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 extern const struct iommu_ops intel_iommu_ops
;
79 static void dmar_register_drhd_unit(struct dmar_drhd_unit
*drhd
)
82 * add INCLUDE_ALL at the tail, so scan the list will find it at
85 if (drhd
->include_all
)
86 list_add_tail_rcu(&drhd
->list
, &dmar_drhd_units
);
88 list_add_rcu(&drhd
->list
, &dmar_drhd_units
);
91 void *dmar_alloc_dev_scope(void *start
, void *end
, int *cnt
)
93 struct acpi_dmar_device_scope
*scope
;
98 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_NAMESPACE
||
99 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
||
100 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
102 else if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_IOAPIC
&&
103 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_HPET
) {
104 pr_warn("Unsupported device scope\n");
106 start
+= scope
->length
;
111 return kcalloc(*cnt
, sizeof(struct dmar_dev_scope
), GFP_KERNEL
);
114 void dmar_free_dev_scope(struct dmar_dev_scope
**devices
, int *cnt
)
117 struct device
*tmp_dev
;
119 if (*devices
&& *cnt
) {
120 for_each_active_dev_scope(*devices
, *cnt
, i
, tmp_dev
)
129 /* Optimize out kzalloc()/kfree() for normal cases */
130 static char dmar_pci_notify_info_buf
[64];
132 static struct dmar_pci_notify_info
*
133 dmar_alloc_pci_notify_info(struct pci_dev
*dev
, unsigned long event
)
138 struct dmar_pci_notify_info
*info
;
140 BUG_ON(dev
->is_virtfn
);
142 /* Only generate path[] for device addition event */
143 if (event
== BUS_NOTIFY_ADD_DEVICE
)
144 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
)
147 size
= sizeof(*info
) + level
* sizeof(struct acpi_dmar_pci_path
);
148 if (size
<= sizeof(dmar_pci_notify_info_buf
)) {
149 info
= (struct dmar_pci_notify_info
*)dmar_pci_notify_info_buf
;
151 info
= kzalloc(size
, GFP_KERNEL
);
153 pr_warn("Out of memory when allocating notify_info "
154 "for %s.\n", pci_name(dev
));
155 if (dmar_dev_scope_status
== 0)
156 dmar_dev_scope_status
= -ENOMEM
;
163 info
->seg
= pci_domain_nr(dev
->bus
);
165 if (event
== BUS_NOTIFY_ADD_DEVICE
) {
166 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
) {
168 info
->path
[level
].bus
= tmp
->bus
->number
;
169 info
->path
[level
].device
= PCI_SLOT(tmp
->devfn
);
170 info
->path
[level
].function
= PCI_FUNC(tmp
->devfn
);
171 if (pci_is_root_bus(tmp
->bus
))
172 info
->bus
= tmp
->bus
->number
;
179 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info
*info
)
181 if ((void *)info
!= dmar_pci_notify_info_buf
)
185 static bool dmar_match_pci_path(struct dmar_pci_notify_info
*info
, int bus
,
186 struct acpi_dmar_pci_path
*path
, int count
)
190 if (info
->bus
!= bus
)
192 if (info
->level
!= count
)
195 for (i
= 0; i
< count
; i
++) {
196 if (path
[i
].device
!= info
->path
[i
].device
||
197 path
[i
].function
!= info
->path
[i
].function
)
209 if (bus
== info
->path
[i
].bus
&&
210 path
[0].device
== info
->path
[i
].device
&&
211 path
[0].function
== info
->path
[i
].function
) {
212 pr_info(FW_BUG
"RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
213 bus
, path
[0].device
, path
[0].function
);
220 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
221 int dmar_insert_dev_scope(struct dmar_pci_notify_info
*info
,
222 void *start
, void*end
, u16 segment
,
223 struct dmar_dev_scope
*devices
,
227 struct device
*tmp
, *dev
= &info
->dev
->dev
;
228 struct acpi_dmar_device_scope
*scope
;
229 struct acpi_dmar_pci_path
*path
;
231 if (segment
!= info
->seg
)
234 for (; start
< end
; start
+= scope
->length
) {
236 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
237 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
240 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
241 level
= (scope
->length
- sizeof(*scope
)) / sizeof(*path
);
242 if (!dmar_match_pci_path(info
, scope
->bus
, path
, level
))
246 * We expect devices with endpoint scope to have normal PCI
247 * headers, and devices with bridge scope to have bridge PCI
248 * headers. However PCI NTB devices may be listed in the
249 * DMAR table with bridge scope, even though they have a
250 * normal PCI header. NTB devices are identified by class
251 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
252 * for this special case.
254 if ((scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
255 info
->dev
->hdr_type
!= PCI_HEADER_TYPE_NORMAL
) ||
256 (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
&&
257 (info
->dev
->hdr_type
== PCI_HEADER_TYPE_NORMAL
&&
258 info
->dev
->class >> 8 != PCI_CLASS_BRIDGE_OTHER
))) {
259 pr_warn("Device scope type does not match for %s\n",
260 pci_name(info
->dev
));
264 for_each_dev_scope(devices
, devices_cnt
, i
, tmp
)
266 devices
[i
].bus
= info
->dev
->bus
->number
;
267 devices
[i
].devfn
= info
->dev
->devfn
;
268 rcu_assign_pointer(devices
[i
].dev
,
272 BUG_ON(i
>= devices_cnt
);
278 int dmar_remove_dev_scope(struct dmar_pci_notify_info
*info
, u16 segment
,
279 struct dmar_dev_scope
*devices
, int count
)
284 if (info
->seg
!= segment
)
287 for_each_active_dev_scope(devices
, count
, index
, tmp
)
288 if (tmp
== &info
->dev
->dev
) {
289 RCU_INIT_POINTER(devices
[index
].dev
, NULL
);
298 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info
*info
)
301 struct dmar_drhd_unit
*dmaru
;
302 struct acpi_dmar_hardware_unit
*drhd
;
304 for_each_drhd_unit(dmaru
) {
305 if (dmaru
->include_all
)
308 drhd
= container_of(dmaru
->hdr
,
309 struct acpi_dmar_hardware_unit
, header
);
310 ret
= dmar_insert_dev_scope(info
, (void *)(drhd
+ 1),
311 ((void *)drhd
) + drhd
->header
.length
,
313 dmaru
->devices
, dmaru
->devices_cnt
);
318 ret
= dmar_iommu_notify_scope_dev(info
);
319 if (ret
< 0 && dmar_dev_scope_status
== 0)
320 dmar_dev_scope_status
= ret
;
325 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info
*info
)
327 struct dmar_drhd_unit
*dmaru
;
329 for_each_drhd_unit(dmaru
)
330 if (dmar_remove_dev_scope(info
, dmaru
->segment
,
331 dmaru
->devices
, dmaru
->devices_cnt
))
333 dmar_iommu_notify_scope_dev(info
);
336 static int dmar_pci_bus_notifier(struct notifier_block
*nb
,
337 unsigned long action
, void *data
)
339 struct pci_dev
*pdev
= to_pci_dev(data
);
340 struct dmar_pci_notify_info
*info
;
342 /* Only care about add/remove events for physical functions.
