powerpc: Add PPC_EMULATED_STATS to powernv_defconfig
[linux-2.6/btrfs-unstable.git] / drivers / iommu / dmar.c
blob57c920c1372d09f927a7dcdeadc25375bf4164a8
1 /*
2 * Copyright (c) 2006, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
17 * Copyright (C) 2006-2008 Intel Corporation
18 * Author: Ashok Raj <ashok.raj@intel.com>
19 * Author: Shaohua Li <shaohua.li@intel.com>
20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
22 * This file implements early detection/parsing of Remapping Devices
23 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
24 * tables.
26 * These routines are used by both DMA-remapping and Interrupt-remapping
29 #define pr_fmt(fmt) "DMAR: " fmt
31 #include <linux/pci.h>
32 #include <linux/dmar.h>
33 #include <linux/iova.h>
34 #include <linux/intel-iommu.h>
35 #include <linux/timer.h>
36 #include <linux/irq.h>
37 #include <linux/interrupt.h>
38 #include <linux/tboot.h>
39 #include <linux/dmi.h>
40 #include <linux/slab.h>
41 #include <linux/iommu.h>
42 #include <asm/irq_remapping.h>
43 #include <asm/iommu_table.h>
45 #include "irq_remapping.h"
47 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
48 struct dmar_res_callback {
49 dmar_res_handler_t cb[ACPI_DMAR_TYPE_RESERVED];
50 void *arg[ACPI_DMAR_TYPE_RESERVED];
51 bool ignore_unhandled;
52 bool print_entry;
56 * Assumptions:
57 * 1) The hotplug framework guarentees that DMAR unit will be hot-added
58 * before IO devices managed by that unit.
59 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
60 * after IO devices managed by that unit.
61 * 3) Hotplug events are rare.
63 * Locking rules for DMA and interrupt remapping related global data structures:
64 * 1) Use dmar_global_lock in process context
65 * 2) Use RCU in interrupt context
67 DECLARE_RWSEM(dmar_global_lock);
68 LIST_HEAD(dmar_drhd_units);
70 struct acpi_table_header * __initdata dmar_tbl;
71 static int dmar_dev_scope_status = 1;
72 static unsigned long dmar_seq_ids[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED)];
74 static int alloc_iommu(struct dmar_drhd_unit *drhd);
75 static void free_iommu(struct intel_iommu *iommu);
77 extern const struct iommu_ops intel_iommu_ops;
79 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
82 * add INCLUDE_ALL at the tail, so scan the list will find it at
83 * the very end.
85 if (drhd->include_all)
86 list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
87 else
88 list_add_rcu(&drhd->list, &dmar_drhd_units);
91 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
93 struct acpi_dmar_device_scope *scope;
95 *cnt = 0;
96 while (start < end) {
97 scope = start;
98 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
99 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
100 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
101 (*cnt)++;
102 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
103 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
104 pr_warn("Unsupported device scope\n");
106 start += scope->length;
108 if (*cnt == 0)
109 return NULL;
111 return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
114 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
116 int i;
117 struct device *tmp_dev;
119 if (*devices && *cnt) {
120 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
121 put_device(tmp_dev);
122 kfree(*devices);
125 *devices = NULL;
126 *cnt = 0;
129 /* Optimize out kzalloc()/kfree() for normal cases */
130 static char dmar_pci_notify_info_buf[64];
132 static struct dmar_pci_notify_info *
133 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
135 int level = 0;
136 size_t size;
137 struct pci_dev *tmp;
138 struct dmar_pci_notify_info *info;
140 BUG_ON(dev->is_virtfn);
142 /* Only generate path[] for device addition event */
143 if (event == BUS_NOTIFY_ADD_DEVICE)
144 for (tmp = dev; tmp; tmp = tmp->bus->self)
145 level++;
147 size = sizeof(*info) + level * sizeof(struct acpi_dmar_pci_path);
148 if (size <= sizeof(dmar_pci_notify_info_buf)) {
149 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
150 } else {
151 info = kzalloc(size, GFP_KERNEL);
152 if (!info) {
153 pr_warn("Out of memory when allocating notify_info "
154 "for %s.\n", pci_name(dev));
155 if (dmar_dev_scope_status == 0)
156 dmar_dev_scope_status = -ENOMEM;
157 return NULL;
161 info->event = event;
162 info->dev = dev;
163 info->seg = pci_domain_nr(dev->bus);
164 info->level = level;
165 if (event == BUS_NOTIFY_ADD_DEVICE) {
166 for (tmp = dev; tmp; tmp = tmp->bus->self) {
167 level--;
168 info->path[level].bus = tmp->bus->number;
169 info->path[level].device = PCI_SLOT(tmp->devfn);
170 info->path[level].function = PCI_FUNC(tmp->devfn);
171 if (pci_is_root_bus(tmp->bus))
172 info->bus = tmp->bus->number;
176 return info;
179 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
181 if ((void *)info != dmar_pci_notify_info_buf)
182 kfree(info);
185 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
186 struct acpi_dmar_pci_path *path, int count)
188 int i;
190 if (info->bus != bus)
191 goto fallback;
192 if (info->level != count)
193 goto fallback;
195 for (i = 0; i < count; i++) {
196 if (path[i].device != info->path[i].device ||
197 path[i].function != info->path[i].function)
198 goto fallback;
201 return true;
203 fallback:
205 if (count != 1)
206 return false;
208 i = info->level - 1;
209 if (bus == info->path[i].bus &&
210 path[0].device == info->path[i].device &&
211 path[0].function == info->path[i].function) {
212 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
213 bus, path[0].device, path[0].function);
214 return true;
217 return false;
220 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
221 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
222 void *start, void*end, u16 segment,
223 struct dmar_dev_scope *devices,
224 int devices_cnt)
226 int i, level;
227 struct device *tmp, *dev = &info->dev->dev;
228 struct acpi_dmar_device_scope *scope;
229 struct acpi_dmar_pci_path *path;
231 if (segment != info->seg)
232 return 0;
234 for (; start < end; start += scope->length) {
235 scope = start;
236 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
237 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
238 continue;
240 path = (struct acpi_dmar_pci_path *)(scope + 1);
241 level = (scope->length - sizeof(*scope)) / sizeof(*path);
242 if (!dmar_match_pci_path(info, scope->bus, path, level))
243 continue;
246 * We expect devices with endpoint scope to have normal PCI
247 * headers, and devices with bridge scope to have bridge PCI
248 * headers. However PCI NTB devices may be listed in the
249 * DMAR table with bridge scope, even though they have a
250 * normal PCI header. NTB devices are identified by class
251 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
252 * for this special case.
