1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2006, Intel Corporation.
5 * Copyright (C) 2006-2008 Intel Corporation
6 * Author: Ashok Raj <ashok.raj@intel.com>
7 * Author: Shaohua Li <shaohua.li@intel.com>
8 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
10 * This file implements early detection/parsing of Remapping Devices
11 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
14 * These routines are used by both DMA-remapping and Interrupt-remapping
17 #define pr_fmt(fmt) "DMAR: " fmt
19 #include <linux/pci.h>
20 #include <linux/dmar.h>
21 #include <linux/iova.h>
22 #include <linux/intel-iommu.h>
23 #include <linux/timer.h>
24 #include <linux/irq.h>
25 #include <linux/interrupt.h>
26 #include <linux/tboot.h>
27 #include <linux/dmi.h>
28 #include <linux/slab.h>
29 #include <linux/iommu.h>
30 #include <linux/numa.h>
31 #include <asm/irq_remapping.h>
32 #include <asm/iommu_table.h>
34 #include "irq_remapping.h"
36 typedef int (*dmar_res_handler_t
)(struct acpi_dmar_header
*, void *);
37 struct dmar_res_callback
{
38 dmar_res_handler_t cb
[ACPI_DMAR_TYPE_RESERVED
];
39 void *arg
[ACPI_DMAR_TYPE_RESERVED
];
40 bool ignore_unhandled
;
46 * 1) The hotplug framework guarentees that DMAR unit will be hot-added
47 * before IO devices managed by that unit.
48 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
49 * after IO devices managed by that unit.
50 * 3) Hotplug events are rare.
52 * Locking rules for DMA and interrupt remapping related global data structures:
53 * 1) Use dmar_global_lock in process context
54 * 2) Use RCU in interrupt context
56 DECLARE_RWSEM(dmar_global_lock
);
57 LIST_HEAD(dmar_drhd_units
);
59 struct acpi_table_header
* __initdata dmar_tbl
;
60 static int dmar_dev_scope_status
= 1;
61 static unsigned long dmar_seq_ids
[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED
)];
63 static int alloc_iommu(struct dmar_drhd_unit
*drhd
);
64 static void free_iommu(struct intel_iommu
*iommu
);
66 extern const struct iommu_ops intel_iommu_ops
;
68 static void dmar_register_drhd_unit(struct dmar_drhd_unit
*drhd
)
71 * add INCLUDE_ALL at the tail, so scan the list will find it at
74 if (drhd
->include_all
)
75 list_add_tail_rcu(&drhd
->list
, &dmar_drhd_units
);
77 list_add_rcu(&drhd
->list
, &dmar_drhd_units
);
80 void *dmar_alloc_dev_scope(void *start
, void *end
, int *cnt
)
82 struct acpi_dmar_device_scope
*scope
;
87 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_NAMESPACE
||
88 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
||
89 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
91 else if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_IOAPIC
&&
92 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_HPET
) {
93 pr_warn("Unsupported device scope\n");
95 start
+= scope
->length
;
100 return kcalloc(*cnt
, sizeof(struct dmar_dev_scope
), GFP_KERNEL
);
103 void dmar_free_dev_scope(struct dmar_dev_scope
**devices
, int *cnt
)
106 struct device
*tmp_dev
;
108 if (*devices
&& *cnt
) {
109 for_each_active_dev_scope(*devices
, *cnt
, i
, tmp_dev
)
118 /* Optimize out kzalloc()/kfree() for normal cases */
119 static char dmar_pci_notify_info_buf
[64];
121 static struct dmar_pci_notify_info
*
122 dmar_alloc_pci_notify_info(struct pci_dev
*dev
, unsigned long event
)
127 struct dmar_pci_notify_info
*info
;
129 BUG_ON(dev
->is_virtfn
);
131 /* Only generate path[] for device addition event */
132 if (event
== BUS_NOTIFY_ADD_DEVICE
)
133 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
)
136 size
= struct_size(info
, path
, level
);
137 if (size
<= sizeof(dmar_pci_notify_info_buf
)) {
138 info
= (struct dmar_pci_notify_info
*)dmar_pci_notify_info_buf
;
140 info
= kzalloc(size
, GFP_KERNEL
);
142 pr_warn("Out of memory when allocating notify_info "
143 "for %s.\n", pci_name(dev
));
144 if (dmar_dev_scope_status
== 0)
145 dmar_dev_scope_status
= -ENOMEM
;
152 info
->seg
= pci_domain_nr(dev
->bus
);
154 if (event
== BUS_NOTIFY_ADD_DEVICE
) {
155 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
) {
157 info
->path
[level
].bus
= tmp
->bus
->number
;
158 info
->path
[level
].device
= PCI_SLOT(tmp
->devfn
);
159 info
->path
[level
].function
= PCI_FUNC(tmp
->devfn
);
160 if (pci_is_root_bus(tmp
->bus
))
161 info
->bus
= tmp
->bus
->number
;
168 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info
*info
)
170 if ((void *)info
!= dmar_pci_notify_info_buf
)
174 static bool dmar_match_pci_path(struct dmar_pci_notify_info
*info
, int bus
,
175 struct acpi_dmar_pci_path
*path
, int count
)
179 if (info
->bus
!= bus
)
181 if (info
->level
!= count
)
184 for (i
= 0; i
< count
; i
++) {
185 if (path
[i
].device
!= info
->path
[i
].device
||
186 path
[i
].function
!= info
->path
[i
].function
)
198 if (bus
== info
->path
[i
].bus
&&
199 path
[0].device
== info
->path
[i
].device
&&
200 path
[0].function
== info
->path
[i
].function
) {
201 pr_info(FW_BUG
"RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
202 bus
, path
[0].device
, path
[0].function
);
209 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
210 int dmar_insert_dev_scope(struct dmar_pci_notify_info
*info
,
211 void *start
, void*end
, u16 segment
,
212 struct dmar_dev_scope
*devices
,
216 struct device
*tmp
, *dev
= &info
->dev
->dev
;
217 struct acpi_dmar_device_scope
*scope
;
218 struct acpi_dmar_pci_path
*path
;
220 if (segment
!= info
->seg
)
223 for (; start
< end
; start
+= scope
->length
) {
225 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
226 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
229 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
230 level
= (scope
->length
- sizeof(*scope
)) / sizeof(*path
);
231 if (!dmar_match_pci_path(info
, scope
->bus
, path
, level
))
235 * We expect devices with endpoint scope to have normal PCI
236 * headers, and devices with bridge scope to have bridge PCI
237 * headers. However PCI NTB devices may be listed in the
238 * DMAR table with bridge scope, even though they have a
239 * normal PCI header. NTB devices are identified by class
240 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
241 * for this special case.