343 * For VFs we actually do the lookup based on the corresponding
344 * PF in device_to_iommu() anyway. */
347 if (action
!= BUS_NOTIFY_ADD_DEVICE
&&
348 action
!= BUS_NOTIFY_REMOVED_DEVICE
)
351 info
= dmar_alloc_pci_notify_info(pdev
, action
);
355 down_write(&dmar_global_lock
);
356 if (action
== BUS_NOTIFY_ADD_DEVICE
)
357 dmar_pci_bus_add_dev(info
);
358 else if (action
== BUS_NOTIFY_REMOVED_DEVICE
)
359 dmar_pci_bus_del_dev(info
);
360 up_write(&dmar_global_lock
);
362 dmar_free_pci_notify_info(info
);
367 static struct notifier_block dmar_pci_bus_nb
= {
368 .notifier_call
= dmar_pci_bus_notifier
,
372 static struct dmar_drhd_unit
*
373 dmar_find_dmaru(struct acpi_dmar_hardware_unit
*drhd
)
375 struct dmar_drhd_unit
*dmaru
;
377 list_for_each_entry_rcu(dmaru
, &dmar_drhd_units
, list
)
378 if (dmaru
->segment
== drhd
->segment
&&
379 dmaru
->reg_base_addr
== drhd
->address
)
386 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
387 * structure which uniquely represent one DMA remapping hardware unit
388 * present in the platform
390 static int dmar_parse_one_drhd(struct acpi_dmar_header
*header
, void *arg
)
392 struct acpi_dmar_hardware_unit
*drhd
;
393 struct dmar_drhd_unit
*dmaru
;
396 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
397 dmaru
= dmar_find_dmaru(drhd
);
401 dmaru
= kzalloc(sizeof(*dmaru
) + header
->length
, GFP_KERNEL
);
406 * If header is allocated from slab by ACPI _DSM method, we need to
407 * copy the content because the memory buffer will be freed on return.
409 dmaru
->hdr
= (void *)(dmaru
+ 1);
410 memcpy(dmaru
->hdr
, header
, header
->length
);
411 dmaru
->reg_base_addr
= drhd
->address
;
412 dmaru
->segment
= drhd
->segment
;
413 dmaru
->include_all
= drhd
->flags
& 0x1; /* BIT0: INCLUDE_ALL */
414 dmaru
->devices
= dmar_alloc_dev_scope((void *)(drhd
+ 1),
415 ((void *)drhd
) + drhd
->header
.length
,
416 &dmaru
->devices_cnt
);
417 if (dmaru
->devices_cnt
&& dmaru
->devices
== NULL
) {
422 ret
= alloc_iommu(dmaru
);
424 dmar_free_dev_scope(&dmaru
->devices
,
425 &dmaru
->devices_cnt
);
429 dmar_register_drhd_unit(dmaru
);
438 static void dmar_free_drhd(struct dmar_drhd_unit
*dmaru
)
440 if (dmaru
->devices
&& dmaru
->devices_cnt
)
441 dmar_free_dev_scope(&dmaru
->devices
, &dmaru
->devices_cnt
);
443 free_iommu(dmaru
->iommu
);
447 static int __init
dmar_parse_one_andd(struct acpi_dmar_header
*header
,
450 struct acpi_dmar_andd
*andd
= (void *)header
;
452 /* Check for NUL termination within the designated length */
453 if (strnlen(andd
->device_name
, header
->length
- 8) == header
->length
- 8) {
454 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND
,
455 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
456 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
457 dmi_get_system_info(DMI_BIOS_VENDOR
),
458 dmi_get_system_info(DMI_BIOS_VERSION
),
459 dmi_get_system_info(DMI_PRODUCT_VERSION
));
462 pr_info("ANDD device: %x name: %s\n", andd
->device_number
,
468 #ifdef CONFIG_ACPI_NUMA
469 static int dmar_parse_one_rhsa(struct acpi_dmar_header
*header
, void *arg
)
471 struct acpi_dmar_rhsa
*rhsa
;
472 struct dmar_drhd_unit
*drhd
;
474 rhsa
= (struct acpi_dmar_rhsa
*)header
;
475 for_each_drhd_unit(drhd
) {
476 if (drhd
->reg_base_addr
== rhsa
->base_address
) {
477 int node
= acpi_map_pxm_to_node(rhsa
->proximity_domain
);
479 if (!node_online(node
))
481 drhd
->iommu
->node
= node
;
486 1, TAINT_FIRMWARE_WORKAROUND
,
487 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
488 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
490 dmi_get_system_info(DMI_BIOS_VENDOR
),
491 dmi_get_system_info(DMI_BIOS_VERSION
),
492 dmi_get_system_info(DMI_PRODUCT_VERSION
));
497 #define dmar_parse_one_rhsa dmar_res_noop
501 dmar_table_print_dmar_entry(struct acpi_dmar_header
*header
)
503 struct acpi_dmar_hardware_unit
*drhd
;
504 struct acpi_dmar_reserved_memory
*rmrr
;
505 struct acpi_dmar_atsr
*atsr
;
506 struct acpi_dmar_rhsa
*rhsa
;
508 switch (header
->type
) {
509 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
510 drhd
= container_of(header
, struct acpi_dmar_hardware_unit
,
512 pr_info("DRHD base: %#016Lx flags: %#x\n",
513 (unsigned long long)drhd
->address
, drhd
->flags
);
515 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
516 rmrr
= container_of(header
, struct acpi_dmar_reserved_memory
,
518 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
519 (unsigned long long)rmrr
->base_address
,
520 (unsigned long long)rmrr
->end_address
);
522 case ACPI_DMAR_TYPE_ROOT_ATS
:
523 atsr
= container_of(header
, struct acpi_dmar_atsr
, header
);
524 pr_info("ATSR flags: %#x\n", atsr
->flags
);
526 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY
:
527 rhsa
= container_of(header
, struct acpi_dmar_rhsa
, header
);
528 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
529 (unsigned long long)rhsa
->base_address
,
530 rhsa
->proximity_domain
);
532 case ACPI_DMAR_TYPE_NAMESPACE
:
533 /* We don't print this here because we need to sanity-check
534 it first. So print it in dmar_parse_one_andd() instead. */
540 * dmar_table_detect - checks to see if the platform supports DMAR devices
542 static int __init
dmar_table_detect(void)
544 acpi_status status
