net/mlx5e: Implement Fragmented Work Queue (WQ)
[linux-2.6/btrfs-unstable.git] / drivers / iommu / dmar.c
blob58470f5ced04a9547428ec538dbdec3c58f9a1f2
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;
245 * We expect devices with endpoint scope to have normal PCI
246 * headers, and devices with bridge scope to have bridge PCI
247 * headers. However PCI NTB devices may be listed in the
248 * DMAR table with bridge scope, even though they have a
249 * normal PCI header. NTB devices are identified by class
250 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
251 * for this special case.
253 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
254 info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
255 (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
256 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
257 info->dev->class >> 8 != PCI_CLASS_BRIDGE_OTHER))) {
258 pr_warn("Device scope type does not match for %s\n",
259 pci_name(info->dev));
260 return -EINVAL;
263 for_each_dev_scope(devices, devices_cnt, i, tmp)
264 if (tmp == NULL) {
265 devices[i].bus = info->dev->bus->number;
266 devices[i].devfn = info->dev->devfn;
267 rcu_assign_pointer(devices[i].dev,
268 get_device(dev));
269 return 1;
271 BUG_ON(i >= devices_cnt);
274 return 0;
277 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
278 struct dmar_dev_scope *devices, int count)
280 int index;
281 struct device *tmp;
283 if (info->seg != segment)
284 return 0;
286 for_each_active_dev_scope(devices, count, index, tmp)
287 if (tmp == &info->dev->dev) {
288 RCU_INIT_POINTER(devices[index].dev, NULL);
289 synchronize_rcu();
290 put_device(tmp);
291 return 1;
294 return 0;
297 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
299 int ret = 0;
300 struct dmar_drhd_unit *dmaru;
301 struct acpi_dmar_hardware_unit *drhd;
303 for_each_drhd_unit(dmaru) {
304 if (dmaru->include_all)
305 continue;
307 drhd = container_of(dmaru->hdr,
308 struct acpi_dmar_hardware_unit, header);
309 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
310 ((void *)drhd) + drhd->header.length,
311 dmaru->segment,
312 dmaru->devices, dmaru->devices_cnt);
313 if (ret != 0)
314 break;
316 if (ret >= 0)
317 ret = dmar_iommu_notify_scope_dev(info);
318 if (ret < 0 && dmar_dev_scope_status == 0)
319 dmar_dev_scope_status = ret;
321 return ret;
324 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
326 struct dmar_drhd_unit *dmaru;
328 for_each_drhd_unit(dmaru)
329 if (dmar_remove_dev_scope(info, dmaru->segment,
330 dmaru->devices, dmaru->devices_cnt))
331 break;
332 dmar_iommu_notify_scope_dev(info);
335 static int dmar_pci_bus_notifier(struct notifier_block *nb,
336 unsigned long action, void *data)
338 struct pci_dev *pdev = to_pci_dev(data);
339 struct dmar_pci_notify_info *info;
341 /* Only care about add/remove events for physical functions */
342 if (pdev->is_virtfn)
343 return NOTIFY_DONE;
344 if (action != BUS_NOTIFY_ADD_DEVICE &&
345 action != BUS_NOTIFY_REMOVED_DEVICE)
346 return NOTIFY_DONE;
348 info = dmar_alloc_pci_notify_info(pdev, action);
349 if (!info)
350 return NOTIFY_DONE;
352 down_write(&dmar_global_lock);
353 if (action == BUS_NOTIFY_ADD_DEVICE)
354 dmar_pci_bus_add_dev(info);
355 else if (action == BUS_NOTIFY_REMOVED_DEVICE)
356 dmar_pci_bus_del_dev(info);
357 up_write(&dmar_global_lock);
359 dmar_free_pci_notify_info(info);
361 return NOTIFY_OK;
364 static struct notifier_block dmar_pci_bus_nb = {
365 .notifier_call = dmar_pci_bus_notifier,
366 .priority = INT_MIN,
369 static struct dmar_drhd_unit *
370 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
372 struct dmar_drhd_unit *dmaru;
374 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
375 if (dmaru->segment == drhd->segment &&
376 dmaru->reg_base_addr == drhd->address)
377 return dmaru;
379 return NULL;
383 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
384 * structure which uniquely represent one DMA remapping hardware unit
385 * present in the platform
387 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
389 struct acpi_dmar_hardware_unit *drhd;
390 struct dmar_drhd_unit *dmaru;
391 int ret = 0;
393 drhd = (struct acpi_dmar_hardware_unit *)header;
394 dmaru = dmar_find_dmaru(drhd);
395 if (dmaru)
396 goto out;
398 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
399 if (!dmaru)
400 return -ENOMEM;
403 * If header is allocated from slab by ACPI _DSM method, we need to
404 * copy the content because the memory buffer will be freed on return.
