmar_pci_notify_info *info;
329 /* Only care about add/remove events for physical functions */
330 if (pdev->is_virtfn)
331 return NOTIFY_DONE;
332 if (action != BUS_NOTIFY_ADD_DEVICE &&
333 action != BUS_NOTIFY_REMOVED_DEVICE)
334 return NOTIFY_DONE;
336 info = dmar_alloc_pci_notify_info(pdev, action);
337 if (!info)
338 return NOTIFY_DONE;
340 down_write(&dmar_global_lock);
341 if (action == BUS_NOTIFY_ADD_DEVICE)
342 dmar_pci_bus_add_dev(info);
343 else if (action == BUS_NOTIFY_REMOVED_DEVICE)
344 dmar_pci_bus_del_dev(info);
345 up_write(&dmar_global_lock);
347 dmar_free_pci_notify_info(info);
349 return NOTIFY_OK;
352 static struct notifier_block dmar_pci_bus_nb = {
353 .notifier_call = dmar_pci_bus_notifier,
354 .priority = INT_MIN,
357 static struct dmar_drhd_unit *
358 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
360 struct dmar_drhd_unit *dmaru;
362 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
363 if (dmaru->segment == drhd->segment &&
364 dmaru->reg_base_addr == drhd->address)
365 return dmaru;
367 return NULL;
371 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
372 * structure which uniquely represent one DMA remapping hardware unit
373 * present in the platform
375 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
377 struct acpi_dmar_hardware_unit *drhd;
378 struct dmar_drhd_unit *dmaru;
379 int ret = 0;
381 drhd = (struct acpi_dmar_hardware_unit *)header;
382 dmaru = dmar_find_dmaru(drhd);
383 if (dmaru)
384 goto out;
386 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
387 if (!dmaru)
388 return -ENOMEM;
391 * If header is allocated from slab by ACPI _DSM method, we need to
392 * copy the content because the memory buffer will be freed on return.
394 dmaru->hdr = (void *)(dmaru + 1);
395 memcpy(dmaru->hdr, header, header->length);
396 dmaru->reg_base_addr = drhd->address;
397 dmaru->segment = drhd->segment;
398 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
399 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
400 ((void *)drhd) + drhd->header.length,
401 &dmaru->devices_cnt);
402 if (dmaru->devices_cnt && dmaru->devices == NULL) {
403 kfree(dmaru);
404 return -ENOMEM;
407 ret = alloc_iommu(dmaru);
408 if (ret) {
409 dmar_free_dev_scope(&dmaru->devices,
410 &dmaru->devices_cnt);
411 kfree(dmaru);
412 return ret;
414 dmar_register_drhd_unit(dmaru);
416 out:
417 if (arg)
418 (*(int *)arg)++;
420 return 0;
423 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
425 if (dmaru->devices && dmaru->devices_cnt)
426 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
427 if (dmaru->iommu)
428 free_iommu(dmaru->iommu);
429 kfree(dmaru);
432 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
433 void *arg)
435 struct acpi_dmar_andd *andd = (void *)header;
437 /* Check for NUL termination within the designated length */
438 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
439 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND,
440 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
441 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
442 dmi_get_system_info(DMI_BIOS_VENDOR),
443 dmi_get_system_info(DMI_BIOS_VERSION),
444 dmi_get_system_info(DMI_PRODUCT_VERSION));
445 return -EINVAL;
447 pr_info("ANDD device: %x name: %s\n", andd->device_number,
448 andd->device_name);
450 return 0;
453 #ifdef CONFIG_ACPI_NUMA
454 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
456 struct acpi_dmar_rhsa *rhsa;
457 struct dmar_drhd_unit *drhd;
459 rhsa = (struct acpi_dmar_rhsa *)header;
460 for_each_drhd_unit(drhd) {
461 if (drhd->reg_base_addr == rhsa->base_address) {
462 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
464 if (!node_online(node))
465 node = -1;
466 drhd->iommu->node = node;
467 return 0;
470 WARN_TAINT(
471 1, TAINT_FIRMWARE_WORKAROUND,
472 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
473 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
474 drhd->reg_base_addr,
475 dmi_get_system_info(DMI_BIOS_VENDOR),
476 dmi_get_system_info(DMI_BIOS_VERSION),
477 dmi_get_system_info(DMI_PRODUCT_VERSION));
479 return 0;
481 #else
482 #define dmar_parse_one_rhsa dmar_res_noop
483 #endif
485 static void __init
486 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
488 struct acpi_dmar_hardware_unit *drhd;
489 struct acpi_dmar_reserved_memory *rmrr;
490 struct acpi_dmar_atsr *atsr;
491 struct acpi_dmar_rhsa *rhsa;
493 switch (header->type) {
494 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
495 drhd = container_of(header, struct acpi_dmar_hardware_unit,
496 header);
497 pr_info("DRHD base: %#016Lx flags: %#x\n",
498 (unsigned long long)drhd->address, drhd->flags);
499 break;
500 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
501 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
502 header);
503 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
504 (unsigned long long)rmrr->base_address,
505 (unsigned long long)rmrr->end_address);
506 break;
507 case ACPI_DMAR_TYPE_ROOT_ATS:
508 atsr = container_of(header, struct acpi_dmar_atsr, header);