243 if ((scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
244 info
->dev
->hdr_type
!= PCI_HEADER_TYPE_NORMAL
) ||
245 (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
&&
246 (info
->dev
->hdr_type
== PCI_HEADER_TYPE_NORMAL
&&
247 info
->dev
->class >> 8 != PCI_CLASS_BRIDGE_OTHER
))) {
248 pr_warn("Device scope type does not match for %s\n",
249 pci_name(info
->dev
));
253 for_each_dev_scope(devices
, devices_cnt
, i
, tmp
)
255 devices
[i
].bus
= info
->dev
->bus
->number
;
256 devices
[i
].devfn
= info
->dev
->devfn
;
257 rcu_assign_pointer(devices
[i
].dev
,
261 BUG_ON(i
>= devices_cnt
);
267 int dmar_remove_dev_scope(struct dmar_pci_notify_info
*info
, u16 segment
,
268 struct dmar_dev_scope
*devices
, int count
)
273 if (info
->seg
!= segment
)
276 for_each_active_dev_scope(devices
, count
, index
, tmp
)
277 if (tmp
== &info
->dev
->dev
) {
278 RCU_INIT_POINTER(devices
[index
].dev
, NULL
);
287 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info
*info
)
290 struct dmar_drhd_unit
*dmaru
;
291 struct acpi_dmar_hardware_unit
*drhd
;
293 for_each_drhd_unit(dmaru
) {
294 if (dmaru
->include_all
)
297 drhd
= container_of(dmaru
->hdr
,
298 struct acpi_dmar_hardware_unit
, header
);
299 ret
= dmar_insert_dev_scope(info
, (void *)(drhd
+ 1),
300 ((void *)drhd
) + drhd
->header
.length
,
302 dmaru
->devices
, dmaru
->devices_cnt
);
307 ret
= dmar_iommu_notify_scope_dev(info
);
308 if (ret
< 0 && dmar_dev_scope_status
== 0)
309 dmar_dev_scope_status
= ret
;
314 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info
*info
)
316 struct dmar_drhd_unit
*dmaru
;
318 for_each_drhd_unit(dmaru
)
319 if (dmar_remove_dev_scope(info
, dmaru
->segment
,
320 dmaru
->devices
, dmaru
->devices_cnt
))
322 dmar_iommu_notify_scope_dev(info
);
325 static int dmar_pci_bus_notifier(struct notifier_block
*nb
,
326 unsigned long action
, void *data
)
328 struct pci_dev
*pdev
= to_pci_dev(data
);
329 struct dmar_pci_notify_info
*info
;
331 /* Only care about add/remove events for physical functions.
332 * For VFs we actually do the lookup based on the corresponding
333 * PF in device_to_iommu() anyway. */
336 if (action
!= BUS_NOTIFY_ADD_DEVICE
&&
337 action
!= BUS_NOTIFY_REMOVED_DEVICE
)
340 info
= dmar_alloc_pci_notify_info(pdev
, action
);
344 down_write(&dmar_global_lock
);
345 if (action
== BUS_NOTIFY_ADD_DEVICE
)
346 dmar_pci_bus_add_dev(info
);
347 else if (action
== BUS_NOTIFY_REMOVED_DEVICE
)
348 dmar_pci_bus_del_dev(info
);
349 up_write(&dmar_global_lock
);
351 dmar_free_pci_notify_info(info
);
356 static struct notifier_block dmar_pci_bus_nb
= {
357 .notifier_call
= dmar_pci_bus_notifier
,
361 static struct dmar_drhd_unit
*
362 dmar_find_dmaru(struct acpi_dmar_hardware_unit
*drhd
)
364 struct dmar_drhd_unit
*dmaru
;
366 list_for_each_entry_rcu(dmaru
, &dmar_drhd_units
, list
)
367 if (dmaru
->segment
== drhd
->segment
&&
368 dmaru
->reg_base_addr
== drhd
->address
)
375 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
376 * structure which uniquely represent one DMA remapping hardware unit
377 * present in the platform
379 static int dmar_parse_one_drhd(struct acpi_dmar_header
*header
, void *arg
)
381 struct acpi_dmar_hardware_unit
*drhd
;
382 struct dmar_drhd_unit
*dmaru
;
385 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
386 dmaru
= dmar_find_dmaru(drhd
);
390 dmaru
= kzalloc(sizeof(*dmaru
) + header
->length
, GFP_KERNEL
);
395 * If header is allocated from slab by ACPI _DSM method, we need to
396 * copy the content because the memory buffer will be freed on return.
398 dmaru
->hdr
= (void *)(dmaru
+ 1);
399 memcpy(dmaru
->hdr
, header
, header
->length
);
400 dmaru
->reg_base_addr
= drhd
->address
;
401 dmaru
->segment
= drhd
->segment
;
402 dmaru
->include_all
= drhd
->flags
& 0x1; /* BIT0: INCLUDE_ALL */
403 dmaru
->devices
= dmar_alloc_dev_scope((void *)(drhd
+ 1),
404 ((void *)drhd
) + drhd
->header
.length
,
405 &dmaru
->devices_cnt
);
406 if (dmaru
->devices_cnt
&& dmaru
->devices
== NULL
) {
411 ret
= alloc_iommu(dmaru
);
413 dmar_free_dev_scope(&dmaru
->devices
,
414 &dmaru
->devices_cnt
);
418 dmar_register_drhd_unit(dmaru
);
427 static void dmar_free_drhd(struct dmar_drhd_unit
*dmaru
)
429 if (dmaru
->devices
&& dmaru
->devices_cnt
)
430 dmar_free_dev_scope(&dmaru
->devices
, &dmaru
->devices_cnt
);
432 free_iommu(dmaru
->iommu
);
436 static int __init
dmar_parse_one_andd(struct acpi_dmar_header
*header
,
439 struct acpi_dmar_andd
*andd
= (void *)header
;
441 /* Check for NUL termination within the designated length */
442 if (strnlen(andd
->device_name
, header
->length
- 8) == header
->length
- 8) {
443 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND
,
444 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
445 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
446 dmi_get_system_info(DMI_BIOS_VENDOR
),
447 dmi_get_system_info(DMI_BIOS_VERSION
),
448 dmi_get_system_info(DMI_PRODUCT_VERSION
));
451 pr_info("ANDD device: %x name: %s\n", andd
->device_number
,
457 #ifdef CONFIG_ACPI_NUMA
458 static int dmar_parse_one_rhsa(struct acpi_dmar_header
*header
, void *arg
)
460 struct acpi_dmar_rhsa
*rhsa
;
461 struct dmar_drhd_unit
*drhd
;
463 rhsa
= (struct acpi_dmar_rhsa
*)header
;
464 for_each_drhd_unit(drhd
) {
465 if (drhd
->reg_base_addr
== rhsa
->base_address
) {
466 int node
= acpi_map_pxm_to_node(rhsa
->proximity_domain
);
468 if (!node_online(node
))
470 drhd
->iommu
->node
= node
;
475 1, TAINT_FIRMWARE_WORKAROUND
,
476 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
477 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
479 dmi_get_system_info(DMI_BIOS_VENDOR
),
480 dmi_get_system_info(DMI_BIOS_VERSION
),
481 dmi_get_system_info(DMI_PRODUCT_VERSION
));
486 #define dmar_parse_one_rhsa dmar_res_noop
490 dmar_table_print_dmar_entry(struct acpi_dmar_header
*header
)
492 struct acpi_dmar_hardware_unit
*drhd
;
493 struct acpi_dmar_reserved_memory
*rmrr
;
494 struct acpi_dmar_atsr
*atsr
;
495 struct acpi_dmar_rhsa
*rhsa
;
497 switch (header
->type
) {
498 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
499 drhd
= container_of(header
, struct acpi_dmar_hardware_unit
,
501 pr_info("DRHD base: %#016Lx flags: %#x\n",
502 (unsigned long long)drhd
->address
, drhd
->flags
);
504 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
505 rmrr
= container_of(header
, struct acpi_dmar_reserved_memory
,
507 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
508 (unsigned long long)rmrr
->base_address
,
509 (unsigned long long)rmrr