= AE_OK
;
546 /* if we could find DMAR table, then there are DMAR devices */
547 status
= acpi_get_table(ACPI_SIG_DMAR
, 0, &dmar_tbl
);
549 if (ACPI_SUCCESS(status
) && !dmar_tbl
) {
550 pr_warn("Unable to map DMAR\n");
551 status
= AE_NOT_FOUND
;
554 return ACPI_SUCCESS(status
) ? 0 : -ENOENT
;
557 static int dmar_walk_remapping_entries(struct acpi_dmar_header
*start
,
558 size_t len
, struct dmar_res_callback
*cb
)
560 struct acpi_dmar_header
*iter
, *next
;
561 struct acpi_dmar_header
*end
= ((void *)start
) + len
;
563 for (iter
= start
; iter
< end
; iter
= next
) {
564 next
= (void *)iter
+ iter
->length
;
565 if (iter
->length
== 0) {
566 /* Avoid looping forever on bad ACPI tables */
567 pr_debug(FW_BUG
"Invalid 0-length structure\n");
569 } else if (next
> end
) {
570 /* Avoid passing table end */
571 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
]) {
585 ret
= cb
->cb
[iter
->type
](iter
, cb
->arg
[iter
->type
]);
588 } else if (!cb
->ignore_unhandled
) {
589 pr_warn("No handler for DMAR structure type %d\n",
598 static inline int dmar_walk_dmar_table(struct acpi_table_dmar
*dmar
,
599 struct dmar_res_callback
*cb
)
601 return dmar_walk_remapping_entries((void *)(dmar
+ 1),
602 dmar
->header
.length
- sizeof(*dmar
), cb
);
606 * parse_dmar_table - parses the DMA reporting table
609 parse_dmar_table(void)
611 struct acpi_table_dmar
*dmar
;
614 struct dmar_res_callback cb
= {
616 .ignore_unhandled
= true,
617 .arg
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &drhd_count
,
618 .cb
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &dmar_parse_one_drhd
,
619 .cb
[ACPI_DMAR_TYPE_RESERVED_MEMORY
] = &dmar_parse_one_rmrr
,
620 .cb
[ACPI_DMAR_TYPE_ROOT_ATS
] = &dmar_parse_one_atsr
,
621 .cb
[ACPI_DMAR_TYPE_HARDWARE_AFFINITY
] = &dmar_parse_one_rhsa
,
622 .cb
[ACPI_DMAR_TYPE_NAMESPACE
] = &dmar_parse_one_andd
,
626 * Do it again, earlier dmar_tbl mapping could be mapped with
632 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
633 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
635 dmar_tbl
= tboot_get_dmar_table(dmar_tbl
);
637 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
641 if (dmar
->width
< PAGE_SHIFT
- 1) {
642 pr_warn("Invalid DMAR haw\n");
646 pr_info("Host address width %d\n", dmar
->width
+ 1);
647 ret
= dmar_walk_dmar_table(dmar
, &cb
);
648 if (ret
== 0 && drhd_count
== 0)
649 pr_warn(FW_BUG
"No DRHD structure found in DMAR table\n");
654 static int dmar_pci_device_match(struct dmar_dev_scope devices
[],
655 int cnt
, struct pci_dev
*dev
)
661 for_each_active_dev_scope(devices
, cnt
, index
, tmp
)
662 if (dev_is_pci(tmp
) && dev
== to_pci_dev(tmp
))
665 /* Check our parent */
666 dev
= dev
->bus
->self
;
672 struct dmar_drhd_unit
*
673 dmar_find_matched_drhd_unit(struct pci_dev
*dev
)
675 struct dmar_drhd_unit
*dmaru
;
676 struct acpi_dmar_hardware_unit
*drhd
;
678 dev
= pci_physfn(dev
);
681 for_each_drhd_unit(dmaru
) {
682 drhd
= container_of(dmaru
->hdr
,
683 struct acpi_dmar_hardware_unit
,
686 if (dmaru
->include_all
&&
687 drhd
->segment
== pci_domain_nr(dev
->bus
))
690 if (dmar_pci_device_match(dmaru
->devices
,
691 dmaru
->devices_cnt
, dev
))
701 static void __init
dmar_acpi_insert_dev_scope(u8 device_number
,
702 struct acpi_device
*adev
)
704 struct dmar_drhd_unit
*dmaru
;
705 struct acpi_dmar_hardware_unit
*drhd
;
706 struct acpi_dmar_device_scope
*scope
;
709 struct acpi_dmar_pci_path
*path
;
711 for_each_drhd_unit(dmaru
) {
712 drhd
= container_of(dmaru
->hdr
,
713 struct acpi_dmar_hardware_unit
,
716 for (scope
= (void *)(drhd
+ 1);
717 (unsigned long)scope
< ((unsigned long)drhd
) + drhd
->header
.length
;
718 scope
= ((void *)scope
) + scope
->length
) {
719 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_NAMESPACE
)
721 if (scope
->enumeration_id
!= device_number
)
724 path
= (void *)(scope
+ 1);
725 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
726 dev_name(&adev
->dev
), dmaru
->reg_base_addr
,
727 scope
->bus
, path
->device
, path
->function
);
728 for_each_dev_scope(dmaru
->devices
, dmaru
->devices_cnt
, i
, tmp
)
730 dmaru
->devices
[i
].bus
= scope
->bus
;
731 dmaru
->devices
[i
].devfn
= PCI_DEVFN(path
->device
,
733 rcu_assign_pointer(dmaru
->devices
[i
].dev
,
734 get_device(&adev
->dev
));
737 BUG_ON(i
>= dmaru
->devices_cnt
);
740 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
741 device_number
, dev_name(&adev
->dev
));
744 static int __init
dmar_acpi_dev_scope_init(void)
746 struct acpi_dmar_andd
*andd
;
748 if (dmar_tbl
== NULL
)
751 for (andd
= (void *)dmar_tbl
+ sizeof(struct acpi_table_dmar
);
752 ((unsigned long)andd
) < ((unsigned long)dmar_tbl
) + dmar_tbl
->length
;
753 andd
= ((void *)andd
) + andd
->header
.length
) {
754 if (andd
->header
.type
== ACPI_DMAR_TYPE_NAMESPACE
) {
756 struct acpi_device
*adev
;
758 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT
,
761 pr_err("Failed to find handle for ACPI object %s\n",
765 if (acpi_bus_get_device(h
, &adev
)) {
766 pr_err("Failed to get device for ACPI object %s\n",
770 dmar_acpi_insert_dev_scope(andd
->device_number
, adev
);
776 int __init
dmar_dev_scope_init(void)
778 struct pci_dev
*dev
= NULL
;
779 struct dmar_pci_notify_info
*info
;
