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 #include <linux/pci.h>
30 #include <linux/dmar.h>
31 #include <linux/iova.h>
32 #include <linux/intel-iommu.h>
33 #include <linux/timer.h>
34 #include <linux/irq.h>
35 #include <linux/interrupt.h>
36 #include <linux/tboot.h>
39 #define PREFIX "DMAR:"
41 /* No locks are needed as DMA remapping hardware unit
42 * list is constructed at boot time and hotplug of
43 * these units are not supported by the architecture.
45 LIST_HEAD(dmar_drhd_units
);
47 static struct acpi_table_header
* __initdata dmar_tbl
;
48 static acpi_size dmar_tbl_size
;
50 static void __init
dmar_register_drhd_unit(struct dmar_drhd_unit
*drhd
)
53 * add INCLUDE_ALL at the tail, so scan the list will find it at
56 if (drhd
->include_all
)
57 list_add_tail(&drhd
->list
, &dmar_drhd_units
);
59 list_add(&drhd
->list
, &dmar_drhd_units
);
62 static int __init
dmar_parse_one_dev_scope(struct acpi_dmar_device_scope
*scope
,
63 struct pci_dev
**dev
, u16 segment
)
66 struct pci_dev
*pdev
= NULL
;
67 struct acpi_dmar_pci_path
*path
;
70 bus
= pci_find_bus(segment
, scope
->bus
);
71 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
72 count
= (scope
->length
- sizeof(struct acpi_dmar_device_scope
))
73 / sizeof(struct acpi_dmar_pci_path
);
79 * Some BIOSes list non-exist devices in DMAR table, just
84 PREFIX
"Device scope bus [%d] not found\n",
88 pdev
= pci_get_slot(bus
, PCI_DEVFN(path
->dev
, path
->fn
));
90 printk(KERN_WARNING PREFIX
91 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
92 segment
, bus
->number
, path
->dev
, path
->fn
);
97 bus
= pdev
->subordinate
;
100 printk(KERN_WARNING PREFIX
101 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
102 segment
, scope
->bus
, path
->dev
, path
->fn
);
106 if ((scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
&& \
107 pdev
->subordinate
) || (scope
->entry_type
== \
108 ACPI_DMAR_SCOPE_TYPE_BRIDGE
&& !pdev
->subordinate
)) {
110 printk(KERN_WARNING PREFIX
111 "Device scope type does not match for %s\n",
119 static int __init
dmar_parse_dev_scope(void *start
, void *end
, int *cnt
,
120 struct pci_dev
***devices
, u16 segment
)
122 struct acpi_dmar_device_scope
*scope
;
128 while (start
< end
) {
130 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
||
131 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
)
134 printk(KERN_WARNING PREFIX
135 "Unsupported device scope\n");
136 start
+= scope
->length
;
141 *devices
= kcalloc(*cnt
, sizeof(struct pci_dev
*), GFP_KERNEL
);
147 while (start
< end
) {
149 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_ENDPOINT
||
150 scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_BRIDGE
) {
151 ret
= dmar_parse_one_dev_scope(scope
,
152 &(*devices
)[index
], segment
);
159 start
+= scope
->length
;
166 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
167 * structure which uniquely represent one DMA remapping hardware unit
168 * present in the platform
171 dmar_parse_one_drhd(struct acpi_dmar_header
*header
)
173 struct acpi_dmar_hardware_unit
*drhd
;
174 struct dmar_drhd_unit
*dmaru
;
177 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
178 if (!drhd
->address
) {
179 /* Promote an attitude of violence to a BIOS engineer today */
180 WARN(1, "Your BIOS is broken; DMAR reported at address zero!\n"
181 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
