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) KBUILD_MODNAME ": " fmt /* has to precede printk.h */
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 <asm/irq_remapping.h>
42 #include <asm/iommu_table.h>
44 #include "irq_remapping.h"
48 * 1) The hotplug framework guarentees that DMAR unit will be hot-added
49 * before IO devices managed by that unit.
50 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
51 * after IO devices managed by that unit.
52 * 3) Hotplug events are rare.
54 * Locking rules for DMA and interrupt remapping related global data structures:
55 * 1) Use dmar_global_lock in process context
56 * 2) Use RCU in interrupt context
58 DECLARE_RWSEM(dmar_global_lock
);
59 LIST_HEAD(dmar_drhd_units
);
61 struct acpi_table_header
* __initdata dmar_tbl
;
62 static acpi_size dmar_tbl_size
;
63 static int dmar_dev_scope_status
= 1;
65 static int alloc_iommu(struct dmar_drhd_unit
*drhd
);
66 static void free_iommu(struct intel_iommu
*iommu
);
68 static void __init
dmar_register_drhd_unit(struct dmar_drhd_unit
*drhd
)
71 * add INCLUDE_ALL at the tail, so scan the list will find it at
74 if (drhd
->include_all
)
75 list_add_tail_rcu(&drhd
->list
, &dmar_drhd_units
);
77 list_add_rcu(&drhd
->list
, &dmar_drhd_units
);
80 void *dmar_alloc_dev_scope(void *start
, void *end
, int *cnt
)
82 struct acpi_dmar_device_scope
*scope
;
87 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ACPI
||
88 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
||
89 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
91 else if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_IOAPIC
&&
92 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_HPET
) {
93 pr_warn("Unsupported device scope\n");
95 start
+= scope
->length
;
100 return kcalloc(*cnt
, sizeof(struct pci_dev
*), GFP_KERNEL
);
103 void dmar_free_dev_scope(struct pci_dev __rcu
***devices
, int *cnt
)
106 struct pci_dev
*tmp_dev
;
108 if (*devices
&& *cnt
) {
109 for_each_active_dev_scope(*devices
, *cnt
, i
, tmp_dev
)
110 pci_dev_put(tmp_dev
);
118 /* Optimize out kzalloc()/kfree() for normal cases */
119 static char dmar_pci_notify_info_buf
[64];
121 static struct dmar_pci_notify_info
*
122 dmar_alloc_pci_notify_info(struct pci_dev
*dev
, unsigned long event
)
127 struct dmar_pci_notify_info
*info
;
129 BUG_ON(dev
->is_virtfn
);
131 /* Only generate path[] for device addition event */
132 if (event
== BUS_NOTIFY_ADD_DEVICE
)
133 for (tmp
= dev
; tmp
; tmp
= tmp
->bus
->self
)
136 size
= sizeof(*info
) + level
* sizeof(struct acpi_dmar_pci_path
);
137 if (size
<= sizeof(dmar_pci_notify_info_buf
)) {
138 info
= (struct dmar_pci_notify_info
*)dmar_pci_notify_info_buf
;
140 info
= kzalloc(size
, GFP_KERNEL
);
142 pr_warn("Out of memory when allocating notify_info "
143 "for %s.\n", pci_name(dev
));
144 if (dmar_dev_scope_status
== 0)
145 dmar_dev_scope_status
= -ENOMEM
;
152 info
->seg
= pci_domain_nr(dev
->bus
);
154 if (event
== BUS_NOTIFY_ADD_DEVICE
) {
155 for (tmp
= dev
, level
--; tmp
; tmp
= tmp
->bus
->self
) {
156 info
->path
[level
].device
= PCI_SLOT(tmp
->devfn
);
157 info
->path
[level
].function
= PCI_FUNC(tmp
->devfn
);
158 if (pci_is_root_bus(tmp
->bus
))
159 info
->bus
= tmp
->bus
->number
;
166 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info
*info
)
168 if ((void *)info
!= dmar_pci_notify_info_buf
)
172 static bool dmar_match_pci_path(struct dmar_pci_notify_info
*info
, int bus
,
173 struct acpi_dmar_pci_path
*path
, int count
)
177 if (info
->bus
!= bus
)
179 if (info
->level
!= count
)
182 for (i
= 0; i
< count
; i
++) {
183 if (path
[i
].device
!= info
->path
[i
].device
||
184 path
[i
].function
!= info
->path
[i
].function
)
191 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
192 int dmar_insert_dev_scope(struct dmar_pci_notify_info
*info
,
193 void *start
, void*end
, u16 segment
,
194 struct pci_dev __rcu
**devices
, int devices_cnt
)
197 struct pci_dev
*tmp
, *dev
= info
->dev
;
198 struct acpi_dmar_device_scope
*scope
;
199 struct acpi_dmar_pci_path
*path
;
201 if (segment
!= info
->seg
)
204 for (; start
< end
; start
+= scope
->length
) {