406 dmaru->hdr = (void *)(dmaru + 1);
407 memcpy(dmaru->hdr, header, header->length);
408 dmaru->reg_base_addr = drhd->address;
409 dmaru->segment = drhd->segment;
410 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
411 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
412 ((void *)drhd) + drhd->header.length,
413 &dmaru->devices_cnt);
414 if (dmaru->devices_cnt && dmaru->devices == NULL) {
415 kfree(dmaru);
416 return -ENOMEM;
419 ret = alloc_iommu(dmaru);
420 if (ret) {
421 dmar_free_dev_scope(&dmaru->devices,
422 &dmaru->devices_cnt);
423 kfree(dmaru);
424 return ret;
426 dmar_register_drhd_unit(dmaru);
428 out:
429 if (arg)
430 (*(int *)arg)++;
432 return 0;
435 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
437 if (dmaru->devices && dmaru->devices_cnt)
438 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
439 if (dmaru->iommu)
440 free_iommu(dmaru->iommu);
441 kfree(dmaru);
444 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
445 void *arg)
447 struct acpi_dmar_andd *andd = (void *)header;
449 /* Check for NUL termination within the designated length */
450 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
451 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND,
452 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
453 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
454 dmi_get_system_info(DMI_BIOS_VENDOR),
455 dmi_get_system_info(DMI_BIOS_VERSION),
456 dmi_get_system_info(DMI_PRODUCT_VERSION));
457 return -EINVAL;
459 pr_info("ANDD device: %x name: %s\n", andd->device_number,
460 andd->device_name);
462 return 0;
465 #ifdef CONFIG_ACPI_NUMA
466 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
468 struct acpi_dmar_rhsa *rhsa;
469 struct dmar_drhd_unit *drhd;
471 rhsa = (struct acpi_dmar_rhsa *)header;
472 for_each_drhd_unit(drhd) {
473 if (drhd->reg_base_addr == rhsa->base_address) {
474 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
476 if (!node_online(node))
477 node = -1;
478 drhd->iommu->node = node;
479 return 0;
482 WARN_TAINT(
483 1, TAINT_FIRMWARE_WORKAROUND,
484 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
485 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
486 drhd->reg_base_addr,
487 dmi_get_system_info(DMI_BIOS_VENDOR),
488 dmi_get_system_info(DMI_BIOS_VERSION),
489 dmi_get_system_info(DMI_PRODUCT_VERSION));
491 return 0;
493 #else
494 #define dmar_parse_one_rhsa dmar_res_noop
495 #endif
497 static void __init
498 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
500 struct acpi_dmar_hardware_unit *drhd;
501 struct acpi_dmar_reserved_memory *rmrr;
502 struct acpi_dmar_atsr *atsr;
503 struct acpi_dmar_rhsa *rhsa;
505 switch (header->type) {
506 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
507 drhd = container_of(header, struct acpi_dmar_hardware_unit,
508 header);
509 pr_info("DRHD base: %#016Lx flags: %#x\n",
510 (unsigned long long)drhd->address, drhd->flags);
511 break;
512 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
513 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
514 header);
515 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
516 (unsigned long long)rmrr->base_address,
517 (unsigned long long)rmrr->end_address);
518 break;
519 case ACPI_DMAR_TYPE_ROOT_ATS:
520 atsr = container_of(header, struct acpi_dmar_atsr, header);
521 pr_info("ATSR flags: %#x\n", atsr->flags);
522 break;
523 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
524 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
525 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
526 (unsigned long long)rhsa->base_address,
527 rhsa->proximity_domain);
528 break;
529 case ACPI_DMAR_TYPE_NAMESPACE:
530 /* We don't print this here because we need to sanity-check
531 it first. So print it in dmar_parse_one_andd() instead. */
532 break;
537 * dmar_table_detect - checks to see if the platform supports DMAR devices
539 static int __init dmar_table_detect(void)
541 acpi_status status = AE_OK;
543 /* if we could find DMAR table, then there are DMAR devices */
544 status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
545 (struct acpi_table_header **)&dmar_tbl,
546 &dmar_tbl_size);
548 if (ACPI_SUCCESS(status) && !dmar_tbl) {
549 pr_warn("Unable to map DMAR\n");
550 status = AE_NOT_FOUND;
553 return (ACPI_SUCCESS(status) ? 1 : 0);
556 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
557 size_t len, struct dmar_res_callback *cb)
559 int ret = 0;
560 struct acpi_dmar_header *iter, *next;
561 struct acpi_dmar_header *end = ((void *)start) + len;
563 for (iter = start; iter < end && ret == 0; 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 ret = -EINVAL;
573 break;
576 if (cb->print_entry)
577 dmar_table_print_dmar_entry(iter);
579 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
580 /* continue for forward compatibility */
581 pr_debug("Unknown DMAR structure type %d\n",
582 iter->type);
583 } else if (cb->cb[iter->type]) {
584 ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
585 } else if (!cb->ignore_unhandled) {
586 pr_warn("No handler for DMAR structure type %d\n",
587 iter->type);
588 ret = -EINVAL;
592 return ret;
595 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
596 struct dmar_res_callback *cb)
598 return dmar_walk_remapping_entries((void *)(dmar + 1),
599 dmar->header.length - sizeof(*dmar), cb);
603 * parse_dmar_table - parses the DMA reporting table
605 static int __init
606 parse_dmar_table(void)
608 struct acpi_table_dmar *dmar;
609 int ret = 0;
610 int drhd_count = 0;
611 struct dmar_res_callback cb = {
612 .print_entry = true,
613 .ignore_unhandled = true,
614 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
615 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
616 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
617 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
618 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
619 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
623 * Do it again, earlier dmar_tbl mapping could be mapped with
624 * fixed map.