->end_address
);
511 case ACPI_DMAR_TYPE_ROOT_ATS
:
512 atsr
= container_of(header
, struct acpi_dmar_atsr
, header
);
513 pr_info("ATSR flags: %#x\n", atsr
->flags
);
515 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY
:
516 rhsa
= container_of(header
, struct acpi_dmar_rhsa
, header
);
517 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
518 (unsigned long long)rhsa
->base_address
,
519 rhsa
->proximity_domain
);
521 case ACPI_DMAR_TYPE_NAMESPACE
:
522 /* We don't print this here because we need to sanity-check
523 it first. So print it in dmar_parse_one_andd() instead. */
529 * dmar_table_detect - checks to see if the platform supports DMAR devices
531 static int __init
dmar_table_detect(void)
533 acpi_status status
= AE_OK
;
535 /* if we could find DMAR table, then there are DMAR devices */
536 status
= acpi_get_table(ACPI_SIG_DMAR
, 0, &dmar_tbl
);
538 if (ACPI_SUCCESS(status
) && !dmar_tbl
) {
539 pr_warn("Unable to map DMAR\n");
540 status
= AE_NOT_FOUND
;
543 return ACPI_SUCCESS(status
) ? 0 : -ENOENT
;
546 static int dmar_walk_remapping_entries(struct acpi_dmar_header
*start
,
547 size_t len
, struct dmar_res_callback
*cb
)
549 struct acpi_dmar_header
*iter
, *next
;
550 struct acpi_dmar_header
*end
= ((void *)start
) + len
;
552 for (iter
= start
; iter
< end
; iter
= next
) {
553 next
= (void *)iter
+ iter
->length
;
554 if (iter
->length
== 0) {
555 /* Avoid looping forever on bad ACPI tables */
556 pr_debug(FW_BUG
"Invalid 0-length structure\n");
558 } else if (next
> end
) {
559 /* Avoid passing table end */
560 pr_warn(FW_BUG
"Record passes table end\n");
565 dmar_table_print_dmar_entry(iter
);
567 if (iter
->type
>= ACPI_DMAR_TYPE_RESERVED
) {
568 /* continue for forward compatibility */
569 pr_debug("Unknown DMAR structure type %d\n",
571 } else if (cb
->cb
[iter
->type
]) {
574 ret
= cb
->cb
[iter
->type
](iter
, cb
->arg
[iter
->type
]);
577 } else if (!cb
->ignore_unhandled
) {
578 pr_warn("No handler for DMAR structure type %d\n",
587 static inline int dmar_walk_dmar_table(struct acpi_table_dmar
*dmar
,
588 struct dmar_res_callback
*cb
)
590 return dmar_walk_remapping_entries((void *)(dmar
+ 1),
591 dmar
->header
.length
- sizeof(*dmar
), cb
);
595 * parse_dmar_table - parses the DMA reporting table
598 parse_dmar_table(void)
600 struct acpi_table_dmar
*dmar
;
603 struct dmar_res_callback cb
= {
605 .ignore_unhandled
= true,
606 .arg
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &drhd_count
,
607 .cb
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &dmar_parse_one_drhd
,
608 .cb
[ACPI_DMAR_TYPE_RESERVED_MEMORY
] = &dmar_parse_one_rmrr
,
609 .cb
[ACPI_DMAR_TYPE_ROOT_ATS
] = &dmar_parse_one_atsr
,
610 .cb
[ACPI_DMAR_TYPE_HARDWARE_AFFINITY
] = &dmar_parse_one_rhsa
,
611 .cb
[ACPI_DMAR_TYPE_NAMESPACE
] = &dmar_parse_one_andd
,
615 * Do it again, earlier dmar_tbl mapping could be mapped with
621 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
622 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
624 dmar_tbl
= tboot_get_dmar_table(dmar_tbl
);
626 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
630 if (dmar
->width
< PAGE_SHIFT
- 1) {
631 pr_warn("Invalid DMAR haw\n");
635 pr_info("Host address width %d\n", dmar
->width
+ 1);
636 ret
= dmar_walk_dmar_table(dmar
, &cb
);
637 if (ret
== 0 && drhd_count
== 0)
638 pr_warn(FW_BUG
"No DRHD structure found in DMAR table\n");
643 static int dmar_pci_device_match(struct dmar_dev_scope devices
[],
644 int cnt
, struct pci_dev
*dev
)
650 for_each_active_dev_scope(devices
, cnt
, index
, tmp
)
651 if (dev_is_pci(tmp
) && dev
== to_pci_dev(tmp
))
654 /* Check our parent */
655 dev
= dev
->bus
->self
;
661 struct dmar_drhd_unit
*
662 dmar_find_matched_drhd_unit(struct pci_dev
*dev
)
664 struct dmar_drhd_unit
*dmaru
;
665 struct acpi_dmar_hardware_unit
*drhd
;
667 dev
= pci_physfn(dev
);
670 for_each_drhd_unit(dmaru
) {
671 drhd
= container_of(dmaru
->hdr
,
672 struct acpi_dmar_hardware_unit
,
675 if (dmaru
->include_all
&&
676 drhd
->segment
== pci_domain_nr(dev
->bus
))
679 if (dmar_pci_device_match(dmaru
->devices
,
680 dmaru
->devices_cnt
, dev
))
690 static void __init
dmar_acpi_insert_dev_scope(u8 device_number
,
691 struct acpi_device
*adev
)
693 struct dmar_drhd_unit
*dmaru
;
694 struct acpi_dmar_hardware_unit
*drhd
;
695 struct acpi_dmar_device_scope
*scope
;
698 struct acpi_dmar_pci_path
*path
;
700 for_each_drhd_unit(dmaru
) {
701 drhd
= container_of(dmaru
->hdr
,
702 struct acpi_dmar_hardware_unit
,
705 for (scope
= (void *)(drhd
+ 1);
706 (unsigned long)scope
< ((unsigned long)drhd
) + drhd
->header
.length
;
707 scope
= ((void *)scope
) + scope
->length
) {
708 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_NAMESPACE
)
710 if (scope
->enumeration_id
!= device_number
)
713 path
= (void *)(scope
+ 1);
714 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
715 dev_name(&adev
->dev
), dmaru
->reg_base_addr
,
716 scope
->bus
, path
->device
, path
->function
);
717 for_each_dev_scope(dmaru
->devices
, dmaru
->devices_cnt
, i
, tmp
)
719 dmaru
->devices
[i
].bus
= scope
->bus
;
720 dmaru
->devices
[i
].devfn
= PCI_DEVFN(path
->device
,
722 rcu_assign_pointer(dmaru
->devices
[i
].dev
,
723 get_device(&adev
->dev
));
726 BUG_ON(i
>= dmaru
->devices_cnt
);
729 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
730 device_number
, dev_name(&adev
->dev
));
733 static int __init
dmar_acpi_dev_scope_init(void)
735 struct acpi_dmar_andd
*andd
;
737 if (dmar_tbl
== NULL
)
740 for (andd
= (void *)dmar_tbl
+ sizeof(struct acpi_table_dmar
);
741 ((unsigned long)andd
) < ((unsigned long)dmar_tbl
) + dmar_tbl
->length
;
742 andd
= ((void *)andd
) + andd
->header
.length
) {
743 if (andd
->header
.type
== ACPI_DMAR_TYPE_NAMESPACE
) {
745 struct acpi_device
*adev
;
747 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT
,
750 pr_err("Failed to find handle for ACPI object %s\n",
754 if (acpi_bus_get_device(h
, &adev
)) {
755 pr_err("Failed to get device for ACPI object %s\n",
759 dmar_acpi_insert_dev_scope(andd
->device_number
, adev
);
765 int __init
dmar_dev_scope_init(void)
767 struct pci_dev
*dev
= NULL
;
768 struct dmar_pci_notify_info
*info
;
770 if (dmar_dev_scope_status
!= 1)
771 return dmar_dev_scope_status
;
773 if (list_empty(&dmar_drhd_units
)) {
774 dmar_dev_scope_status
= -ENODEV
;
776 dmar_dev_scope_status
= 0;
778 dmar_acpi_dev_scope_init();
780 for_each_pci_dev(dev
) {
784 info
= dmar_alloc_pci_notify_info(dev
,
785 BUS_NOTIFY_ADD_DEVICE
);
787 return dmar_dev_scope_status
;