206 if (scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&&
207 scope
->entry_type
!= ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
210 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
211 level
= (scope
->length
- sizeof(*scope
)) / sizeof(*path
);
212 if (!dmar_match_pci_path(info
, scope
->bus
, path
, level
))
215 if ((scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
) ^
216 (dev
->hdr_type
== PCI_HEADER_TYPE_NORMAL
)) {
217 pr_warn("Device scope type does not match for %s\n",
222 for_each_dev_scope(devices
, devices_cnt
, i
, tmp
)
224 rcu_assign_pointer(devices
[i
],
228 BUG_ON(i
>= devices_cnt
);
234 int dmar_remove_dev_scope(struct dmar_pci_notify_info
*info
, u16 segment
,
235 struct pci_dev __rcu
**devices
, int count
)
240 if (info
->seg
!= segment
)
243 for_each_active_dev_scope(devices
, count
, index
, tmp
)
244 if (tmp
== info
->dev
) {
245 rcu_assign_pointer(devices
[index
], NULL
);
254 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info
*info
)
257 struct dmar_drhd_unit
*dmaru
;
258 struct acpi_dmar_hardware_unit
*drhd
;
260 for_each_drhd_unit(dmaru
) {
261 if (dmaru
->include_all
)
264 drhd
= container_of(dmaru
->hdr
,
265 struct acpi_dmar_hardware_unit
, header
);
266 ret
= dmar_insert_dev_scope(info
, (void *)(drhd
+ 1),
267 ((void *)drhd
) + drhd
->header
.length
,
269 dmaru
->devices
, dmaru
->devices_cnt
);
274 ret
= dmar_iommu_notify_scope_dev(info
);
275 if (ret
< 0 && dmar_dev_scope_status
== 0)
276 dmar_dev_scope_status
= ret
;
281 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info
*info
)
283 struct dmar_drhd_unit
*dmaru
;
285 for_each_drhd_unit(dmaru
)
286 if (dmar_remove_dev_scope(info
, dmaru
->segment
,
287 dmaru
->devices
, dmaru
->devices_cnt
))
289 dmar_iommu_notify_scope_dev(info
);
292 static int dmar_pci_bus_notifier(struct notifier_block
*nb
,
293 unsigned long action
, void *data
)
295 struct pci_dev
*pdev
= to_pci_dev(data
);
296 struct dmar_pci_notify_info
*info
;
298 /* Only care about add/remove events for physical functions */
301 if (action
!= BUS_NOTIFY_ADD_DEVICE
&& action
!= BUS_NOTIFY_DEL_DEVICE
)
304 info
= dmar_alloc_pci_notify_info(pdev
, action
);
308 down_write(&dmar_global_lock
);
309 if (action
== BUS_NOTIFY_ADD_DEVICE
)
310 dmar_pci_bus_add_dev(info
);
311 else if (action
== BUS_NOTIFY_DEL_DEVICE
)
312 dmar_pci_bus_del_dev(info
);
313 up_write(&dmar_global_lock
);
315 dmar_free_pci_notify_info(info
);
320 static struct notifier_block dmar_pci_bus_nb
= {
321 .notifier_call
= dmar_pci_bus_notifier
,
326 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
327 * structure which uniquely represent one DMA remapping hardware unit
328 * present in the platform
331 dmar_parse_one_drhd(struct acpi_dmar_header
*header
)
333 struct acpi_dmar_hardware_unit
*drhd
;
334 struct dmar_drhd_unit
*dmaru
;
337 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
338 dmaru
= kzalloc(sizeof(*dmaru
), GFP_KERNEL
);
343 dmaru
->reg_base_addr
= drhd
->address
;
344 dmaru
->segment
= drhd
->segment
;
345 dmaru
->include_all
= drhd
->flags
& 0x1; /* BIT0: INCLUDE_ALL */
346 dmaru
->devices
= dmar_alloc_dev_scope((void *)(drhd
+ 1),
347 ((void *)drhd
) + drhd
->header
.length
,
348 &dmaru
->devices_cnt
);
349 if (dmaru
->devices_cnt
&& dmaru
->devices
== NULL
) {
354 ret
= alloc_iommu(dmaru
);
356 dmar_free_dev_scope(&dmaru
->devices
,
357 &dmaru
->devices_cnt
);
361 dmar_register_drhd_unit(dmaru
);
365 static void dmar_free_drhd(struct dmar_drhd_unit
*dmaru
)
367 if (dmaru
->devices
&& dmaru
->devices_cnt
)
368 dmar_free_dev_scope(&dmaru
->devices
, &dmaru
->devices_cnt
);
370 free_iommu(dmaru
->iommu
);
374 static int __init
dmar_parse_one_andd(struct acpi_dmar_header
*header
)
376 struct acpi_dmar_andd
*andd
= (void *)header
;
378 /* Check for NUL termination within the designated length */
379 if (strnlen(andd
->object_name
, header
->length
- 8) == header
->length
- 8) {
380 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND
,
381 "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
382 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
383 dmi_get_system_info(DMI_BIOS_VENDOR