781 if (dmar_dev_scope_status
!= 1)
782 return dmar_dev_scope_status
;
784 if (list_empty(&dmar_drhd_units
)) {
785 dmar_dev_scope_status
= -ENODEV
;
787 dmar_dev_scope_status
= 0;
789 dmar_acpi_dev_scope_init();
791 for_each_pci_dev(dev
) {
795 info
= dmar_alloc_pci_notify_info(dev
,
796 BUS_NOTIFY_ADD_DEVICE
);
798 return dmar_dev_scope_status
;
800 dmar_pci_bus_add_dev(info
);
801 dmar_free_pci_notify_info(info
);
806 return dmar_dev_scope_status
;
809 void dmar_register_bus_notifier(void)
811 bus_register_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
815 int __init
dmar_table_init(void)
817 static int dmar_table_initialized
;
820 if (dmar_table_initialized
== 0) {
821 ret
= parse_dmar_table();
824 pr_info("Parse DMAR table failure.\n");
825 } else if (list_empty(&dmar_drhd_units
)) {
826 pr_info("No DMAR devices found\n");
831 dmar_table_initialized
= ret
;
833 dmar_table_initialized
= 1;
836 return dmar_table_initialized
< 0 ? dmar_table_initialized
: 0;
839 static void warn_invalid_dmar(u64 addr
, const char *message
)
842 1, TAINT_FIRMWARE_WORKAROUND
,
843 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
844 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
846 dmi_get_system_info(DMI_BIOS_VENDOR
),
847 dmi_get_system_info(DMI_BIOS_VERSION
),
848 dmi_get_system_info(DMI_PRODUCT_VERSION
));
852 dmar_validate_one_drhd(struct acpi_dmar_header
*entry
, void *arg
)
854 struct acpi_dmar_hardware_unit
*drhd
;
858 drhd
= (void *)entry
;
859 if (!drhd
->address
) {
860 warn_invalid_dmar(0, "");
865 addr
= ioremap(drhd
->address
, VTD_PAGE_SIZE
);
867 addr
= early_ioremap(drhd
->address
, VTD_PAGE_SIZE
);
869 pr_warn("Can't validate DRHD address: %llx\n", drhd
->address
);
873 cap
= dmar_readq(addr
+ DMAR_CAP_REG
);
874 ecap
= dmar_readq(addr
+ DMAR_ECAP_REG
);
879 early_iounmap(addr
, VTD_PAGE_SIZE
);
881 if (cap
== (uint64_t)-1 && ecap
== (uint64_t)-1) {
882 warn_invalid_dmar(drhd
->address
, " returns all ones");
889 int __init
detect_intel_iommu(void)
892 struct dmar_res_callback validate_drhd_cb
= {
893 .cb
[ACPI_DMAR_TYPE_HARDWARE_UNIT
] = &dmar_validate_one_drhd
,
894 .ignore_unhandled
= true,
897 down_write(&dmar_global_lock
);
898 ret
= dmar_table_detect();
900 ret
= dmar_walk_dmar_table((struct acpi_table_dmar
*)dmar_tbl
,
902 if (!ret
&& !no_iommu
&& !iommu_detected
&& !dmar_disabled
) {
904 /* Make sure ACS will be enabled */
910 x86_init
.iommu
.iommu_init
= intel_iommu_init
;
914 acpi_put_table(dmar_tbl
);
917 up_write(&dmar_global_lock
);
919 return ret
? ret
: 1;
922 static void unmap_iommu(struct intel_iommu
*iommu
)
925 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
929 * map_iommu: map the iommu's registers
930 * @iommu: the iommu to map
931 * @phys_addr: the physical address of the base resgister
933 * Memory map the iommu's registers. Start w/ a single page, and
934 * possibly expand if that turns out to be insufficent.
936 static int map_iommu(struct intel_iommu
*iommu
, u64 phys_addr
)
940 iommu
->reg_phys
= phys_addr
;
941 iommu
->reg_size
= VTD_PAGE_SIZE
;
943 if (!request_mem_region(iommu
->reg_phys
, iommu
->reg_size
, iommu
->name
)) {
944 pr_err("Can't reserve memory\n");
949 iommu
->reg
= ioremap(iommu
->reg_phys
, iommu
->reg_size
);
951 pr_err("Can't map the region\n");
956 iommu
->cap
= dmar_readq(iommu
->reg
+ DMAR_CAP_REG
);
957 iommu
->ecap
= dmar_readq(iommu
->reg
+ DMAR_ECAP_REG
);
959 if (iommu
->cap
== (uint64_t)-1 && iommu
->ecap
== (uint64_t)-1) {
961 warn_invalid_dmar(phys_addr
, " returns all ones");
965 /* the registers might be more than one page */
966 map_size
= max_t(int, ecap_max_iotlb_offset(iommu
->ecap
),
967 cap_max_fault_reg_offset(iommu
->cap
));
968 map_size
= VTD_PAGE_ALIGN(map_size
);
969 if (map_size
> iommu
->reg_size
) {
971 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
972 iommu
->reg_size
= map_size
;
973 if (!request_mem_region(iommu
->reg_phys
, iommu
->reg_size
,
975 pr_err("Can't reserve memory\n");
979 iommu
->reg
= ioremap(iommu
->reg_phys
, iommu
->reg_size
);
981 pr_err("Can't map the region\n");
992 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
997 static int dmar_alloc_seq_id(struct intel_iommu
*iommu
)
999 iommu
->seq_id
= find_first_zero_bit(dmar_seq_ids
,
1000 DMAR_UNITS_SUPPORTED
);
1001 if (iommu
->seq_id
>= DMAR_UNITS_SUPPORTED
) {
1004 set_bit(iommu
->seq_id
, dmar_seq_ids
);
1005 sprintf(iommu
->name
, "dmar%d", iommu
->seq_id
);
1008 return iommu
->seq_id
;
1011 static void dmar_free_seq_id(struct intel_iommu
*iommu
)
1013 if (iommu
->seq_id
>= 0) {
1014 clear_bit(iommu
->seq_id
, dmar_seq_ids
);
1019 static int alloc_iommu(struct dmar_drhd_unit
*drhd
)
1021 struct intel_iommu
*iommu
;
1027 if (!drhd
->reg_base_addr
) {
1028 warn_invalid_dmar(0, "");
1032 iommu
= kzalloc(sizeof(*iommu
), GFP_KERNEL
);
1036 if (dmar_alloc_seq_id(iommu
) < 0) {
1037 pr_err("Failed to allocate seq_id\n");
1042 err
= map_iommu(iommu
, drhd
->reg_base_addr
);
1044 pr_err("Failed to map %s\n", iommu
->name
);
1045 goto error_free_seq_id
;
1049 agaw
= iommu_calculate_agaw(iommu
);
1051 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1055 msagaw
= iommu_calculate_max_sagaw(iommu
);
1057 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1062 iommu