626 dmar_table_detect();
629 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
630 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
632 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
634 dmar = (struct acpi_table_dmar *)dmar_tbl;
635 if (!dmar)
636 return -ENODEV;
638 if (dmar->width < PAGE_SHIFT - 1) {
639 pr_warn("Invalid DMAR haw\n");
640 return -EINVAL;
643 pr_info("Host address width %d\n", dmar->width + 1);
644 ret = dmar_walk_dmar_table(dmar, &cb);
645 if (ret == 0 && drhd_count == 0)
646 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
648 return ret;
651 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
652 int cnt, struct pci_dev *dev)
654 int index;
655 struct device *tmp;
657 while (dev) {
658 for_each_active_dev_scope(devices, cnt, index, tmp)
659 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
660 return 1;
662 /* Check our parent */
663 dev = dev->bus->self;
666 return 0;
669 struct dmar_drhd_unit *
670 dmar_find_matched_drhd_unit(struct pci_dev *dev)
672 struct dmar_drhd_unit *dmaru;
673 struct acpi_dmar_hardware_unit *drhd;
675 dev = pci_physfn(dev);
677 rcu_read_lock();
678 for_each_drhd_unit(dmaru) {
679 drhd = container_of(dmaru->hdr,
680 struct acpi_dmar_hardware_unit,
681 header);
683 if (dmaru->include_all &&
684 drhd->segment == pci_domain_nr(dev->bus))
685 goto out;
687 if (dmar_pci_device_match(dmaru->devices,
688 dmaru->devices_cnt, dev))
689 goto out;
691 dmaru = NULL;
692 out:
693 rcu_read_unlock();
695 return dmaru;
698 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
699 struct acpi_device *adev)
701 struct dmar_drhd_unit *dmaru;
702 struct acpi_dmar_hardware_unit *drhd;
703 struct acpi_dmar_device_scope *scope;
704 struct device *tmp;
705 int i;
706 struct acpi_dmar_pci_path *path;
708 for_each_drhd_unit(dmaru) {
709 drhd = container_of(dmaru->hdr,
710 struct acpi_dmar_hardware_unit,
711 header);
713 for (scope = (void *)(drhd + 1);
714 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
715 scope = ((void *)scope) + scope->length) {
716 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
717 continue;
718 if (scope->enumeration_id != device_number)
719 continue;
721 path = (void *)(scope + 1);
722 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
723 dev_name(&adev->dev), dmaru->reg_base_addr,
724 scope->bus, path->device, path->function);
725 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
726 if (tmp == NULL) {
727 dmaru->devices[i].bus = scope->bus;
728 dmaru->devices[i].devfn = PCI_DEVFN(path->device,
729 path->function);
730 rcu_assign_pointer(dmaru->devices[i].dev,
731 get_device(&adev->dev));
732 return;
734 BUG_ON(i >= dmaru->devices_cnt);
737 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
738 device_number, dev_name(&adev->dev));
741 static int __init dmar_acpi_dev_scope_init(void)
743 struct acpi_dmar_andd *andd;
745 if (dmar_tbl == NULL)
746 return -ENODEV;
748 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
749 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
750 andd = ((void *)andd) + andd->header.length) {
751 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
752 acpi_handle h;
753 struct acpi_device *adev;
755 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
756 andd->device_name,
757 &h))) {
758 pr_err("Failed to find handle for ACPI object %s\n",
759 andd->device_name);
760 continue;
762 if (acpi_bus_get_device(h, &adev)) {
763 pr_err("Failed to get device for ACPI object %s\n",
764 andd->device_name);
765 continue;
767 dmar_acpi_insert_dev_scope(andd->device_number, adev);
770 return 0;
773 int __init dmar_dev_scope_init(void)
775 struct pci_dev *dev = NULL;
776 struct dmar_pci_notify_info *info;
778 if (dmar_dev_scope_status != 1)
779 return dmar_dev_scope_status;
781 if (list_empty(&dmar_drhd_units)) {
782 dmar_dev_scope_status = -ENODEV;
783 } else {
784 dmar_dev_scope_status = 0;
786 dmar_acpi_dev_scope_init();
788 for_each_pci_dev(dev) {
789 if (dev->is_virtfn)
790 continue;
792 info = dmar_alloc_pci_notify_info(dev,
793 BUS_NOTIFY_ADD_DEVICE);
794 if (!info) {
795 return dmar_dev_scope_status;
796 } else {
797 dmar_pci_bus_add_dev(info);
798 dmar_free_pci_notify_info(info);
802 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
805 return dmar_dev_scope_status;
809 int __init dmar_table_init(void)
811 static int dmar_table_initialized;
812 int ret;
814 if (dmar_table_initialized == 0) {
815 ret = parse_dmar_table();
816 if (ret < 0) {
817 if (ret != -ENODEV)
818 pr_info("Parse DMAR table failure.\n");
819 } else if (list_empty(&dmar_drhd_units)) {
820 pr_info("No DMAR devices found\n");
821 ret = -ENODEV;
824 if (ret < 0)
825 dmar_table_initialized = ret;
826 else
827 dmar_table_initialized = 1;
830 return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
833 static void warn_invalid_dmar(u64 addr, const char *message)
835 WARN_TAINT_ONCE(
836 1, TAINT_FIRMWARE_WORKAROUND,
837 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
838 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
839 addr, message,
840 dmi_get_system_info(DMI_BIOS_VENDOR),
841 dmi_get_system_info(DMI_BIOS_VERSION),
842 dmi_get_system_info(DMI_PRODUCT_VERSION));
845 static int __ref
846 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
848 struct acpi_dmar_hardware_unit *drhd;
849 void __iomem *addr;
850 u64 cap, ecap;
852 drhd = (void *)entry;
853 if (!drhd->address) {
854 warn_invalid_dmar(0, "");
855 return -EINVAL;
858 if (arg)
859 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
860 else
861 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
862 if (!addr) {
863 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
864 return -EINVAL;
867 cap = dmar_readq(addr + DMAR_CAP_REG);
868 ecap = dmar_readq(addr + DMAR_ECAP_REG);
870 if (arg)
871 iounmap(addr);
872 else
873 early_iounmap(addr, VTD_PAGE_SIZE);
875 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
876 warn_invalid_dmar(drhd->address, " returns all ones");
877 return -EINVAL;
880 return 0;
883 int __init detect_intel_iommu(void)
885 int ret;
886 struct dmar_res_callback validate_drhd_cb = {
887 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
888 .ignore_unhandled = true,
891 down_write(&dmar_global_lock);
892 ret = dmar_table_detect();
893 if (ret)
894 ret = !dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
895 &validate_drhd_cb);
896 if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
897 iommu_detected = 1;
898 /* Make sure ACS will be enabled */
899 pci_request_acs();
902 #ifdef CONFIG_X86
903 if (ret)
904 x86_init.iommu.iommu_init = intel_iommu_init;
905 #endif
907 early_acpi_os_unmap_memory((void __iomem *)dmar_tbl, dmar_tbl_size);
908 dmar_tbl = NULL;
909 up_write(&dmar_global_lock);
911 return ret ? 1 : -ENODEV;
915 static void unmap_iommu(struct intel_iommu *iommu)
917 iounmap(iommu->reg);
918 release_mem_region(iommu->reg_phys, iommu->reg_size);
922 * map_iommu: map the iommu's registers
923 * @iommu: the iommu to map
924 * @phys_addr: the physical address of the base resgister
926 * Memory map the iommu's registers. Start w/ a single page, and
927 * possibly expand if that turns out to be insufficent.
