2 * Copyright (c) 2006, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
17 * Copyright (C) 2006-2008 Intel Corporation
18 * Author: Ashok Raj <ashok.raj@intel.com>
19 * Author: Shaohua Li <shaohua.li@intel.com>
20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
22 * This file implements early detection/parsing of Remapping Devices
23 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
26 * These routines are used by both DMA-remapping and Interrupt-remapping
29 #define pr_fmt(fmt) "DMAR: " fmt
31 #include <linux/pci.h>
32 #include <linux/dmar.h>
33 #include <linux/iova.h>
34 #include <linux/intel-iommu.h>
35 #include <linux/timer.h>
36 #include <linux/irq.h>
37 #include <linux/interrupt.h>
38 #include <linux/tboot.h>
39 #include <linux/dmi.h>
40 #include <linux/slab.h>
41 #include <linux/iommu.h>
42 #include <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
;
57 * 1) The hotplug framework guarentees that DMAR unit will be hot-added
58 * before IO devices managed by that unit.
59 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
60 * after IO devices managed by that unit.
61 * 3) Hotplug events are rare.
63 * Locking rules for DMA and interrupt remapping related global data structures:
64 * 1) Use dmar_global_lock in process context
65 * 2) Use RCU in interrupt context
67 DECLARE_RWSEM(dmar_global_lock
);
68 LIST_HEAD(dmar_drhd_units
);
70 struct acpi_table_header
* __initdata dmar_tbl
;
71 static int dmar_dev_scope_status
= 1;
72 static unsigned long dmar_seq_ids
[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED
)];
74 static int alloc_iommu(struct dmar_drhd_unit
*drhd
);
75 static void free_iommu(struct intel_iommu
*iommu
);
77 static void dmar_register_drhd_unit(struct dmar_drhd_unit
*drhd
)
80 * add INCLUDE_ALL at the tail, so scan the list will find it at
83 if (drhd
->include_all
)
84 list_add_tail_rcu(&drhd
->list
, &dmar_drhd_units
);
86 list_add_rcu(&drhd
->list
, &dmar_drhd_units
);
89 void *dmar_alloc_dev_scope(void *start
, void *end
, int *cnt
)
91 struct acpi_dmar_device_scope
*scope
;
96 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_NAMESPACE
||
97 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
||
98 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
100 else if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_IOAPIC
&&
101 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_HPET
) {
102 pr_warn("Unsupported device scope\n");
104 start
+= scope
->length
;
109 return kcalloc(*cnt
, sizeof(struct dmar_dev_scope
), GFP_KERNEL
);
112 void dmar_free_dev_scope(struct dmar_dev_scope
**devices
, int *cnt
)
115 struct device
*tmp_dev
;
117 if (*devices
&& *cnt
) {
118 for_each_active_dev_scope(*devices
, *cnt
, i
, tmp_dev
)
127 /* Optimize out kzalloc()/kfree() for normal cases */
128 static char dmar_pci_notify_info_buf
[64];
130 static struct dmar_pci_notify_info
*
131 dmar_alloc_pci_notify_info(struct pci_dev
*dev
, unsigned long event
)
136 struct dmar_pci_notify_info
*info
;
138 BUG_ON(dev
->is_virtfn
);
140 /* Only generate path[] for device addition event */
141 if (event
== BUS_NOTIFY_ADD_DEVICE
)
142 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
)
145 size
= sizeof(*info
) + level
* sizeof(struct acpi_dmar_pci_path
);
146 if (size
<= sizeof(dmar_pci_notify_info_buf
)) {
147 info
= (struct dmar_pci_notify_info
*)dmar_pci_notify_info_buf
;
149 info
= kzalloc(size
, GFP_KERNEL
);
151 pr_warn("Out of memory when allocating notify_info "
152 "for %s.\n", pci_name(dev
));
153 if (dmar_dev_scope_status
== 0)
154 dmar_dev_scope_status
= -ENOMEM
;
161 info
->seg
= pci_domain_nr(dev
->bus
);
163 if (event
== BUS_NOTIFY_ADD_DEVICE
) {
164 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
) {
166 info
->path
[level
].bus
= tmp
->bus
->number
;
167 info
->path
[level
].device
= PCI_SLOT(tmp
->devfn
);
168 info
->path
[level
].function
= PCI_FUNC(tmp
->devfn
);
169 if (pci_is_root_bus(tmp
->bus
))
170 info
->bus
= tmp
->bus
->number
;
177 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info
*info
)
179 if ((void *)info
!= dmar_pci_notify_info_buf
)
183 static bool dmar_match_pci_path(struct dmar_pci_notify_info
*info
, int bus
,
184 struct acpi_dmar_pci_path
*path
, int count
)
188 if (info
->bus
!= bus
)
190 if (info
->level
!= count
)
193 for (i
= 0; i
< count
; i
++) {
194 if (path
[i
].device
!= info
->path
[i
].device
||
195 path
[i
].function
!= info
->path
[i
].function
)
207 if (bus
== info
->path
[i
].bus
&&
208 path
[0].device
== info
->path
[i
].device
&&
209 path
[0].function
== info
->path
[i
].function
) {
210 pr_info(FW_BUG
"RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
211 bus
, path
[0].device
, path
[0].function
);
218 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
219 int dmar_insert_dev_scope(struct dmar_pci_notify_info
*info
,
220 void *start
, void*end
, u16 segment
,
221 struct dmar_dev_scope
*devices
,
225 struct device
*tmp
, *dev
= &info
->dev
->dev
;
226 struct acpi_dmar_device_scope
*scope
;
227 struct acpi_dmar_pci_path
*path
;
229 if (segment
!= info
->seg
)
232 for (; start
< end
; start
+= scope
->length
) {
234 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
235 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
238 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
239 level
= (scope
->length
- sizeof(*scope
)) / sizeof(*path
);
240 if (!dmar_match_pci_path(info
, scope
->bus
, path
, level
))
244 * We expect devices with endpoint scope to have normal PCI
245 * headers, and devices with bridge scope to have bridge PCI
246 * headers. However PCI NTB devices may be listed in the
247 * DMAR table with bridge scope, even though they have a
248 * normal PCI header. NTB devices are identified by class
249 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
250 * for this special case.