254 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
255 info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
256 (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
257 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
258 info->dev->class >> 8 != PCI_CLASS_BRIDGE_OTHER))) {
259 pr_warn("Device scope type does not match for %s\n",
260 pci_name(info->dev));
261 return -EINVAL;
264 for_each_dev_scope(devices, devices_cnt, i, tmp)
265 if (tmp == NULL) {
266 devices[i].bus = info->dev->bus->number;
267 devices[i].devfn = info->dev->devfn;
268 rcu_assign_pointer(devices[i].dev,
269 get_device(dev));
270 return 1;
272 BUG_ON(i >= devices_cnt);
275 return 0;
278 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
279 struct dmar_dev_scope *devices, int count)
281 int index;
282 struct device *tmp;
284 if (info->seg != segment)
285 return 0;
287 for_each_active_dev_scope(devices, count, index, tmp)
288 if (tmp == &info->dev->dev) {
289 RCU_INIT_POINTER(devices[index].dev, NULL);
290 synchronize_rcu();
291 put_device(tmp);
292 return 1;
295 return 0;
298 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
300 int ret = 0;
301 struct dmar_drhd_unit *dmaru;
302 struct acpi_dmar_hardware_unit *drhd;
304 for_each_drhd_unit(dmaru) {
305 if (dmaru->include_all)
306 continue;
308 drhd = container_of(dmaru->hdr,
309 struct acpi_dmar_hardware_unit, header);
310 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
311 ((void *)drhd) + drhd->header.length,
312 dmaru->segment,
313 dmaru->devices, dmaru->devices_cnt);
314 if (ret)
315 break;
317 if (ret >= 0)
318 ret = dmar_iommu_notify_scope_dev(info);
319 if (ret < 0 && dmar_dev_scope_status == 0)
320 dmar_dev_scope_status = ret;
322 return ret;
325 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
327 struct dmar_drhd_unit *dmaru;
329 for_each_drhd_unit(dmaru)
330 if (dmar_remove_dev_scope(info, dmaru->segment,
331 dmaru->devices, dmaru->devices_cnt))
332 break;
333 dmar_iommu_notify_scope_dev(info);
336 static int dmar_pci_bus_notifier(struct notifier_block *nb,
337 unsigned long action, void *data)
339 struct pci_dev *pdev = to_pci_dev(data);
340 struct dmar_pci_notify_info *info;
342 /* Only care about add/remove events for physical functions.
343 * For VFs we actually do the lookup based on the corresponding
344 * PF in device_to_iommu() anyway. */
345 if (pdev->is_virtfn)
346 return NOTIFY_DONE;
347 if (action != BUS_NOTIFY_ADD_DEVICE &&
348 action != BUS_NOTIFY_REMOVED_DEVICE)
349 return NOTIFY_DONE;
351 info = dmar_alloc_pci_notify_info(pdev, action);
352 if (!info)
353 return NOTIFY_DONE;
355 down_write(&dmar_global_lock);
356 if (action == BUS_NOTIFY_ADD_DEVICE)
357 dmar_pci_bus_add_dev(info);
358 else if (action == BUS_NOTIFY_REMOVED_DEVICE)
359 dmar_pci_bus_del_dev(info);
360 up_write(&dmar_global_lock);
362 dmar_free_pci_notify_info(info);
364 return NOTIFY_OK;
367 static struct notifier_block dmar_pci_bus_nb = {
368 .notifier_call = dmar_pci_bus_notifier,
369 .priority = INT_MIN,
372 static struct dmar_drhd_unit *
373 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
375 struct dmar_drhd_unit *dmaru;
377 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
378 if (dmaru->segment == drhd->segment &&
379 dmaru->reg_base_addr == drhd->address)
380 return dmaru;
382 return NULL;
386 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
387 * structure which uniquely represent one DMA remapping hardware unit
388 * present in the platform
390 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
392 struct acpi_dmar_hardware_unit *drhd;
393 struct dmar_drhd_unit *dmaru;
394 int ret;
396 drhd = (struct acpi_dmar_hardware_unit *)header;
397 dmaru = dmar_find_dmaru(drhd);
398 if (dmaru)
399 goto out;
401 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
402 if (!dmaru)
403 return -ENOMEM;
406 * If header is allocated from slab by ACPI _DSM method, we need to
407 * copy the content because the memory buffer will be freed on return.
409 dmaru->hdr = (void *)(dmaru + 1);
410 memcpy(dmaru->hdr, header, header->length);
411 dmaru->reg_base_addr = drhd->address;
412 dmaru->segment = drhd->segment;
413 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
414 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
415 ((void *)drhd) + drhd->header.length,
416 &dmaru->devices_cnt);
417 if (dmaru->devices_cnt && dmaru->devices == NULL) {
418 kfree(dmaru);
419 return -ENOMEM;
422 ret = alloc_iommu(dmaru);
423 if (ret) {
424 dmar_free_dev_scope(&dmaru->devices,
425 &dmaru->devices_cnt);
426 kfree(dmaru);
427 return ret;
429 dmar_register_drhd_unit(dmaru);
431 out:
432 if (arg)
433 (*(int *)arg)++;
435 return 0;
438 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
440 if (dmaru->devices && dmaru->devices_cnt)
441 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
442 if (dmaru->iommu)
443 free_iommu(dmaru->iommu);
444 kfree(dmaru);
447 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
448 void *arg)
450 struct acpi_dmar_andd *andd = (void *)header;
452 /* Check for NUL termination within the designated length */
453 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
454 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND,
455 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
456 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
457 dmi_get_system_info(DMI_BIOS_VENDOR),
458 dmi_get_system_info(DMI_BIOS_VERSION),
459 dmi_get_system_info(DMI_PRODUCT_VERSION));
460 return -EINVAL;
462 pr_info("ANDD device: %x name: %s\n", andd->device_number,
463 andd->device_name);
465 return 0;
468 #ifdef CONFIG_ACPI_NUMA
469 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
471 struct acpi_dmar_rhsa *rhsa;
472 struct dmar_drhd_unit *drhd;
474 rhsa = (struct acpi_dmar_rhsa *)header;