789 dmar_pci_bus_add_dev(info
);
790 dmar_free_pci_notify_info(info
);
795 return dmar_dev_scope_status
;
798 void __init
dmar_register_bus_notifier(void)
800 bus_register_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
804 int __init
dmar_table_init(void)
806 static int dmar_table_initialized
;
809 if (dmar_table_initialized
== 0) {
810 ret
= parse_dmar_table();
813 pr_info("Parse DMAR table failure.\n");
814 } else if (list_empty(&dmar_drhd_units
)) {
815 pr_info("No DMAR devices found\n");
820 dmar_table_initialized
= ret
;
822 dmar_table_initialized
= 1;
825 return dmar_table_initialized
< 0 ? dmar_table_initialized
: 0;
828 static void warn_invalid_dmar(u64 addr
, const char *message
)
831 1, TAINT_FIRMWARE_WORKAROUND
,
832 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
833 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
835 dmi_get_system_info(DMI_BIOS_VENDOR
),
836 dmi_get_system_info(DMI_BIOS_VERSION
),
837 dmi_get_system_info(DMI_PRODUCT_VERSION
));
841 dmar_validate_one_drhd(struct acpi_dmar_header
*entry
, void *arg
)
843 struct acpi_dmar_hardware_unit
*drhd
;
847 drhd
= (void *)entry
;
848 if (!drhd
->address
) {
849 warn_invalid_dmar(0, "");
854 addr
= ioremap(drhd
->address
, VTD_PAGE_SIZE
);
856 addr
= early_ioremap(drhd
->address
, VTD_PAGE_SIZE
);
858 pr_warn("Can't validate DRHD address: %llx\n", drhd
->address
);
862 cap
= dmar_readq(addr
+ DMAR_CAP_REG
);
863 ecap
= dmar_readq(addr
+ DMAR_ECAP_REG
);
868 early_iounmap(addr
, VTD_PAGE_SIZE
);
870 if (cap
== (uint64_t)-1 && ecap
== (uint64_t)-1) {
871 warn_invalid_dmar(drhd
->address
, " returns all ones");
878 int __init
detect_intel_iommu(void)
881 struct dmar_res_callback validate_drhd_cb
= {
882 .cb
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &dmar_validate_one_drhd
,
883 .ignore_unhandled
= true,
886 down_write(&dmar_global_lock
);
887 ret
= dmar_table_detect();
889 ret
= dmar_walk_dmar_table((struct acpi_table_dmar
*)dmar_tbl
,
891 if (!ret
&& !no_iommu
&& !iommu_detected
&& !dmar_disabled
) {
893 /* Make sure ACS will be enabled */
899 x86_init
.iommu
.iommu_init
= intel_iommu_init
;
900 x86_platform
.iommu_shutdown
= intel_iommu_shutdown
;
906 acpi_put_table(dmar_tbl
);
909 up_write(&dmar_global_lock
);
911 return ret
? ret
: 1;
914 static void unmap_iommu(struct intel_iommu
*iommu
)
917 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
921 * map_iommu: map the iommu's registers
922 * @iommu: the iommu to map
923 * @phys_addr: the physical address of the base resgister
925 * Memory map the iommu's registers. Start w/ a single page, and
926 * possibly expand if that turns out to be insufficent.
928 static int map_iommu(struct intel_iommu
*iommu
, u64 phys_addr
)
932 iommu
->reg_phys
= phys_addr
;
933 iommu
->reg_size
= VTD_PAGE_SIZE
;
935 if (!request_mem_region(iommu
->reg_phys
, iommu
->reg_size
, iommu
->name
)) {
936 pr_err("Can't reserve memory\n");
941 iommu
->reg
= ioremap(iommu
->reg_phys
, iommu
->reg_size
);
943 pr_err("Can't map the region\n");
948 iommu
->cap
= dmar_readq(iommu
->reg
+ DMAR_CAP_REG
);
949 iommu
->ecap
= dmar_readq(iommu
->reg
+ DMAR_ECAP_REG
);
951 if (iommu
->cap
== (uint64_t)-1 && iommu
->ecap
== (uint64_t)-1) {
953 warn_invalid_dmar(phys_addr
, " returns all ones");
957 /* the registers might be more than one page */
958 map_size
= max_t(int, ecap_max_iotlb_offset(iommu
->ecap
),
959 cap_max_fault_reg_offset(iommu
->cap
));
960 map_size
= VTD_PAGE_ALIGN(map_size
);
961 if (map_size
> iommu
->reg_size
) {
963 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
964 iommu
->reg_size
= map_size
;
965 if (!request_mem_region(iommu
->reg_phys
, iommu
->reg_size
,
967 pr_err("Can't reserve memory\n");
971 iommu
->reg
= ioremap(iommu
->reg_phys
, iommu
->reg_size
);
973 pr_err("Can't map the region\n");
984 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
989 static int dmar_alloc_seq_id(struct intel_iommu
*iommu
)
991 iommu
->seq_id
= find_first_zero_bit(dmar_seq_ids
,
992 DMAR_UNITS_SUPPORTED
);
993 if (iommu
->seq_id
>= DMAR_UNITS_SUPPORTED
) {
996 set_bit(iommu
->seq_id
, dmar_seq_ids
);
997 sprintf(iommu
->name
, "dmar%d", iommu
->seq_id
);
1000 return iommu
->seq_id
;
1003 static void dmar_free_seq_id(struct intel_iommu
*iommu
)
1005 if (iommu
->seq_id
>= 0) {
1006 clear_bit(iommu
->seq_id
, dmar_seq_ids
);
1011 static int alloc_iommu(struct dmar_drhd_unit
*drhd
)
1013 struct intel_iommu
*iommu
;
1019 if (!drhd
->reg_base_addr
) {
1020 warn_invalid_dmar(0, "");
1024 iommu
= kzalloc(sizeof(*iommu
), GFP_KERNEL
);
1028 if (dmar_alloc_seq_id(iommu
) < 0) {
1029 pr_err("Failed to allocate seq_id\n");
1034 err
= map_iommu(iommu
, drhd
->reg_base_addr
);
1036 pr_err("Failed to map %s\n", iommu
->name
);
1037 goto error_free_seq_id
;
1041 agaw
= iommu_calculate_agaw(iommu
);
1043 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1047 msagaw
= iommu_calculate_max_sagaw(iommu
);
1049 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1054 iommu
->msagaw
= msagaw
;
1055 iommu
->segment
= drhd
->segment
;
1057 iommu
->node
= NUMA_NO_NODE
;
1059 ver
= readl(iommu
->reg
+ DMAR_VER_REG
);
1060 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1062 (unsigned long long)drhd
->reg_base_addr
,
1063 DMAR_VER_MAJOR(ver
), DMAR_VER_MINOR(ver
),
1064 (unsigned long long)iommu
->cap
,
1065 (unsigned long long)iommu
->ecap
);
1067 /* Reflect status in gcmd */
1068 sts
= readl(iommu
->reg
+ DMAR_GSTS_REG
);
1069 if (sts
& DMA_GSTS_IRES
)
1070 iommu
->gcmd
|= DMA_GCMD_IRE
;
1071 if (sts
& DMA_GSTS_TES
)
1072 iommu
->gcmd
|= DMA_GCMD_TE
;
1073 if (sts
& DMA_GSTS_QIES
)
1074 iommu
->gcmd
|= DMA_GCMD_QIE
;
1076 raw_spin_lock_init(&iommu
->register_lock
);
1078 if (intel_iommu_enabled
) {
1079 err
= iommu_device_sysfs_add(&iommu
->iommu
, NULL
,
1085 iommu_device_set_ops(&iommu
->iommu
, &intel_iommu_ops
);
1087 err
= iommu_device_register(&iommu
->iommu
);
1092 drhd
->iommu
= iommu
;
1099 dmar_free_seq_id(iommu
);
1105 static void free_iommu(struct intel_iommu
*iommu
)
1107 if (intel_iommu_enabled
) {
1108 iommu_device_unregister(&iommu
->iommu
);
1109 iommu_device_sysfs_remove(&iommu
->iommu
);
1113 if (iommu
->pr_irq
) {
1114 free_irq(iommu
->pr_irq
, iommu
);
1115 dmar_free_hwirq(iommu
->pr_irq
);
1118 free_irq(iommu
->irq
, iommu
);
1119 dmar_free_hwirq(iommu
->irq
);
1124 free_page((unsigned long)iommu
->qi
->desc
);
1125 kfree(iommu
->qi
->desc_status
);
1132 dmar_free_seq_id(iommu
);
1137 * Reclaim all the submitted descriptors which have completed its work.