252 if ((scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
253 info
->dev
->hdr_type
!= PCI_HEADER_TYPE_NORMAL
) ||
254 (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
&&
255 (info
->dev
->hdr_type
== PCI_HEADER_TYPE_NORMAL
&&
256 info
->dev
->class >> 8 != PCI_CLASS_BRIDGE_OTHER
))) {
257 pr_warn("Device scope type does not match for %s\n",
258 pci_name(info
->dev
));
262 for_each_dev_scope(devices
, devices_cnt
, i
, tmp
)
264 devices
[i
].bus
= info
->dev
->bus
->number
;
265 devices
[i
].devfn
= info
->dev
->devfn
;
266 rcu_assign_pointer(devices
[i
].dev
,
270 BUG_ON(i
>= devices_cnt
);
276 int dmar_remove_dev_scope(struct dmar_pci_notify_info
*info
, u16 segment
,
277 struct dmar_dev_scope
*devices
, int count
)
282 if (info
->seg
!= segment
)
285 for_each_active_dev_scope(devices
, count
, index
, tmp
)
286 if (tmp
== &info
->dev
->dev
) {
287 RCU_INIT_POINTER(devices
[index
].dev
, NULL
);
296 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info
*info
)
299 struct dmar_drhd_unit
*dmaru
;
300 struct acpi_dmar_hardware_unit
*drhd
;
302 for_each_drhd_unit(dmaru
) {
303 if (dmaru
->include_all
)
306 drhd
= container_of(dmaru
->hdr
,
307 struct acpi_dmar_hardware_unit
, header
);
308 ret
= dmar_insert_dev_scope(info
, (void *)(drhd
+ 1),
309 ((void *)drhd
) + drhd
->header
.length
,
311 dmaru
->devices
, dmaru
->devices_cnt
);
316 ret
= dmar_iommu_notify_scope_dev(info
);
317 if (ret
< 0 && dmar_dev_scope_status
== 0)
318 dmar_dev_scope_status
= ret
;
323 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info
*info
)
325 struct dmar_drhd_unit
*dmaru
;
327 for_each_drhd_unit(dmaru
)
328 if (dmar_remove_dev_scope(info
, dmaru
->segment
,
329 dmaru
->devices
, dmaru
->devices_cnt
))
331 dmar_iommu_notify_scope_dev(info
);
334 static int dmar_pci_bus_notifier(struct notifier_block
*nb
,
335 unsigned long action
, void *data
)
337 struct pci_dev
*pdev
= to_pci_dev(data
);
338 struct dmar_pci_notify_info
*info
;
340 /* Only care about add/remove events for physical functions.
341 * For VFs we actually do the lookup based on the corresponding
342 * PF in device_to_iommu() anyway. */
345 if (action
!= BUS_NOTIFY_ADD_DEVICE
&&
346 action
!= BUS_NOTIFY_REMOVED_DEVICE
)
349 info
= dmar_alloc_pci_notify_info(pdev
, action
);
353 down_write(&dmar_global_lock
);
354 if (action
== BUS_NOTIFY_ADD_DEVICE
)
355 dmar_pci_bus_add_dev(info
);
356 else if (action
== BUS_NOTIFY_REMOVED_DEVICE
)
357 dmar_pci_bus_del_dev(info
);
358 up_write(&dmar_global_lock
);
360 dmar_free_pci_notify_info(info
);
365 static struct notifier_block dmar_pci_bus_nb
= {
366 .notifier_call
= dmar_pci_bus_notifier
,
370 static struct dmar_drhd_unit
*
371 dmar_find_dmaru(struct acpi_dmar_hardware_unit
*drhd
)
373 struct dmar_drhd_unit
*dmaru
;
375 list_for_each_entry_rcu(dmaru
, &dmar_drhd_units
, list
)
376 if (dmaru
->segment
== drhd
->segment
&&
377 dmaru
->reg_base_addr
== drhd
->address
)
384 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
385 * structure which uniquely represent one DMA remapping hardware unit
386 * present in the platform
388 static int dmar_parse_one_drhd(struct acpi_dmar_header
*header
, void *arg
)
390 struct acpi_dmar_hardware_unit
*drhd
;
391 struct dmar_drhd_unit
*dmaru
;
394 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
395 dmaru
= dmar_find_dmaru(drhd
);
399 dmaru
= kzalloc(sizeof(*dmaru
) + header
->length
, GFP_KERNEL
);
404 * If header is allocated from slab by ACPI _DSM method, we need to
405 * copy the content because the memory buffer will be freed on return.
407 dmaru
->hdr
= (void *)(dmaru
+ 1);
408 memcpy(dmaru
->hdr
, header
, header
->length
);
409 dmaru
->reg_base_addr
= drhd
->address
;
410 dmaru
->segment
= drhd
->segment
;
411 dmaru
->include_all
= drhd
->flags
& 0x1; /* BIT0: INCLUDE_ALL */
412 dmaru
->devices
= dmar_alloc_dev_scope((void *)(drhd
+ 1),
413 ((void *)drhd
) + drhd
->header
.length
,
414 &dmaru
->devices_cnt
);
415 if (dmaru
->devices_cnt
&& dmaru
->devices
== NULL
) {
420 ret
= alloc_iommu(dmaru
);
422 dmar_free_dev_scope(&dmaru
->devices
,
423 &dmaru
->devices_cnt
);
427 dmar_register_drhd_unit(dmaru
);
436 static void dmar_free_drhd(struct dmar_drhd_unit
*dmaru
)
438 if (dmaru
->devices
&& dmaru
->devices_cnt
)
439 dmar_free_dev_scope(&dmaru
->devices
, &dmaru
->devices_cnt
);
441 free_iommu(dmaru
->iommu
);
445 static int __init
dmar_parse_one_andd(struct acpi_dmar_header
*header
,
448 struct acpi_dmar_andd
*andd
= (void *)header
;
450 /* Check for NUL termination within the designated length */
451 if (strnlen(andd
->device_name
, header
->length
- 8) == header
->length
- 8) {
452 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND
,
453 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
454 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
455 dmi_get_system_info(DMI_BIOS_VENDOR
),
456 dmi_get_system_info(DMI_BIOS_VERSION
),
457 dmi_get_system_info(DMI_PRODUCT_VERSION
));
460 pr_info("ANDD device: %x name: %s\n", andd
->device_number
,
466 #ifdef CONFIG_ACPI_NUMA
467 static int dmar_parse_one_rhsa(struct acpi_dmar_header
*header
, void *arg
)
469 struct acpi_dmar_rhsa
*rhsa
;
470 struct dmar_drhd_unit
*drhd
;
472 rhsa
= (struct acpi_dmar_rhsa
*)header
;
473 for_each_drhd_unit(drhd
) {
474 if (drhd
->reg_base_addr
== rhsa
->base_address
) {
475 int node