182 dmi_get_system_info(DMI_BIOS_VENDOR
),
183 dmi_get_system_info(DMI_BIOS_VERSION
),
184 dmi_get_system_info(DMI_PRODUCT_VERSION
));
187 dmaru
= kzalloc(sizeof(*dmaru
), GFP_KERNEL
);
192 dmaru
->reg_base_addr
= drhd
->address
;
193 dmaru
->segment
= drhd
->segment
;
194 dmaru
->include_all
= drhd
->flags
& 0x1; /* BIT0: INCLUDE_ALL */
196 ret
= alloc_iommu(dmaru
);
201 dmar_register_drhd_unit(dmaru
);
205 static int __init
dmar_parse_dev(struct dmar_drhd_unit
*dmaru
)
207 struct acpi_dmar_hardware_unit
*drhd
;
210 drhd
= (struct acpi_dmar_hardware_unit
*) dmaru
->hdr
;
212 if (dmaru
->include_all
)
215 ret
= dmar_parse_dev_scope((void *)(drhd
+ 1),
216 ((void *)drhd
) + drhd
->header
.length
,
217 &dmaru
->devices_cnt
, &dmaru
->devices
,
220 list_del(&dmaru
->list
);
227 LIST_HEAD(dmar_rmrr_units
);
229 static void __init
dmar_register_rmrr_unit(struct dmar_rmrr_unit
*rmrr
)
231 list_add(&rmrr
->list
, &dmar_rmrr_units
);
236 dmar_parse_one_rmrr(struct acpi_dmar_header
*header
)
238 struct acpi_dmar_reserved_memory
*rmrr
;
239 struct dmar_rmrr_unit
*rmrru
;
241 rmrru
= kzalloc(sizeof(*rmrru
), GFP_KERNEL
);
246 rmrr
= (struct acpi_dmar_reserved_memory
*)header
;
247 rmrru
->base_address
= rmrr
->base_address
;
248 rmrru
->end_address
= rmrr
->end_address
;
250 dmar_register_rmrr_unit(rmrru
);
255 rmrr_parse_dev(struct dmar_rmrr_unit
*rmrru
)
257 struct acpi_dmar_reserved_memory
*rmrr
;
260 rmrr
= (struct acpi_dmar_reserved_memory
*) rmrru
->hdr
;
261 ret
= dmar_parse_dev_scope((void *)(rmrr
+ 1),
262 ((void *)rmrr
) + rmrr
->header
.length
,
263 &rmrru
->devices_cnt
, &rmrru
->devices
, rmrr
->segment
);
265 if (ret
|| (rmrru
->devices_cnt
== 0)) {
266 list_del(&rmrru
->list
);
272 static LIST_HEAD(dmar_atsr_units
);
274 static int __init
dmar_parse_one_atsr(struct acpi_dmar_header
*hdr
)
276 struct acpi_dmar_atsr
*atsr
;
277 struct dmar_atsr_unit
*atsru
;
279 atsr
= container_of(hdr
, struct acpi_dmar_atsr
, header
);
280 atsru
= kzalloc(sizeof(*atsru
), GFP_KERNEL
);
285 atsru
->include_all
= atsr
->flags
& 0x1;
287 list_add(&atsru
->list
, &dmar_atsr_units
);
292 static int __init
atsr_parse_dev(struct dmar_atsr_unit
*atsru
)
295 struct acpi_dmar_atsr
*atsr
;
297 if (atsru
->include_all
)
300 atsr
= container_of(atsru
->hdr
, struct acpi_dmar_atsr
, header
);
301 rc
= dmar_parse_dev_scope((void *)(atsr
+ 1),
302 (void *)atsr
+ atsr
->header
.length
,
303 &atsru
->devices_cnt
, &atsru
->devices
,
305 if (rc
|| !atsru
->devices_cnt
) {
306 list_del(&atsru
->list
);
313 int dmar_find_matched_atsr_unit(struct pci_dev
*dev
)
317 struct acpi_dmar_atsr
*atsr
;
318 struct dmar_atsr_unit
*atsru
;
320 list_for_each_entry(atsru
, &dmar_atsr_units
, list
) {
321 atsr
= container_of(atsru
->hdr
, struct acpi_dmar_atsr
, header
);
322 if (atsr
->segment
== pci_domain_nr(dev
->bus
))
329 for (bus
= dev
->bus
; bus
; bus
= bus
->parent
) {
330 struct pci_dev
*bridge
= bus
->self
;
332 if (!bridge
|| !bridge
->is_pcie
||
333 bridge
->pcie_type
== PCI_EXP_TYPE_PCI_BRIDGE
)
336 if (bridge
->pcie_type
== PCI_EXP_TYPE_ROOT_PORT
) {
337 for (i
= 0; i
< atsru
->devices_cnt
; i
++)
338 if (atsru
->devices
[i
] == bridge
)
344 if (atsru
->include_all
)
352 dmar_table_print_dmar_entry(struct acpi_dmar_header
*header
)
354 struct acpi_dmar_hardware_unit
*drhd
;