1139 static inline void reclaim_free_desc(struct q_inval
*qi
)
1141 while (qi
->desc_status
[qi
->free_tail
] == QI_DONE
||
1142 qi
->desc_status
[qi
->free_tail
] == QI_ABORT
) {
1143 qi
->desc_status
[qi
->free_tail
] = QI_FREE
;
1144 qi
->free_tail
= (qi
->free_tail
+ 1) % QI_LENGTH
;
1149 static int qi_check_fault(struct intel_iommu
*iommu
, int index
)
1153 struct q_inval
*qi
= iommu
->qi
;
1154 int wait_index
= (index
+ 1) % QI_LENGTH
;
1155 int shift
= qi_shift(iommu
);
1157 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
1160 fault
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1163 * If IQE happens, the head points to the descriptor associated
1164 * with the error. No new descriptors are fetched until the IQE
1167 if (fault
& DMA_FSTS_IQE
) {
1168 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
1169 if ((head
>> shift
) == index
) {
1170 struct qi_desc
*desc
= qi
->desc
+ head
;
1173 * desc->qw2 and desc->qw3 are either reserved or
1174 * used by software as private data. We won't print
1175 * out these two qw's for security consideration.
1177 pr_err("VT-d detected invalid descriptor: qw0 = %llx, qw1 = %llx\n",
1178 (unsigned long long)desc
->qw0
,
1179 (unsigned long long)desc
->qw1
);
1180 memcpy(desc
, qi
->desc
+ (wait_index
<< shift
),
1182 writel(DMA_FSTS_IQE
, iommu
->reg
+ DMAR_FSTS_REG
);
1188 * If ITE happens, all pending wait_desc commands are aborted.
1189 * No new descriptors are fetched until the ITE is cleared.
1191 if (fault
& DMA_FSTS_ITE
) {
1192 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
1193 head
= ((head
>> shift
) - 1 + QI_LENGTH
) % QI_LENGTH
;
1195 tail
= readl(iommu
->reg
+ DMAR_IQT_REG
);
1196 tail
= ((tail
>> shift
) - 1 + QI_LENGTH
) % QI_LENGTH
;
1198 writel(DMA_FSTS_ITE
, iommu
->reg
+ DMAR_FSTS_REG
);
1201 if (qi
->desc_status
[head
] == QI_IN_USE
)
1202 qi
->desc_status
[head
] = QI_ABORT
;
1203 head
= (head
- 2 + QI_LENGTH
) % QI_LENGTH
;
1204 } while (head
!= tail
);
1206 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
1210 if (fault
& DMA_FSTS_ICE
)
1211 writel(DMA_FSTS_ICE
, iommu
->reg
+ DMAR_FSTS_REG
);
1217 * Submit the queued invalidation descriptor to the remapping
1218 * hardware unit and wait for its completion.
1220 int qi_submit_sync(struct qi_desc
*desc
, struct intel_iommu
*iommu
)
1223 struct q_inval
*qi
= iommu
->qi
;
1224 int offset
, shift
, length
;
1225 struct qi_desc wait_desc
;
1226 int wait_index
, index
;
1227 unsigned long flags
;
1235 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1236 while (qi
->free_cnt
< 3) {
1237 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1239 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1242 index
= qi
->free_head
;
1243 wait_index
= (index
+ 1) % QI_LENGTH
;
1244 shift
= qi_shift(iommu
);
1245 length
= 1 << shift
;
1247 qi
->desc_status
[index
] = qi
->desc_status
[wait_index
] = QI_IN_USE
;
1249 offset
= index
<< shift
;
1250 memcpy(qi
->desc
+ offset
, desc
, length
);
1251 wait_desc
.qw0
= QI_IWD_STATUS_DATA(QI_DONE
) |
1252 QI_IWD_STATUS_WRITE
| QI_IWD_TYPE
;
1253 wait_desc
.qw1
= virt_to_phys(&qi
->desc_status
[wait_index
]);
1257 offset
= wait_index
<< shift
;
1258 memcpy(qi
->desc
+ offset
, &wait_desc
, length
);
1260 qi
->free_head
= (qi
->free_head
+ 2) % QI_LENGTH
;
1264 * update the HW tail register indicating the presence of
1267 writel(qi
->free_head
<< shift
, iommu
->reg
+ DMAR_IQT_REG
);
1269 while (qi
->desc_status
[wait_index
] != QI_DONE
) {
1271 * We will leave the interrupts disabled, to prevent interrupt
1272 * context to queue another cmd while a cmd is already submitted
1273 * and waiting for completion on this cpu. This is to avoid
1274 * a deadlock where the interrupt context can wait indefinitely
1275 * for free slots in the queue.
1277 rc
= qi_check_fault(iommu
, index
);
1281 raw_spin_unlock(&qi
->q_lock
);
1283 raw_spin_lock(&qi
->q_lock
);
1286 qi
->desc_status
[index
] = QI_DONE
;
1288 reclaim_free_desc(qi
);
1289 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1298 * Flush the global interrupt entry cache.
1300 void qi_global_iec(struct intel_iommu
*iommu
)
1302 struct qi_desc desc
;
1304 desc
.qw0
= QI_IEC_TYPE
;
1309 /* should never fail */
1310 qi_submit_sync(&desc
, iommu
);
1313 void qi_flush_context(struct intel_iommu
*iommu
, u16 did
, u16 sid
, u8 fm
,
1316 struct qi_desc desc
;
1318 desc
.qw0
= QI_CC_FM(fm
) | QI_CC_SID(sid
) | QI_CC_DID(did
)
1319 | QI_CC_GRAN(type
) | QI_CC_TYPE
;
1324 qi_submit_sync(&desc
, iommu
);
1327 void qi_flush_iotlb(struct intel_iommu
*iommu
, u16 did
, u64 addr
,
1328 unsigned int size_order
, u64 type
)
1332 struct qi_desc desc
;
1335 if (cap_write_drain(iommu
->cap
))
1338 if (cap_read_drain(iommu
->cap
))
1341 desc
.qw0
= QI_IOTLB_DID(did
) | QI_IOTLB_DR(dr
) | QI_IOTLB_DW(dw
)
1342 | QI_IOTLB_GRAN(type
) | QI_IOTLB_TYPE
;
1343 desc
.qw1
= QI_IOTLB_ADDR(addr
) | QI_IOTLB_IH(ih
)
1344 | QI_IOTLB_AM(size_order
);
1348 qi_submit_sync(&desc
, iommu
);
1351 void qi_flush_dev_iotlb(struct intel_iommu
*iommu
, u16 sid
, u16 pfsid
,
1352 u16 qdep
, u64 addr
, unsigned mask
)
1354 struct qi_desc desc
;
1357 WARN_ON_ONCE(addr
& ((1ULL << (VTD_PAGE_SHIFT
+ mask
)) - 1));
1358 addr
|= (1ULL << (VTD_PAGE_SHIFT
+ mask
- 1)) - 1;
1359 desc
.qw1
= QI_DEV_IOTLB_ADDR(addr
) | QI_DEV_IOTLB_SIZE
;
1361 desc
.qw1
= QI_DEV_IOTLB_ADDR(addr
);
1363 if (qdep
>= QI_DEV_IOTLB_MAX_INVS
)
1366 desc
.qw0
= QI_DEV_IOTLB_SID(sid
) | QI_DEV_IOTLB_QDEP(qdep
) |
1367 QI_DIOTLB_TYPE
| QI_DEV_IOTLB_PFSID(pfsid
);
1371 qi_submit_sync(&desc
, iommu
);
1375 * Disable Queued Invalidation interface.