),
384 dmi_get_system_info(DMI_BIOS_VERSION
),
385 dmi_get_system_info(DMI_PRODUCT_VERSION
));
388 pr_info("ANDD device: %x name: %s\n", andd
->device_number
,
394 #ifdef CONFIG_ACPI_NUMA
396 dmar_parse_one_rhsa(struct acpi_dmar_header
*header
)
398 struct acpi_dmar_rhsa
*rhsa
;
399 struct dmar_drhd_unit
*drhd
;
401 rhsa
= (struct acpi_dmar_rhsa
*)header
;
402 for_each_drhd_unit(drhd
) {
403 if (drhd
->reg_base_addr
== rhsa
->base_address
) {
404 int node
= acpi_map_pxm_to_node(rhsa
->proximity_domain
);
406 if (!node_online(node
))
408 drhd
->iommu
->node
= node
;
413 1, TAINT_FIRMWARE_WORKAROUND
,
414 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
415 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
417 dmi_get_system_info(DMI_BIOS_VENDOR
),
418 dmi_get_system_info(DMI_BIOS_VERSION
),
419 dmi_get_system_info(DMI_PRODUCT_VERSION
));
426 dmar_table_print_dmar_entry(struct acpi_dmar_header
*header
)
428 struct acpi_dmar_hardware_unit
*drhd
;
429 struct acpi_dmar_reserved_memory
*rmrr
;
430 struct acpi_dmar_atsr
*atsr
;
431 struct acpi_dmar_rhsa
*rhsa
;
433 switch (header
->type
) {
434 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
435 drhd
= container_of(header
, struct acpi_dmar_hardware_unit
,
437 pr_info("DRHD base: %#016Lx flags: %#x\n",
438 (unsigned long long)drhd
->address
, drhd
->flags
);
440 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
441 rmrr
= container_of(header
, struct acpi_dmar_reserved_memory
,
443 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
444 (unsigned long long)rmrr
->base_address
,
445 (unsigned long long)rmrr
->end_address
);
447 case ACPI_DMAR_TYPE_ATSR
:
448 atsr
= container_of(header
, struct acpi_dmar_atsr
, header
);
449 pr_info("ATSR flags: %#x\n", atsr
->flags
);
451 case ACPI_DMAR_HARDWARE_AFFINITY
:
452 rhsa
= container_of(header
, struct acpi_dmar_rhsa
, header
);
453 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
454 (unsigned long long)rhsa
->base_address
,
455 rhsa
->proximity_domain
);
457 case ACPI_DMAR_TYPE_ANDD
:
458 /* We don't print this here because we need to sanity-check
459 it first. So print it in dmar_parse_one_andd() instead. */
465 * dmar_table_detect - checks to see if the platform supports DMAR devices
467 static int __init
dmar_table_detect(void)
469 acpi_status status
= AE_OK
;
471 /* if we could find DMAR table, then there are DMAR devices */
472 status
= acpi_get_table_with_size(ACPI_SIG_DMAR
, 0,
473 (struct acpi_table_header
**)&dmar_tbl
,
476 if (ACPI_SUCCESS(status
) && !dmar_tbl
) {
477 pr_warn("Unable to map DMAR\n");
478 status
= AE_NOT_FOUND
;
481 return (ACPI_SUCCESS(status
) ? 1 : 0);
485 * parse_dmar_table - parses the DMA reporting table
488 parse_dmar_table(void)
490 struct acpi_table_dmar
*dmar
;
491 struct acpi_dmar_header
*entry_header
;
496 * Do it again, earlier dmar_tbl mapping could be mapped with
502 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
503 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
505 dmar_tbl
= tboot_get_dmar_table(dmar_tbl
);
507 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
511 if (dmar
->width
< PAGE_SHIFT
- 1) {
512 pr_warn("Invalid DMAR haw\n");
516 pr_info("Host address width %d\n", dmar
->width
+ 1);
518 entry_header
= (struct acpi_dmar_header
*)(dmar
+ 1);
519 while (((unsigned long)entry_header
) <
520 (((unsigned long)dmar
) + dmar_tbl
->length
)) {
521 /* Avoid looping forever on bad ACPI tables */
522 if (entry_header
->length
== 0) {
523 pr_warn("Invalid 0-length structure\n");
528 dmar_table_print_dmar_entry(entry_header
);
530 switch (entry_header
->type
) {
531 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
533 ret
= dmar_parse_one_drhd(entry_header
);
535 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
536 ret
= dmar_parse_one_rmrr(entry_header
);
538 case ACPI_DMAR_TYPE_ATSR
:
539 ret
= dmar_parse_one_atsr(entry_header
);
541 case ACPI_DMAR_HARDWARE_AFFINITY
:
542 #ifdef CONFIG_ACPI_NUMA
543 ret
= dmar_parse_one_rhsa(entry_header
);
546 case ACPI_DMAR_TYPE_ANDD
:
547 ret
= dmar_parse_one_andd(entry_header
);
550 pr_warn("Unknown DMAR structure type %d\n",