355 struct acpi_dmar_reserved_memory
*rmrr
;
356 struct acpi_dmar_atsr
*atsr
;
358 switch (header
->type
) {
359 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
360 drhd
= container_of(header
, struct acpi_dmar_hardware_unit
,
362 printk (KERN_INFO PREFIX
363 "DRHD base: %#016Lx flags: %#x\n",
364 (unsigned long long)drhd
->address
, drhd
->flags
);
366 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
367 rmrr
= container_of(header
, struct acpi_dmar_reserved_memory
,
369 printk (KERN_INFO PREFIX
370 "RMRR base: %#016Lx end: %#016Lx\n",
371 (unsigned long long)rmrr
->base_address
,
372 (unsigned long long)rmrr
->end_address
);
374 case ACPI_DMAR_TYPE_ATSR
:
375 atsr
= container_of(header
, struct acpi_dmar_atsr
, header
);
376 printk(KERN_INFO PREFIX
"ATSR flags: %#x\n", atsr
->flags
);
382 * dmar_table_detect - checks to see if the platform supports DMAR devices
384 static int __init
dmar_table_detect(void)
386 acpi_status status
= AE_OK
;
388 /* if we could find DMAR table, then there are DMAR devices */
389 status
= acpi_get_table_with_size(ACPI_SIG_DMAR
, 0,
390 (struct acpi_table_header
**)&dmar_tbl
,
393 if (ACPI_SUCCESS(status
) && !dmar_tbl
) {
394 printk (KERN_WARNING PREFIX
"Unable to map DMAR\n");
395 status
= AE_NOT_FOUND
;
398 return (ACPI_SUCCESS(status
) ? 1 : 0);
402 * parse_dmar_table - parses the DMA reporting table
405 parse_dmar_table(void)
407 struct acpi_table_dmar
*dmar
;
408 struct acpi_dmar_header
*entry_header
;
412 * Do it again, earlier dmar_tbl mapping could be mapped with
418 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
419 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
421 dmar_tbl
= tboot_get_dmar_table(dmar_tbl
);
423 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
427 if (dmar
->width
< PAGE_SHIFT
- 1) {
428 printk(KERN_WARNING PREFIX
"Invalid DMAR haw\n");
432 printk (KERN_INFO PREFIX
"Host address width %d\n",
435 entry_header
= (struct acpi_dmar_header
*)(dmar
+ 1);
436 while (((unsigned long)entry_header
) <
437 (((unsigned long)dmar
) + dmar_tbl
->length
)) {
438 /* Avoid looping forever on bad ACPI tables */
439 if (entry_header
->length
== 0) {
440 printk(KERN_WARNING PREFIX
441 "Invalid 0-length structure\n");
446 dmar_table_print_dmar_entry(entry_header
);
448 switch (entry_header
->type
) {
449 case ACPI_DMAR_TYPE_HARDWARE_UNIT
:
450 ret
= dmar_parse_one_drhd(entry_header
);
452 case ACPI_DMAR_TYPE_RESERVED_MEMORY
:
454 ret
= dmar_parse_one_rmrr(entry_header
);
457 case ACPI_DMAR_TYPE_ATSR
:
459 ret
= dmar_parse_one_atsr(entry_header
);
463 printk(KERN_WARNING PREFIX
464 "Unknown DMAR structure type\n");
465 ret
= 0; /* for forward compatibility */
471 entry_header
= ((void *)entry_header
+ entry_header
->length
);
476 int dmar_pci_device_match(struct pci_dev
*devices
[], int cnt
,
482 for (index
= 0; index
< cnt
; index
++)
483 if (dev
== devices
[index
])
486 /* Check our parent */
487 dev
= dev
->bus
->self
;
493 struct dmar_drhd_unit
*
494 dmar_find_matched_drhd_unit(struct pci_dev
*dev
)
496 struct dmar_drhd_unit
*dmaru
= NULL
;
497 struct acpi_dmar_hardware_unit
*drhd
;
499 list_for_each_entry(dmaru
, &dmar_drhd_units
, list
) {
500 drhd
= container_of(dmaru
->hdr
,
501 struct acpi_dmar_hardware_unit
,
504 if (dmaru
->include_all
&&
505 drhd
->segment
== pci_domain_nr(dev
->bus
))
508 if (dmar_pci_device_match(dmaru