475 for_each_drhd_unit(drhd) {
476 if (drhd->reg_base_addr == rhsa->base_address) {
477 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
479 if (!node_online(node))
480 node = -1;
481 drhd->iommu->node = node;
482 return 0;
485 WARN_TAINT(
486 1, TAINT_FIRMWARE_WORKAROUND,
487 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
488 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
489 drhd->reg_base_addr,
490 dmi_get_system_info(DMI_BIOS_VENDOR),
491 dmi_get_system_info(DMI_BIOS_VERSION),
492 dmi_get_system_info(DMI_PRODUCT_VERSION));
494 return 0;
496 #else
497 #define dmar_parse_one_rhsa dmar_res_noop
498 #endif
500 static void
501 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
503 struct acpi_dmar_hardware_unit *drhd;
504 struct acpi_dmar_reserved_memory *rmrr;
505 struct acpi_dmar_atsr *atsr;
506 struct acpi_dmar_rhsa *rhsa;
508 switch (header->type) {
509 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
510 drhd = container_of(header, struct acpi_dmar_hardware_unit,
511 header);
512 pr_info("DRHD base: %#016Lx flags: %#x\n",
513 (unsigned long long)drhd->address, drhd->flags);
514 break;
515 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
516 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
517 header);
518 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
519 (unsigned long long)rmrr->base_address,
520 (unsigned long long)rmrr->end_address);
521 break;
522 case ACPI_DMAR_TYPE_ROOT_ATS:
523 atsr = container_of(header, struct acpi_dmar_atsr, header);
524 pr_info("ATSR flags: %#x\n", atsr->flags);
525 break;
526 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
527 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
528 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
529 (unsigned long long)rhsa->base_address,
530 rhsa->proximity_domain);
531 break;
532 case ACPI_DMAR_TYPE_NAMESPACE:
533 /* We don't print this here because we need to sanity-check
534 it first. So print it in dmar_parse_one_andd() instead. */
535 break;
540 * dmar_table_detect - checks to see if the platform supports DMAR devices
542 static int __init dmar_table_detect(void)
544 acpi_status status = AE_OK;
546 /* if we could find DMAR table, then there are DMAR devices */
547 status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
549 if (ACPI_SUCCESS(status) && !dmar_tbl) {
550 pr_warn("Unable to map DMAR\n");
551 status = AE_NOT_FOUND;
554 return ACPI_SUCCESS(status) ? 0 : -ENOENT;
557 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
558 size_t len, struct dmar_res_callback *cb)
560 struct acpi_dmar_header *iter, *next;
561 struct acpi_dmar_header *end = ((void *)start) + len;
563 for (iter = start; iter < end; iter = next) {
564 next = (void *)iter + iter->length;
565 if (iter->length == 0) {
566 /* Avoid looping forever on bad ACPI tables */
567 pr_debug(FW_BUG "Invalid 0-length structure\n");
568 break;
569 } else if (next > end) {
570 /* Avoid passing table end */
571 pr_warn(FW_BUG "Record passes table end\n");
572 return -EINVAL;
575 if (cb->print_entry)
576 dmar_table_print_dmar_entry(iter);
578 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
579 /* continue for forward compatibility */
580 pr_debug("Unknown DMAR structure type %d\n",
581 iter->type);
582 } else if (cb->cb[iter->type]) {
583 int ret;
585 ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
586 if (ret)
587 return ret;
588 } else if (!cb->ignore_unhandled) {
589 pr_warn("No handler for DMAR structure type %d\n",
590 iter->type);
591 return -EINVAL;
595 return 0;
598 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
599 struct dmar_res_callback *cb)
601 return dmar_walk_remapping_entries((void *)(dmar + 1),
602 dmar->header.length - sizeof(*dmar), cb);
606 * parse_dmar_table - parses the DMA reporting table
608 static int __init
609 parse_dmar_table(void)
611 struct acpi_table_dmar *dmar;
612 int drhd_count = 0;
613 int ret;
614 struct dmar_res_callback cb = {
615 .print_entry = true,
616 .ignore_unhandled = true,
617 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
618 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
619 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
620 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
621 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
622 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
626 * Do it again, earlier dmar_tbl mapping could be mapped with
627 * fixed map.
629 dmar_table_detect();
632 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
633 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
635 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
637 dmar = (struct acpi_table_dmar *)dmar_tbl;
638 if (!dmar)
639 return -ENODEV;
641 if (dmar->width < PAGE_SHIFT - 1) {
642 pr_warn("Invalid DMAR haw\n");
643 return -EINVAL;
646 pr_info("Host address width %d\n", dmar->width + 1);
647 ret = dmar_walk_dmar_table(dmar, &cb);
648 if (ret == 0 && drhd_count == 0)
649 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
651 return ret;
654 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
655 int cnt, struct pci_dev *dev)
657 int index;
658 struct device *tmp;
660 while (dev) {
661 for_each_active_dev_scope(devices, cnt, index, tmp)
662 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
663 return 1;
665 /* Check our parent */
666 dev = dev->bus->self;
669 return 0;
672 struct dmar_drhd_unit *
673 dmar_find_matched_drhd_unit(struct pci_dev *dev)
675 struct dmar_drhd_unit *dmaru;
676 struct acpi_dmar_hardware_unit *drhd;
678 dev = pci_physfn(dev);
680 rcu_read_lock();
681 for_each_drhd_unit(dmaru) {
682 drhd = container_of(dmaru->hdr,
683 struct acpi_dmar_hardware_unit,
684 header);
686 if (dmaru->include_all &&
687 drhd->segment == pci_domain_nr(dev->bus))
688 goto out;