552 ret
= 0; /* for forward compatibility */
558 entry_header
= ((void *)entry_header
+ entry_header
->length
);
561 pr_warn(FW_BUG
"No DRHD structure found in DMAR table\n");
565 static int dmar_pci_device_match(struct pci_dev __rcu
*devices
[], int cnt
,
572 for_each_active_dev_scope(devices
, cnt
, index
, tmp
)
576 /* Check our parent */
577 dev
= dev
->bus
->self
;
583 struct dmar_drhd_unit
*
584 dmar_find_matched_drhd_unit(struct pci_dev
*dev
)
586 struct dmar_drhd_unit
*dmaru
;
587 struct acpi_dmar_hardware_unit
*drhd
;
589 dev
= pci_physfn(dev
);
592 for_each_drhd_unit(dmaru
) {
593 drhd
= container_of(dmaru
->hdr
,
594 struct acpi_dmar_hardware_unit
,
597 if (dmaru
->include_all
&&
598 drhd
->segment
== pci_domain_nr(dev
->bus
))
601 if (dmar_pci_device_match(dmaru
->devices
,
602 dmaru
->devices_cnt
, dev
))
612 int __init
dmar_dev_scope_init(void)
614 struct pci_dev
*dev
= NULL
;
615 struct dmar_pci_notify_info
*info
;
617 if (dmar_dev_scope_status
!= 1)
618 return dmar_dev_scope_status
;
620 if (list_empty(&dmar_drhd_units
)) {
621 dmar_dev_scope_status
= -ENODEV
;
623 dmar_dev_scope_status
= 0;
625 for_each_pci_dev(dev
) {
629 info
= dmar_alloc_pci_notify_info(dev
,
630 BUS_NOTIFY_ADD_DEVICE
);
632 return dmar_dev_scope_status
;
634 dmar_pci_bus_add_dev(info
);
635 dmar_free_pci_notify_info(info
);
639 bus_register_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
642 return dmar_dev_scope_status
;
646 int __init
dmar_table_init(void)
648 static int dmar_table_initialized
;
651 if (dmar_table_initialized
== 0) {
652 ret
= parse_dmar_table();
655 pr_info("parse DMAR table failure.\n");
656 } else if (list_empty(&dmar_drhd_units
)) {
657 pr_info("No DMAR devices found\n");
662 dmar_table_initialized
= ret
;
664 dmar_table_initialized
= 1;
667 return dmar_table_initialized
< 0 ? dmar_table_initialized
: 0;
670 static void warn_invalid_dmar(u64 addr
, const char *message
)
673 1, TAINT_FIRMWARE_WORKAROUND
,
674 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
675 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
677 dmi_get_system_info(DMI_BIOS_VENDOR
),
678 dmi_get_system_info(DMI_BIOS_VERSION
),
679 dmi_get_system_info(DMI_PRODUCT_VERSION
));
682 static int __init
check_zero_address(void)
684 struct acpi_table_dmar
*dmar
;
685 struct acpi_dmar_header
*entry_header
;
686 struct acpi_dmar_hardware_unit
*drhd
;
688 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
689 entry_header
= (struct acpi_dmar_header
*)(dmar
+ 1);
691 while (((unsigned long)entry_header
) <
692 (((unsigned long)dmar
) + dmar_tbl
->length
)) {
693 /* Avoid looping forever on bad ACPI tables */
694 if (entry_header
->length
== 0) {
695 pr_warn("Invalid 0-length structure\n");
699 if (entry_header
->type
== ACPI_DMAR_TYPE_HARDWARE_UNIT
) {
703 drhd
= (void *)entry_header
;
704 if (!drhd
->address
) {
705 warn_invalid_dmar(0, "");
709 addr
= early_ioremap(drhd
->address
, VTD_PAGE_SIZE
);
711 printk("IOMMU: can't validate: %llx\n", drhd
->address
);
714 cap
= dmar_readq(addr
+ DMAR_CAP_REG
);
715 ecap
= dmar_readq(addr
+ DMAR_ECAP_REG
);
716 early_iounmap(addr
, VTD_PAGE_SIZE
);
717 if (cap
== (uint64_t)-1 && ecap
== (uint64_t)-1) {
718 warn_invalid_dmar(drhd
->address
,
719 " returns all ones");
724 entry_header
= ((void *)entry_header
+ entry_header
->length
);
732 int __init
detect_intel_iommu(void)
736 down_write(&dmar_global_lock
);
737 ret
= dmar_table_detect();
739 ret
= check_zero_address();
741 if (ret
&& !no_iommu
&& !iommu_detected
&& !dmar_disabled
) {
743 /* Make sure ACS will be enabled */
749 x86_init
.iommu
.iommu_init
= intel_iommu_init
;
752 early_acpi_os_unmap_memory((void __iomem
*)dmar_tbl
, dmar_tbl_size
);
754 up_write(&dmar_global_lock
);
756 return ret
? 1 : -ENODEV
;
760 static void unmap_iommu(struct intel_iommu
*iommu
)
763 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
767 * map_iommu: map the iommu's registers
768 * @iommu: the iommu to map
769 * @phys_addr: the physical address of the base resgister
771 * Memory map the iommu's registers. Start w/ a single page, and
772 * possibly expand if that turns out to be insufficent.