929 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
931 int map_size, err=0;
933 iommu->reg_phys = phys_addr;
934 iommu->reg_size = VTD_PAGE_SIZE;
936 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
937 pr_err("Can't reserve memory\n");
938 err = -EBUSY;
939 goto out;
942 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
943 if (!iommu->reg) {
944 pr_err("Can't map the region\n");
945 err = -ENOMEM;
946 goto release;
949 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
950 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
952 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
953 err = -EINVAL;
954 warn_invalid_dmar(phys_addr, " returns all ones");
955 goto unmap;
958 /* the registers might be more than one page */
959 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
960 cap_max_fault_reg_offset(iommu->cap));
961 map_size = VTD_PAGE_ALIGN(map_size);
962 if (map_size > iommu->reg_size) {
963 iounmap(iommu->reg);
964 release_mem_region(iommu->reg_phys, iommu->reg_size);
965 iommu->reg_size = map_size;
966 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
967 iommu->name)) {
968 pr_err("Can't reserve memory\n");
969 err = -EBUSY;
970 goto out;
972 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
973 if (!iommu->reg) {
974 pr_err("Can't map the region\n");
975 err = -ENOMEM;
976 goto release;
979 err = 0;
980 goto out;
982 unmap:
983 iounmap(iommu->reg);
984 release:
985 release_mem_region(iommu->reg_phys, iommu->reg_size);
986 out:
987 return err;
990 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
992 iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
993 DMAR_UNITS_SUPPORTED);
994 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
995 iommu->seq_id = -1;
996 } else {
997 set_bit(iommu->seq_id, dmar_seq_ids);
998 sprintf(iommu->name, "dmar%d", iommu->seq_id);
1001 return iommu->seq_id;
1004 static void dmar_free_seq_id(struct intel_iommu *iommu)
1006 if (iommu->seq_id >= 0) {
1007 clear_bit(iommu->seq_id, dmar_seq_ids);
1008 iommu->seq_id = -1;
1012 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1014 struct intel_iommu *iommu;
1015 u32 ver, sts;
1016 int agaw = 0;
1017 int msagaw = 0;
1018 int err;
1020 if (!drhd->reg_base_addr) {
1021 warn_invalid_dmar(0, "");
1022 return -EINVAL;
1025 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1026 if (!iommu)
1027 return -ENOMEM;
1029 if (dmar_alloc_seq_id(iommu) < 0) {
1030 pr_err("Failed to allocate seq_id\n");
1031 err = -ENOSPC;
1032 goto error;
1035 err = map_iommu(iommu, drhd->reg_base_addr);
1036 if (err) {
1037 pr_err("Failed to map %s\n", iommu->name);
1038 goto error_free_seq_id;
1041 err = -EINVAL;
1042 agaw = iommu_calculate_agaw(iommu);
1043 if (agaw < 0) {
1044 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1045 iommu->seq_id);
1046 goto err_unmap;
1048 msagaw = iommu_calculate_max_sagaw(iommu);
1049 if (msagaw < 0) {
1050 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1051 iommu->seq_id);
1052 goto err_unmap;
1054 iommu->agaw = agaw;
1055 iommu->msagaw = msagaw;
1056 iommu->segment = drhd->segment;
1058 iommu->node = -1;
1060 ver = readl(iommu->reg + DMAR_VER_REG);
1061 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1062 iommu->name,
1063 (unsigned long long)drhd->reg_base_addr,
1064 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1065 (unsigned long long)iommu->cap,
1066 (unsigned long long)iommu->ecap);
1068 /* Reflect status in gcmd */
1069 sts = readl(iommu->reg + DMAR_GSTS_REG);
1070 if (sts & DMA_GSTS_IRES)
1071 iommu->gcmd |= DMA_GCMD_IRE;
1072 if (sts & DMA_GSTS_TES)
1073 iommu->gcmd |= DMA_GCMD_TE;
1074 if (sts & DMA_GSTS_QIES)
1075 iommu->gcmd |= DMA_GCMD_QIE;
1077 raw_spin_lock_init(&iommu->register_lock);
1079 if (intel_iommu_enabled) {
1080 iommu->iommu_dev = iommu_device_create(NULL, iommu,
1081 intel_iommu_groups,
1082 "%s", iommu->name);
1084 if (IS_ERR(iommu->iommu_dev)) {
1085 err = PTR_ERR(iommu->iommu_dev);
1086 goto err_unmap;
1090 drhd->iommu = iommu;
1092 return 0;
1094 err_unmap:
1095 unmap_iommu(iommu);
1096 error_free_seq_id:
1097 dmar_free_seq_id(iommu);
1098 error:
1099 kfree(iommu);
1100 return err;
1103 static void free_iommu(struct intel_iommu *iommu)
1105 iommu_device_destroy(iommu->iommu_dev);
1107 if (iommu->irq) {
1108 if (iommu->pr_irq) {
1109 free_irq(iommu->pr_irq, iommu);
1110 dmar_free_hwirq(iommu->pr_irq);
1111 iommu->pr_irq = 0;
1113 free_irq(iommu->irq, iommu);
1114 dmar_free_hwirq(iommu->irq);
1115 iommu->irq = 0;
1118 if (iommu->qi) {
1119 free_page((unsigned long)iommu->qi->desc);
1120 kfree(iommu->qi->desc_status);
1121 kfree(iommu->qi);
1124 if (iommu->reg)
1125 unmap_iommu(iommu);
1127 dmar_free_seq_id(iommu);
1128 kfree(iommu);
1132 * Reclaim all the submitted descriptors which have completed its work.