509 pr_info("ATSR flags: %#x\n", atsr->flags);
510 break;
511 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
512 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
513 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
514 (unsigned long long)rhsa->base_address,
515 rhsa->proximity_domain);
516 break;
517 case ACPI_DMAR_TYPE_NAMESPACE:
518 /* We don't print this here because we need to sanity-check
519 it first. So print it in dmar_parse_one_andd() instead. */
520 break;
525 * dmar_table_detect - checks to see if the platform supports DMAR devices
527 static int __init dmar_table_detect(void)
529 acpi_status status = AE_OK;
531 /* if we could find DMAR table, then there are DMAR devices */
532 status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
533 (struct acpi_table_header **)&dmar_tbl,
534 &dmar_tbl_size);
536 if (ACPI_SUCCESS(status) && !dmar_tbl) {
537 pr_warn("Unable to map DMAR\n");
538 status = AE_NOT_FOUND;
541 return (ACPI_SUCCESS(status) ? 1 : 0);
544 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
545 size_t len, struct dmar_res_callback *cb)
547 int ret = 0;
548 struct acpi_dmar_header *iter, *next;
549 struct acpi_dmar_header *end = ((void *)start) + len;
551 for (iter = start; iter < end && ret == 0; iter = next) {
552 next = (void *)iter + iter->length;
553 if (iter->length == 0) {
554 /* Avoid looping forever on bad ACPI tables */
555 pr_debug(FW_BUG "Invalid 0-length structure\n");
556 break;
557 } else if (next > end) {
558 /* Avoid passing table end */
559 pr_warn(FW_BUG "Record passes table end\n");
560 ret = -EINVAL;
561 break;
564 if (cb->print_entry)
565 dmar_table_print_dmar_entry(iter);
567 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
568 /* continue for forward compatibility */
569 pr_debug("Unknown DMAR structure type %d\n",
570 iter->type);
571 } else if (cb->cb[iter->type]) {
572 ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
573 } else if (!cb->ignore_unhandled) {
574 pr_warn("No handler for DMAR structure type %d\n",
575 iter->type);
576 ret = -EINVAL;
580 return ret;
583 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
584 struct dmar_res_callback *cb)
586 return dmar_walk_remapping_entries((void *)(dmar + 1),
587 dmar->header.length - sizeof(*dmar), cb);
591 * parse_dmar_table - parses the DMA reporting table
593 static int __init
594 parse_dmar_table(void)
596 struct acpi_table_dmar *dmar;
597 int ret = 0;
598 int drhd_count = 0;
599 struct dmar_res_callback cb = {
600 .print_entry = true,
601 .ignore_unhandled = true,
602 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
603 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
604 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
605 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
606 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
607 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
611 * Do it again, earlier dmar_tbl mapping could be mapped with
612 * fixed map.
614 dmar_table_detect();
617 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
618 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
620 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
622 dmar = (struct acpi_table_dmar *)dmar_tbl;
623 if (!dmar)
624 return -ENODEV;
626 if (dmar->width < PAGE_SHIFT - 1) {
627 pr_warn("Invalid DMAR haw\n");
628 return -EINVAL;
631 pr_info("Host address width %d\n", dmar->width + 1);
632 ret = dmar_walk_dmar_table(dmar, &cb);
633 if (ret == 0 && drhd_count == 0)
634 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
636 return ret;
639 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
640 int cnt, struct pci_dev *dev)
642 int index;
643 struct device *tmp;
645 while (dev) {
646 for_each_active_dev_scope(devices, cnt, index, tmp)
647 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
648 return 1;
650 /* Check our parent */
651 dev = dev->bus->self;
654 return 0;
657 struct dmar_drhd_unit *
658 dmar_find_matched_drhd_unit(struct pci_dev *dev)
660 struct dmar_drhd_unit *dmaru;
661 struct acpi_dmar_hardware_unit *drhd;
663 dev = pci_physfn(dev);
665 rcu_read_lock();
666 for_each_drhd_unit(dmaru) {
667 drhd = container_of(dmaru->hdr,
668 struct acpi_dmar_hardware_unit,
669 header);
671 if (dmaru->include_all &&
672 drhd->segment == pci_domain_nr(dev->bus))
673 goto out;
675 if (dmar_pci_device_match(dmaru->devices,
676 dmaru->devices_cnt, dev))
677 goto out;
679 dmaru = NULL;
680 out:
681 rcu_read_unlock();
683 return dmaru;
686 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
687 struct acpi_device *adev)
689 struct dmar_drhd_unit *dmaru;
690 struct acpi_dmar_hardware_unit *drhd;
691 struct acpi_dmar_device_scope *scope;
692 struct device *tmp;
693 int i;
694 struct acpi_dmar_pci_path *path;
696 for_each_drhd_unit(dmaru) {
697 drhd = container_of(dmaru->hdr,
698 struct acpi_dmar_hardware_unit,
699 header);
701 for (scope = (void *)(drhd + 1);
702 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
703 scope = ((void *)scope) + scope->length) {
704 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