774 static int map_iommu(struct intel_iommu
*iommu
, u64 phys_addr
)
778 iommu
->reg_phys
= phys_addr
;
779 iommu
->reg_size
= VTD_PAGE_SIZE
;
781 if (!request_mem_region(iommu
->reg_phys
, iommu
->reg_size
, iommu
->name
)) {
782 pr_err("IOMMU: can't reserve memory\n");
787 iommu
->reg
= ioremap(iommu
->reg_phys
, iommu
->reg_size
);
789 pr_err("IOMMU: can't map the region\n");
794 iommu
->cap
= dmar_readq(iommu
->reg
+ DMAR_CAP_REG
);
795 iommu
->ecap
= dmar_readq(iommu
->reg
+ DMAR_ECAP_REG
);
797 if (iommu
->cap
== (uint64_t)-1 && iommu
->ecap
== (uint64_t)-1) {
799 warn_invalid_dmar(phys_addr
, " returns all ones");
803 /* the registers might be more than one page */
804 map_size
= max_t(int, ecap_max_iotlb_offset(iommu
->ecap
),
805 cap_max_fault_reg_offset(iommu
->cap
));
806 map_size
= VTD_PAGE_ALIGN(map_size
);
807 if (map_size
> iommu
->reg_size
) {
809 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
810 iommu
->reg_size
= map_size
;
811 if (!request_mem_region(iommu
->reg_phys
, iommu
->reg_size
,
813 pr_err("IOMMU: can't reserve memory\n");
817 iommu
->reg
= ioremap(iommu
->reg_phys
, iommu
->reg_size
);
819 pr_err("IOMMU: can't map the region\n");
830 release_mem_region(iommu
->reg_phys
, iommu
->reg_size
);
835 static int alloc_iommu(struct dmar_drhd_unit
*drhd
)
837 struct intel_iommu
*iommu
;
839 static int iommu_allocated
= 0;
844 if (!drhd
->reg_base_addr
) {
845 warn_invalid_dmar(0, "");
849 iommu
= kzalloc(sizeof(*iommu
), GFP_KERNEL
);
853 iommu
->seq_id
= iommu_allocated
++;
854 sprintf (iommu
->name
, "dmar%d", iommu
->seq_id
);
856 err
= map_iommu(iommu
, drhd
->reg_base_addr
);
858 pr_err("IOMMU: failed to map %s\n", iommu
->name
);
863 agaw
= iommu_calculate_agaw(iommu
);
865 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
869 msagaw
= iommu_calculate_max_sagaw(iommu
);
871 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
876 iommu
->msagaw
= msagaw
;
880 ver
= readl(iommu
->reg
+ DMAR_VER_REG
);
881 pr_info("IOMMU %d: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
883 (unsigned long long)drhd
->reg_base_addr
,
884 DMAR_VER_MAJOR(ver
), DMAR_VER_MINOR(ver
),
885 (unsigned long long)iommu
->cap
,
886 (unsigned long long)iommu
->ecap
);
888 /* Reflect status in gcmd */
889 sts
= readl(iommu
->reg
+ DMAR_GSTS_REG
);
890 if (sts
& DMA_GSTS_IRES
)
891 iommu
->gcmd
|= DMA_GCMD_IRE
;
892 if (sts
& DMA_GSTS_TES
)
893 iommu
->gcmd
|= DMA_GCMD_TE
;
894 if (sts
& DMA_GSTS_QIES
)
895 iommu
->gcmd
|= DMA_GCMD_QIE
;
897 raw_spin_lock_init(&iommu
->register_lock
);
909 static void free_iommu(struct intel_iommu
*iommu
)
912 free_irq(iommu
->irq
, iommu
);
913 irq_set_handler_data(iommu
->irq
, NULL
);
914 destroy_irq(iommu
->irq
);
918 free_page((unsigned long)iommu
->qi
->desc
);
919 kfree(iommu
->qi
->desc_status
);
930 * Reclaim all the submitted descriptors which have completed its work.
932 static inline void reclaim_free_desc(struct q_inval
*qi
)
934 while (qi
->desc_status
[qi
->free_tail
] == QI_DONE
||
935 qi
->desc_status
[qi
->free_tail
] == QI_ABORT
) {
936 qi
->desc_status
[qi
->free_tail
] = QI_FREE
;
937 qi
->free_tail
= (qi
->free_tail
+ 1) % QI_LENGTH
;
942 static int qi_check_fault(struct intel_iommu
*iommu
, int index
)
946 struct q_inval
*qi
= iommu
->qi
;
947 int wait_index
= (index
+ 1) % QI_LENGTH
;
949 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
952 fault
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
955 * If IQE happens, the head points to the descriptor associated
956 * with the error. No new descriptors are fetched until the IQE
959 if (fault
& DMA_FSTS_IQE
) {
960 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
961 if ((head
>> DMAR_IQ_SHIFT
) == index
) {
962 pr_err("VT-d detected invalid descriptor: "
963 "low=%llx, high=%llx\n",
964 (unsigned long long)qi
->desc
[index
].low
,
965 (unsigned long long)qi
->desc
[index
].high
);
966 memcpy(&qi
->desc
[index
], &qi
->desc
[wait_index
],
967 sizeof(struct qi_desc
));
968 __iommu_flush_cache(iommu
, &qi
->desc
[index
],
969 sizeof(struct qi_desc
));
970 writel(DMA_FSTS_IQE
, iommu
->reg
+ DMAR_FSTS_REG
);
976 * If ITE happens, all pending wait_desc commands are aborted.
977 * No new descriptors are fetched until the ITE is cleared.
979 if (fault
& DMA_FSTS_ITE
) {
980 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
981 head
= ((head
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
983 tail
= readl(iommu
->reg
+ DMAR_IQT_REG
);
984 tail
= ((tail
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
986 writel(DMA_FSTS_ITE
, iommu
->reg
+ DMAR_FSTS_REG
);
989 if (qi
->desc_status
[head
] == QI_IN_USE
)
990 qi
->desc_status
[head
] = QI_ABORT
;
991 head
= (head
- 2 + QI_LENGTH
) % QI_LENGTH
;
992 } while (head
!= tail
);
994 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
998 if (fault
& DMA_FSTS_ICE
)
999 writel(DMA_FSTS_ICE
, iommu
->reg
+ DMAR_FSTS_REG
);
1005 * Submit the queued invalidation descriptor to the remapping
1006 * hardware unit and wait for its completion.