= acpi_map_pxm_to_node(rhsa
->proximity_domain
);
477 if (!node_online(node
))
479 drhd
->iommu
->node
= node
;
484 1, TAINT_FIRMWARE_WORKAROUND
,
485 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
486 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
488 dmi_get_system_info(DMI_BIOS_VENDOR
),
489 dmi_get_system_info(DMI_BIOS_VERSION
),
490 dmi_get_system_info(DMI_PRODUCT_VERSION
));
495 #define dmar_parse_one_rhsa dmar_res_noop
499 dmar_table_print_dmar_entry(struct acpi_dmar_header
*header
)
501 struct acpi_dmar_hardware_unit
*drhd
;
502 struct acpi_dmar_reserved_memory
*rmrr
;
503 struct acpi_dmar_atsr
*atsr
;
504 struct acpi_dmar_rhsa
*rhsa
;
506 switch (header
->type
) {
507 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
508 drhd
= container_of(header
, struct acpi_dmar_hardware_unit
,
510 pr_info("DRHD base: %#016Lx flags: %#x\n",
511 (unsigned long long)drhd
->address
, drhd
->flags
);
513 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
514 rmrr
= container_of(header
, struct acpi_dmar_reserved_memory
,
516 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
517 (unsigned long long)rmrr
->base_address
,
518 (unsigned long long)rmrr
->end_address
);
520 case ACPI_DMAR_TYPE_ROOT_ATS
:
521 atsr
= container_of(header
, struct acpi_dmar_atsr
, header
);
522 pr_info("ATSR flags: %#x\n", atsr
->flags
);
524 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY
:
525 rhsa
= container_of(header
, struct acpi_dmar_rhsa
, header
);
526 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
527 (unsigned long long)rhsa
->base_address
,
528 rhsa
->proximity_domain
);
530 case ACPI_DMAR_TYPE_NAMESPACE
:
531 /* We don't print this here because we need to sanity-check
532 it first. So print it in dmar_parse_one_andd() instead. */
538 * dmar_table_detect - checks to see if the platform supports DMAR devices
540 static int __init
dmar_table_detect(void)
542 acpi_status status
= AE_OK
;
544 /* if we could find DMAR table, then there are DMAR devices */
545 status
= acpi_get_table(ACPI_SIG_DMAR
, 0, &dmar_tbl
);
547 if (ACPI_SUCCESS(status
) && !dmar_tbl
) {
548 pr_warn("Unable to map DMAR\n");
549 status
= AE_NOT_FOUND
;
552 return (ACPI_SUCCESS(status
) ? 1 : 0);
555 static int dmar_walk_remapping_entries(struct acpi_dmar_header
*start
,
556 size_t len
, struct dmar_res_callback
*cb
)
559 struct acpi_dmar_header
*iter
, *next
;
560 struct acpi_dmar_header
*end
= ((void *)start
) + len
;
562 for (iter
= start
; iter
< end
&& ret
== 0; iter
= next
) {
563 next
= (void *)iter
+ iter
->length
;
564 if (iter
->length
== 0) {
565 /* Avoid looping forever on bad ACPI tables */
566 pr_debug(FW_BUG
"Invalid 0-length structure\n");
568 } else if (next
> end
) {
569 /* Avoid passing table end */
570 pr_warn(FW_BUG
"Record passes table end\n");
576 dmar_table_print_dmar_entry(iter
);
578 if (iter
->type
>= ACPI_DMAR_TYPE_RESERVED
) {
579 /* continue for forward compatibility */
580 pr_debug("Unknown DMAR structure type %d\n",
582 } else if (cb
->cb
[iter
->type
]) {
583 ret
= cb
->cb
[iter
->type
](iter
, cb
->arg
[iter
->type
]);
584 } else if (!cb
->ignore_unhandled
) {
585 pr_warn("No handler for DMAR structure type %d\n",
594 static inline int dmar_walk_dmar_table(struct acpi_table_dmar
*dmar
,
595 struct dmar_res_callback
*cb
)
597 return dmar_walk_remapping_entries((void *)(dmar
+ 1),
598 dmar
->header
.length
- sizeof(*dmar
), cb
);
602 * parse_dmar_table - parses the DMA reporting table
605 parse_dmar_table(void)
607 struct acpi_table_dmar
*dmar
;
610 struct dmar_res_callback cb
= {
612 .ignore_unhandled
= true,
613 .arg
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &drhd_count
,
614 .cb
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &dmar_parse_one_drhd
,
615 .cb
[ACPI_DMAR_TYPE_RESERVED_MEMORY
] = &dmar_parse_one_rmrr
,
616 .cb
[ACPI_DMAR_TYPE_ROOT_ATS
] = &dmar_parse_one_atsr
,
617 .cb
[ACPI_DMAR_TYPE_HARDWARE_AFFINITY
] = &dmar_parse_one_rhsa
,
618 .cb
[ACPI_DMAR_TYPE_NAMESPACE
] = &dmar_parse_one_andd
,
622 * Do it again, earlier dmar_tbl mapping could be mapped with
628 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
629 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
631 dmar_tbl
= tboot_get_dmar_table(dmar_tbl
);
633 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
637 if (dmar
->width
< PAGE_SHIFT
- 1) {
638 pr_warn("Invalid DMAR haw\n");
642 pr_info("Host address width %d\n", dmar
->width
+ 1);
643 ret
= dmar_walk_dmar_table(dmar
, &cb
);
644 if (ret
== 0 && drhd_count
== 0)
645 pr_warn(FW_BUG
"No DRHD structure found in DMAR table\n");
650 static int dmar_pci_device_match(struct dmar_dev_scope devices
[],
651 int cnt
, struct pci_dev
*dev
)
657 for_each_active_dev_scope(devices
, cnt
, index
, tmp
)
658 if (dev_is_pci(tmp
) && dev
== to_pci_dev(tmp
))
661 /* Check our parent */
662 dev
= dev
->bus
->self
;
668 struct dmar_drhd_unit
*
669 dmar_find_matched_drhd_unit(struct pci_dev
*dev
)
671 struct dmar_drhd_unit
*dmaru
;
672 struct acpi_dmar_hardware_unit
*drhd
;
674 dev
= pci_physfn(dev
);
677 for_each_drhd_unit(dmaru
) {
678 drhd
= container_of(dmaru
->hdr
,
679 struct acpi_dmar_hardware_unit
,
682 if (dmaru
->include_all
&&
683 drhd
->segment
== pci_domain_nr(dev
->bus
))
686 if (dmar_pci_device_match(dmaru
->devices
,
687 dmaru
->devices_cnt
, dev
))
697 static void __init