705 continue;
706 if (scope->enumeration_id != device_number)
707 continue;
709 path = (void *)(scope + 1);
710 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
711 dev_name(&adev->dev), dmaru->reg_base_addr,
712 scope->bus, path->device, path->function);
713 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
714 if (tmp == NULL) {
715 dmaru->devices[i].bus = scope->bus;
716 dmaru->devices[i].devfn = PCI_DEVFN(path->device,
717 path->function);
718 rcu_assign_pointer(dmaru->devices[i].dev,
719 get_device(&adev->dev));
720 return;
722 BUG_ON(i >= dmaru->devices_cnt);
725 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
726 device_number, dev_name(&adev->dev));
729 static int __init dmar_acpi_dev_scope_init(void)
731 struct acpi_dmar_andd *andd;
733 if (dmar_tbl == NULL)
734 return -ENODEV;
736 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
737 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
738 andd = ((void *)andd) + andd->header.length) {
739 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
740 acpi_handle h;
741 struct acpi_device *adev;
743 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
744 andd->device_name,
745 &h))) {
746 pr_err("Failed to find handle for ACPI object %s\n",
747 andd->device_name);
748 continue;
750 if (acpi_bus_get_device(h, &adev)) {
751 pr_err("Failed to get device for ACPI object %s\n",
752 andd->device_name);
753 continue;
755 dmar_acpi_insert_dev_scope(andd->device_number, adev);
758 return 0;
761 int __init dmar_dev_scope_init(void)
763 struct pci_dev *dev = NULL;
764 struct dmar_pci_notify_info *info;
766 if (dmar_dev_scope_status != 1)
767 return dmar_dev_scope_status;
769 if (list_empty(&dmar_drhd_units)) {
770 dmar_dev_scope_status = -ENODEV;
771 } else {
772 dmar_dev_scope_status = 0;
774 dmar_acpi_dev_scope_init();
776 for_each_pci_dev(dev) {
777 if (dev->is_virtfn)
778 continue;
780 info = dmar_alloc_pci_notify_info(dev,
781 BUS_NOTIFY_ADD_DEVICE);
782 if (!info) {
783 return dmar_dev_scope_status;
784 } else {
785 dmar_pci_bus_add_dev(info);
786 dmar_free_pci_notify_info(info);
790 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
793 return dmar_dev_scope_status;
797 int __init dmar_table_init(void)
799 static int dmar_table_initialized;
800 int ret;
802 if (dmar_table_initialized == 0) {
803 ret = parse_dmar_table();
804 if (ret < 0) {
805 if (ret != -ENODEV)
806 pr_info("Parse DMAR table failure.\n");
807 } else if (list_empty(&dmar_drhd_units)) {
808 pr_info("No DMAR devices found\n");
809 ret = -ENODEV;
812 if (ret < 0)
813 dmar_table_initialized = ret;
814 else
815 dmar_table_initialized = 1;
818 return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
821 static void warn_invalid_dmar(u64 addr, const char *message)
823 WARN_TAINT_ONCE(
824 1, TAINT_FIRMWARE_WORKAROUND,
825 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
826 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
827 addr, message,
828 dmi_get_system_info(DMI_BIOS_VENDOR),
829 dmi_get_system_info(DMI_BIOS_VERSION),
830 dmi_get_system_info(DMI_PRODUCT_VERSION));
833 static int __ref
834 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
836 struct acpi_dmar_hardware_unit *drhd;
837 void __iomem *addr;
838 u64 cap, ecap;
840 drhd = (void *)entry;
841 if (!drhd->address) {
842 warn_invalid_dmar(0, "");
843 return -EINVAL;
846 if (arg)
847 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
848 else
849 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
850 if (!addr) {
851 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
852 return -EINVAL;
855 cap = dmar_readq(addr + DMAR_CAP_REG);
856 ecap = dmar_readq(addr + DMAR_ECAP_REG);
858 if (arg)
859 iounmap(addr);
860 else
861 early_iounmap(addr, VTD_PAGE_SIZE);
863 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
864 warn_invalid_dmar(drhd->address, " returns all ones");
865 return -EINVAL;
868 return 0;
871 int __init detect_intel_iommu(void)
873 int ret;
874 struct dmar_res_callback validate_drhd_cb = {
875 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
876 .ignore_unhandled = true,
879 down_write(&dmar_global_lock);
880 ret = dmar_table_detect();
881 if (ret)
882 ret = !dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
883 &validate_drhd_cb);
884 if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
885 iommu_detected = 1;
886 /* Make sure ACS will be enabled */
887 pci_request_acs();
890 #ifdef CONFIG_X86
891 if (ret)
892 x86_init.iommu.iommu_init = intel_iommu_init;
893 #endif
895 early_acpi_os_unmap_memory((void __iomem *)dmar_tbl, dmar_tbl_size);
896 dmar_tbl = NULL;
897 up_write(&dmar_global_lock);
899 return ret ? 1 : -ENODEV;
903 static void unmap_iommu(struct intel_iommu *iommu)
905 iounmap(iommu->reg);
906 release_mem_region(iommu->reg_phys, iommu->reg_size);
910 * map_iommu: map the iommu's registers
911 * @iommu: the iommu to map
912 * @phys_addr: the physical address of the base resgister
914 * Memory map the iommu's registers. Start w/ a single page, and
915 * possibly expand if that turns out to be insufficent.