1008 int qi_submit_sync(struct qi_desc
*desc
, struct intel_iommu
*iommu
)
1011 struct q_inval
*qi
= iommu
->qi
;
1012 struct qi_desc
*hw
, wait_desc
;
1013 int wait_index
, index
;
1014 unsigned long flags
;
1024 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1025 while (qi
->free_cnt
< 3) {
1026 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1028 raw_spin_lock_irqsave(&qi
->q_lock
, flags
);
1031 index
= qi
->free_head
;
1032 wait_index
= (index
+ 1) % QI_LENGTH
;
1034 qi
->desc_status
[index
] = qi
->desc_status
[wait_index
] = QI_IN_USE
;
1038 wait_desc
.low
= QI_IWD_STATUS_DATA(QI_DONE
) |
1039 QI_IWD_STATUS_WRITE
| QI_IWD_TYPE
;
1040 wait_desc
.high
= virt_to_phys(&qi
->desc_status
[wait_index
]);
1042 hw
[wait_index
] = wait_desc
;
1044 __iommu_flush_cache(iommu
, &hw
[index
], sizeof(struct qi_desc
));
1045 __iommu_flush_cache(iommu
, &hw
[wait_index
], sizeof(struct qi_desc
));
1047 qi
->free_head
= (qi
->free_head
+ 2) % QI_LENGTH
;
1051 * update the HW tail register indicating the presence of
1054 writel(qi
->free_head
<< DMAR_IQ_SHIFT
, iommu
->reg
+ DMAR_IQT_REG
);
1056 while (qi
->desc_status
[wait_index
] != QI_DONE
) {
1058 * We will leave the interrupts disabled, to prevent interrupt
1059 * context to queue another cmd while a cmd is already submitted
1060 * and waiting for completion on this cpu. This is to avoid
1061 * a deadlock where the interrupt context can wait indefinitely
1062 * for free slots in the queue.
1064 rc
= qi_check_fault(iommu
, index
);
1068 raw_spin_unlock(&qi
->q_lock
);
1070 raw_spin_lock(&qi
->q_lock
);
1073 qi
->desc_status
[index
] = QI_DONE
;
1075 reclaim_free_desc(qi
);
1076 raw_spin_unlock_irqrestore(&qi
->q_lock
, flags
);
1085 * Flush the global interrupt entry cache.
1087 void qi_global_iec(struct intel_iommu
*iommu
)
1089 struct qi_desc desc
;
1091 desc
.low
= QI_IEC_TYPE
;
1094 /* should never fail */
1095 qi_submit_sync(&desc
, iommu
);
1098 void qi_flush_context(struct intel_iommu
*iommu
, u16 did
, u16 sid
, u8 fm
,
1101 struct qi_desc desc
;
1103 desc
.low
= QI_CC_FM(fm
) | QI_CC_SID(sid
) | QI_CC_DID(did
)
1104 | QI_CC_GRAN(type
) | QI_CC_TYPE
;
1107 qi_submit_sync(&desc
, iommu
);
1110 void qi_flush_iotlb(struct intel_iommu
*iommu
, u16 did
, u64 addr
,
1111 unsigned int size_order
, u64 type
)
1115 struct qi_desc desc
;
1118 if (cap_write_drain(iommu
->cap
))
1121 if (cap_read_drain(iommu
->cap
))
1124 desc
.low
= QI_IOTLB_DID(did
) | QI_IOTLB_DR(dr
) | QI_IOTLB_DW(dw
)
1125 | QI_IOTLB_GRAN(type
) | QI_IOTLB_TYPE
;
1126 desc
.high
= QI_IOTLB_ADDR(addr
) | QI_IOTLB_IH(ih
)
1127 | QI_IOTLB_AM(size_order
);
1129 qi_submit_sync(&desc
, iommu
);
1132 void qi_flush_dev_iotlb(struct intel_iommu
*iommu
, u16 sid
, u16 qdep
,
1133 u64 addr
, unsigned mask
)
1135 struct qi_desc desc
;
1138 BUG_ON(addr
& ((1 << (VTD_PAGE_SHIFT
+ mask
)) - 1));
1139 addr
|= (1 << (VTD_PAGE_SHIFT
+ mask
- 1)) - 1;
1140 desc
.high
= QI_DEV_IOTLB_ADDR(addr
) | QI_DEV_IOTLB_SIZE
;
1142 desc
.high
= QI_DEV_IOTLB_ADDR(addr
);
1144 if (qdep
>= QI_DEV_IOTLB_MAX_INVS
)
1147 desc
.low
= QI_DEV_IOTLB_SID(sid
) | QI_DEV_IOTLB_QDEP(qdep
) |
1150 qi_submit_sync(&desc
, iommu
);
1154 * Disable Queued Invalidation interface.
1156 void dmar_disable_qi(struct intel_iommu
*iommu
)
1158 unsigned long flags
;
1160 cycles_t start_time
= get_cycles();
1162 if (!ecap_qis(iommu
->ecap
))
1165 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1167 sts
= dmar_readq(iommu
->reg
+ DMAR_GSTS_REG
);
1168 if (!(sts
& DMA_GSTS_QIES
))
1172 * Give a chance to HW to complete the pending invalidation requests.
1174 while ((readl(iommu
->reg
+ DMAR_IQT_REG
) !=
1175 readl(iommu
->reg
+ DMAR_IQH_REG
)) &&
1176 (DMAR_OPERATION_TIMEOUT
> (get_cycles() - start_time
)))
1179 iommu
->gcmd
&= ~DMA_GCMD_QIE
;
1180 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1182 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
,
1183 !(sts
& DMA_GSTS_QIES
), sts
);
1185 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1189 * Enable queued invalidation.