690 if (dmar_pci_device_match(dmaru->devices,
691 dmaru->devices_cnt, dev))
692 goto out;
694 dmaru = NULL;
695 out:
696 rcu_read_unlock();
698 return dmaru;
701 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
702 struct acpi_device *adev)
704 struct dmar_drhd_unit *dmaru;
705 struct acpi_dmar_hardware_unit *drhd;
706 struct acpi_dmar_device_scope *scope;
707 struct device *tmp;
708 int i;
709 struct acpi_dmar_pci_path *path;
711 for_each_drhd_unit(dmaru) {
712 drhd = container_of(dmaru->hdr,
713 struct acpi_dmar_hardware_unit,
714 header);
716 for (scope = (void *)(drhd + 1);
717 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
718 scope = ((void *)scope) + scope->length) {
719 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
720 continue;
721 if (scope->enumeration_id != device_number)
722 continue;
724 path = (void *)(scope + 1);
725 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
726 dev_name(&adev->dev), dmaru->reg_base_addr,
727 scope->bus, path->device, path->function);
728 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
729 if (tmp == NULL) {
730 dmaru->devices[i].bus = scope->bus;
731 dmaru->devices[i].devfn = PCI_DEVFN(path->device,
732 path->function);
733 rcu_assign_pointer(dmaru->devices[i].dev,
734 get_device(&adev->dev));
735 return;
737 BUG_ON(i >= dmaru->devices_cnt);
740 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
741 device_number, dev_name(&adev->dev));
744 static int __init dmar_acpi_dev_scope_init(void)
746 struct acpi_dmar_andd *andd;
748 if (dmar_tbl == NULL)
749 return -ENODEV;
751 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
752 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
753 andd = ((void *)andd) + andd->header.length) {
754 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
755 acpi_handle h;
756 struct acpi_device *adev;
758 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
759 andd->device_name,
760 &h))) {
761 pr_err("Failed to find handle for ACPI object %s\n",
762 andd->device_name);
763 continue;
765 if (acpi_bus_get_device(h, &adev)) {
766 pr_err("Failed to get device for ACPI object %s\n",
767 andd->device_name);
768 continue;
770 dmar_acpi_insert_dev_scope(andd->device_number, adev);
773 return 0;
776 int __init dmar_dev_scope_init(void)
778 struct pci_dev *dev = NULL;
779 struct dmar_pci_notify_info *info;
781 if (dmar_dev_scope_status != 1)
782 return dmar_dev_scope_status;
784 if (list_empty(&dmar_drhd_units)) {
785 dmar_dev_scope_status = -ENODEV;
786 } else {
787 dmar_dev_scope_status = 0;
789 dmar_acpi_dev_scope_init();
791 for_each_pci_dev(dev) {
792 if (dev->is_virtfn)
793 continue;
795 info = dmar_alloc_pci_notify_info(dev,
796 BUS_NOTIFY_ADD_DEVICE);
797 if (!info) {
798 return dmar_dev_scope_status;
799 } else {
800 dmar_pci_bus_add_dev(info);
801 dmar_free_pci_notify_info(info);
805 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
808 return dmar_dev_scope_status;
812 int __init dmar_table_init(void)
814 static int dmar_table_initialized;
815 int ret;
817 if (dmar_table_initialized == 0) {
818 ret = parse_dmar_table();
819 if (ret < 0) {
820 if (ret != -ENODEV)
821 pr_info("Parse DMAR table failure.\n");
822 } else if (list_empty(&dmar_drhd_units)) {
823 pr_info("No DMAR devices found\n");
824 ret = -ENODEV;
827 if (ret < 0)
828 dmar_table_initialized = ret;
829 else
830 dmar_table_initialized = 1;
833 return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
836 static void warn_invalid_dmar(u64 addr, const char *message)
838 WARN_TAINT_ONCE(
839 1, TAINT_FIRMWARE_WORKAROUND,
840 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
841 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
842 addr, message,
843 dmi_get_system_info(DMI_BIOS_VENDOR),
844 dmi_get_system_info(DMI_BIOS_VERSION),
845 dmi_get_system_info(DMI_PRODUCT_VERSION));
848 static int __ref
849 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
851 struct acpi_dmar_hardware_unit *drhd;
852 void __iomem *addr;
853 u64 cap, ecap;
855 drhd = (void *)entry;
856 if (!drhd->address) {
857 warn_invalid_dmar(0, "");
858 return -EINVAL;
861 if (arg)
862 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
863 else
864 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
865 if (!addr) {
866 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
867 return -EINVAL;
870 cap = dmar_readq(addr + DMAR_CAP_REG);
871 ecap = dmar_readq(addr + DMAR_ECAP_REG);
873 if (arg)
874 iounmap(addr);
875 else
876 early_iounmap(addr, VTD_PAGE_SIZE);
878 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
879 warn_invalid_dmar(drhd->address, " returns all ones");
880 return -EINVAL;
883 return 0;
886 int __init detect_intel_iommu(void)
888 int ret;
889 struct dmar_res_callback validate_drhd_cb = {
890 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
891 .ignore_unhandled = true,
894 down_write(&dmar_global_lock);
895 ret = dmar_table_detect();
896 if (!ret)
897 ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
898 &validate_drhd_cb);
899 if (!ret && !no_iommu && !iommu_detected && !dmar_disabled) {
900 iommu_detected = 1;
901 /* Make sure ACS will be enabled */
902 pci_request_acs();
905 #ifdef CONFIG_X86
906 if (!ret)
907 x86_init.iommu.iommu_init = intel_iommu_init;
908 #endif
910 if (dmar_tbl) {
911 acpi_put_table(dmar_tbl);
912 dmar_tbl = NULL;
914 up_write(&dmar_global_lock);
916 return ret ? ret : 1;
919 static void unmap_iommu(struct intel_iommu *iommu)
921 iounmap(iommu->reg);
922 release_mem_region(iommu->reg_phys, iommu->reg_size);
926 * map_iommu: map the iommu's registers
927 * @iommu: the iommu to map
928 * @phys_addr: the physical address of the base resgister
930 * Memory map the iommu's registers. Start w/ a single page, and
931 * possibly expand if that turns out to be insufficent.