dmar_acpi_insert_dev_scope(u8 device_number
,
698 struct acpi_device
*adev
)
700 struct dmar_drhd_unit
*dmaru
;
701 struct acpi_dmar_hardware_unit
*drhd
;
702 struct acpi_dmar_device_scope
*scope
;
705 struct acpi_dmar_pci_path
*path
;
707 for_each_drhd_unit(dmaru
) {
708 drhd
= container_of(dmaru
->hdr
,
709 struct acpi_dmar_hardware_unit
,
712 for (scope
= (void *)(drhd
+ 1);
713 (unsigned long)scope
< ((unsigned long)drhd
) + drhd
->header
.length
;
714 scope
= ((void *)scope
) + scope
->length
) {
715 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_NAMESPACE
)
717 if (scope
->enumeration_id
!= device_number
)
720 path
= (void *)(scope
+ 1);
721 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
722 dev_name(&adev
->dev
), dmaru
->reg_base_addr
,
723 scope
->bus
, path
->device
, path
->function
);
724 for_each_dev_scope(dmaru
->devices
, dmaru
->devices_cnt
, i
, tmp
)
726 dmaru
->devices
[i
].bus
= scope
->bus
;
727 dmaru
->devices
[i
].devfn
= PCI_DEVFN(path
->device
,
729 rcu_assign_pointer(dmaru
->devices
[i
].dev
,
730 get_device(&adev
->dev
));
733 BUG_ON(i
>= dmaru
->devices_cnt
);
736 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
737 device_number
, dev_name(&adev
->dev
));
740 static int __init
dmar_acpi_dev_scope_init(void)
742 struct acpi_dmar_andd
*andd
;
744 if (dmar_tbl
== NULL
)
747 for (andd
= (void *)dmar_tbl
+ sizeof(struct acpi_table_dmar
);
748 ((unsigned long)andd
) < ((unsigned long)dmar_tbl
) + dmar_tbl
->length
;
749 andd
= ((void *)andd
) + andd
->header
.length
) {
750 if (andd
->header
.type
== ACPI_DMAR_TYPE_NAMESPACE
) {
752 struct acpi_device
*adev
;
754 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT
,
757 pr_err("Failed to find handle for ACPI object %s\n",
761 if (acpi_bus_get_device(h
, &adev
)) {
762 pr_err("Failed to get device for ACPI object %s\n",
766 dmar_acpi_insert_dev_scope(andd
->device_number
, adev
);
772 int __init
dmar_dev_scope_init(void)
774 struct pci_dev
*dev
= NULL
;
775 struct dmar_pci_notify_info
*info
;
777 if (dmar_dev_scope_status
!= 1)
778 return dmar_dev_scope_status
;
780 if (list_empty(&dmar_drhd_units
)) {
781 dmar_dev_scope_status
= -ENODEV
;
783 dmar_dev_scope_status
= 0;
785 dmar_acpi_dev_scope_init();
787 for_each_pci_dev(dev
) {
791 info
= dmar_alloc_pci_notify_info(dev
,
792 BUS_NOTIFY_ADD_DEVICE
);
794 return dmar_dev_scope_status
;
796 dmar_pci_bus_add_dev(info
);
797 dmar_free_pci_notify_info(info
);
801 bus_register_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
804 return dmar_dev_scope_status
;
808 int __init
dmar_table_init(void)
810 static int dmar_table_initialized
;
813 if (dmar_table_initialized
== 0) {
814 ret
= parse_dmar_table();
817 pr_info("Parse DMAR table failure.\n");
818 } else if (list_empty(&dmar_drhd_units
)) {
819 pr_info("No DMAR devices found\n");
824 dmar_table_initialized
= ret
;
826 dmar_table_initialized
= 1;
829 return dmar_table_initialized
< 0 ? dmar_table_initialized
: 0;
832 static void warn_invalid_dmar(u64 addr
, const char *message
)
835 1, TAINT_FIRMWARE_WORKAROUND
,
836 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
837 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
839 dmi_get_system_info(DMI_BIOS_VENDOR
),
840 dmi_get_system_info(DMI_BIOS_VERSION
),
841 dmi_get_system_info(DMI_PRODUCT_VERSION
));
845 dmar_validate_one_drhd(struct acpi_dmar_header
*entry
, void *arg
)
847 struct acpi_dmar_hardware_unit
*drhd
;
851 drhd
= (void *)entry
;
852 if (!drhd
->address
) {
853 warn_invalid_dmar(0, "");
858 addr
= ioremap(drhd
->address
, VTD_PAGE_SIZE
);
860 addr
= early_ioremap(drhd
->address
, VTD_PAGE_SIZE
);
862 pr_warn("Can't validate DRHD address: %llx\n", drhd
->address
);
866 cap
= dmar_readq(addr
+ DMAR_CAP_REG
);
867 ecap
= dmar_readq(addr
+ DMAR_ECAP_REG
);
872 early_iounmap(addr
, VTD_PAGE_SIZE
);
874 if (cap
== (uint64_t)-1 && ecap
== (uint64_t)-1) {
875 warn_invalid_dmar(drhd
->address
, " returns all ones");
882 int __init
detect_intel_iommu(void)
885 struct dmar_res_callback validate_drhd_cb
= {
886 .cb
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &dmar_validate_one_drhd
,
887 .ignore_unhandled
= true,
890 down_write(&dmar_global_lock
);
891 ret
= dmar_table_detect();
893 ret
= !dmar_walk_dmar_table((struct acpi_table_dmar
*)dmar_tbl
,
895 if (ret
&& !no_iommu
&& !iommu_detected
&& !dmar_disabled
) {
897 /* Make sure ACS will be enabled */
903 x86_init
.iommu
.iommu_init
= intel_iommu_init
;
906 acpi_put_table(dmar_tbl
);
908 up_write(&dmar_global_lock
);
910 return ret
? 1 : -ENODEV
;
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
;
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 iommu
->iommu_dev
= iommu_device_create(NULL
, iommu
,
1083 if (IS_ERR(iommu
->iommu_dev
)) {
1084 err
= PTR_ERR(iommu
->iommu_dev
);
1089 drhd
->iommu
= iommu
;
1096 dmar_free_seq_id(iommu
);
1102 static void free_iommu(struct intel_iommu
*iommu
)
1104 iommu_device_destroy(iommu
->iommu_dev
);
1107 if (iommu
->pr_irq
) {
1108 free_irq(iommu
->pr_irq
, iommu
);
1109 dmar_free_hwirq(iommu
->pr_irq
);
1112 free_irq(iommu
->irq
, iommu
);
1113 dmar_free_hwirq(iommu
->irq
);
1118 free_page((unsigned long)iommu
->qi
->desc
);
1119 kfree(iommu
->qi
->desc_status
);
1126 dmar_free_seq_id(iommu
);
1131 * Reclaim all the submitted descriptors which have completed its work.