->devices
,
509 dmaru
->devices_cnt
, dev
))
516 int __init
dmar_dev_scope_init(void)
518 struct dmar_drhd_unit
*drhd
, *drhd_n
;
521 list_for_each_entry_safe(drhd
, drhd_n
, &dmar_drhd_units
, list
) {
522 ret
= dmar_parse_dev(drhd
);
529 struct dmar_rmrr_unit
*rmrr
, *rmrr_n
;
530 struct dmar_atsr_unit
*atsr
, *atsr_n
;
532 list_for_each_entry_safe(rmrr
, rmrr_n
, &dmar_rmrr_units
, list
) {
533 ret
= rmrr_parse_dev(rmrr
);
538 list_for_each_entry_safe(atsr
, atsr_n
, &dmar_atsr_units
, list
) {
539 ret
= atsr_parse_dev(atsr
);
550 int __init
dmar_table_init(void)
552 static int dmar_table_initialized
;
555 if (dmar_table_initialized
)
558 dmar_table_initialized
= 1;
560 ret
= parse_dmar_table();
563 printk(KERN_INFO PREFIX
"parse DMAR table failure.\n");
567 if (list_empty(&dmar_drhd_units
)) {
568 printk(KERN_INFO PREFIX
"No DMAR devices found\n");
573 if (list_empty(&dmar_rmrr_units
))
574 printk(KERN_INFO PREFIX
"No RMRR found\n");
576 if (list_empty(&dmar_atsr_units
))
577 printk(KERN_INFO PREFIX
"No ATSR found\n");
580 #ifdef CONFIG_INTR_REMAP
581 parse_ioapics_under_ir();
586 void __init
detect_intel_iommu(void)
590 ret
= dmar_table_detect();
593 #ifdef CONFIG_INTR_REMAP
594 struct acpi_table_dmar
*dmar
;
596 * for now we will disable dma-remapping when interrupt
597 * remapping is enabled.
598 * When support for queued invalidation for IOTLB invalidation
599 * is added, we will not need this any more.
601 dmar
= (struct acpi_table_dmar
*) dmar_tbl
;
602 if (ret
&& cpu_has_x2apic
&& dmar
->flags
& 0x1)
604 "Queued invalidation will be enabled to support "
605 "x2apic and Intr-remapping.\n");
608 if (ret
&& !no_iommu
&& !iommu_detected
&& !swiotlb
&&
613 early_acpi_os_unmap_memory(dmar_tbl
, dmar_tbl_size
);
618 int alloc_iommu(struct dmar_drhd_unit
*drhd
)
620 struct intel_iommu
*iommu
;
623 static int iommu_allocated
= 0;
627 iommu
= kzalloc(sizeof(*iommu
), GFP_KERNEL
);
631 iommu
->seq_id
= iommu_allocated
++;
632 sprintf (iommu
->name
, "dmar%d", iommu
->seq_id
);
634 iommu
->reg
= ioremap(drhd
->reg_base_addr
, VTD_PAGE_SIZE
);
636 printk(KERN_ERR
"IOMMU: can't map the region\n");
639 iommu
->cap
= dmar_readq(iommu
->reg
+ DMAR_CAP_REG
);
640 iommu
->ecap
= dmar_readq(iommu
->reg
+ DMAR_ECAP_REG
);
643 agaw
= iommu_calculate_agaw(iommu
);
646 "Cannot get a valid agaw for iommu (seq_id = %d)\n",
650 msagaw
= iommu_calculate_max_sagaw(iommu
);
653 "Cannot get a valid max agaw for iommu (seq_id = %d)\n",
659 iommu
->msagaw
= msagaw
;
661 /* the registers might be more than one page */
662 map_size
= max_t(int, ecap_max_iotlb_offset(iommu
->ecap
),
663 cap_max_fault_reg_offset(iommu
->cap
));
664 map_size
= VTD_PAGE_ALIGN(map_size
);
665 if (map_size
> VTD_PAGE_SIZE
) {
667 iommu
->reg
= ioremap(drhd
->reg_base_addr
, map_size
);
669 printk(KERN_ERR
"IOMMU: can't map the region\n");
674 ver
= readl(iommu
->reg
+ DMAR_VER_REG
);
675 pr_debug("IOMMU %llx: ver %d:%d cap %llx ecap %llx\n",
676 (unsigned long long)drhd
->reg_base_addr
,
677 DMAR_VER_MAJOR(ver
), DMAR_VER_MINOR(ver
),
678 (unsigned long long)iommu
->cap
,
679 (unsigned long long)iommu
->ecap
);
681 spin_lock_init(&iommu
->register_lock
);
690 void free_iommu(struct intel_iommu
*iommu
)
696 free_dmar_iommu(iommu
);
705 * Reclaim all the submitted descriptors which have completed its work.