933 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
935 int map_size, err=0;
937 iommu->reg_phys = phys_addr;
938 iommu->reg_size = VTD_PAGE_SIZE;
940 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
941 pr_err("Can't reserve memory\n");
942 err = -EBUSY;
943 goto out;
946 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
947 if (!iommu->reg) {
948 pr_err("Can't map the region\n");
949 err = -ENOMEM;
950 goto release;
953 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
954 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
956 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
957 err = -EINVAL;
958 warn_invalid_dmar(phys_addr, " returns all ones");
959 goto unmap;
962 /* the registers might be more than one page */
963 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
964 cap_max_fault_reg_offset(iommu->cap));
965 map_size = VTD_PAGE_ALIGN(map_size);
966 if (map_size > iommu->reg_size) {
967 iounmap(iommu->reg);
968 release_mem_region(iommu->reg_phys, iommu->reg_size);
969 iommu->reg_size = map_size;
970 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
971 iommu->name)) {
972 pr_err("Can't reserve memory\n");
973 err = -EBUSY;
974 goto out;
976 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
977 if (!iommu->reg) {
978 pr_err("Can't map the region\n");
979 err = -ENOMEM;
980 goto release;
983 err = 0;
984 goto out;
986 unmap:
987 iounmap(iommu->reg);
988 release:
989 release_mem_region(iommu->reg_phys, iommu->reg_size);
990 out:
991 return err;
994 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
996 iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
997 DMAR_UNITS_SUPPORTED);
998 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
999 iommu->seq_id = -1;
1000 } else {
1001 set_bit(iommu->seq_id, dmar_seq_ids);
1002 sprintf(iommu->name, "dmar%d", iommu->seq_id);
1005 return iommu->seq_id;
1008 static void dmar_free_seq_id(struct intel_iommu *iommu)
1010 if (iommu->seq_id >= 0) {
1011 clear_bit(iommu->seq_id, dmar_seq_ids);
1012 iommu->seq_id = -1;
1016 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1018 struct intel_iommu *iommu;
1019 u32 ver, sts;
1020 int agaw = 0;
1021 int msagaw = 0;
1022 int err;
1024 if (!drhd->reg_base_addr) {
1025 warn_invalid_dmar(0, "");
1026 return -EINVAL;
1029 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1030 if (!iommu)
1031 return -ENOMEM;
1033 if (dmar_alloc_seq_id(iommu) < 0) {
1034 pr_err("Failed to allocate seq_id\n");
1035 err = -ENOSPC;
1036 goto error;
1039 err = map_iommu(iommu, drhd->reg_base_addr);
1040 if (err) {
1041 pr_err("Failed to map %s\n", iommu->name);
1042 goto error_free_seq_id;
1045 err = -EINVAL;
1046 agaw = iommu_calculate_agaw(iommu);
1047 if (agaw < 0) {
1048 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1049 iommu->seq_id);
1050 goto err_unmap;
1052 msagaw = iommu_calculate_max_sagaw(iommu);
1053 if (msagaw < 0) {
1054 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1055 iommu->seq_id);
1056 goto err_unmap;
1058 iommu->agaw = agaw;
1059 iommu->msagaw = msagaw;
1060 iommu->segment = drhd->segment;
1062 iommu->node = -1;
1064 ver = readl(iommu->reg + DMAR_VER_REG);
1065 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1066 iommu->name,
1067 (unsigned long long)drhd->reg_base_addr,
1068 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1069 (unsigned long long)iommu->cap,
1070 (unsigned long long)iommu->ecap);
1072 /* Reflect status in gcmd */
1073 sts = readl(iommu->reg + DMAR_GSTS_REG);
1074 if (sts & DMA_GSTS_IRES)
1075 iommu->gcmd |= DMA_GCMD_IRE;
1076 if (sts & DMA_GSTS_TES)
1077 iommu->gcmd |= DMA_GCMD_TE;
1078 if (sts & DMA_GSTS_QIES)
1079 iommu->gcmd |= DMA_GCMD_QIE;
1081 raw_spin_lock_init(&iommu->register_lock);
1083 if (intel_iommu_enabled) {
1084 err = iommu_device_sysfs_add(&iommu->iommu, NULL,
1085 intel_iommu_groups,
1086 "%s", iommu->name);
1087 if (err)
1088 goto err_unmap;
1090 iommu_device_set_ops(&iommu->iommu, &intel_iommu_ops);
1092 err = iommu_device_register(&iommu->iommu);
1093 if (err)
1094 goto err_unmap;
1097 drhd->iommu = iommu;
1099 return 0;
1101 err_unmap:
1102 unmap_iommu(iommu);
1103 error_free_seq_id:
1104 dmar_free_seq_id(iommu);
1105 error:
1106 kfree(iommu);
1107 return err;
1110 static void free_iommu(struct intel_iommu *iommu)
1112 if (intel_iommu_enabled) {
1113 iommu_device_unregister(&iommu->iommu);
1114 iommu_device_sysfs_remove(&iommu->iommu);
1117 if (iommu->irq) {
1118 if (iommu->pr_irq) {
1119 free_irq(iommu->pr_irq, iommu);
1120 dmar_free_hwirq(iommu->pr_irq);
1121 iommu->pr_irq = 0;
1123 free_irq(iommu->irq, iommu);
1124 dmar_free_hwirq(iommu->irq);
1125 iommu->irq = 0;
1128 if (iommu->qi) {
1129 free_page((unsigned long)iommu->qi->desc);
1130 kfree(iommu->qi->desc_status);
1131 kfree(iommu->qi);
1134 if (iommu->reg)
1135 unmap_iommu(iommu);
1137 dmar_free_seq_id(iommu);
1138 kfree(iommu);
1142 * Reclaim all the submitted descriptors which have completed its work.
1144 static inline void reclaim_free_desc(struct q_inval *qi)
1146 while (qi->desc_status[qi->free_tail] == QI_DONE ||
1147 qi->desc_status[qi->free_tail] == QI_ABORT) {
1148 qi->desc_status[qi->free_tail] = QI_FREE;
1149 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1150 qi->free_cnt++;
1154 static int qi_check_fault(struct intel_iommu *iommu, int index)
1156 u32 fault;
1157 int head, tail;
1158 struct q_inval *qi = iommu->qi;
1159 int wait_index = (index + 1) % QI_LENGTH;
1161 if (qi->desc_status[wait_index] == QI_ABORT)
1162 return -EAGAIN;
1164 fault = readl(iommu->reg + DMAR_FSTS_REG);
1167 * If IQE happens, the head points to the descriptor associated
1168 * with the error. No new descriptors are fetched until the IQE
1169 * is cleared.
1171 if (fault & DMA_FSTS_IQE) {
1172 head = readl(iommu->reg + DMAR_IQH_REG);
1173 if ((head >> DMAR_IQ_SHIFT) == index) {
1174 pr_err("VT-d detected invalid descriptor: "
1175 "low=%llx, high=%llx\n",
1176 (unsigned long long)qi->desc[index].low,
1177 (unsigned long long)qi->desc[index].high);
1178 memcpy(&qi->desc[index], &qi->desc[wait_index],
1179 sizeof(struct qi_desc));
1180 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1181 return -EINVAL;
1186 * If ITE happens, all pending wait_desc commands are aborted.
1187 * No new descriptors are fetched until the ITE is cleared.
1189 if (fault & DMA_FSTS_ITE) {
1190 head = readl(iommu->reg + DMAR_IQH_REG);
1191 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1192 head |= 1;
1193 tail = readl(iommu->reg + DMAR_IQT_REG);
1194 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1196 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1198 do {
1199 if (qi->desc_status[head] == QI_IN_USE)
1200 qi->desc_status[head] = QI_ABORT;
1201 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1202 } while (head != tail);
1204 if (qi->desc_status[wait_index] == QI_ABORT)
1205 return -EAGAIN;
1208 if (fault & DMA_FSTS_ICE)
1209 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1211 return 0;
1215 * Submit the queued invalidation descriptor to the remapping
1216 * hardware unit and wait for its completion.