1377 void dmar_disable_qi(struct intel_iommu
*iommu
)
1379 unsigned long flags
;
1381 cycles_t start_time
= get_cycles();
1383 if (!ecap_qis(iommu
->ecap
))
1386 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1388 sts
= readl(iommu
->reg
+ DMAR_GSTS_REG
);
1389 if (!(sts
& DMA_GSTS_QIES
))
1393 * Give a chance to HW to complete the pending invalidation requests.
1395 while ((readl(iommu
->reg
+ DMAR_IQT_REG
) !=
1396 readl(iommu
->reg
+ DMAR_IQH_REG
)) &&
1397 (DMAR_OPERATION_TIMEOUT
> (get_cycles() - start_time
)))
1400 iommu
->gcmd
&= ~DMA_GCMD_QIE
;
1401 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1403 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
,
1404 !(sts
& DMA_GSTS_QIES
), sts
);
1406 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1410 * Enable queued invalidation.
1412 static void __dmar_enable_qi(struct intel_iommu
*iommu
)
1415 unsigned long flags
;
1416 struct q_inval
*qi
= iommu
->qi
;
1417 u64 val
= virt_to_phys(qi
->desc
);
1419 qi
->free_head
= qi
->free_tail
= 0;
1420 qi
->free_cnt
= QI_LENGTH
;
1423 * Set DW=1 and QS=1 in IQA_REG when Scalable Mode capability
1426 if (ecap_smts(iommu
->ecap
))
1427 val
|= (1 << 11) | 1;
1429 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1431 /* write zero to the tail reg */
1432 writel(0, iommu
->reg
+ DMAR_IQT_REG
);
1434 dmar_writeq(iommu
->reg
+ DMAR_IQA_REG
, val
);
1436 iommu
->gcmd
|= DMA_GCMD_QIE
;
1437 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1439 /* Make sure hardware complete it */
1440 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
, (sts
& DMA_GSTS_QIES
), sts
);
1442 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1446 * Enable Queued Invalidation interface. This is a must to support
1447 * interrupt-remapping. Also used by DMA-remapping, which replaces
1448 * register based IOTLB invalidation.
1450 int dmar_enable_qi(struct intel_iommu
*iommu
)
1453 struct page
*desc_page
;
1455 if (!ecap_qis(iommu
->ecap
))
1459 * queued invalidation is already setup and enabled.
1464 iommu
->qi
= kmalloc(sizeof(*qi
), GFP_ATOMIC
);
1471 * Need two pages to accommodate 256 descriptors of 256 bits each
1472 * if the remapping hardware supports scalable mode translation.
1474 desc_page
= alloc_pages_node(iommu
->node
, GFP_ATOMIC
| __GFP_ZERO
,
1475 !!ecap_smts(iommu
->ecap
));
1482 qi
->desc
= page_address(desc_page
);
1484 qi
->desc_status
= kcalloc(QI_LENGTH
, sizeof(int), GFP_ATOMIC
);
1485 if (!qi
->desc_status
) {
1486 free_page((unsigned long) qi
->desc
);
1492 raw_spin_lock_init(&qi
->q_lock
);
1494 __dmar_enable_qi(iommu
);
1499 /* iommu interrupt handling. Most stuff are MSI-like. */
1507 static const char *dma_remap_fault_reasons
[] =
1510 "Present bit in root entry is clear",
1511 "Present bit in context entry is clear",
1512 "Invalid context entry",
1513 "Access beyond MGAW",
1514 "PTE Write access is not set",
1515 "PTE Read access is not set",
1516 "Next page table ptr is invalid",
1517 "Root table address invalid",
1518 "Context table ptr is invalid",
1519 "non-zero reserved fields in RTP",
1520 "non-zero reserved fields in CTP",
1521 "non-zero reserved fields in PTE",
1522 "PCE for translation request specifies blocking",
1525 static const char * const dma_remap_sm_fault_reasons
[] = {
1526 "SM: Invalid Root Table Address",
1527 "SM: TTM 0 for request with PASID",
1528 "SM: TTM 0 for page group request",
1529 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x33-0x37 */
1530 "SM: Error attempting to access Root Entry",
1531 "SM: Present bit in Root Entry is clear",
1532 "SM: Non-zero reserved field set in Root Entry",
1533 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x3B-0x3F */
1534 "SM: Error attempting to access Context Entry",
1535 "SM: Present bit in Context Entry is clear",
1536 "SM: Non-zero reserved field set in the Context Entry",
1537 "SM: Invalid Context Entry",
1538 "SM: DTE field in Context Entry is clear",
1539 "SM: PASID Enable field in Context Entry is clear",
1540 "SM: PASID is larger than the max in Context Entry",
1541 "SM: PRE field in Context-Entry is clear",
1542 "SM: RID_PASID field error in Context-Entry",
1543 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x49-0x4F */
1544 "SM: Error attempting to access the PASID Directory Entry",
1545 "SM: Present bit in Directory Entry is clear",
1546 "SM: Non-zero reserved field set in PASID Directory Entry",
1547 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x53-0x57 */
1548 "SM: Error attempting to access PASID Table Entry",
1549 "SM: Present bit in PASID Table Entry is clear",
1550 "SM: Non-zero reserved field set in PASID Table Entry",
1551 "SM: Invalid Scalable-Mode PASID Table Entry",
1552 "SM: ERE field is clear in PASID Table Entry",
1553 "SM: SRE field is clear in PASID Table Entry",
1554 "Unknown", "Unknown",/* 0x5E-0x5F */
1555 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x60-0x67 */
1556 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x68-0x6F */
1557 "SM: Error attempting to access first-level paging entry",
1558 "SM: Present bit in first-level paging entry is clear",
1559 "SM: Non-zero reserved field set in first-level paging entry",
1560 "SM: Error attempting to access FL-PML4 entry",
1561 "SM: First-level entry address beyond MGAW in Nested translation",
1562 "SM: Read permission error in FL-PML4 entry in Nested translation",
1563 "SM: Read permission error in first-level paging entry in Nested translation",
1564 "SM: Write permission error in first-level paging entry in Nested translation",
1565 "SM: Error attempting to access second-level paging entry",
1566 "SM: Read/Write permission error in second-level paging entry",
1567 "SM: Non-zero reserved field set in second-level paging entry",
1568 "SM: Invalid second-level page table pointer",
1569 "SM: A/D bit update needed in second-level entry when set up in no snoop",
1570 "Unknown", "Unknown", "Unknown", /* 0x7D-0x7F */
1571 "SM: Address in first-level translation is not canonical",
1572 "SM: U/S set 0 for first-level translation with user privilege",
1573 "SM: No execute permission for request with PASID and ER=1",
1574 "SM: Address beyond the DMA hardware max",
1575 "SM: Second-level entry address beyond the max",
1576 "SM: No write permission for Write/AtomicOp request",
1577 "SM: No read permission for Read/AtomicOp request",
1578 "SM: Invalid address-interrupt address",
1579 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x88-0x8F */
1580 "SM: A/D bit update needed in first-level entry when set up in no snoop",
1583 static const char *irq_remap_fault_reasons
[] =
1585 "Detected reserved fields in the decoded interrupt-remapped request",
1586 "Interrupt index exceeded the interrupt-remapping table size",
1587 "Present field in the IRTE entry is clear",