1133 static inline void reclaim_free_desc(struct q_inval
*qi
)
1135 while (qi
->desc_status
[qi
->free_tail
] == QI_DONE
||
1136 qi
->desc_status
[qi
->free_tail
] == QI_ABORT
) {
1137 qi
->desc_status
[qi
->free_tail
] = QI_FREE
;
1138 qi
->free_tail
= (qi
->free_tail
+ 1) % QI_LENGTH
;
1143 static int qi_check_fault(struct intel_iommu
*iommu
, int index
)
1147 struct q_inval
*qi
= iommu
->qi
;
1148 int wait_index
= (index
+ 1) % QI_LENGTH
;
1150 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
1153 fault
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1156 * If IQE happens, the head points to the descriptor associated
1157 * with the error. No new descriptors are fetched until the IQE
1160 if (fault
& DMA_FSTS_IQE
) {
1161 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
1162 if ((head
>> DMAR_IQ_SHIFT
) == index
) {
1163 pr_err("VT-d detected invalid descriptor: "
1164 "low=%llx, high=%llx\n",
1165 (unsigned long long)qi
->desc
[index
].low
,
1166 (unsigned long long)qi
->desc
[index
].high
);
1167 memcpy(&qi
->desc
[index
], &qi
->desc
[wait_index
],
1168 sizeof(struct qi_desc
));
1169 writel(DMA_FSTS_IQE
, iommu
->reg
+ DMAR_FSTS_REG
);
1175 * If ITE happens, all pending wait_desc commands are aborted.
1176 * No new descriptors are fetched until the ITE is cleared.
1178 if (fault
& DMA_FSTS_ITE
) {
1179 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
1180 head
= ((head
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
1182 tail
= readl(iommu
->reg
+ DMAR_IQT_REG
);
1183 tail
= ((tail
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
1185 writel(DMA_FSTS_ITE
, iommu
->reg
+ DMAR_FSTS_REG
);
1188 if (qi
->desc_status
[head
] == QI_IN_USE
)
1189 qi
->desc_status
[head
] = QI_ABORT
;
1190 head
= (head
- 2 + QI_LENGTH
) % QI_LENGTH
;
1191 } while (head
!= tail
);
1193 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
1197 if (fault
& DMA_FSTS_ICE
)
1198 writel(DMA_FSTS_ICE
, iommu
->reg
+ DMAR_FSTS_REG
);
1204 * Submit the queued invalidation descriptor to the remapping
1205 * hardware unit and wait for its completion.
1207 int qi_submit_sync(struct qi_desc
*desc
, struct intel_iommu
*iommu
)
1210 struct q_inval
*qi
= iommu
->qi
;
1211 struct qi_desc
*hw
, wait_desc
;
1212 int wait_index
, index
;
1213 unsigned long flags
;
1223 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1224 while (qi
->free_cnt
< 3) {
1225 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1227 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1230 index
= qi
->free_head
;
1231 wait_index
= (index
+ 1) % QI_LENGTH
;
1233 qi
->desc_status
[index
] = qi
->desc_status
[wait_index
] = QI_IN_USE
;
1237 wait_desc
.low
= QI_IWD_STATUS_DATA(QI_DONE
) |
1238 QI_IWD_STATUS_WRITE
| QI_IWD_TYPE
;
1239 wait_desc
.high
= virt_to_phys(&qi
->desc_status
[wait_index
]);
1241 hw
[wait_index
] = wait_desc
;
1243 qi
->free_head
= (qi
->free_head
+ 2) % QI_LENGTH
;
1247 * update the HW tail register indicating the presence of
1250 writel(qi
->free_head
<< DMAR_IQ_SHIFT
, iommu
->reg
+ DMAR_IQT_REG
);
1252 while (qi
->desc_status
[wait_index
] != QI_DONE
) {
1254 * We will leave the interrupts disabled, to prevent interrupt
1255 * context to queue another cmd while a cmd is already submitted
1256 * and waiting for completion on this cpu. This is to avoid
1257 * a deadlock where the interrupt context can wait indefinitely
1258 * for free slots in the queue.
1260 rc
= qi_check_fault(iommu
, index
);
1264 raw_spin_unlock(&qi
->q_lock
);
1266 raw_spin_lock(&qi
->q_lock
);
1269 qi
->desc_status
[index
] = QI_DONE
;
1271 reclaim_free_desc(qi
);
1272 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1281 * Flush the global interrupt entry cache.