1218 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1220 int rc;
1221 struct q_inval *qi = iommu->qi;
1222 struct qi_desc *hw, wait_desc;
1223 int wait_index, index;
1224 unsigned long flags;
1226 if (!qi)
1227 return 0;
1229 hw = qi->desc;
1231 restart:
1232 rc = 0;
1234 raw_spin_lock_irqsave(&qi->q_lock, flags);
1235 while (qi->free_cnt < 3) {
1236 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1237 cpu_relax();
1238 raw_spin_lock_irqsave(&qi->q_lock, flags);
1241 index = qi->free_head;
1242 wait_index = (index + 1) % QI_LENGTH;
1244 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1246 hw[index] = *desc;
1248 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
1249 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1250 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
1252 hw[wait_index] = wait_desc;
1254 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1255 qi->free_cnt -= 2;
1258 * update the HW tail register indicating the presence of
1259 * new descriptors.
1261 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
1263 while (qi->desc_status[wait_index] != QI_DONE) {
1265 * We will leave the interrupts disabled, to prevent interrupt
1266 * context to queue another cmd while a cmd is already submitted
1267 * and waiting for completion on this cpu. This is to avoid
1268 * a deadlock where the interrupt context can wait indefinitely
1269 * for free slots in the queue.
1271 rc = qi_check_fault(iommu, index);
1272 if (rc)
1273 break;
1275 raw_spin_unlock(&qi->q_lock);
1276 cpu_relax();
1277 raw_spin_lock(&qi->q_lock);
1280 qi->desc_status[index] = QI_DONE;
1282 reclaim_free_desc(qi);
1283 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1285 if (rc == -EAGAIN)
1286 goto restart;
1288 return rc;
1292 * Flush the global interrupt entry cache.
1294 void qi_global_iec(struct intel_iommu *iommu)
1296 struct qi_desc desc;
1298 desc.low = QI_IEC_TYPE;
1299 desc.high = 0;
1301 /* should never fail */
1302 qi_submit_sync(&desc, iommu);
1305 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1306 u64 type)
1308 struct qi_desc desc;
1310 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1311 | QI_CC_GRAN(type) | QI_CC_TYPE;
1312 desc.high = 0;
1314 qi_submit_sync(&desc, iommu);
1317 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1318 unsigned int size_order, u64 type)
1320 u8 dw = 0, dr = 0;
1322 struct qi_desc desc;
1323 int ih = 0;
1325 if (cap_write_drain(iommu->cap))
1326 dw = 1;
1328 if (cap_read_drain(iommu->cap))
1329 dr = 1;
1331 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1332 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1333 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1334 | QI_IOTLB_AM(size_order);
1336 qi_submit_sync(&desc, iommu);
1339 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
1340 u64 addr, unsigned mask)
1342 struct qi_desc desc;
1344 if (mask) {
1345 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
1346 addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1347 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1348 } else
1349 desc.high = QI_DEV_IOTLB_ADDR(addr);
1351 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1352 qdep = 0;
1354 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1355 QI_DIOTLB_TYPE;
1357 qi_submit_sync(&desc, iommu);
1361 * Disable Queued Invalidation interface.
1363 void dmar_disable_qi(struct intel_iommu *iommu)
1365 unsigned long flags;
1366 u32 sts;
1367 cycles_t start_time = get_cycles();
1369 if (!ecap_qis(iommu->ecap))
1370 return;
1372 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1374 sts = readl(iommu->reg + DMAR_GSTS_REG);
1375 if (!(sts & DMA_GSTS_QIES))
1376 goto end;
1379 * Give a chance to HW to complete the pending invalidation requests.
1381 while ((readl(iommu->reg + DMAR_IQT_REG) !=
1382 readl(iommu->reg + DMAR_IQH_REG)) &&
1383 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1384 cpu_relax();
1386 iommu->gcmd &= ~DMA_GCMD_QIE;
1387 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1389 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1390 !(sts & DMA_GSTS_QIES), sts);
1391 end:
1392 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1396 * Enable queued invalidation.
1398 static void __dmar_enable_qi(struct intel_iommu *iommu)
1400 u32 sts;
1401 unsigned long flags;
1402 struct q_inval *qi = iommu->qi;
1404 qi->free_head = qi->free_tail = 0;
1405 qi->free_cnt = QI_LENGTH;
1407 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1409 /* write zero to the tail reg */
1410 writel(0, iommu->reg + DMAR_IQT_REG);
1412 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1414 iommu->gcmd |= DMA_GCMD_QIE;
1415 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1417 /* Make sure hardware complete it */
1418 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1420 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1424 * Enable Queued Invalidation interface. This is a must to support
1425 * interrupt-remapping. Also used by DMA-remapping, which replaces
1426 * register based IOTLB invalidation.
1428 int dmar_enable_qi(struct intel_iommu *iommu)
1430 struct q_inval *qi;
1431 struct page *desc_page;
1433 if (!ecap_qis(iommu->ecap))
1434 return -ENOENT;
1437 * queued invalidation is already setup and enabled.
1439 if (iommu->qi)
1440 return 0;
1442 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1443 if (!iommu->qi)
1444 return -ENOMEM;
1446 qi = iommu->qi;
1449 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1450 if (!desc_page) {
1451 kfree(qi);
1452 iommu->qi = NULL;
1453 return -ENOMEM;
1456 qi->desc = page_address(desc_page);
1458 qi->desc_status = kzalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1459 if (!qi->desc_status) {
1460 free_page((unsigned long) qi->desc);
1461 kfree(qi);
1462 iommu->qi = NULL;
1463 return -ENOMEM;
1466 raw_spin_lock_init(&qi->q_lock);
1468 __dmar_enable_qi(iommu);
1470 return 0;
1473 /* iommu interrupt handling. Most stuff are MSI-like. */
1475 enum faulttype {
1476 DMA_REMAP,
1477 INTR_REMAP,
1478 UNKNOWN,
1481 static const char *dma_remap_fault_reasons[] =
1483 "Software",
1484 "Present bit in root entry is clear",
1485 "Present bit in context entry is clear",
1486 "Invalid context entry",
1487 "Access beyond MGAW",
1488 "PTE Write access is not set",
1489 "PTE Read access is not set",
1490 "Next page table ptr is invalid",
1491 "Root table address invalid",
1492 "Context table ptr is invalid",
1493 "non-zero reserved fields in RTP",
1494 "non-zero reserved fields in CTP",
1495 "non-zero reserved fields in PTE",
1496 "PCE for translation request specifies blocking",
1499 static const char *irq_remap_fault_reasons[] =
1501 "Detected reserved fields in the decoded interrupt-remapped request",
1502 "Interrupt index exceeded the interrupt-remapping table size",
1503 "Present field in the IRTE entry is clear",
1504 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1505 "Detected reserved fields in the IRTE entry",