1134 static inline void reclaim_free_desc(struct q_inval *qi)
1136 while (qi->desc_status[qi->free_tail] == QI_DONE ||
1137 qi->desc_status[qi->free_tail] == QI_ABORT) {
1138 qi->desc_status[qi->free_tail] = QI_FREE;
1139 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1140 qi->free_cnt++;
1144 static int qi_check_fault(struct intel_iommu *iommu, int index)
1146 u32 fault;
1147 int head, tail;
1148 struct q_inval *qi = iommu->qi;
1149 int wait_index = (index + 1) % QI_LENGTH;
1151 if (qi->desc_status[wait_index] == QI_ABORT)
1152 return -EAGAIN;
1154 fault = readl(iommu->reg + DMAR_FSTS_REG);
1157 * If IQE happens, the head points to the descriptor associated
1158 * with the error. No new descriptors are fetched until the IQE
1159 * is cleared.
1161 if (fault & DMA_FSTS_IQE) {
1162 head = readl(iommu->reg + DMAR_IQH_REG);
1163 if ((head >> DMAR_IQ_SHIFT) == index) {
1164 pr_err("VT-d detected invalid descriptor: "
1165 "low=%llx, high=%llx\n",
1166 (unsigned long long)qi->desc[index].low,
1167 (unsigned long long)qi->desc[index].high);
1168 memcpy(&qi->desc[index], &qi->desc[wait_index],
1169 sizeof(struct qi_desc));
1170 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1171 return -EINVAL;
1176 * If ITE happens, all pending wait_desc commands are aborted.
1177 * No new descriptors are fetched until the ITE is cleared.
1179 if (fault & DMA_FSTS_ITE) {
1180 head = readl(iommu->reg + DMAR_IQH_REG);
1181 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1182 head |= 1;
1183 tail = readl(iommu->reg + DMAR_IQT_REG);
1184 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1186 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1188 do {
1189 if (qi->desc_status[head] == QI_IN_USE)
1190 qi->desc_status[head] = QI_ABORT;
1191 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1192 } while (head != tail);
1194 if (qi->desc_status[wait_index] == QI_ABORT)
1195 return -EAGAIN;
1198 if (fault & DMA_FSTS_ICE)
1199 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1201 return 0;
1205 * Submit the queued invalidation descriptor to the remapping
1206 * hardware unit and wait for its completion.
1208 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1210 int rc;
1211 struct q_inval *qi = iommu->qi;
1212 struct qi_desc *hw, wait_desc;
1213 int wait_index, index;
1214 unsigned long flags;
1216 if (!qi)
1217 return 0;
1219 hw = qi->desc;
1221 restart:
1222 rc = 0;
1224 raw_spin_lock_irqsave(&qi->q_lock, flags);
1225 while (qi->free_cnt < 3) {
1226 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1227 cpu_relax();
1228 raw_spin_lock_irqsave(&qi->q_lock, flags);
1231 index = qi->free_head;
1232 wait_index = (index + 1) % QI_LENGTH;
1234 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1236 hw[index] = *desc;
1238 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
1239 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1240 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
1242 hw[wait_index] = wait_desc;
1244 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1245 qi->free_cnt -= 2;
1248 * update the HW tail register indicating the presence of
1249 * new descriptors.
1251 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
1253 while (qi->desc_status[wait_index] != QI_DONE) {
1255 * We will leave the interrupts disabled, to prevent interrupt
1256 * context to queue another cmd while a cmd is already submitted
1257 * and waiting for completion on this cpu. This is to avoid
1258 * a deadlock where the interrupt context can wait indefinitely
1259 * for free slots in the queue.
1261 rc = qi_check_fault(iommu, index);
1262 if (rc)
1263 break;
1265 raw_spin_unlock(&qi->q_lock);
1266 cpu_relax();
1267 raw_spin_lock(&qi->q_lock);
1270 qi->desc_status[index] = QI_DONE;
1272 reclaim_free_desc(qi);
1273 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1275 if (rc == -EAGAIN)
1276 goto restart;
1278 return rc;
1282 * Flush the global interrupt entry cache.
1284 void qi_global_iec(struct intel_iommu *iommu)
1286 struct qi_desc desc;
1288 desc.low = QI_IEC_TYPE;
1289 desc.high = 0;
1291 /* should never fail */
1292 qi_submit_sync(&desc, iommu);
1295 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1296 u64 type)
1298 struct qi_desc desc;
1300 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1301 | QI_CC_GRAN(type) | QI_CC_TYPE;
1302 desc.high = 0;
1304 qi_submit_sync(&desc, iommu);
1307 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1308 unsigned int size_order, u64 type)
1310 u8 dw = 0, dr = 0;
1312 struct qi_desc desc;
1313 int ih = 0;
1315 if (cap_write_drain(iommu->cap))
1316 dw = 1;
1318 if (cap_read_drain(iommu->cap))
1319 dr = 1;
1321 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1322 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1323 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1324 | QI_IOTLB_AM(size_order);
1326 qi_submit_sync(&desc, iommu);
1329 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
1330 u64 addr, unsigned mask)
1332 struct qi_desc desc;
1334 if (mask) {
1335 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
1336 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1337 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1338 } else
1339 desc.high = QI_DEV_IOTLB_ADDR(addr);
1341 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1342 qdep = 0;
1344 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1345 QI_DIOTLB_TYPE;
1347 qi_submit_sync(&desc, iommu);
1351 * Disable Queued Invalidation interface.