1191 static void __dmar_enable_qi(struct intel_iommu
*iommu
)
1194 unsigned long flags
;
1195 struct q_inval
*qi
= iommu
->qi
;
1197 qi
->free_head
= qi
->free_tail
= 0;
1198 qi
->free_cnt
= QI_LENGTH
;
1200 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
1202 /* write zero to the tail reg */
1203 writel(0, iommu
->reg
+ DMAR_IQT_REG
);
1205 dmar_writeq(iommu
->reg
+ DMAR_IQA_REG
, virt_to_phys(qi
->desc
));
1207 iommu
->gcmd
|= DMA_GCMD_QIE
;
1208 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
1210 /* Make sure hardware complete it */
1211 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
, (sts
& DMA_GSTS_QIES
), sts
);
1213 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
1217 * Enable Queued Invalidation interface. This is a must to support
1218 * interrupt-remapping. Also used by DMA-remapping, which replaces
1219 * register based IOTLB invalidation.
1221 int dmar_enable_qi(struct intel_iommu
*iommu
)
1224 struct page
*desc_page
;
1226 if (!ecap_qis(iommu
->ecap
))
1230 * queued invalidation is already setup and enabled.
1235 iommu
->qi
= kmalloc(sizeof(*qi
), GFP_ATOMIC
);
1242 desc_page
= alloc_pages_node(iommu
->node
, GFP_ATOMIC
| __GFP_ZERO
, 0);
1249 qi
->desc
= page_address(desc_page
);
1251 qi
->desc_status
= kzalloc(QI_LENGTH
* sizeof(int), GFP_ATOMIC
);
1252 if (!qi
->desc_status
) {
1253 free_page((unsigned long) qi
->desc
);
1259 qi
->free_head
= qi
->free_tail
= 0;
1260 qi
->free_cnt
= QI_LENGTH
;
1262 raw_spin_lock_init(&qi
->q_lock
);
1264 __dmar_enable_qi(iommu
);
1269 /* iommu interrupt handling. Most stuff are MSI-like. */
1277 static const char *dma_remap_fault_reasons
[] =
1280 "Present bit in root entry is clear",
1281 "Present bit in context entry is clear",
1282 "Invalid context entry",
1283 "Access beyond MGAW",
1284 "PTE Write access is not set",
1285 "PTE Read access is not set",
1286 "Next page table ptr is invalid",
1287 "Root table address invalid",
1288 "Context table ptr is invalid",
1289 "non-zero reserved fields in RTP",
1290 "non-zero reserved fields in CTP",
1291 "non-zero reserved fields in PTE",
1292 "PCE for translation request specifies blocking",
1295 static const char *irq_remap_fault_reasons
[] =
1297 "Detected reserved fields in the decoded interrupt-remapped request",
1298 "Interrupt index exceeded the interrupt-remapping table size",
1299 "Present field in the IRTE entry is clear",
1300 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1301 "Detected reserved fields in the IRTE entry",
1302 "Blocked a compatibility format interrupt request",
1303 "Blocked an interrupt request due to source-id verification failure",
1306 static const char *dmar_get_fault_reason(u8 fault_reason
, int *fault_type
)
1308 if (fault_reason
>= 0x20 && (fault_reason
- 0x20 <
1309 ARRAY_SIZE(irq_remap_fault_reasons
))) {
1310 *fault_type
= INTR_REMAP
;
1311 return irq_remap_fault_reasons
[fault_reason
- 0x20];
1312 } else if (fault_reason
< ARRAY_SIZE(dma_remap_fault_reasons
)) {
1313 *fault_type
= DMA_REMAP
;
1314 return dma_remap_fault_reasons
[fault_reason
];
1316 *fault_type
= UNKNOWN
;
1321 void dmar_msi_unmask(struct irq_data
*data
)
1323 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1327 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1328 writel(0, iommu
->reg
+ DMAR_FECTL_REG
);
1329 /* Read a reg to force flush the post write */
1330 readl(iommu
->reg
+ DMAR_FECTL_REG
);
1331 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1334 void dmar_msi_mask(struct irq_data
*data
)
1337 struct intel_iommu
*iommu
= irq_data_get_irq_handler_data(data
);
1340 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1341 writel(DMA_FECTL_IM
, iommu
->reg
+ DMAR_FECTL_REG
);
1342 /* Read a reg to force flush the post write */
1343 readl(iommu
->reg
+ DMAR_FECTL_REG
);
1344 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1347 void dmar_msi_write(int irq
, struct msi_msg
*msg
)
1349 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1352 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1353 writel(msg
->data
, iommu
->reg
+ DMAR_FEDATA_REG
);
1354 writel(msg
->address_lo
, iommu
->reg
+ DMAR_FEADDR_REG
);
1355 writel(msg
->address_hi
, iommu
->reg
+ DMAR_FEUADDR_REG
);
1356 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1359 void dmar_msi_read(int irq
, struct msi_msg
*msg
)
1361 struct intel_iommu
*iommu
= irq_get_handler_data(irq
);
1364 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1365 msg
->data
= readl(iommu
->reg
+ DMAR_FEDATA_REG
);
1366 msg
->address_lo
= readl(iommu
->reg
+ DMAR_FEADDR_REG
);
1367 msg
->address_hi
= readl(iommu
->reg