->msagaw
= msagaw
;
1063 iommu
->segment
= drhd
->segment
;
1067 ver
= readl(iommu
->reg
+ DMAR_VER_REG
);
1068 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1070 (unsigned long long)drhd
->reg_base_addr
,
1071 DMAR_VER_MAJOR(ver
), DMAR_VER_MINOR(ver
),
1072 (unsigned long long)iommu
->cap
,
1073 (unsigned long long)iommu
->ecap
);
1075 /* Reflect status in gcmd */
1076 sts
= readl(iommu
->reg
+ DMAR_GSTS_REG
);
1077 if (sts
& DMA_GSTS_IRES
)
1078 iommu
->gcmd
|= DMA_GCMD_IRE
;
1079 if (sts
& DMA_GSTS_TES
)
1080 iommu
->gcmd
|= DMA_GCMD_TE
;
1081 if (sts
& DMA_GSTS_QIES
)
1082 iommu
->gcmd
|= DMA_GCMD_QIE
;
1084 raw_spin_lock_init(&iommu
->register_lock
);
1086 if (intel_iommu_enabled
) {
1087 err
= iommu_device_sysfs_add(&iommu
->iommu
, NULL
,
1093 iommu_device_set_ops(&iommu
->iommu
, &intel_iommu_ops
);
1095 err
= iommu_device_register(&iommu
->iommu
);
1100 drhd
->iommu
= iommu
;
1107 dmar_free_seq_id(iommu
);
1113 static void free_iommu(struct intel_iommu
*iommu
)
1115 if (intel_iommu_enabled
) {
1116 iommu_device_unregister(&iommu
->iommu
);
1117 iommu_device_sysfs_remove(&iommu
->iommu
);
1121 if (iommu
->pr_irq
) {
1122 free_irq(iommu
->pr_irq
, iommu
);
1123 dmar_free_hwirq(iommu
->pr_irq
);
1126 free_irq(iommu
->irq
, iommu
);
1127 dmar_free_hwirq(iommu
->irq
);
1132 free_page((unsigned long)iommu
->qi
->desc
);
1133 kfree(iommu
->qi
->desc_status
);
1140 dmar_free_seq_id(iommu
);
1145 * Reclaim all the submitted descriptors which have completed its work.
1147 static inline void reclaim_free_desc(struct q_inval
*qi
)
1149 while (qi
->desc_status
[qi
->free_tail
] == QI_DONE
||
1150 qi
->desc_status
[qi
->free_tail
] == QI_ABORT
) {
1151 qi
->desc_status
[qi
->free_tail
] = QI_FREE
;
1152 qi
->free_tail
= (qi
->free_tail
+ 1) % QI_LENGTH
;
1157 static int qi_check_fault(struct intel_iommu
*iommu
, int index
)
1161 struct q_inval
*qi
= iommu
->qi
;
1162 int wait_index
= (index
+ 1) % QI_LENGTH
;
1164 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
1167 fault
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1170 * If IQE happens, the head points to the descriptor associated
1171 * with the error. No new descriptors are fetched until the IQE
1174 if (fault
& DMA_FSTS_IQE
) {
1175 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
1176 if ((head
>> DMAR_IQ_SHIFT
) == index
) {
1177 pr_err("VT-d detected invalid descriptor: "
1178 "low=%llx, high=%llx\n",
1179 (unsigned long long)qi
->desc
[index
].low
,
1180 (unsigned long long)qi
->desc
[index
].high
);
1181 memcpy(&qi
->desc
[index
], &qi
->desc
[wait_index
],
1182 sizeof(struct qi_desc
));
1183 writel(DMA_FSTS_IQE
, iommu
->reg
+ DMAR_FSTS_REG
);
1189 * If ITE happens, all pending wait_desc commands are aborted.
1190 * No new descriptors are fetched until the ITE is cleared.
1192 if (fault
& DMA_FSTS_ITE
) {
1193 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
1194 head
= ((head
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
1196 tail
= readl(iommu
->reg
+ DMAR_IQT_REG
);
1197 tail
= ((tail
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
1199 writel(DMA_FSTS_ITE
, iommu
->reg
+ DMAR_FSTS_REG
);
1202 if (qi
->desc_status
[head
] == QI_IN_USE
)
1203 qi
->desc_status
[head
] = QI_ABORT
;
1204 head
= (head
- 2 + QI_LENGTH
) % QI_LENGTH
;
1205 } while (head
!= tail
);
1207 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
1211 if (fault
& DMA_FSTS_ICE
)
1212 writel(DMA_FSTS_ICE
, iommu
->reg
+ DMAR_FSTS_REG
);
1218 * Submit the queued invalidation descriptor to the remapping
1219 * hardware unit and wait for its completion.
1221 int qi_submit_sync(struct qi_desc
*desc
, struct intel_iommu
*iommu
)
1224 struct q_inval
*qi
= iommu
->qi
;
1225 struct qi_desc
*hw
, wait_desc
;
1226 int wait_index
, index
;
1227 unsigned long flags
;
1237 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1238 while (qi
->free_cnt
< 3) {
1239 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1241 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1244 index
= qi
->free_head
;
1245 wait_index
= (index
+ 1) % QI_LENGTH
;
1247 qi
->desc_status
[index
] = qi
->desc_status
[wait_index
] = QI_IN_USE
;
1251 wait_desc
.low
= QI_IWD_STATUS_DATA(QI_DONE
) |
1252 QI_IWD_STATUS_WRITE
| QI_IWD_TYPE
;
1253 wait_desc
.high
= virt_to_phys(&qi
->desc_status
[wait_index
]);
1255 hw
[wait_index
] = wait_desc
;
1257 qi
->free_head
= (qi
->free_head
+ 2) % QI_LENGTH
;
1261 * update the HW tail register indicating the presence of
1264 writel(qi
->free_head
<< DMAR_IQ_SHIFT
, iommu
->reg
+ DMAR_IQT_REG
);
1266 while (qi
->desc_status
[wait_index
] != QI_DONE
) {
1268 * We will leave the interrupts disabled, to prevent interrupt
1269 * context to queue another cmd while a cmd is already submitted
1270 * and waiting for completion on this cpu. This is to avoid
1271 * a deadlock where the interrupt context can wait indefinitely
1272 * for free slots in the queue.
1274 rc
= qi_check_fault(iommu
, index
);
1278 raw_spin_unlock(&qi
->q_lock
);
1280 raw_spin_lock(&qi
->q_lock
);
1283 qi
->desc_status
[index
] = QI_DONE
;
1285 reclaim_free_desc(qi
);
1286 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1295 * Flush the global interrupt entry cache.