917 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
919 int map_size, err=0;
921 iommu->reg_phys = phys_addr;
922 iommu->reg_size = VTD_PAGE_SIZE;
924 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
925 pr_err("Can't reserve memory\n");
926 err = -EBUSY;
927 goto out;
930 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
931 if (!iommu->reg) {
932 pr_err("Can't map the region\n");
933 err = -ENOMEM;
934 goto release;
937 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
938 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
940 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
941 err = -EINVAL;
942 warn_invalid_dmar(phys_addr, " returns all ones");
943 goto unmap;
946 /* the registers might be more than one page */
947 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
948 cap_max_fault_reg_offset(iommu->cap));
949 map_size = VTD_PAGE_ALIGN(map_size);
950 if (map_size > iommu->reg_size) {
951 iounmap(iommu->reg);
952 release_mem_region(iommu->reg_phys, iommu->reg_size);
953 iommu->reg_size = map_size;
954 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
955 iommu->name)) {
956 pr_err("Can't reserve memory\n");
957 err = -EBUSY;
958 goto out;
960 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
961 if (!iommu->reg) {
962 pr_err("Can't map the region\n");
963 err = -ENOMEM;
964 goto release;
967 err = 0;
968 goto out;
970 unmap:
971 iounmap(iommu->reg);
972 release:
973 release_mem_region(iommu->reg_phys, iommu->reg_size);
974 out:
975 return err;
978 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
980 iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
981 DMAR_UNITS_SUPPORTED);
982 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
983 iommu->seq_id = -1;
984 } else {
985 set_bit(iommu->seq_id, dmar_seq_ids);
986 sprintf(iommu->name, "dmar%d", iommu->seq_id);
989 return iommu->seq_id;
992 static void dmar_free_seq_id(struct intel_iommu *iommu)
994 if (iommu->seq_id >= 0) {
995 clear_bit(iommu->seq_id, dmar_seq_ids);
996 iommu->seq_id = -1;
1000 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1002 struct intel_iommu *iommu;
1003 u32 ver, sts;
1004 int agaw = 0;
1005 int msagaw = 0;
1006 int err;
1008 if (!drhd->reg_base_addr) {
1009 warn_invalid_dmar(0, "");
1010 return -EINVAL;
1013 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1014 if (!iommu)
1015 return -ENOMEM;
1017 if (dmar_alloc_seq_id(iommu) < 0) {
1018 pr_err("Failed to allocate seq_id\n");
1019 err = -ENOSPC;
1020 goto error;
1023 err = map_iommu(iommu, drhd->reg_base_addr);
1024 if (err) {
1025 pr_err("Failed to map %s\n", iommu->name);
1026 goto error_free_seq_id;
1029 err = -EINVAL;
1030 agaw = iommu_calculate_agaw(iommu);
1031 if (agaw < 0) {
1032 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1033 iommu->seq_id);
1034 goto err_unmap;
1036 msagaw = iommu_calculate_max_sagaw(iommu);
1037 if (msagaw < 0) {
1038 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1039 iommu->seq_id);
1040 goto err_unmap;
1042 iommu->agaw = agaw;
1043 iommu->msagaw = msagaw;
1044 iommu->segment = drhd->segment;
1046 iommu->node = -1;
1048 ver = readl(iommu->reg + DMAR_VER_REG);
1049 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1050 iommu->name,
1051 (unsigned long long)drhd->reg_base_addr,
1052 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1053 (unsigned long long)iommu->cap,
1054 (unsigned long long)iommu->ecap);
1056 /* Reflect status in gcmd */
1057 sts = readl(iommu->reg + DMAR_GSTS_REG);
1058 if (sts & DMA_GSTS_IRES)
1059 iommu->gcmd |= DMA_GCMD_IRE;
1060 if (sts & DMA_GSTS_TES)
1061 iommu->gcmd |= DMA_GCMD_TE;
1062 if (sts & DMA_GSTS_QIES)
1063 iommu->gcmd |= DMA_GCMD_QIE;
1065 raw_spin_lock_init(&iommu->register_lock);
1067 if (intel_iommu_enabled) {
1068 iommu->iommu_dev = iommu_device_create(NULL, iommu,
1069 intel_iommu_groups,
1070 "%s", iommu->name);
1072 if (IS_ERR(iommu->iommu_dev)) {
1073 err = PTR_ERR(iommu->iommu_dev);
1074 goto err_unmap;
1078 drhd->iommu = iommu;
1080 return 0;
1082 err_unmap:
1083 unmap_iommu(iommu);
1084 error_free_seq_id:
1085 dmar_free_seq_id(iommu);
1086 error:
1087 kfree(iommu);
1088 return err;
1091 static void free_iommu(struct intel_iommu *iommu)
1093 iommu_device_destroy(iommu->iommu_dev);
1095 if (iommu->irq) {
1096 if (iommu->pr_irq) {
1097 free_irq(iommu->pr_irq, iommu);
1098 dmar_free_hwirq(iommu->pr_irq);
1099 iommu->pr_irq = 0;
1101 free_irq(iommu->irq, iommu);
1102 dmar_free_hwirq(iommu->irq);
1103 iommu->irq = 0;
1106 if (iommu->qi) {
1107 free_page((unsigned long)iommu->qi->desc);
1108 kfree(iommu->qi->desc_status);
1109 kfree(iommu->qi);
1112 if (iommu->reg)
1113 unmap_iommu(iommu);
1115 dmar_free_seq_id(iommu);
1116 kfree(iommu);
1120 * Reclaim all the submitted descriptors which have completed its work.
1122 static inline void reclaim_free_desc(struct q_inval *qi)
1124 while (qi->desc_status[qi->free_tail] == QI_DONE ||
1125 qi->desc_status[qi->free_tail] == QI_ABORT) {
1126 qi->desc_status[qi->free_tail] = QI_FREE;
1127 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1128 qi->free_cnt++;
1132 static int qi_check_fault(struct intel_iommu *iommu, int index)
1134 u32 fault;
1135 int head, tail;
1136 struct q_inval *qi = iommu->qi;
1137 int wait_index = (index + 1) % QI_LENGTH;
1139 if (qi->desc_status[wait_index] == QI_ABORT)
1140 return -EAGAIN;
1142 fault = readl(iommu->reg + DMAR_FSTS_REG);
1145 * If IQE happens, the head points to the descriptor associated
1146 * with the error. No new descriptors are fetched until the IQE
1147 * is cleared.