1283 void qi_global_iec(struct intel_iommu
*iommu
)
1285 struct qi_desc desc
;
1287 desc
.low
= QI_IEC_TYPE
;
1290 /* should never fail */
1291 qi_submit_sync(&desc
, iommu
);
1294 void qi_flush_context(struct intel_iommu
*iommu
, u16 did
, u16 sid
, u8 fm
,
1297 struct qi_desc desc
;
1299 desc
.low
= QI_CC_FM(fm
) | QI_CC_SID(sid
) | QI_CC_DID(did
)
1300 | QI_CC_GRAN(type
) | QI_CC_TYPE
;
1303 qi_submit_sync(&desc
, iommu
);
1306 void qi_flush_iotlb(struct intel_iommu
*iommu
, u16 did
, u64 addr
,
1307 unsigned int size_order
, u64 type
)
1311 struct qi_desc desc
;
1314 if (cap_write_drain(iommu
->cap
))
1317 if (cap_read_drain(iommu
->cap
))
1320 desc
.low
= QI_IOTLB_DID(did
) | QI_IOTLB_DR(dr
) | QI_IOTLB_DW(dw
)
1321 | QI_IOTLB_GRAN(type
) | QI_IOTLB_TYPE
;
1322 desc
.high
= QI_IOTLB_ADDR(addr
) | QI_IOTLB_IH(ih
)
1323 | QI_IOTLB_AM(size_order
);
1325 qi_submit_sync(&desc
, iommu
);
1328 void qi_flush_dev_iotlb(struct intel_iommu
*iommu
, u16 sid
, u16 qdep
,
1329 u64 addr
, unsigned mask
)
1331 struct qi_desc desc
;
1334 BUG_ON(addr
& ((1 << (VTD_PAGE_SHIFT
+ mask
)) - 1));
1335 addr
|= (1 << (VTD_PAGE_SHIFT
+ mask
- 1)) - 1;
1336 desc
.high
= QI_DEV_IOTLB_ADDR(addr
) | QI_DEV_IOTLB_SIZE
;
1338 desc
.high
= QI_DEV_IOTLB_ADDR(addr
);
1340 if (qdep
>= QI_DEV_IOTLB_MAX_INVS
)
1343 desc
.low
= QI_DEV_IOTLB_SID(sid
) | QI_DEV_IOTLB_QDEP(qdep
) |
1346 qi_submit_sync(&desc
, iommu
);
1350 * Disable Queued Invalidation interface.
1352 void dmar_disable_qi(struct intel_iommu
*iommu
)
1354 unsigned long flags
;
1356 cycles_t start_time
= get_cycles();
1358 if (!ecap_qis(iommu
->ecap
))
1361 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1363 sts
= readl(iommu
->reg
+ DMAR_GSTS_REG
);
1364 if (!(sts
& DMA_GSTS_QIES
))
1368 * Give a chance to HW to complete the pending invalidation requests.
1370 while ((readl(iommu
->reg
+ DMAR_IQT_REG
) !=
1371 readl(iommu
->reg
+ DMAR_IQH_REG
)) &&
1372 (DMAR_OPERATION_TIMEOUT
> (get_cycles() - start_time
)))
1375 iommu
->gcmd
&= ~DMA_GCMD_QIE
;
1376 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1378 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
,
1379 !(sts
& DMA_GSTS_QIES
), sts
);
1381 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1385 * Enable queued invalidation.
1387 static void __dmar_enable_qi(struct intel_iommu
*iommu
)
1390 unsigned long flags
;
1391 struct q_inval
*qi
= iommu
->qi
;
1393 qi
->free_head
= qi
->free_tail
= 0;
1394 qi
->free_cnt
= QI_LENGTH
;
1396 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1398 /* write zero to the tail reg */
1399 writel(0, iommu
->reg
+ DMAR_IQT_REG
);
1401 dmar_writeq(iommu
->reg
+ DMAR_IQA_REG
, virt_to_phys(qi
->desc
));
1403 iommu
->gcmd
|= DMA_GCMD_QIE
;
1404 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1406 /* Make sure hardware complete it */
1407 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
, (sts
& DMA_GSTS_QIES
), sts
);
1409 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1413 * Enable Queued Invalidation interface. This is a must to support
1414 * interrupt-remapping. Also used by DMA-remapping, which replaces
1415 * register based IOTLB invalidation.
1417 int dmar_enable_qi(struct intel_iommu
*iommu
)
1420 struct page
*desc_page
;
1422 if (!ecap_qis(iommu
->ecap
))
1426 * queued invalidation is already setup and enabled.
1431 iommu
->qi
= kmalloc(sizeof(*qi
), GFP_ATOMIC
);
1438 desc_page
= alloc_pages_node(iommu
->node
, GFP_ATOMIC
| __GFP_ZERO
, 0);
1445 qi
->desc
= page_address(desc_page
);
1447 qi
->desc_status
= kzalloc(QI_LENGTH
* sizeof(int), GFP_ATOMIC
);
1448 if (!qi
->desc_status
) {
1449 free_page((unsigned long) qi
->desc
);
1455 raw_spin_lock_init(&qi
->q_lock
);
1457 __dmar_enable_qi(iommu
);
1462 /* iommu interrupt handling. Most stuff are MSI-like. */
1470 static const char *dma_remap_fault_reasons
[] =
1473 "Present bit in root entry is clear",
1474 "Present bit in context entry is clear",
1475 "Invalid context entry",
1476 "Access beyond MGAW",
1477 "PTE Write access is not set",
1478 "PTE Read access is not set",
1479 "Next page table ptr is invalid",
1480 "Root table address invalid",
1481 "Context table ptr is invalid",
1482 "non-zero reserved fields in RTP",
1483 "non-zero reserved fields in CTP",
1484 "non-zero reserved fields in PTE",
1485 "PCE for translation request specifies blocking",
1488 static const char *irq_remap_fault_reasons
[] =
1490 "Detected reserved fields in the decoded interrupt-remapped request",
1491 "Interrupt index exceeded the interrupt-remapping table size",
1492 "Present field in the IRTE entry is clear",
1493 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1494 "Detected reserved fields in the IRTE entry",
1495 "Blocked a compatibility format interrupt request",
1496 "Blocked an interrupt request due to source-id verification failure",
1499 static const char *dmar_get_fault_reason(u8 fault_reason
, int *fault_type
)
1501 if (fault_reason
>= 0x20 && (fault_reason
- 0x20 <
1502 ARRAY_SIZE(irq_remap_fault_reasons
))) {
1503 *fault_type
= INTR_REMAP
;
1504 return irq_remap_fault_reasons
[fault_reason
- 0x20];
1505 } else if (fault_reason
< ARRAY_SIZE(dma_remap_fault_reasons
)) {
1506 *fault_type
= DMA_REMAP
;
1507 return dma_remap_fault_reasons
[fault_reason
];
1509 *fault_type
= UNKNOWN
;
1515 static inline int dmar_msi_reg(struct intel_iommu
*iommu
, int irq
)
1517 if (iommu
->irq
== irq
)
1518 return DMAR_FECTL_REG
;
1519 else if (iommu
->pr_irq
== irq
)
1520 return DMAR_PECTL_REG
;
1525 void dmar_msi_unmask(struct irq_data
*data
)
1527 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1528 int reg
= dmar_msi_reg(iommu
, data
->irq
);
1532 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1533 writel(0, iommu
->reg
+ reg
);
1534 /* Read a reg to force flush the post write */
1535 readl(iommu
->reg
+ reg
);
1536 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1539 void dmar_msi_mask(struct irq_data
*data
)
1541 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1542 int reg