1353 void dmar_disable_qi(struct intel_iommu *iommu)
1355 unsigned long flags;
1356 u32 sts;
1357 cycles_t start_time = get_cycles();
1359 if (!ecap_qis(iommu->ecap))
1360 return;
1362 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1364 sts = readl(iommu->reg + DMAR_GSTS_REG);
1365 if (!(sts & DMA_GSTS_QIES))
1366 goto end;
1369 * Give a chance to HW to complete the pending invalidation requests.
1371 while ((readl(iommu->reg + DMAR_IQT_REG) !=
1372 readl(iommu->reg + DMAR_IQH_REG)) &&
1373 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1374 cpu_relax();
1376 iommu->gcmd &= ~DMA_GCMD_QIE;
1377 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1379 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1380 !(sts & DMA_GSTS_QIES), sts);
1381 end:
1382 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1386 * Enable queued invalidation.
1388 static void __dmar_enable_qi(struct intel_iommu *iommu)
1390 u32 sts;
1391 unsigned long flags;
1392 struct q_inval *qi = iommu->qi;
1394 qi->free_head = qi->free_tail = 0;
1395 qi->free_cnt = QI_LENGTH;
1397 raw_spin_lock_irqsave(&iommu->register_lock, flags);
1399 /* write zero to the tail reg */
1400 writel(0, iommu->reg + DMAR_IQT_REG);
1402 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1404 iommu->gcmd |= DMA_GCMD_QIE;
1405 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1407 /* Make sure hardware complete it */
1408 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1410 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1414 * Enable Queued Invalidation interface. This is a must to support
1415 * interrupt-remapping. Also used by DMA-remapping, which replaces
1416 * register based IOTLB invalidation.
1418 int dmar_enable_qi(struct intel_iommu *iommu)
1420 struct q_inval *qi;
1421 struct page *desc_page;
1423 if (!ecap_qis(iommu->ecap))
1424 return -ENOENT;
1427 * queued invalidation is already setup and enabled.
1429 if (iommu->qi)
1430 return 0;
1432 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1433 if (!iommu->qi)
1434 return -ENOMEM;
1436 qi = iommu->qi;
1439 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1440 if (!desc_page) {
1441 kfree(qi);
1442 iommu->qi = NULL;
1443 return -ENOMEM;
1446 qi->desc = page_address(desc_page);
1448 qi->desc_status = kzalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1449 if (!qi->desc_status) {
1450 free_page((unsigned long) qi->desc);
1451 kfree(qi);
1452 iommu->qi = NULL;
1453 return -ENOMEM;
1456 raw_spin_lock_init(&qi->q_lock);
1458 __dmar_enable_qi(iommu);
1460 return 0;
1463 /* iommu interrupt handling. Most stuff are MSI-like. */
1465 enum faulttype {
1466 DMA_REMAP,
1467 INTR_REMAP,
1468 UNKNOWN,
1471 static const char *dma_remap_fault_reasons[] =
1473 "Software",
1474 "Present bit in root entry is clear",
1475 "Present bit in context entry is clear",
1476 "Invalid context entry",
1477 "Access beyond MGAW",
1478 "PTE Write access is not set",
1479 "PTE Read access is not set",
1480 "Next page table ptr is invalid",
1481 "Root table address invalid",
1482 "Context table ptr is invalid",
1483 "non-zero reserved fields in RTP",
1484 "non-zero reserved fields in CTP",
1485 "non-zero reserved fields in PTE",
1486 "PCE for translation request specifies blocking",
1489 static const char *irq_remap_fault_reasons[] =
1491 "Detected reserved fields in the decoded interrupt-remapped request",
1492 "Interrupt index exceeded the interrupt-remapping table size",
1493 "Present field in the IRTE entry is clear",
1494 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1495 "Detected reserved fields in the IRTE entry",
1496 "Blocked a compatibility format interrupt request",
1497 "Blocked an interrupt request due to source-id verification failure",
1500 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1502 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1503 ARRAY_SIZE(irq_remap_fault_reasons))) {
1504 *fault_type = INTR_REMAP;
1505 return irq_remap_fault_reasons[fault_reason - 0x20];
1506 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1507 *fault_type = DMA_REMAP;
1508 return dma_remap_fault_reasons[fault_reason];
1509 } else {
1510 *fault_type = UNKNOWN;
1511 return "Unknown";
1516 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1518 if (iommu->irq == irq)
1519 return DMAR_FECTL_REG;
1520 else if (iommu->pr_irq == irq)
1521 return DMAR_PECTL_REG;
1522 else
1523 BUG();
1526 void dmar_msi_unmask(struct irq_data *data)
1528 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1529 int reg = dmar_msi_reg(iommu, data->irq);
1530 unsigned long flag;
1532 /* unmask it */
1533 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1534 writel(0, iommu->reg + reg);
1535 /* Read a reg to force flush the post write */
1536 readl(iommu->reg + reg);
1537 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1540 void dmar_msi_mask(struct irq_data *data)
1542 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1543 int reg = dmar_msi_reg(iommu, data->irq);
1544 unsigned long flag;
1546 /* mask it */
1547 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1548 writel(DMA_FECTL_IM, iommu->reg + reg);
1549 /* Read a reg to force flush the post write */
1550 readl(iommu->reg + reg);
1551 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1554 void dmar_msi_write(int irq, struct msi_msg *msg)
1556 struct intel_iommu *iommu = irq_get_handler_data(irq);
1557 int reg = dmar_msi_reg(iommu, irq);
1558 unsigned long flag;
1560 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1561 writel(msg->data, iommu->reg + reg + 4);
1562 writel(msg->address_lo, iommu->reg + reg + 8);
1563 writel(msg->address_hi, iommu->reg + reg + 12);
1564 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1567 void dmar_msi_read(int irq, struct msi_msg *msg)
1569 struct intel_iommu *iommu = irq_get_handler_data(irq);