1149 if (fault & DMA_FSTS_IQE) {
1150 head = readl(iommu->reg + DMAR_IQH_REG);
1151 if ((head >> DMAR_IQ_SHIFT) == index) {
1152 pr_err("VT-d detected invalid descriptor: "
1153 "low=%llx, high=%llx\n",
1154 (unsigned long long)qi->desc[index].low,
1155 (unsigned long long)qi->desc[index].high);
1156 memcpy(&qi->desc[index], &qi->desc[wait_index],
1157 sizeof(struct qi_desc));
1158 __iommu_flush_cache(iommu, &qi->desc[index],
1159 sizeof(struct qi_desc));
1160 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1161 return -EINVAL;
1166 * If ITE happens, all pending wait_desc commands are aborted.
1167 * No new descriptors are fetched until the ITE is cleared.
1169 if (fault & DMA_FSTS_ITE) {
1170 head = readl(iommu->reg + DMAR_IQH_REG);
1171 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1172 head |= 1;
1173 tail = readl(iommu->reg + DMAR_IQT_REG);
1174 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1176 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1178 do {
1179 if (qi->desc_status[head] == QI_IN_USE)
1180 qi->desc_status[head] = QI_ABORT;
1181 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1182 } while (head != tail);
1184 if (qi->desc_status[wait_index] == QI_ABORT)
1185 return -EAGAIN;
1188 if (fault & DMA_FSTS_ICE)
1189 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1191 return 0;
1195 * Submit the queued invalidation descriptor to the remapping
1196 * hardware unit and wait for its completion.
1198 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1200 int rc;
1201 struct q_inval *qi = iommu->qi;
1202 struct qi_desc *hw, wait_desc;
1203 int wait_index, index;
1204 unsigned long flags;
1206 if (!qi)
1207 return 0;
1209 hw = qi->desc;
1211 restart:
1212 rc = 0;
1214 raw_spin_lock_irqsave(&qi->q_lock, flags);
1215 while (qi->free_cnt < 3) {
1216 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1217 cpu_relax();
1218 raw_spin_lock_irqsave(&qi->q_lock, flags);
1221 index = qi->free_head;
1222 wait_index = (index + 1) % QI_LENGTH;
1224 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1226 hw[index] = *desc;
1228 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
1229 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1230 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
1232 hw[wait_index] = wait_desc;
1234 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
1235 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
1237 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1238 qi->free_cnt -= 2;
1241 * update the HW tail register indicating the presence of
1242 * new descriptors.
1244 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
1246 while (qi->desc_status[wait_index] != QI_DONE) {
1248 * We will leave the interrupts disabled, to prevent interrupt
1249 * context to queue another cmd while a cmd is already submitted
1250 * and waiting for completion on this cpu. This is to avoid
1251 * a deadlock where the interrupt context can wait indefinitely
1252 * for free slots in the queue.
1254 rc = qi_check_fault(iommu, index);
1255 if (rc)
1256 break;
1258 raw_spin_unlock(&qi->q_lock);
1259 cpu_relax();
1260 raw_spin_lock(&qi->q_lock);
1263 qi->desc_status[index] = QI_DONE;
1265 reclaim_free_desc(qi);
1266 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1268 if (rc == -EAGAIN)
1269 goto restart;
1271 return rc;
1275 * Flush the global interrupt entry cache.
1277 void qi_global_iec(struct intel_iommu *iommu)
1279 struct qi_desc desc;
1281 desc.low = QI_IEC_TYPE;
1282 desc.high = 0;
1284 /* should never fail */
1285 qi_submit_sync(&desc, iommu);
1288 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1289 u64 type)
1291 struct qi_desc desc;
1293 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1294 | QI_CC_GRAN(type) | QI_CC_TYPE;
1295 desc.high = 0;
1297 qi_submit_sync(&desc, iommu);
1300 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1301 unsigned int size_order, u64 type)
1303 u8 dw = 0, dr = 0;
1305 struct qi_desc desc;
1306 int ih = 0;
1308 if (cap_write_drain(iommu->cap))
1309 dw = 1;
1311 if (cap_read_drain(iommu->cap))
1312 dr = 1;
1314 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1315 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1316 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1317 | QI_IOTLB_AM(size_order);
1319 qi_submit_sync(&desc, iommu);
1322 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
1323 u64 addr, unsigned mask)
1325 struct qi_desc desc;
1327 if (mask) {
1328 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
1329 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1330 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1331 } else
1332 desc.high = QI_DEV_IOTLB_ADDR(addr);
1334 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1335 qdep = 0;
1337 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1338 QI_DIOTLB_TYPE;
1340 qi_submit_sync(&desc, iommu);
1344 * Disable Queued Invalidation interface.
1346 void dmar_disable_qi(struct intel_iommu *iommu)
1348 unsigned long flags;
1349 u32 sts;
1350 cycles_t start_time = get_cycles();
1352 if (!ecap_qis(iommu->ecap))
1353 return;
1355 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1357 sts = readl(iommu->reg + DMAR_GSTS_REG);
1358 if (!(sts & DMA_GSTS_QIES))
1359 goto end;
1362 * Give a chance to HW to complete the pending invalidation requests.
1364 while ((readl(iommu->reg + DMAR_IQT_REG) !=
1365 readl(iommu->reg + DMAR_IQH_REG)) &&
1366 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1367 cpu_relax();
1369 iommu->gcmd &= ~DMA_GCMD_QIE;
1370 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1372 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1373 !(sts & DMA_GSTS_QIES), sts);
1374 end:
1375 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1379 * Enable queued invalidation.