= dmar_msi_reg(iommu
, data
->irq
);
1546 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1547 writel(DMA_FECTL_IM
, iommu
->reg
+ reg
);
1548 /* Read a reg to force flush the post write */
1549 readl(iommu
->reg
+ reg
);
1550 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1553 void dmar_msi_write(int irq
, struct msi_msg
*msg
)
1555 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1556 int reg
= dmar_msi_reg(iommu
, irq
);
1559 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1560 writel(msg
->data
, iommu
->reg
+ reg
+ 4);
1561 writel(msg
->address_lo
, iommu
->reg
+ reg
+ 8);
1562 writel(msg
->address_hi
, iommu
->reg
+ reg
+ 12);
1563 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1566 void dmar_msi_read(int irq
, struct msi_msg
*msg
)
1568 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1569 int reg
= dmar_msi_reg(iommu
, irq
);
1572 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1573 msg
->data
= readl(iommu
->reg
+ reg
+ 4);
1574 msg
->address_lo
= readl(iommu
->reg
+ reg
+ 8);
1575 msg
->address_hi
= readl(iommu
->reg
+ reg
+ 12);
1576 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1579 static int dmar_fault_do_one(struct intel_iommu
*iommu
, int type
,
1580 u8 fault_reason
, u16 source_id
, unsigned long long addr
)
1585 reason
= dmar_get_fault_reason(fault_reason
, &fault_type
);
1587 if (fault_type
== INTR_REMAP
)
1588 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1589 source_id
>> 8, PCI_SLOT(source_id
& 0xFF),
1590 PCI_FUNC(source_id
& 0xFF), addr
>> 48,
1591 fault_reason
, reason
);
1593 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1594 type
? "DMA Read" : "DMA Write",
1595 source_id
>> 8, PCI_SLOT(source_id
& 0xFF),
1596 PCI_FUNC(source_id
& 0xFF), addr
, fault_reason
, reason
);
1600 #define PRIMARY_FAULT_REG_LEN (16)
1601 irqreturn_t
dmar_fault(int irq
, void *dev_id
)
1603 struct intel_iommu
*iommu
= dev_id
;
1604 int reg
, fault_index
;
1608 static DEFINE_RATELIMIT_STATE(rs
,
1609 DEFAULT_RATELIMIT_INTERVAL
,
1610 DEFAULT_RATELIMIT_BURST
);
1612 /* Disable printing, simply clear the fault when ratelimited */
1613 ratelimited
= !__ratelimit(&rs
);
1615 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1616 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1617 if (fault_status
&& !ratelimited
)
1618 pr_err("DRHD: handling fault status reg %x\n", fault_status
);
1620 /* TBD: ignore advanced fault log currently */
1621 if (!(fault_status
& DMA_FSTS_PPF
))
1624 fault_index
= dma_fsts_fault_record_index(fault_status
);
1625 reg
= cap_fault_reg_offset(iommu
->cap
);
1633 /* highest 32 bits */
1634 data
= readl(iommu
->reg
+ reg
+
1635 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1636 if (!(data
& DMA_FRCD_F
))
1640 fault_reason
= dma_frcd_fault_reason(data
);
1641 type
= dma_frcd_type(data
);
1643 data
= readl(iommu
->reg
+ reg
+
1644 fault_index
* PRIMARY_FAULT_REG_LEN
+ 8);
1645 source_id
= dma_frcd_source_id(data
);
1647 guest_addr
= dmar_readq(iommu
->reg
+ reg
+
1648 fault_index
* PRIMARY_FAULT_REG_LEN
);
1649 guest_addr
= dma_frcd_page_addr(guest_addr
);
1652 /* clear the fault */
1653 writel(DMA_FRCD_F
, iommu
->reg
+ reg
+
1654 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1656 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1659 dmar_fault_do_one(iommu
, type
, fault_reason
,
1660 source_id
, guest_addr
);
1663 if (fault_index
>= cap_num_fault_regs(iommu
->cap
))
1665 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1668 writel(DMA_FSTS_PFO
| DMA_FSTS_PPF
, iommu
->reg
+ DMAR_FSTS_REG
);
1671 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1675 int dmar_set_interrupt(struct intel_iommu
*iommu
)
1680 * Check if the fault interrupt is already initialized.
1685 irq
= dmar_alloc_hwirq(iommu
->seq_id
, iommu
->node
, iommu
);
1689 pr_err("No free IRQ vectors\n");
1693 ret
= request_irq(irq
, dmar_fault
, IRQF_NO_THREAD
, iommu
->name
, iommu
);
1695 pr_err("Can't request irq\n");
1699 int __init
enable_drhd_fault_handling(void)
1701 struct dmar_drhd_unit
*drhd
;
1702 struct intel_iommu
*iommu
;
1705 * Enable fault control interrupt.
1707 for_each_iommu(iommu
, drhd
) {
1709 int ret
= dmar_set_interrupt(iommu
);
1712 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1713 (unsigned long long)drhd
->reg_base_addr
, ret
);
1718 * Clear any previous faults.
1720 dmar_fault(iommu
->irq
, iommu
);
1721 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1722 writel(fault_status
, iommu
->reg
+ DMAR_FSTS_REG
);
1729 * Re-enable Queued Invalidation interface.
1731 int dmar_reenable_qi(struct intel_iommu
*iommu
)
1733 if (!ecap_qis(iommu
->ecap
))
1740 * First disable queued invalidation.
1742 dmar_disable_qi(iommu
);
1744 * Then enable queued invalidation again. Since there is no pending
1745 * invalidation requests now, it's safe to re-enable queued
1748 __dmar_enable_qi(iommu
);
1754 * Check interrupt remapping support in DMAR table description.
1756 int __init
dmar_ir_support(void)
1758 struct acpi_table_dmar
*dmar
;
1759 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
1762 return dmar
->flags
& 0x1;
1765 /* Check whether DMAR units are in use */
1766 static inline bool dmar_in_use(void)
1768 return irq_remapping_enabled
|| intel_iommu_enabled
;
1771 static int __init
dmar_free_unused_resources(void)
1773 struct dmar_drhd_unit
*dmaru
, *dmaru_n
;
1778 if (dmar_dev_scope_status
!= 1 && !list_empty(&dmar_drhd_units
))
1779 bus_unregister_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
1781 down_write(&dmar_global_lock
);
1782 list_for_each_entry_safe(dmaru
, dmaru_n
, &dmar_drhd_units
, list
) {
1783 list_del(&dmaru
->list
);
1784 dmar_free_drhd(dmaru
);
1786 up_write(&dmar_global_lock
);
1791 late_initcall(dmar_free_unused_resources
);
1792 IOMMU_INIT_POST(detect_intel_iommu
);
1795 * DMAR Hotplug Support
1796 * For more details, please refer to Intel(R) Virtualization Technology
1797 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1798 * "Remapping Hardware Unit Hot Plug".