1570 int reg = dmar_msi_reg(iommu, irq);
1571 unsigned long flag;
1573 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1574 msg->data = readl(iommu->reg + reg + 4);
1575 msg->address_lo = readl(iommu->reg + reg + 8);
1576 msg->address_hi = readl(iommu->reg + reg + 12);
1577 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1580 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1581 u8 fault_reason, u16 source_id, unsigned long long addr)
1583 const char *reason;
1584 int fault_type;
1586 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1588 if (fault_type == INTR_REMAP)
1589 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1590 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1591 PCI_FUNC(source_id & 0xFF), addr >> 48,
1592 fault_reason, reason);
1593 else
1594 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1595 type ? "DMA Read" : "DMA Write",
1596 source_id >> 8, PCI_SLOT(source_id & 0xFF),
1597 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1598 return 0;
1601 #define PRIMARY_FAULT_REG_LEN (16)
1602 irqreturn_t dmar_fault(int irq, void *dev_id)
1604 struct intel_iommu *iommu = dev_id;
1605 int reg, fault_index;
1606 u32 fault_status;
1607 unsigned long flag;
1608 bool ratelimited;
1609 static DEFINE_RATELIMIT_STATE(rs,
1610 DEFAULT_RATELIMIT_INTERVAL,
1611 DEFAULT_RATELIMIT_BURST);
1613 /* Disable printing, simply clear the fault when ratelimited */
1614 ratelimited = !__ratelimit(&rs);
1616 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1617 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1618 if (fault_status && !ratelimited)
1619 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1621 /* TBD: ignore advanced fault log currently */
1622 if (!(fault_status & DMA_FSTS_PPF))
1623 goto unlock_exit;
1625 fault_index = dma_fsts_fault_record_index(fault_status);
1626 reg = cap_fault_reg_offset(iommu->cap);
1627 while (1) {
1628 u8 fault_reason;
1629 u16 source_id;
1630 u64 guest_addr;
1631 int type;
1632 u32 data;
1634 /* highest 32 bits */
1635 data = readl(iommu->reg + reg +
1636 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1637 if (!(data & DMA_FRCD_F))
1638 break;
1640 if (!ratelimited) {
1641 fault_reason = dma_frcd_fault_reason(data);
1642 type = dma_frcd_type(data);
1644 data = readl(iommu->reg + reg +
1645 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1646 source_id = dma_frcd_source_id(data);
1648 guest_addr = dmar_readq(iommu->reg + reg +
1649 fault_index * PRIMARY_FAULT_REG_LEN);
1650 guest_addr = dma_frcd_page_addr(guest_addr);
1653 /* clear the fault */
1654 writel(DMA_FRCD_F, iommu->reg + reg +
1655 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1657 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1659 if (!ratelimited)
1660 dmar_fault_do_one(iommu, type, fault_reason,
1661 source_id, guest_addr);
1663 fault_index++;
1664 if (fault_index >= cap_num_fault_regs(iommu->cap))
1665 fault_index = 0;
1666 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1669 writel(DMA_FSTS_PFO | DMA_FSTS_PPF, iommu->reg + DMAR_FSTS_REG);
1671 unlock_exit:
1672 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1673 return IRQ_HANDLED;
1676 int dmar_set_interrupt(struct intel_iommu *iommu)
1678 int irq, ret;
1681 * Check if the fault interrupt is already initialized.
1683 if (iommu->irq)
1684 return 0;
1686 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1687 if (irq > 0) {
1688 iommu->irq = irq;
1689 } else {
1690 pr_err("No free IRQ vectors\n");
1691 return -EINVAL;
1694 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1695 if (ret)
1696 pr_err("Can't request irq\n");
1697 return ret;
1700 int __init enable_drhd_fault_handling(void)
1702 struct dmar_drhd_unit *drhd;
1703 struct intel_iommu *iommu;
1706 * Enable fault control interrupt.
1708 for_each_iommu(iommu, drhd) {
1709 u32 fault_status;
1710 int ret = dmar_set_interrupt(iommu);
1712 if (ret) {
1713 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1714 (unsigned long long)drhd->reg_base_addr, ret);
1715 return -1;
1719 * Clear any previous faults.
1721 dmar_fault(iommu->irq, iommu);
1722 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1723 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1726 return 0;
1730 * Re-enable Queued Invalidation interface.
1732 int dmar_reenable_qi(struct intel_iommu *iommu)
1734 if (!ecap_qis(iommu->ecap))
1735 return -ENOENT;
1737 if (!iommu->qi)
1738 return -ENOENT;
1741 * First disable queued invalidation.
1743 dmar_disable_qi(iommu);
1745 * Then enable queued invalidation again. Since there is no pending
1746 * invalidation requests now, it's safe to re-enable queued
1747 * invalidation.
1749 __dmar_enable_qi(iommu);
1751 return 0;
1755 * Check interrupt remapping support in DMAR table description.
1757 int __init dmar_ir_support(void)
1759 struct acpi_table_dmar *dmar;
1760 dmar = (struct acpi_table_dmar *)dmar_tbl;
1761 if (!dmar)
1762 return 0;
1763 return dmar->flags & 0x1;
1766 /* Check whether DMAR units are in use */
1767 static inline bool dmar_in_use(void)
1769 return irq_remapping_enabled || intel_iommu_enabled;
1772 static int __init dmar_free_unused_resources(void)
1774 struct dmar_drhd_unit *dmaru, *dmaru_n;
1776 if (dmar_in_use())
1777 return 0;
1779 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1780 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1782 down_write(&dmar_global_lock);
1783 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1784 list_del(&dmaru->list);
1785 dmar_free_drhd(dmaru);
1787 up_write(&dmar_global_lock);
1789 return 0;
1792 late_initcall(dmar_free_unused_resources);
1793 IOMMU_INIT_POST(detect_intel_iommu);
1796 * DMAR Hotplug Support
1797 * For more details, please refer to Intel(R) Virtualization Technology
1798 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1799 * "Remapping Hardware Unit Hot Plug".