1297 void qi_global_iec(struct intel_iommu
*iommu
)
1299 struct qi_desc desc
;
1301 desc
.low
= QI_IEC_TYPE
;
1304 /* should never fail */
1305 qi_submit_sync(&desc
, iommu
);
1308 void qi_flush_context(struct intel_iommu
*iommu
, u16 did
, u16 sid
, u8 fm
,
1311 struct qi_desc desc
;
1313 desc
.low
= QI_CC_FM(fm
) | QI_CC_SID(sid
) | QI_CC_DID(did
)
1314 | QI_CC_GRAN(type
) | QI_CC_TYPE
;
1317 qi_submit_sync(&desc
, iommu
);
1320 void qi_flush_iotlb(struct intel_iommu
*iommu
, u16 did
, u64 addr
,
1321 unsigned int size_order
, u64 type
)
1325 struct qi_desc desc
;
1328 if (cap_write_drain(iommu
->cap
))
1331 if (cap_read_drain(iommu
->cap
))
1334 desc
.low
= QI_IOTLB_DID(did
) | QI_IOTLB_DR(dr
) | QI_IOTLB_DW(dw
)
1335 | QI_IOTLB_GRAN(type
) | QI_IOTLB_TYPE
;
1336 desc
.high
= QI_IOTLB_ADDR(addr
) | QI_IOTLB_IH(ih
)
1337 | QI_IOTLB_AM(size_order
);
1339 qi_submit_sync(&desc
, iommu
);
1342 void qi_flush_dev_iotlb(struct intel_iommu
*iommu
, u16 sid
, u16 qdep
,
1343 u64 addr
, unsigned mask
)
1345 struct qi_desc desc
;
1348 BUG_ON(addr
& ((1 << (VTD_PAGE_SHIFT
+ mask
)) - 1));
1349 addr
|= (1ULL << (VTD_PAGE_SHIFT
+ mask
- 1)) - 1;
1350 desc
.high
= QI_DEV_IOTLB_ADDR(addr
) | QI_DEV_IOTLB_SIZE
;
1352 desc
.high
= QI_DEV_IOTLB_ADDR(addr
);
1354 if (qdep
>= QI_DEV_IOTLB_MAX_INVS
)
1357 desc
.low
= QI_DEV_IOTLB_SID(sid
) | QI_DEV_IOTLB_QDEP(qdep
) |
1360 qi_submit_sync(&desc
, iommu
);
1364 * Disable Queued Invalidation interface.
1366 void dmar_disable_qi(struct intel_iommu
*iommu
)
1368 unsigned long flags
;
1370 cycles_t start_time
= get_cycles();
1372 if (!ecap_qis(iommu
->ecap
))
1375 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1377 sts
= readl(iommu
->reg
+ DMAR_GSTS_REG
);
1378 if (!(sts
& DMA_GSTS_QIES
))
1382 * Give a chance to HW to complete the pending invalidation requests.
1384 while ((readl(iommu
->reg
+ DMAR_IQT_REG
) !=
1385 readl(iommu
->reg
+ DMAR_IQH_REG
)) &&
1386 (DMAR_OPERATION_TIMEOUT
> (get_cycles() - start_time
)))
1389 iommu
->gcmd
&= ~DMA_GCMD_QIE
;
1390 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1392 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
,
1393 !(sts
& DMA_GSTS_QIES
), sts
);
1395 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1399 * Enable queued invalidation.
1401 static void __dmar_enable_qi(struct intel_iommu
*iommu
)
1404 unsigned long flags
;
1405 struct q_inval
*qi
= iommu
->qi
;
1407 qi
->free_head
= qi
->free_tail
= 0;
1408 qi
->free_cnt
= QI_LENGTH
;
1410 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1412 /* write zero to the tail reg */
1413 writel(0, iommu
->reg
+ DMAR_IQT_REG
);
1415 dmar_writeq(iommu
->reg
+ DMAR_IQA_REG
, virt_to_phys(qi
->desc
));
1417 iommu
->gcmd
|= DMA_GCMD_QIE
;
1418 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1420 /* Make sure hardware complete it */
1421 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
, (sts
& DMA_GSTS_QIES
), sts
);
1423 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1427 * Enable Queued Invalidation interface. This is a must to support
1428 * interrupt-remapping. Also used by DMA-remapping, which replaces
1429 * register based IOTLB invalidation.
1431 int dmar_enable_qi(struct intel_iommu
*iommu
)
1434 struct page
*desc_page
;
1436 if (!ecap_qis(iommu
->ecap
))
1440 * queued invalidation is already setup and enabled.
1445 iommu
->qi
= kmalloc(sizeof(*qi
), GFP_ATOMIC
);
1452 desc_page
= alloc_pages_node(iommu
->node
, GFP_ATOMIC
| __GFP_ZERO
, 0);
1459 qi
->desc
= page_address(desc_page
);
1461 qi
->desc_status
= kzalloc(QI_LENGTH
* sizeof(int), GFP_ATOMIC
);
1462 if (!qi
->desc_status
) {
1463 free_page((unsigned long) qi
->desc
);
1469 raw_spin_lock_init(&qi
->q_lock
);
1471 __dmar_enable_qi(iommu
);
1476 /* iommu interrupt handling. Most stuff are MSI-like. */
1484 static const char *dma_remap_fault_reasons
[] =
1487 "Present bit in root entry is clear",
1488 "Present bit in context entry is clear",
1489 "Invalid context entry",
1490 "Access beyond MGAW",
1491 "PTE Write access is not set",
1492 "PTE Read access is not set",
1493 "Next page table ptr is invalid",
1494 "Root table address invalid",
1495 "Context table ptr is invalid",
1496 "non-zero reserved fields in RTP",
1497 "non-zero reserved fields in CTP",
1498 "non-zero reserved fields in PTE",
1499 "PCE for translation request specifies blocking",
1502 static const char *irq_remap_fault_reasons
[] =
1504 "Detected reserved fields in the decoded interrupt-remapped request",
1505 "Interrupt index exceeded the interrupt-remapping table size",
1506 "Present field in the IRTE entry is clear",
1507 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1508 "Detected reserved fields in the IRTE entry",
1509 "Blocked a compatibility format interrupt request",
1510 "Blocked an interrupt request due to source-id verification failure",
1513 static const char *dmar_get_fault_reason(u8 fault_reason
, int *fault_type
)
1515 if (fault_reason
>= 0x20 && (fault_reason
- 0x20 <
1516 ARRAY_SIZE(irq_remap_fault_reasons
))) {
1517 *fault_type
= INTR_REMAP
;
1518 return irq_remap_fault_reasons
[fault_reason
- 0x20];
1519 } else if (fault_reason
< ARRAY_SIZE(dma_remap_fault_reasons
)) {
1520 *fault_type
= DMA_REMAP
;
1521 return dma_remap_fault_reasons
[fault_reason
];
1523 *fault_type
= UNKNOWN
;
1529 static inline int dmar_msi_reg(struct intel_iommu