1381 static void __dmar_enable_qi(struct intel_iommu *iommu)
1383 u32 sts;
1384 unsigned long flags;
1385 struct q_inval *qi = iommu->qi;
1387 qi->free_head = qi->free_tail = 0;
1388 qi->free_cnt = QI_LENGTH;
1390 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1392 /* write zero to the tail reg */
1393 writel(0, iommu->reg + DMAR_IQT_REG);
1395 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1397 iommu->gcmd |= DMA_GCMD_QIE;
1398 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1400 /* Make sure hardware complete it */
1401 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1403 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1407 * Enable Queued Invalidation interface. This is a must to support
1408 * interrupt-remapping. Also used by DMA-remapping, which replaces
1409 * register based IOTLB invalidation.
1411 int dmar_enable_qi(struct intel_iommu *iommu)
1413 struct q_inval *qi;
1414 struct page *desc_page;
1416 if (!ecap_qis(iommu->ecap))
1417 return -ENOENT;
1420 * queued invalidation is already setup and enabled.
1422 if (iommu->qi)
1423 return 0;
1425 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1426 if (!iommu->qi)
1427 return -ENOMEM;
1429 qi = iommu->qi;
1432 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1433 if (!desc_page) {
1434 kfree(qi);
1435 iommu->qi = NULL;
1436 return -ENOMEM;
1439 qi->desc = page_address(desc_page);
1441 qi->desc_status = kzalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1442 if (!qi->desc_status) {
1443 free_page((unsigned long) qi->desc);
1444 kfree(qi);
1445 iommu->qi = NULL;
1446 return -ENOMEM;
1449 raw_spin_lock_init(&qi->q_lock);
1451 __dmar_enable_qi(iommu);
1453 return 0;
1456 /* iommu interrupt handling. Most stuff are MSI-like. */
1458 enum faulttype {
1459 DMA_REMAP,
1460 INTR_REMAP,
1461 UNKNOWN,
1464 static const char *dma_remap_fault_reasons[] =
1466 "Software",
1467 "Present bit in root entry is clear",
1468 "Present bit in context entry is clear",
1469 "Invalid context entry",
1470 "Access beyond MGAW",
1471 "PTE Write access is not set",
1472 "PTE Read access is not set",
1473 "Next page table ptr is invalid",
1474 "Root table address invalid",
1475 "Context table ptr is invalid",
1476 "non-zero reserved fields in RTP",
1477 "non-zero reserved fields in CTP",
1478 "non-zero reserved fields in PTE",
1479 "PCE for translation request specifies blocking",
1482 static const char *irq_remap_fault_reasons[] =
1484 "Detected reserved fields in the decoded interrupt-remapped request",
1485 "Interrupt index exceeded the interrupt-remapping table size",
1486 "Present field in the IRTE entry is clear",
1487 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1488 "Detected reserved fields in the IRTE entry",
1489 "Blocked a compatibility format interrupt request",
1490 "Blocked an interrupt request due to source-id verification failure",
1493 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1495 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1496 ARRAY_SIZE(irq_remap_fault_reasons))) {
1497 *fault_type = INTR_REMAP;
1498 return irq_remap_fault_reasons[fault_reason - 0x20];
1499 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1500 *fault_type = DMA_REMAP;
1501 return dma_remap_fault_reasons[fault_reason];
1502 } else {
1503 *fault_type = UNKNOWN;
1504 return "Unknown";
1509 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1511 if (iommu->irq == irq)
1512 return DMAR_FECTL_REG;
1513 else if (iommu->pr_irq == irq)
1514 return DMAR_PECTL_REG;
1515 else
1516 BUG();
1519 void dmar_msi_unmask(struct irq_data *data)
1521 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1522 int reg = dmar_msi_reg(iommu, data->irq);
1523 unsigned long flag;
1525 /* unmask it */
1526 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1527 writel(0, iommu->reg + reg);
1528 /* Read a reg to force flush the post write */
1529 readl(iommu->reg + reg);
1530 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1533 void dmar_msi_mask(struct irq_data *data)
1535 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1536 int reg = dmar_msi_reg(iommu, data->irq);
1537 unsigned long flag;
1539 /* mask it */
1540 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1541 writel(DMA_FECTL_IM, iommu->reg + reg);
1542 /* Read a reg to force flush the post write */
1543 readl(iommu->reg + reg);
1544 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1547 void dmar_msi_write(int irq, struct msi_msg *msg)
1549 struct intel_iommu *iommu = irq_get_handler_data(irq);
1550 int reg = dmar_msi_reg(iommu, irq);
1551 unsigned long flag;
1553 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1554 writel(msg->data, iommu->reg + reg + 4);
1555 writel(msg->address_lo, iommu->reg + reg + 8);
1556 writel(msg->address_hi, iommu->reg + reg + 12);
1557 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1560 void dmar_msi_read(int irq, struct msi_msg *msg)
1562 struct intel_iommu *iommu = irq_get_handler_data(irq);
1563 int reg = dmar_msi_reg(iommu, irq);
1564 unsigned long flag;
1566 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1567 msg->data = readl(iommu->reg + reg + 4);
1568 msg->address_lo = readl(iommu->reg + reg + 8);
1569 msg->address_hi = readl(iommu->reg + reg + 12);
1570 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1573 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1574 u8 fault_reason, u16 source_id, unsigned long long addr)
1576 const char *reason;
1577 int fault_type;