1800 static u8 dmar_hp_uuid
[] = {
1801 /* 0000 */ 0xA6, 0xA3, 0xC1, 0xD8, 0x9B, 0xBE, 0x9B, 0x4C,
1802 /* 0008 */ 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF
1806 * Currently there's only one revision and BIOS will not check the revision id,
1807 * so use 0 for safety.
1809 #define DMAR_DSM_REV_ID 0
1810 #define DMAR_DSM_FUNC_DRHD 1
1811 #define DMAR_DSM_FUNC_ATSR 2
1812 #define DMAR_DSM_FUNC_RHSA 3
1814 static inline bool dmar_detect_dsm(acpi_handle handle
, int func
)
1816 return acpi_check_dsm(handle
, dmar_hp_uuid
, DMAR_DSM_REV_ID
, 1 << func
);
1819 static int dmar_walk_dsm_resource(acpi_handle handle
, int func
,
1820 dmar_res_handler_t handler
, void *arg
)
1823 union acpi_object
*obj
;
1824 struct acpi_dmar_header
*start
;
1825 struct dmar_res_callback callback
;
1826 static int res_type
[] = {
1827 [DMAR_DSM_FUNC_DRHD
] = ACPI_DMAR_TYPE_HARDWARE_UNIT
,
1828 [DMAR_DSM_FUNC_ATSR
] = ACPI_DMAR_TYPE_ROOT_ATS
,
1829 [DMAR_DSM_FUNC_RHSA
] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY
,
1832 if (!dmar_detect_dsm(handle
, func
))
1835 obj
= acpi_evaluate_dsm_typed(handle
, dmar_hp_uuid
, DMAR_DSM_REV_ID
,
1836 func
, NULL
, ACPI_TYPE_BUFFER
);
1840 memset(&callback
, 0, sizeof(callback
));
1841 callback
.cb
[res_type
[func
]] = handler
;
1842 callback
.arg
[res_type
[func
]] = arg
;
1843 start
= (struct acpi_dmar_header
*)obj
->buffer
.pointer
;
1844 ret
= dmar_walk_remapping_entries(start
, obj
->buffer
.length
, &callback
);
1851 static int dmar_hp_add_drhd(struct acpi_dmar_header
*header
, void *arg
)
1854 struct dmar_drhd_unit
*dmaru
;
1856 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1860 ret
= dmar_ir_hotplug(dmaru
, true);
1862 ret
= dmar_iommu_hotplug(dmaru
, true);
1867 static int dmar_hp_remove_drhd(struct acpi_dmar_header
*header
, void *arg
)
1871 struct dmar_drhd_unit
*dmaru
;
1873 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1878 * All PCI devices managed by this unit should have been destroyed.
1880 if (!dmaru
->include_all
&& dmaru
->devices
&& dmaru
->devices_cnt
) {
1881 for_each_active_dev_scope(dmaru
->devices
,
1882 dmaru
->devices_cnt
, i
, dev
)
1886 ret
= dmar_ir_hotplug(dmaru
, false);
1888 ret
= dmar_iommu_hotplug(dmaru
, false);
1893 static int dmar_hp_release_drhd(struct acpi_dmar_header
*header
, void *arg
)
1895 struct dmar_drhd_unit
*dmaru
;
1897 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1899 list_del_rcu(&dmaru
->list
);
1901 dmar_free_drhd(dmaru
);
1907 static int dmar_hotplug_insert(acpi_handle handle
)
1912 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1913 &dmar_validate_one_drhd
, (void *)1);
1917 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1918 &dmar_parse_one_drhd
, (void *)&drhd_count
);
1919 if (ret
== 0 && drhd_count
== 0) {
1920 pr_warn(FW_BUG
"No DRHD structures in buffer returned by _DSM method\n");
1926 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_RHSA
,
1927 &dmar_parse_one_rhsa
, NULL
);
1931 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1932 &dmar_parse_one_atsr
, NULL
);
1936 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1937 &dmar_hp_add_drhd
, NULL
);
1941 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1942 &dmar_hp_remove_drhd
, NULL
);
1944 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1945 &dmar_release_one_atsr
, NULL
);
1947 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1948 &dmar_hp_release_drhd
, NULL
);
1953 static int dmar_hotplug_remove(acpi_handle handle
)
1957 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1958 &dmar_check_one_atsr
, NULL
);
1962 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1963 &dmar_hp_remove_drhd
, NULL
);
1965 WARN_ON(dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1966 &dmar_release_one_atsr
, NULL
));
1967 WARN_ON(dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1968 &dmar_hp_release_drhd
, NULL
));
1970 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1971 &dmar_hp_add_drhd
, NULL
);
1977 static acpi_status
dmar_get_dsm_handle(acpi_handle handle
, u32 lvl
,
1978 void *context
, void **retval
)
1980 acpi_handle
*phdl
= retval
;
1982 if (dmar_detect_dsm(handle
, DMAR_DSM_FUNC_DRHD
)) {
1984 return AE_CTRL_TERMINATE
;
1990 static int dmar_device_hotplug(acpi_handle handle
, bool insert
)
1993 acpi_handle tmp
= NULL
;
1999 if (dmar_detect_dsm(handle
, DMAR_DSM_FUNC_DRHD
)) {
2002 status
= acpi_walk_namespace(ACPI_TYPE_DEVICE
, handle
,
2004 dmar_get_dsm_handle
,
2006 if (ACPI_FAILURE(status
)) {
2007 pr_warn("Failed to locate _DSM method.\n");
2014 down_write(&dmar_global_lock
);
2016 ret
= dmar_hotplug_insert(tmp
);
2018 ret
= dmar_hotplug_remove(tmp
);
2019 up_write(&dmar_global_lock
);
2024 int dmar_device_add(acpi_handle handle
)
2026 return dmar_device_hotplug(handle
, true);
2029 int dmar_device_remove(acpi_handle handle
)
2031 return dmar_device_hotplug(handle
, false);