1588 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1589 "Detected reserved fields in the IRTE entry",
1590 "Blocked a compatibility format interrupt request",
1591 "Blocked an interrupt request due to source-id verification failure",
1594 static const char *dmar_get_fault_reason(u8 fault_reason
, int *fault_type
)
1596 if (fault_reason
>= 0x20 && (fault_reason
- 0x20 <
1597 ARRAY_SIZE(irq_remap_fault_reasons
))) {
1598 *fault_type
= INTR_REMAP
;
1599 return irq_remap_fault_reasons
[fault_reason
- 0x20];
1600 } else if (fault_reason
>= 0x30 && (fault_reason
- 0x30 <
1601 ARRAY_SIZE(dma_remap_sm_fault_reasons
))) {
1602 *fault_type
= DMA_REMAP
;
1603 return dma_remap_sm_fault_reasons
[fault_reason
- 0x30];
1604 } else if (fault_reason
< ARRAY_SIZE(dma_remap_fault_reasons
)) {
1605 *fault_type
= DMA_REMAP
;
1606 return dma_remap_fault_reasons
[fault_reason
];
1608 *fault_type
= UNKNOWN
;
1614 static inline int dmar_msi_reg(struct intel_iommu
*iommu
, int irq
)
1616 if (iommu
->irq
== irq
)
1617 return DMAR_FECTL_REG
;
1618 else if (iommu
->pr_irq
== irq
)
1619 return DMAR_PECTL_REG
;
1624 void dmar_msi_unmask(struct irq_data
*data
)
1626 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1627 int reg
= dmar_msi_reg(iommu
, data
->irq
);
1631 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1632 writel(0, iommu
->reg
+ reg
);
1633 /* Read a reg to force flush the post write */
1634 readl(iommu
->reg
+ reg
);
1635 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1638 void dmar_msi_mask(struct irq_data
*data
)
1640 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1641 int reg
= dmar_msi_reg(iommu
, data
->irq
);
1645 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1646 writel(DMA_FECTL_IM
, iommu
->reg
+ reg
);
1647 /* Read a reg to force flush the post write */
1648 readl(iommu
->reg
+ reg
);
1649 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1652 void dmar_msi_write(int irq
, struct msi_msg
*msg
)
1654 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1655 int reg
= dmar_msi_reg(iommu
, irq
);
1658 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1659 writel(msg
->data
, iommu
->reg
+ reg
+ 4);
1660 writel(msg
->address_lo
, iommu
->reg
+ reg
+ 8);
1661 writel(msg
->address_hi
, iommu
->reg
+ reg
+ 12);
1662 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1665 void dmar_msi_read(int irq
, struct msi_msg
*msg
)
1667 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1668 int reg
= dmar_msi_reg(iommu
, irq
);
1671 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1672 msg
->data
= readl(iommu
->reg
+ reg
+ 4);
1673 msg
->address_lo
= readl(iommu
->reg
+ reg
+ 8);
1674 msg
->address_hi
= readl(iommu
->reg
+ reg
+ 12);
1675 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1678 static int dmar_fault_do_one(struct intel_iommu
*iommu
, int type
,
1679 u8 fault_reason
, int pasid
, u16 source_id
,
1680 unsigned long long addr
)
1685 reason
= dmar_get_fault_reason(fault_reason
, &fault_type
);
1687 if (fault_type
== INTR_REMAP
)
1688 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1689 source_id
>> 8, PCI_SLOT(source_id
& 0xFF),
1690 PCI_FUNC(source_id
& 0xFF), addr
>> 48,
1691 fault_reason
, reason
);
1693 pr_err("[%s] Request device [%02x:%02x.%d] PASID %x fault addr %llx [fault reason %02d] %s\n",
1694 type
? "DMA Read" : "DMA Write",
1695 source_id
>> 8, PCI_SLOT(source_id
& 0xFF),
1696 PCI_FUNC(source_id
& 0xFF), pasid
, addr
,
1697 fault_reason
, reason
);
1701 #define PRIMARY_FAULT_REG_LEN (16)
1702 irqreturn_t
dmar_fault(int irq
, void *dev_id
)
1704 struct intel_iommu
*iommu
= dev_id
;
1705 int reg
, fault_index
;
1708 static DEFINE_RATELIMIT_STATE(rs
,
1709 DEFAULT_RATELIMIT_INTERVAL
,
1710 DEFAULT_RATELIMIT_BURST
);
1712 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1713 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1714 if (fault_status
&& __ratelimit(&rs
))
1715 pr_err("DRHD: handling fault status reg %x\n", fault_status
);
1717 /* TBD: ignore advanced fault log currently */
1718 if (!(fault_status
& DMA_FSTS_PPF
))
1721 fault_index
= dma_fsts_fault_record_index(fault_status
);
1722 reg
= cap_fault_reg_offset(iommu
->cap
);
1724 /* Disable printing, simply clear the fault when ratelimited */
1725 bool ratelimited
= !__ratelimit(&rs
);
1733 /* highest 32 bits */
1734 data
= readl(iommu
->reg
+ reg
+
1735 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1736 if (!(data
& DMA_FRCD_F
))
1740 fault_reason
= dma_frcd_fault_reason(data
);
1741 type
= dma_frcd_type(data
);
1743 pasid
= dma_frcd_pasid_value(data
);
1744 data
= readl(iommu
->reg
+ reg
+
1745 fault_index
* PRIMARY_FAULT_REG_LEN
+ 8);
1746 source_id
= dma_frcd_source_id(data
);
1748 pasid_present
= dma_frcd_pasid_present(data
);
1749 guest_addr
= dmar_readq(iommu
->reg
+ reg
+
1750 fault_index
* PRIMARY_FAULT_REG_LEN
);
1751 guest_addr
= dma_frcd_page_addr(guest_addr
);
1754 /* clear the fault */
1755 writel(DMA_FRCD_F
, iommu
->reg
+ reg
+
1756 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1758 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1761 /* Using pasid -1 if pasid is not present */
1762 dmar_fault_do_one(iommu
, type
, fault_reason
,
1763 pasid_present
? pasid
: -1,
1764 source_id
, guest_addr
);
1767 if (fault_index
>= cap_num_fault_regs(iommu
->cap
))
1769 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1772 writel(DMA_FSTS_PFO
| DMA_FSTS_PPF
| DMA_FSTS_PRO
,
1773 iommu
->reg
+ DMAR_FSTS_REG
);
1776 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1780 int dmar_set_interrupt(struct intel_iommu
*iommu
)
1785 * Check if the fault interrupt is already initialized.
1790 irq
= dmar_alloc_hwirq(iommu
->seq_id
, iommu
->node
, iommu
);
1794 pr_err("No free IRQ vectors\n");
1798 ret
= request_irq(irq
, dmar_fault
, IRQF_NO_THREAD
, iommu
->name
, iommu
);
1800 pr_err("Can't request irq\n");
1804 int __init
enable_drhd_fault_handling(void)
1806 struct dmar_drhd_unit
*drhd
;
1807 struct intel_iommu
*iommu
;
1810 * Enable fault control interrupt.
1812 for_each_iommu(iommu
, drhd
) {
1814 int ret
= dmar_set_interrupt(iommu
);
1817 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1818 (unsigned long long)drhd
->reg_base_addr
, ret
);
1823 * Clear any previous faults.
1825 dmar_fault(iommu
->irq
, iommu
);
1826 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1827 writel(fault_status
, iommu
->reg
+ DMAR_FSTS_REG
);
1834 * Re-enable Queued Invalidation interface.
1836 int dmar_reenable_qi(struct intel_iommu
*iommu
)
1838 if (!ecap_qis(iommu
->ecap
))
1845 * First disable queued invalidation.
1847 dmar_disable_qi(iommu
);
1849 * Then enable queued invalidation again. Since there is no pending
1850 * invalidation requests now, it's safe to re-enable queued
1853 __dmar_enable_qi(iommu
);
1859 * Check interrupt remapping support in DMAR table description.