*iommu
, int irq
)
1531 if (iommu
->irq
== irq
)
1532 return DMAR_FECTL_REG
;
1533 else if (iommu
->pr_irq
== irq
)
1534 return DMAR_PECTL_REG
;
1539 void dmar_msi_unmask(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(0, 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_mask(struct irq_data
*data
)
1555 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1556 int reg
= dmar_msi_reg(iommu
, data
->irq
);
1560 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1561 writel(DMA_FECTL_IM
, iommu
->reg
+ reg
);
1562 /* Read a reg to force flush the post write */
1563 readl(iommu
->reg
+ reg
);
1564 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1567 void dmar_msi_write(int irq
, struct msi_msg
*msg
)
1569 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1570 int reg
= dmar_msi_reg(iommu
, irq
);
1573 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1574 writel(msg
->data
, iommu
->reg
+ reg
+ 4);
1575 writel(msg
->address_lo
, iommu
->reg
+ reg
+ 8);
1576 writel(msg
->address_hi
, iommu
->reg
+ reg
+ 12);
1577 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1580 void dmar_msi_read(int irq
, struct msi_msg
*msg
)
1582 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1583 int reg
= dmar_msi_reg(iommu
, irq
);
1586 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1587 msg
->data
= readl(iommu
->reg
+ reg
+ 4);
1588 msg
->address_lo
= readl(iommu
->reg
+ reg
+ 8);
1589 msg
->address_hi
= readl(iommu
->reg
+ reg
+ 12);
1590 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1593 static int dmar_fault_do_one(struct intel_iommu
*iommu
, int type
,
1594 u8 fault_reason
, u16 source_id
, unsigned long long addr
)
1599 reason
= dmar_get_fault_reason(fault_reason
, &fault_type
);
1601 if (fault_type
== INTR_REMAP
)
1602 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1603 source_id
>> 8, PCI_SLOT(source_id
& 0xFF),
1604 PCI_FUNC(source_id
& 0xFF), addr
>> 48,
1605 fault_reason
, reason
);
1607 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1608 type
? "DMA Read" : "DMA Write",
1609 source_id
>> 8, PCI_SLOT(source_id
& 0xFF),
1610 PCI_FUNC(source_id
& 0xFF), addr
, fault_reason
, reason
);
1614 #define PRIMARY_FAULT_REG_LEN (16)
1615 irqreturn_t
dmar_fault(int irq
, void *dev_id
)
1617 struct intel_iommu
*iommu
= dev_id
;
1618 int reg
, fault_index
;
1622 static DEFINE_RATELIMIT_STATE(rs
,
1623 DEFAULT_RATELIMIT_INTERVAL
,
1624 DEFAULT_RATELIMIT_BURST
);
1626 /* Disable printing, simply clear the fault when ratelimited */
1627 ratelimited
= !__ratelimit(&rs
);
1629 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1630 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1631 if (fault_status
&& !ratelimited
)
1632 pr_err("DRHD: handling fault status reg %x\n", fault_status
);
1634 /* TBD: ignore advanced fault log currently */
1635 if (!(fault_status
& DMA_FSTS_PPF
))
1638 fault_index
= dma_fsts_fault_record_index(fault_status
);
1639 reg
= cap_fault_reg_offset(iommu
->cap
);
1647 /* highest 32 bits */
1648 data
= readl(iommu
->reg
+ reg
+
1649 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1650 if (!(data
& DMA_FRCD_F
))
1654 fault_reason
= dma_frcd_fault_reason(data
);
1655 type
= dma_frcd_type(data
);
1657 data
= readl(iommu
->reg
+ reg
+
1658 fault_index
* PRIMARY_FAULT_REG_LEN
+ 8);
1659 source_id
= dma_frcd_source_id(data
);
1661 guest_addr
= dmar_readq(iommu
->reg
+ reg
+
1662 fault_index
* PRIMARY_FAULT_REG_LEN
);
1663 guest_addr
= dma_frcd_page_addr(guest_addr
);
1666 /* clear the fault */
1667 writel(DMA_FRCD_F
, iommu
->reg
+ reg
+
1668 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1670 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1673 dmar_fault_do_one(iommu
, type
, fault_reason
,
1674 source_id
, guest_addr
);
1677 if (fault_index
>= cap_num_fault_regs(iommu
->cap
))
1679 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1682 writel(DMA_FSTS_PFO
| DMA_FSTS_PPF
| DMA_FSTS_PRO
,
1683 iommu
->reg
+ DMAR_FSTS_REG
);
1686 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1690 int dmar_set_interrupt(struct intel_iommu
*iommu
)
1695 * Check if the fault interrupt is already initialized.
1700 irq
= dmar_alloc_hwirq(iommu
->seq_id
, iommu
->node
, iommu
);
1704 pr_err("No free IRQ vectors\n");
1708 ret
= request_irq(irq
, dmar_fault
, IRQF_NO_THREAD
, iommu
->name
, iommu
);
1710 pr_err("Can't request irq\n");
1714 int __init
enable_drhd_fault_handling(void)
1716 struct dmar_drhd_unit
*drhd
;
1717 struct intel_iommu
*iommu
;
1720 * Enable fault control interrupt.
1722 for_each_iommu(iommu
, drhd
) {
1724 int ret
= dmar_set_interrupt(iommu
);
1727 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1728 (unsigned long long)drhd
->reg_base_addr
, ret
);
1733 * Clear any previous faults.
1735 dmar_fault(iommu
->irq
, iommu
);
1736 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1737 writel(fault_status
, iommu
->reg
+ DMAR_FSTS_REG
);
1744 * Re-enable Queued Invalidation interface.
1746 int dmar_reenable_qi(struct intel_iommu
*iommu
)
1748 if (!ecap_qis(iommu
->ecap
))
1755 * First disable queued invalidation.
1757 dmar_disable_qi(iommu
);
1759 * Then enable queued invalidation again. Since there is no pending
1760 * invalidation requests now, it's safe to re-enable queued
1763 __dmar_enable_qi(iommu
);
1769 * Check interrupt remapping support in DMAR table description.