1579 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1581 if (fault_type == INTR_REMAP)
1582 pr_err("INTR-REMAP: Request device [[%02x:%02x.%d] "
1583 "fault index %llx\n"
1584 "INTR-REMAP:[fault reason %02d] %s\n",
1585 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1586 PCI_FUNC(source_id & 0xFF), addr >> 48,
1587 fault_reason, reason);
1588 else
1589 pr_err("DMAR:[%s] Request device [%02x:%02x.%d] "
1590 "fault addr %llx \n"
1591 "DMAR:[fault reason %02d] %s\n",
1592 (type ? "DMA Read" : "DMA Write"),
1593 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1594 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1595 return 0;
1598 #define PRIMARY_FAULT_REG_LEN (16)
1599 irqreturn_t dmar_fault(int irq, void *dev_id)
1601 struct intel_iommu *iommu = dev_id;
1602 int reg, fault_index;
1603 u32 fault_status;
1604 unsigned long flag;
1606 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1607 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1608 if (fault_status)
1609 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1611 /* TBD: ignore advanced fault log currently */
1612 if (!(fault_status & DMA_FSTS_PPF))
1613 goto unlock_exit;
1615 fault_index = dma_fsts_fault_record_index(fault_status);
1616 reg = cap_fault_reg_offset(iommu->cap);
1617 while (1) {
1618 u8 fault_reason;
1619 u16 source_id;
1620 u64 guest_addr;
1621 int type;
1622 u32 data;
1624 /* highest 32 bits */
1625 data = readl(iommu->reg + reg +
1626 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1627 if (!(data & DMA_FRCD_F))
1628 break;
1630 fault_reason = dma_frcd_fault_reason(data);
1631 type = dma_frcd_type(data);
1633 data = readl(iommu->reg + reg +
1634 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1635 source_id = dma_frcd_source_id(data);
1637 guest_addr = dmar_readq(iommu->reg + reg +
1638 fault_index * PRIMARY_FAULT_REG_LEN);
1639 guest_addr = dma_frcd_page_addr(guest_addr);
1640 /* clear the fault */
1641 writel(DMA_FRCD_F, iommu->reg + reg +
1642 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1644 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1646 dmar_fault_do_one(iommu, type, fault_reason,
1647 source_id, guest_addr);
1649 fault_index++;
1650 if (fault_index >= cap_num_fault_regs(iommu->cap))
1651 fault_index = 0;
1652 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1655 writel(DMA_FSTS_PFO | DMA_FSTS_PPF, iommu->reg + DMAR_FSTS_REG);
1657 unlock_exit:
1658 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1659 return IRQ_HANDLED;
1662 int dmar_set_interrupt(struct intel_iommu *iommu)
1664 int irq, ret;
1667 * Check if the fault interrupt is already initialized.
1669 if (iommu->irq)
1670 return 0;
1672 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1673 if (irq > 0) {
1674 iommu->irq = irq;
1675 } else {
1676 pr_err("No free IRQ vectors\n");
1677 return -EINVAL;
1680 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1681 if (ret)
1682 pr_err("Can't request irq\n");
1683 return ret;
1686 int __init enable_drhd_fault_handling(void)
1688 struct dmar_drhd_unit *drhd;
1689 struct intel_iommu *iommu;
1692 * Enable fault control interrupt.
1694 for_each_iommu(iommu, drhd) {
1695 u32 fault_status;
1696 int ret = dmar_set_interrupt(iommu);
1698 if (ret) {
1699 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1700 (unsigned long long)drhd->reg_base_addr, ret);
1701 return -1;
1705 * Clear any previous faults.
1707 dmar_fault(iommu->irq, iommu);
1708 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1709 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1712 return 0;
1716 * Re-enable Queued Invalidation interface.
1718 int dmar_reenable_qi(struct intel_iommu *iommu)
1720 if (!ecap_qis(iommu->ecap))
1721 return -ENOENT;
1723 if (!iommu->qi)
1724 return -ENOENT;
1727 * First disable queued invalidation.
1729 dmar_disable_qi(iommu);
1731 * Then enable queued invalidation again. Since there is no pending
1732 * invalidation requests now, it's safe to re-enable queued
1733 * invalidation.
1735 __dmar_enable_qi(iommu);
1737 return 0;
1741 * Check interrupt remapping support in DMAR table description.
1743 int __init dmar_ir_support(void)
1745 struct acpi_table_dmar *dmar;
1746 dmar = (struct acpi_table_dmar *)dmar_tbl;
1747 if (!dmar)
1748 return 0;
1749 return dmar->flags & 0x1;
1752 /* Check whether DMAR units are in use */
1753 static inline bool dmar_in_use(void)
1755 return irq_remapping_enabled || intel_iommu_enabled;
1758 static int __init dmar_free_unused_resources(void)
1760 struct dmar_drhd_unit *dmaru, *dmaru_n;
1762 if (dmar_in_use())
1763 return 0;
1765 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1766 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1768 down_write(&dmar_global_lock);
1769 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1770 list_del(&dmaru->list);
1771 dmar_free_drhd(dmaru);
1773 up_write(&dmar_global_lock);
1775 return 0;
1778 late_initcall(dmar_free_unused_resources);
1779 IOMMU_INIT_POST(detect_intel_iommu);
1782 * DMAR Hotplug Support
1783 * For more details, please refer to Intel(R) Virtualization Technology
1784 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1785 * "Remapping Hardware Unit Hot Plug".
1787 static u8 dmar_hp_uuid[] = {
1788 /* 0000 */ 0xA6, 0xA3, 0xC1, 0xD8, 0x9B, 0xBE, 0x9B, 0x4C,
1789 /* 0008 */ 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF
1793 * Currently there's only one revision and BIOS will not check the revision id,
1794 * so use 0 for safety.