1861 int __init
dmar_ir_support(void)
1863 struct acpi_table_dmar
*dmar
;
1864 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
1867 return dmar
->flags
& 0x1;
1870 /* Check whether DMAR units are in use */
1871 static inline bool dmar_in_use(void)
1873 return irq_remapping_enabled
|| intel_iommu_enabled
;
1876 static int __init
dmar_free_unused_resources(void)
1878 struct dmar_drhd_unit
*dmaru
, *dmaru_n
;
1883 if (dmar_dev_scope_status
!= 1 && !list_empty(&dmar_drhd_units
))
1884 bus_unregister_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
1886 down_write(&dmar_global_lock
);
1887 list_for_each_entry_safe(dmaru
, dmaru_n
, &dmar_drhd_units
, list
) {
1888 list_del(&dmaru
->list
);
1889 dmar_free_drhd(dmaru
);
1891 up_write(&dmar_global_lock
);
1896 late_initcall(dmar_free_unused_resources
);
1897 IOMMU_INIT_POST(detect_intel_iommu
);
1900 * DMAR Hotplug Support
1901 * For more details, please refer to Intel(R) Virtualization Technology
1902 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1903 * "Remapping Hardware Unit Hot Plug".
1905 static guid_t dmar_hp_guid
=
1906 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
1907 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
1910 * Currently there's only one revision and BIOS will not check the revision id,
1911 * so use 0 for safety.
1913 #define DMAR_DSM_REV_ID 0
1914 #define DMAR_DSM_FUNC_DRHD 1
1915 #define DMAR_DSM_FUNC_ATSR 2
1916 #define DMAR_DSM_FUNC_RHSA 3
1918 static inline bool dmar_detect_dsm(acpi_handle handle
, int func
)
1920 return acpi_check_dsm(handle
, &dmar_hp_guid
, DMAR_DSM_REV_ID
, 1 << func
);
1923 static int dmar_walk_dsm_resource(acpi_handle handle
, int func
,
1924 dmar_res_handler_t handler
, void *arg
)
1927 union acpi_object
*obj
;
1928 struct acpi_dmar_header
*start
;
1929 struct dmar_res_callback callback
;
1930 static int res_type
[] = {
1931 [DMAR_DSM_FUNC_DRHD
] = ACPI_DMAR_TYPE_HARDWARE_UNIT
,
1932 [DMAR_DSM_FUNC_ATSR
] = ACPI_DMAR_TYPE_ROOT_ATS
,
1933 [DMAR_DSM_FUNC_RHSA
] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY
,
1936 if (!dmar_detect_dsm(handle
, func
))
1939 obj
= acpi_evaluate_dsm_typed(handle
, &dmar_hp_guid
, DMAR_DSM_REV_ID
,
1940 func
, NULL
, ACPI_TYPE_BUFFER
);
1944 memset(&callback
, 0, sizeof(callback
));
1945 callback
.cb
[res_type
[func
]] = handler
;
1946 callback
.arg
[res_type
[func
]] = arg
;
1947 start
= (struct acpi_dmar_header
*)obj
->buffer
.pointer
;
1948 ret
= dmar_walk_remapping_entries(start
, obj
->buffer
.length
, &callback
);
1955 static int dmar_hp_add_drhd(struct acpi_dmar_header
*header
, void *arg
)
1958 struct dmar_drhd_unit
*dmaru
;
1960 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1964 ret
= dmar_ir_hotplug(dmaru
, true);
1966 ret
= dmar_iommu_hotplug(dmaru
, true);
1971 static int dmar_hp_remove_drhd(struct acpi_dmar_header
*header
, void *arg
)
1975 struct dmar_drhd_unit
*dmaru
;
1977 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1982 * All PCI devices managed by this unit should have been destroyed.
1984 if (!dmaru
->include_all
&& dmaru
->devices
&& dmaru
->devices_cnt
) {
1985 for_each_active_dev_scope(dmaru
->devices
,
1986 dmaru
->devices_cnt
, i
, dev
)
1990 ret
= dmar_ir_hotplug(dmaru
, false);
1992 ret
= dmar_iommu_hotplug(dmaru
, false);
1997 static int dmar_hp_release_drhd(struct acpi_dmar_header
*header
, void *arg
)
1999 struct dmar_drhd_unit
*dmaru
;
2001 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
2003 list_del_rcu(&dmaru
->list
);
2005 dmar_free_drhd(dmaru
);
2011 static int dmar_hotplug_insert(acpi_handle handle
)
2016 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2017 &dmar_validate_one_drhd
, (void *)1);
2021 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2022 &dmar_parse_one_drhd
, (void *)&drhd_count
);
2023 if (ret
== 0 && drhd_count
== 0) {
2024 pr_warn(FW_BUG
"No DRHD structures in buffer returned by _DSM method\n");
2030 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_RHSA
,
2031 &dmar_parse_one_rhsa
, NULL
);
2035 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
2036 &dmar_parse_one_atsr
, NULL
);
2040 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2041 &dmar_hp_add_drhd
, NULL
);
2045 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2046 &dmar_hp_remove_drhd
, NULL
);
2048 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
2049 &dmar_release_one_atsr
, NULL
);
2051 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2052 &dmar_hp_release_drhd
, NULL
);
2057 static int dmar_hotplug_remove(acpi_handle handle
)
2061 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
2062 &dmar_check_one_atsr
, NULL
);
2066 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2067 &dmar_hp_remove_drhd
, NULL
);
2069 WARN_ON(dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
2070 &dmar_release_one_atsr
, NULL
));
2071 WARN_ON(dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2072 &dmar_hp_release_drhd
, NULL
));
2074 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
2075 &dmar_hp_add_drhd
, NULL
);
2081 static acpi_status
dmar_get_dsm_handle(acpi_handle handle
, u32 lvl
,
2082 void *context
, void **retval
)
2084 acpi_handle
*phdl
= retval
;
2086 if (dmar_detect_dsm(handle
, DMAR_DSM_FUNC_DRHD
)) {
2088 return AE_CTRL_TERMINATE
;
2094 static int dmar_device_hotplug(acpi_handle handle
, bool insert
)
2097 acpi_handle tmp
= NULL
;
2103 if (dmar_detect_dsm(handle
, DMAR_DSM_FUNC_DRHD
)) {
2106 status
= acpi_walk_namespace(ACPI_TYPE_DEVICE
, handle
,
2108 dmar_get_dsm_handle
,
2110 if (ACPI_FAILURE(status
)) {
2111 pr_warn("Failed to locate _DSM method.\n");
2118 down_write(&dmar_global_lock
);
2120 ret
= dmar_hotplug_insert(tmp
);
2122 ret
= dmar_hotplug_remove(tmp
);
2123 up_write(&dmar_global_lock
);
2128 int dmar_device_add(acpi_handle handle
)
2130 return dmar_device_hotplug(handle
, true);
2133 int dmar_device_remove(acpi_handle handle
)
2135 return dmar_device_hotplug(handle
, false);
2139 * dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table
2141 * Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in
2142 * the ACPI DMAR table. This means that the platform boot firmware has made
2143 * sure no device can issue DMA outside of RMRR regions.
2145 bool dmar_platform_optin(void)
2147 struct acpi_table_dmar
*dmar
;
2151 status
= acpi_get_table(ACPI_SIG_DMAR
, 0,
2152 (struct acpi_table_header
**)&dmar
);
2153 if (ACPI_FAILURE(status
))
2156 ret
= !!(dmar
->flags
& DMAR_PLATFORM_OPT_IN
);
2157 acpi_put_table((struct acpi_table_header
*)dmar
);
2161 EXPORT_SYMBOL_GPL(dmar_platform_optin
);