1506 "Blocked a compatibility format interrupt request",
1507 "Blocked an interrupt request due to source-id verification failure",
1510 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1512 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1513 ARRAY_SIZE(irq_remap_fault_reasons))) {
1514 *fault_type = INTR_REMAP;
1515 return irq_remap_fault_reasons[fault_reason - 0x20];
1516 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1517 *fault_type = DMA_REMAP;
1518 return dma_remap_fault_reasons[fault_reason];
1519 } else {
1520 *fault_type = UNKNOWN;
1521 return "Unknown";
1526 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1528 if (iommu->irq == irq)
1529 return DMAR_FECTL_REG;
1530 else if (iommu->pr_irq == irq)
1531 return DMAR_PECTL_REG;
1532 else
1533 BUG();
1536 void dmar_msi_unmask(struct irq_data *data)
1538 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1539 int reg = dmar_msi_reg(iommu, data->irq);
1540 unsigned long flag;
1542 /* unmask it */
1543 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1544 writel(0, iommu->reg + reg);
1545 /* Read a reg to force flush the post write */
1546 readl(iommu->reg + reg);
1547 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1550 void dmar_msi_mask(struct irq_data *data)
1552 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1553 int reg = dmar_msi_reg(iommu, data->irq);
1554 unsigned long flag;
1556 /* mask it */
1557 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1558 writel(DMA_FECTL_IM, iommu->reg + reg);
1559 /* Read a reg to force flush the post write */
1560 readl(iommu->reg + reg);
1561 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1564 void dmar_msi_write(int irq, struct msi_msg *msg)
1566 struct intel_iommu *iommu = irq_get_handler_data(irq);
1567 int reg = dmar_msi_reg(iommu, irq);
1568 unsigned long flag;
1570 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1571 writel(msg->data, iommu->reg + reg + 4);
1572 writel(msg->address_lo, iommu->reg + reg + 8);
1573 writel(msg->address_hi, iommu->reg + reg + 12);
1574 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1577 void dmar_msi_read(int irq, struct msi_msg *msg)
1579 struct intel_iommu *iommu = irq_get_handler_data(irq);
1580 int reg = dmar_msi_reg(iommu, irq);
1581 unsigned long flag;
1583 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1584 msg->data = readl(iommu->reg + reg + 4);
1585 msg->address_lo = readl(iommu->reg + reg + 8);
1586 msg->address_hi = readl(iommu->reg + reg + 12);
1587 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1590 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1591 u8 fault_reason, u16 source_id, unsigned long long addr)
1593 const char *reason;
1594 int fault_type;
1596 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1598 if (fault_type == INTR_REMAP)
1599 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1600 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1601 PCI_FUNC(source_id & 0xFF), addr >> 48,
1602 fault_reason, reason);
1603 else
1604 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1605 type ? "DMA Read" : "DMA Write",
1606 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1607 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1608 return 0;
1611 #define PRIMARY_FAULT_REG_LEN (16)
1612 irqreturn_t dmar_fault(int irq, void *dev_id)
1614 struct intel_iommu *iommu = dev_id;
1615 int reg, fault_index;
1616 u32 fault_status;
1617 unsigned long flag;
1618 bool ratelimited;
1619 static DEFINE_RATELIMIT_STATE(rs,
1620 DEFAULT_RATELIMIT_INTERVAL,
1621 DEFAULT_RATELIMIT_BURST);
1623 /* Disable printing, simply clear the fault when ratelimited */
1624 ratelimited = !__ratelimit(&rs);
1626 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1627 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1628 if (fault_status && !ratelimited)
1629 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1631 /* TBD: ignore advanced fault log currently */
1632 if (!(fault_status & DMA_FSTS_PPF))
1633 goto unlock_exit;
1635 fault_index = dma_fsts_fault_record_index(fault_status);
1636 reg = cap_fault_reg_offset(iommu->cap);
1637 while (1) {
1638 u8 fault_reason;
1639 u16 source_id;
1640 u64 guest_addr;
1641 int type;
1642 u32 data;
1644 /* highest 32 bits */
1645 data = readl(iommu->reg + reg +
1646 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1647 if (!(data & DMA_FRCD_F))
1648 break;
1650 if (!ratelimited) {
1651 fault_reason = dma_frcd_fault_reason(data);
1652 type = dma_frcd_type(data);
1654 data = readl(iommu->reg + reg +
1655 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1656 source_id = dma_frcd_source_id(data);
1658 guest_addr = dmar_readq(iommu->reg + reg +
1659 fault_index * PRIMARY_FAULT_REG_LEN);
1660 guest_addr = dma_frcd_page_addr(guest_addr);
1663 /* clear the fault */
1664 writel(DMA_FRCD_F, iommu->reg + reg +
1665 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1667 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1669 if (!ratelimited)
1670 dmar_fault_do_one(iommu, type, fault_reason,
1671 source_id, guest_addr);
1673 fault_index++;
1674 if (fault_index >= cap_num_fault_regs(iommu->cap))
1675 fault_index = 0;
1676 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1679 writel(DMA_FSTS_PFO | DMA_FSTS_PPF, iommu->reg + DMAR_FSTS_REG);
1681 unlock_exit:
1682 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1683 return IRQ_HANDLED;
1686 int dmar_set_interrupt(struct intel_iommu *iommu)
1688 int irq, ret;
1691 * Check if the fault interrupt is already initialized.
1693 if (iommu->irq)
1694 return 0;
1696 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1697 if (irq > 0) {
1698 iommu->irq = irq;
1699 } else {
1700 pr_err("No free IRQ vectors\n");
1701 return -EINVAL;
1704 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1705 if (ret)
1706 pr_err("Can't request irq\n");
1707 return ret;
1710 int __init enable_drhd_fault_handling(void)
1712 struct dmar_drhd_unit *drhd;
1713 struct intel_iommu *iommu;
1716 * Enable fault control interrupt.
1718 for_each_iommu(iommu, drhd) {
1719 u32 fault_status;
1720 int ret = dmar_set_interrupt(iommu);
1722 if (ret) {
1723 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1724 (unsigned long long)drhd->reg_base_addr, ret);
1725 return -1;
1729 * Clear any previous faults.
1731 dmar_fault(iommu->irq, iommu);
1732 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1733 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1736 return 0;
1740 * Re-enable Queued Invalidation interface.
1742 int dmar_reenable_qi(struct intel_iommu *iommu)
1744 if (!ecap_qis(iommu->ecap))
1745 return -ENOENT;
1747 if (!iommu->qi)
1748 return -ENOENT;
1751 * First disable queued invalidation.
1753 dmar_disable_qi(iommu);
1755 * Then enable queued invalidation again. Since there is no pending
1756 * invalidation requests now, it's safe to re-enable queued
1757 * invalidation.