+ DMAR_FEUADDR_REG
);
1368 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1371 static int dmar_fault_do_one(struct intel_iommu
*iommu
, int type
,
1372 u8 fault_reason
, u16 source_id
, unsigned long long addr
)
1377 reason
= dmar_get_fault_reason(fault_reason
, &fault_type
);
1379 if (fault_type
== INTR_REMAP
)
1380 pr_err("INTR-REMAP: Request device [[%02x:%02x.%d] "
1381 "fault index %llx\n"
1382 "INTR-REMAP:[fault reason %02d] %s\n",
1383 (source_id
>> 8), PCI_SLOT(source_id
& 0xFF),
1384 PCI_FUNC(source_id
& 0xFF), addr
>> 48,
1385 fault_reason
, reason
);
1387 pr_err("DMAR:[%s] Request device [%02x:%02x.%d] "
1388 "fault addr %llx \n"
1389 "DMAR:[fault reason %02d] %s\n",
1390 (type
? "DMA Read" : "DMA Write"),
1391 (source_id
>> 8), PCI_SLOT(source_id
& 0xFF),
1392 PCI_FUNC(source_id
& 0xFF), addr
, fault_reason
, reason
);
1396 #define PRIMARY_FAULT_REG_LEN (16)
1397 irqreturn_t
dmar_fault(int irq
, void *dev_id
)
1399 struct intel_iommu
*iommu
= dev_id
;
1400 int reg
, fault_index
;
1404 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1405 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1407 pr_err("DRHD: handling fault status reg %x\n", fault_status
);
1409 /* TBD: ignore advanced fault log currently */
1410 if (!(fault_status
& DMA_FSTS_PPF
))
1413 fault_index
= dma_fsts_fault_record_index(fault_status
);
1414 reg
= cap_fault_reg_offset(iommu
->cap
);
1422 /* highest 32 bits */
1423 data
= readl(iommu
->reg
+ reg
+
1424 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1425 if (!(data
& DMA_FRCD_F
))
1428 fault_reason
= dma_frcd_fault_reason(data
);
1429 type
= dma_frcd_type(data
);
1431 data
= readl(iommu
->reg
+ reg
+
1432 fault_index
* PRIMARY_FAULT_REG_LEN
+ 8);
1433 source_id
= dma_frcd_source_id(data
);
1435 guest_addr
= dmar_readq(iommu
->reg
+ reg
+
1436 fault_index
* PRIMARY_FAULT_REG_LEN
);
1437 guest_addr
= dma_frcd_page_addr(guest_addr
);
1438 /* clear the fault */
1439 writel(DMA_FRCD_F
, iommu
->reg
+ reg
+
1440 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1442 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1444 dmar_fault_do_one(iommu
, type
, fault_reason
,
1445 source_id
, guest_addr
);
1448 if (fault_index
>= cap_num_fault_regs(iommu
->cap
))
1450 raw_spin_lock_irqsave(&iommu
->register_lock
, flag
);
1453 writel(DMA_FSTS_PFO
| DMA_FSTS_PPF
, iommu
->reg
+ DMAR_FSTS_REG
);
1456 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1460 int dmar_set_interrupt(struct intel_iommu
*iommu
)
1465 * Check if the fault interrupt is already initialized.
1472 pr_err("IOMMU: no free vectors\n");
1476 irq_set_handler_data(irq
, iommu
);
1479 ret
= arch_setup_dmar_msi(irq
);
1481 irq_set_handler_data(irq
, NULL
);
1487 ret
= request_irq(irq
, dmar_fault
, IRQF_NO_THREAD
, iommu
->name
, iommu
);
1489 pr_err("IOMMU: can't request irq\n");
1493 int __init
enable_drhd_fault_handling(void)
1495 struct dmar_drhd_unit
*drhd
;
1496 struct intel_iommu
*iommu
;
1499 * Enable fault control interrupt.
1501 for_each_iommu(iommu
, drhd
) {
1503 int ret
= dmar_set_interrupt(iommu
);
1506 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1507 (unsigned long long)drhd
->reg_base_addr
, ret
);
1512 * Clear any previous faults.
1514 dmar_fault(iommu
->irq
, iommu
);
1515 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1516 writel(fault_status
, iommu
->reg
+ DMAR_FSTS_REG
);
1523 * Re-enable Queued Invalidation interface.
1525 int dmar_reenable_qi(struct intel_iommu
*iommu
)
1527 if (!ecap_qis(iommu
->ecap
))
1534 * First disable queued invalidation.
1536 dmar_disable_qi(iommu
);
1538 * Then enable queued invalidation again. Since there is no pending
1539 * invalidation requests now, it's safe to re-enable queued
1542 __dmar_enable_qi(iommu
);
1548 * Check interrupt remapping support in DMAR table description.
1550 int __init
dmar_ir_support(void)
1552 struct acpi_table_dmar
*dmar
;
1553 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
1556 return dmar
->flags
& 0x1;
1559 static int __init
dmar_free_unused_resources(void)
1561 struct dmar_drhd_unit
*dmaru
, *dmaru_n
;
1563 /* DMAR units are in use */
1564 if (irq_remapping_enabled
|| intel_iommu_enabled
)
1567 if (dmar_dev_scope_status
!= 1 && !list_empty(&dmar_drhd_units
))
1568 bus_unregister_notifier(&pci_bus_type
, &dmar_pci_bus_nb
);
1570 down_write(&dmar_global_lock
);
1571 list_for_each_entry_safe(dmaru
, dmaru_n
, &dmar_drhd_units
, list
) {
1572 list_del(&dmaru
->list
);
1573 dmar_free_drhd(dmaru
);
1575 up_write(&dmar_global_lock
);
1580 late_initcall(dmar_free_unused_resources
);
1581 IOMMU_INIT_POST(detect_intel_iommu
);