1796 #define DMAR_DSM_REV_ID 0
1797 #define DMAR_DSM_FUNC_DRHD 1
1798 #define DMAR_DSM_FUNC_ATSR 2
1799 #define DMAR_DSM_FUNC_RHSA 3
1801 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1803 return acpi_check_dsm(handle, dmar_hp_uuid, DMAR_DSM_REV_ID, 1 << func);
1806 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1807 dmar_res_handler_t handler, void *arg)
1809 int ret = -ENODEV;
1810 union acpi_object *obj;
1811 struct acpi_dmar_header *start;
1812 struct dmar_res_callback callback;
1813 static int res_type[] = {
1814 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1815 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1816 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1819 if (!dmar_detect_dsm(handle, func))
1820 return 0;
1822 obj = acpi_evaluate_dsm_typed(handle, dmar_hp_uuid, DMAR_DSM_REV_ID,
1823 func, NULL, ACPI_TYPE_BUFFER);
1824 if (!obj)
1825 return -ENODEV;
1827 memset(&callback, 0, sizeof(callback));
1828 callback.cb[res_type[func]] = handler;
1829 callback.arg[res_type[func]] = arg;
1830 start = (struct acpi_dmar_header *)obj->buffer.pointer;
1831 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1833 ACPI_FREE(obj);
1835 return ret;
1838 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1840 int ret;
1841 struct dmar_drhd_unit *dmaru;
1843 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1844 if (!dmaru)
1845 return -ENODEV;
1847 ret = dmar_ir_hotplug(dmaru, true);
1848 if (ret == 0)
1849 ret = dmar_iommu_hotplug(dmaru, true);
1851 return ret;
1854 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
1856 int i, ret;
1857 struct device *dev;
1858 struct dmar_drhd_unit *dmaru;
1860 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1861 if (!dmaru)
1862 return 0;
1865 * All PCI devices managed by this unit should have been destroyed.
1867 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt)
1868 for_each_active_dev_scope(dmaru->devices,
1869 dmaru->devices_cnt, i, dev)
1870 return -EBUSY;
1872 ret = dmar_ir_hotplug(dmaru, false);
1873 if (ret == 0)
1874 ret = dmar_iommu_hotplug(dmaru, false);
1876 return ret;
1879 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
1881 struct dmar_drhd_unit *dmaru;
1883 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1884 if (dmaru) {
1885 list_del_rcu(&dmaru->list);
1886 synchronize_rcu();
1887 dmar_free_drhd(dmaru);
1890 return 0;
1893 static int dmar_hotplug_insert(acpi_handle handle)
1895 int ret;
1896 int drhd_count = 0;
1898 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1899 &dmar_validate_one_drhd, (void *)1);
1900 if (ret)
1901 goto out;
1903 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1904 &dmar_parse_one_drhd, (void *)&drhd_count);
1905 if (ret == 0 && drhd_count == 0) {
1906 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
1907 goto out;
1908 } else if (ret) {
1909 goto release_drhd;
1912 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
1913 &dmar_parse_one_rhsa, NULL);
1914 if (ret)
1915 goto release_drhd;
1917 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1918 &dmar_parse_one_atsr, NULL);
1919 if (ret)
1920 goto release_atsr;
1922 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1923 &dmar_hp_add_drhd, NULL);
1924 if (!ret)
1925 return 0;
1927 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1928 &dmar_hp_remove_drhd, NULL);
1929 release_atsr:
1930 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1931 &dmar_release_one_atsr, NULL);
1932 release_drhd:
1933 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1934 &dmar_hp_release_drhd, NULL);
1935 out:
1936 return ret;
1939 static int dmar_hotplug_remove(acpi_handle handle)
1941 int ret;
1943 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1944 &dmar_check_one_atsr, NULL);
1945 if (ret)
1946 return ret;
1948 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1949 &dmar_hp_remove_drhd, NULL);
1950 if (ret == 0) {
1951 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1952 &dmar_release_one_atsr, NULL));
1953 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1954 &dmar_hp_release_drhd, NULL));
1955 } else {
1956 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1957 &dmar_hp_add_drhd, NULL);
1960 return ret;
1963 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
1964 void *context, void **retval)
1966 acpi_handle *phdl = retval;
1968 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
1969 *phdl = handle;
1970 return AE_CTRL_TERMINATE;
1973 return AE_OK;
1976 static int dmar_device_hotplug(acpi_handle handle, bool insert)
1978 int ret;
1979 acpi_handle tmp = NULL;
1980 acpi_status status;
1982 if (!dmar_in_use())
1983 return 0;
1985 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
1986 tmp = handle;
1987 } else {
1988 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
1989 ACPI_UINT32_MAX,
1990 dmar_get_dsm_handle,
1991 NULL, NULL, &tmp);
1992 if (ACPI_FAILURE(status)) {
1993 pr_warn("Failed to locate _DSM method.\n");
1994 return -ENXIO;
1997 if (tmp == NULL)
1998 return 0;
2000 down_write(&dmar_global_lock);
2001 if (insert)
2002 ret = dmar_hotplug_insert(tmp);
2003 else
2004 ret = dmar_hotplug_remove(tmp);
2005 up_write(&dmar_global_lock);
2007 return ret;
2010 int dmar_device_add(acpi_handle handle)
2012 return dmar_device_hotplug(handle, true);
2015 int dmar_device_remove(acpi_handle handle)
2017 return dmar_device_hotplug(handle, false);