1771 int __init
dmar_ir_support(void)
1773 struct acpi_table_dmar
*dmar
;
1774 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
1777 return dmar
->flags
& 0x1;
1780 /* Check whether DMAR units are in use */
1781 static inline bool dmar_in_use(void)
1783 return irq_remapping_enabled
|| intel_iommu_enabled
;
1786 static int __init
dmar_free_unused_resources(void)
1788 struct dmar_drhd_unit
*dmaru
, *dmaru_n
;
1793 if (dmar_dev_scope_status
!= 1 && !list_empty(&dmar_drhd_units
))
1794 bus_unregister_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
1796 down_write(&dmar_global_lock
);
1797 list_for_each_entry_safe(dmaru
, dmaru_n
, &dmar_drhd_units
, list
) {
1798 list_del(&dmaru
->list
);
1799 dmar_free_drhd(dmaru
);
1801 up_write(&dmar_global_lock
);
1806 late_initcall(dmar_free_unused_resources
);
1807 IOMMU_INIT_POST(detect_intel_iommu
);
1810 * DMAR Hotplug Support
1811 * For more details, please refer to Intel(R) Virtualization Technology
1812 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1813 * "Remapping Hardware Unit Hot Plug".
1815 static guid_t dmar_hp_guid
=
1816 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
1817 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
1820 * Currently there's only one revision and BIOS will not check the revision id,
1821 * so use 0 for safety.
1823 #define DMAR_DSM_REV_ID 0
1824 #define DMAR_DSM_FUNC_DRHD 1
1825 #define DMAR_DSM_FUNC_ATSR 2
1826 #define DMAR_DSM_FUNC_RHSA 3
1828 static inline bool dmar_detect_dsm(acpi_handle handle
, int func
)
1830 return acpi_check_dsm(handle
, &dmar_hp_guid
, DMAR_DSM_REV_ID
, 1 << func
);
1833 static int dmar_walk_dsm_resource(acpi_handle handle
, int func
,
1834 dmar_res_handler_t handler
, void *arg
)
1837 union acpi_object
*obj
;
1838 struct acpi_dmar_header
*start
;
1839 struct dmar_res_callback callback
;
1840 static int res_type
[] = {
1841 [DMAR_DSM_FUNC_DRHD
] = ACPI_DMAR_TYPE_HARDWARE_UNIT
,
1842 [DMAR_DSM_FUNC_ATSR
] = ACPI_DMAR_TYPE_ROOT_ATS
,
1843 [DMAR_DSM_FUNC_RHSA
] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY
,
1846 if (!dmar_detect_dsm(handle
, func
))
1849 obj
= acpi_evaluate_dsm_typed(handle
, &dmar_hp_guid
, DMAR_DSM_REV_ID
,
1850 func
, NULL
, ACPI_TYPE_BUFFER
);
1854 memset(&callback
, 0, sizeof(callback
));
1855 callback
.cb
[res_type
[func
]] = handler
;
1856 callback
.arg
[res_type
[func
]] = arg
;
1857 start
= (struct acpi_dmar_header
*)obj
->buffer
.pointer
;
1858 ret
= dmar_walk_remapping_entries(start
, obj
->buffer
.length
, &callback
);
1865 static int dmar_hp_add_drhd(struct acpi_dmar_header
*header
, void *arg
)
1868 struct dmar_drhd_unit
*dmaru
;
1870 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1874 ret
= dmar_ir_hotplug(dmaru
, true);
1876 ret
= dmar_iommu_hotplug(dmaru
, true);
1881 static int dmar_hp_remove_drhd(struct acpi_dmar_header
*header
, void *arg
)
1885 struct dmar_drhd_unit
*dmaru
;
1887 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1892 * All PCI devices managed by this unit should have been destroyed.
1894 if (!dmaru
->include_all
&& dmaru
->devices
&& dmaru
->devices_cnt
) {
1895 for_each_active_dev_scope(dmaru
->devices
,
1896 dmaru
->devices_cnt
, i
, dev
)
1900 ret
= dmar_ir_hotplug(dmaru
, false);
1902 ret
= dmar_iommu_hotplug(dmaru
, false);
1907 static int dmar_hp_release_drhd(struct acpi_dmar_header
*header
, void *arg
)
1909 struct dmar_drhd_unit
*dmaru
;
1911 dmaru
= dmar_find_dmaru((struct acpi_dmar_hardware_unit
*)header
);
1913 list_del_rcu(&dmaru
->list
);
1915 dmar_free_drhd(dmaru
);
1921 static int dmar_hotplug_insert(acpi_handle handle
)
1926 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1927 &dmar_validate_one_drhd
, (void *)1);
1931 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1932 &dmar_parse_one_drhd
, (void *)&drhd_count
);
1933 if (ret
== 0 && drhd_count
== 0) {
1934 pr_warn(FW_BUG
"No DRHD structures in buffer returned by _DSM method\n");
1940 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_RHSA
,
1941 &dmar_parse_one_rhsa
, NULL
);
1945 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1946 &dmar_parse_one_atsr
, NULL
);
1950 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1951 &dmar_hp_add_drhd
, NULL
);
1955 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1956 &dmar_hp_remove_drhd
, NULL
);
1958 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1959 &dmar_release_one_atsr
, NULL
);
1961 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1962 &dmar_hp_release_drhd
, NULL
);
1967 static int dmar_hotplug_remove(acpi_handle handle
)
1971 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1972 &dmar_check_one_atsr
, NULL
);
1976 ret
= dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1977 &dmar_hp_remove_drhd
, NULL
);
1979 WARN_ON(dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_ATSR
,
1980 &dmar_release_one_atsr
, NULL
));
1981 WARN_ON(dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1982 &dmar_hp_release_drhd
, NULL
));
1984 dmar_walk_dsm_resource(handle
, DMAR_DSM_FUNC_DRHD
,
1985 &dmar_hp_add_drhd
, NULL
);
1991 static acpi_status
dmar_get_dsm_handle(acpi_handle handle
, u32 lvl
,
1992 void *context
, void **retval
)
1994 acpi_handle
*phdl
= retval
;
1996 if (dmar_detect_dsm(handle
, DMAR_DSM_FUNC_DRHD
)) {
1998 return AE_CTRL_TERMINATE
;
2004 static int dmar_device_hotplug(acpi_handle handle
, bool insert
)
2007 acpi_handle tmp
= NULL
;
2013 if (dmar_detect_dsm(handle
, DMAR_DSM_FUNC_DRHD
)) {
2016 status
= acpi_walk_namespace(ACPI_TYPE_DEVICE
, handle
,
2018 dmar_get_dsm_handle
,
2020 if (ACPI_FAILURE(status
)) {
2021 pr_warn("Failed to locate _DSM method.\n");
2028 down_write(&dmar_global_lock
);
2030 ret
= dmar_hotplug_insert(tmp
);
2032 ret
= dmar_hotplug_remove(tmp
);
2033 up_write(&dmar_global_lock
);
2038 int dmar_device_add(acpi_handle handle
)
2040 return dmar_device_hotplug(handle
, true);
2043 int dmar_device_remove(acpi_handle handle
)
2045 return dmar_device_hotplug(handle
, false);