1759 __dmar_enable_qi(iommu);
1761 return 0;
1765 * Check interrupt remapping support in DMAR table description.
1767 int __init dmar_ir_support(void)
1769 struct acpi_table_dmar *dmar;
1770 dmar = (struct acpi_table_dmar *)dmar_tbl;
1771 if (!dmar)
1772 return 0;
1773 return dmar->flags & 0x1;
1776 /* Check whether DMAR units are in use */
1777 static inline bool dmar_in_use(void)
1779 return irq_remapping_enabled || intel_iommu_enabled;
1782 static int __init dmar_free_unused_resources(void)
1784 struct dmar_drhd_unit *dmaru, *dmaru_n;
1786 if (dmar_in_use())
1787 return 0;
1789 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1790 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1792 down_write(&dmar_global_lock);
1793 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1794 list_del(&dmaru->list);
1795 dmar_free_drhd(dmaru);
1797 up_write(&dmar_global_lock);
1799 return 0;
1802 late_initcall(dmar_free_unused_resources);
1803 IOMMU_INIT_POST(detect_intel_iommu);
1806 * DMAR Hotplug Support
1807 * For more details, please refer to Intel(R) Virtualization Technology
1808 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1809 * "Remapping Hardware Unit Hot Plug".
1811 static guid_t dmar_hp_guid =
1812 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
1813 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
1816 * Currently there's only one revision and BIOS will not check the revision id,
1817 * so use 0 for safety.
1819 #define DMAR_DSM_REV_ID 0
1820 #define DMAR_DSM_FUNC_DRHD 1
1821 #define DMAR_DSM_FUNC_ATSR 2
1822 #define DMAR_DSM_FUNC_RHSA 3
1824 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1826 return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func);
1829 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1830 dmar_res_handler_t handler, void *arg)
1832 int ret = -ENODEV;
1833 union acpi_object *obj;
1834 struct acpi_dmar_header *start;
1835 struct dmar_res_callback callback;
1836 static int res_type[] = {
1837 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1838 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1839 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1842 if (!dmar_detect_dsm(handle, func))
1843 return 0;
1845 obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID,
1846 func, NULL, ACPI_TYPE_BUFFER);
1847 if (!obj)
1848 return -ENODEV;
1850 memset(&callback, 0, sizeof(callback));
1851 callback.cb[res_type[func]] = handler;
1852 callback.arg[res_type[func]] = arg;
1853 start = (struct acpi_dmar_header *)obj->buffer.pointer;
1854 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1856 ACPI_FREE(obj);
1858 return ret;
1861 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1863 int ret;
1864 struct dmar_drhd_unit *dmaru;
1866 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1867 if (!dmaru)
1868 return -ENODEV;
1870 ret = dmar_ir_hotplug(dmaru, true);
1871 if (ret == 0)
1872 ret = dmar_iommu_hotplug(dmaru, true);
1874 return ret;
1877 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
1879 int i, ret;
1880 struct device *dev;
1881 struct dmar_drhd_unit *dmaru;
1883 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1884 if (!dmaru)
1885 return 0;
1888 * All PCI devices managed by this unit should have been destroyed.
1890 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
1891 for_each_active_dev_scope(dmaru->devices,
1892 dmaru->devices_cnt, i, dev)
1893 return -EBUSY;
1896 ret = dmar_ir_hotplug(dmaru, false);
1897 if (ret == 0)
1898 ret = dmar_iommu_hotplug(dmaru, false);
1900 return ret;
1903 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
1905 struct dmar_drhd_unit *dmaru;
1907 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1908 if (dmaru) {
1909 list_del_rcu(&dmaru->list);
1910 synchronize_rcu();
1911 dmar_free_drhd(dmaru);
1914 return 0;
1917 static int dmar_hotplug_insert(acpi_handle handle)
1919 int ret;
1920 int drhd_count = 0;
1922 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1923 &dmar_validate_one_drhd, (void *)1);
1924 if (ret)
1925 goto out;
1927 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1928 &dmar_parse_one_drhd, (void *)&drhd_count);
1929 if (ret == 0 && drhd_count == 0) {
1930 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
1931 goto out;
1932 } else if (ret) {
1933 goto release_drhd;
1936 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
1937 &dmar_parse_one_rhsa, NULL);
1938 if (ret)
1939 goto release_drhd;
1941 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1942 &dmar_parse_one_atsr, NULL);
1943 if (ret)
1944 goto release_atsr;
1946 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1947 &dmar_hp_add_drhd, NULL);
1948 if (!ret)
1949 return 0;
1951 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1952 &dmar_hp_remove_drhd, NULL);
1953 release_atsr:
1954 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1955 &dmar_release_one_atsr, NULL);
1956 release_drhd:
1957 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1958 &dmar_hp_release_drhd, NULL);
1959 out:
1960 return ret;
1963 static int dmar_hotplug_remove(acpi_handle handle)
1965 int ret;
1967 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1968 &dmar_check_one_atsr, NULL);
1969 if (ret)
1970 return ret;
1972 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1973 &dmar_hp_remove_drhd, NULL);
1974 if (ret == 0) {
1975 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1976 &dmar_release_one_atsr, NULL));
1977 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1978 &dmar_hp_release_drhd, NULL));
1979 } else {
1980 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1981 &dmar_hp_add_drhd, NULL);
1984 return ret;
1987 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
1988 void *context, void **retval)
1990 acpi_handle *phdl = retval;
1992 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
1993 *phdl = handle;
1994 return AE_CTRL_TERMINATE;
1997 return AE_OK;
2000 static int dmar_device_hotplug(acpi_handle handle, bool insert)
2002 int ret;
2003 acpi_handle tmp = NULL;
2004 acpi_status status;
2006 if (!dmar_in_use())
2007 return 0;
2009 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2010 tmp = handle;
2011 } else {
2012 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2013 ACPI_UINT32_MAX,
2014 dmar_get_dsm_handle,
2015 NULL, NULL, &tmp);
2016 if (ACPI_FAILURE(status)) {
2017 pr_warn("Failed to locate _DSM method.\n");
2018 return -ENXIO;
2021 if (tmp == NULL)
2022 return 0;
2024 down_write(&dmar_global_lock);
2025 if (insert)
2026 ret = dmar_hotplug_insert(tmp);
2027 else
2028 ret = dmar_hotplug_remove(tmp);
2029 up_write(&dmar_global_lock);
2031 return ret;
2034 int dmar_device_add(acpi_handle handle)
2036 return dmar_device_hotplug(handle, true);
2039 int dmar_device_remove(acpi_handle handle)
2041 return dmar_device_hotplug(handle, false);