1801 static u8 dmar_hp_uuid[] = {
1802 /* 0000 */ 0xA6, 0xA3, 0xC1, 0xD8, 0x9B, 0xBE, 0x9B, 0x4C,
1803 /* 0008 */ 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF
1807 * Currently there's only one revision and BIOS will not check the revision id,
1808 * so use 0 for safety.
1810 #define DMAR_DSM_REV_ID 0
1811 #define DMAR_DSM_FUNC_DRHD 1
1812 #define DMAR_DSM_FUNC_ATSR 2
1813 #define DMAR_DSM_FUNC_RHSA 3
1815 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1817 return acpi_check_dsm(handle, dmar_hp_uuid, DMAR_DSM_REV_ID, 1 << func);
1820 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1821 dmar_res_handler_t handler, void *arg)
1823 int ret = -ENODEV;
1824 union acpi_object *obj;
1825 struct acpi_dmar_header *start;
1826 struct dmar_res_callback callback;
1827 static int res_type[] = {
1828 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1829 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1830 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1833 if (!dmar_detect_dsm(handle, func))
1834 return 0;
1836 obj = acpi_evaluate_dsm_typed(handle, dmar_hp_uuid, DMAR_DSM_REV_ID,
1837 func, NULL, ACPI_TYPE_BUFFER);
1838 if (!obj)
1839 return -ENODEV;
1841 memset(&callback, 0, sizeof(callback));
1842 callback.cb[res_type[func]] = handler;
1843 callback.arg[res_type[func]] = arg;
1844 start = (struct acpi_dmar_header *)obj->buffer.pointer;
1845 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1847 ACPI_FREE(obj);
1849 return ret;
1852 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1854 int ret;
1855 struct dmar_drhd_unit *dmaru;
1857 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1858 if (!dmaru)
1859 return -ENODEV;
1861 ret = dmar_ir_hotplug(dmaru, true);
1862 if (ret == 0)
1863 ret = dmar_iommu_hotplug(dmaru, true);
1865 return ret;
1868 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
1870 int i, ret;
1871 struct device *dev;
1872 struct dmar_drhd_unit *dmaru;
1874 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1875 if (!dmaru)
1876 return 0;
1879 * All PCI devices managed by this unit should have been destroyed.
1881 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
1882 for_each_active_dev_scope(dmaru->devices,
1883 dmaru->devices_cnt, i, dev)
1884 return -EBUSY;
1887 ret = dmar_ir_hotplug(dmaru, false);
1888 if (ret == 0)
1889 ret = dmar_iommu_hotplug(dmaru, false);
1891 return ret;
1894 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
1896 struct dmar_drhd_unit *dmaru;
1898 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1899 if (dmaru) {
1900 list_del_rcu(&dmaru->list);
1901 synchronize_rcu();
1902 dmar_free_drhd(dmaru);
1905 return 0;
1908 static int dmar_hotplug_insert(acpi_handle handle)
1910 int ret;
1911 int drhd_count = 0;
1913 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1914 &dmar_validate_one_drhd, (void *)1);
1915 if (ret)
1916 goto out;
1918 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1919 &dmar_parse_one_drhd, (void *)&drhd_count);
1920 if (ret == 0 && drhd_count == 0) {
1921 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
1922 goto out;
1923 } else if (ret) {
1924 goto release_drhd;
1927 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
1928 &dmar_parse_one_rhsa, NULL);
1929 if (ret)
1930 goto release_drhd;
1932 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1933 &dmar_parse_one_atsr, NULL);
1934 if (ret)
1935 goto release_atsr;
1937 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1938 &dmar_hp_add_drhd, NULL);
1939 if (!ret)
1940 return 0;
1942 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1943 &dmar_hp_remove_drhd, NULL);
1944 release_atsr:
1945 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1946 &dmar_release_one_atsr, NULL);
1947 release_drhd:
1948 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1949 &dmar_hp_release_drhd, NULL);
1950 out:
1951 return ret;
1954 static int dmar_hotplug_remove(acpi_handle handle)
1956 int ret;
1958 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1959 &dmar_check_one_atsr, NULL);
1960 if (ret)
1961 return ret;
1963 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1964 &dmar_hp_remove_drhd, NULL);
1965 if (ret == 0) {
1966 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1967 &dmar_release_one_atsr, NULL));
1968 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1969 &dmar_hp_release_drhd, NULL));
1970 } else {
1971 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1972 &dmar_hp_add_drhd, NULL);
1975 return ret;
1978 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
1979 void *context, void **retval)
1981 acpi_handle *phdl = retval;
1983 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
1984 *phdl = handle;
1985 return AE_CTRL_TERMINATE;
1988 return AE_OK;
1991 static int dmar_device_hotplug(acpi_handle handle, bool insert)
1993 int ret;
1994 acpi_handle tmp = NULL;
1995 acpi_status status;
1997 if (!dmar_in_use())
1998 return 0;
2000 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2001 tmp = handle;
2002 } else {
2003 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2004 ACPI_UINT32_MAX,
2005 dmar_get_dsm_handle,
2006 NULL, NULL, &tmp);
2007 if (ACPI_FAILURE(status)) {
2008 pr_warn("Failed to locate _DSM method.\n");
2009 return -ENXIO;
2012 if (tmp == NULL)
2013 return 0;
2015 down_write(&dmar_global_lock);
2016 if (insert)
2017 ret = dmar_hotplug_insert(tmp);
2018 else
2019 ret = dmar_hotplug_remove(tmp);
2020 up_write(&dmar_global_lock);
2022 return ret;
2025 int dmar_device_add(acpi_handle handle)
2027 return dmar_device_hotplug(handle, true);
2030 int dmar_device_remove(acpi_handle handle)
2032 return dmar_device_hotplug(handle, false);