707 static inline void reclaim_free_desc(struct q_inval
*qi
)
709 while (qi
->desc_status
[qi
->free_tail
] == QI_DONE
||
710 qi
->desc_status
[qi
->free_tail
] == QI_ABORT
) {
711 qi
->desc_status
[qi
->free_tail
] = QI_FREE
;
712 qi
->free_tail
= (qi
->free_tail
+ 1) % QI_LENGTH
;
717 static int qi_check_fault(struct intel_iommu
*iommu
, int index
)
721 struct q_inval
*qi
= iommu
->qi
;
722 int wait_index
= (index
+ 1) % QI_LENGTH
;
724 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
727 fault
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
730 * If IQE happens, the head points to the descriptor associated
731 * with the error. No new descriptors are fetched until the IQE
734 if (fault
& DMA_FSTS_IQE
) {
735 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
736 if ((head
>> DMAR_IQ_SHIFT
) == index
) {
737 printk(KERN_ERR
"VT-d detected invalid descriptor: "
738 "low=%llx, high=%llx\n",
739 (unsigned long long)qi
->desc
[index
].low
,
740 (unsigned long long)qi
->desc
[index
].high
);
741 memcpy(&qi
->desc
[index
], &qi
->desc
[wait_index
],
742 sizeof(struct qi_desc
));
743 __iommu_flush_cache(iommu
, &qi
->desc
[index
],
744 sizeof(struct qi_desc
));
745 writel(DMA_FSTS_IQE
, iommu
->reg
+ DMAR_FSTS_REG
);
751 * If ITE happens, all pending wait_desc commands are aborted.
752 * No new descriptors are fetched until the ITE is cleared.
754 if (fault
& DMA_FSTS_ITE
) {
755 head
= readl(iommu
->reg
+ DMAR_IQH_REG
);
756 head
= ((head
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
758 tail
= readl(iommu
->reg
+ DMAR_IQT_REG
);
759 tail
= ((tail
>> DMAR_IQ_SHIFT
) - 1 + QI_LENGTH
) % QI_LENGTH
;
761 writel(DMA_FSTS_ITE
, iommu
->reg
+ DMAR_FSTS_REG
);
764 if (qi
->desc_status
[head
] == QI_IN_USE
)
765 qi
->desc_status
[head
] = QI_ABORT
;
766 head
= (head
- 2 + QI_LENGTH
) % QI_LENGTH
;
767 } while (head
!= tail
);
769 if (qi
->desc_status
[wait_index
] == QI_ABORT
)
773 if (fault
& DMA_FSTS_ICE
)
774 writel(DMA_FSTS_ICE
, iommu
->reg
+ DMAR_FSTS_REG
);
780 * Submit the queued invalidation descriptor to the remapping
781 * hardware unit and wait for its completion.
783 int qi_submit_sync(struct qi_desc
*desc
, struct intel_iommu
*iommu
)
786 struct q_inval
*qi
= iommu
->qi
;
787 struct qi_desc
*hw
, wait_desc
;
788 int wait_index
, index
;
799 spin_lock_irqsave(&qi
->q_lock
, flags
);
800 while (qi
->free_cnt
< 3) {
801 spin_unlock_irqrestore(&qi
->q_lock
, flags
);
803 spin_lock_irqsave(&qi
->q_lock
, flags
);
806 index
= qi
->free_head
;
807 wait_index
= (index
+ 1) % QI_LENGTH
;
809 qi
->desc_status
[index
] = qi
->desc_status
[wait_index
] = QI_IN_USE
;
813 wait_desc
.low
= QI_IWD_STATUS_DATA(QI_DONE
) |
814 QI_IWD_STATUS_WRITE
| QI_IWD_TYPE
;
815 wait_desc
.high
= virt_to_phys(&qi
->desc_status
[wait_index
]);
817 hw
[wait_index
] = wait_desc
;
819 __iommu_flush_cache(iommu
, &hw
[index
], sizeof(struct qi_desc
));
820 __iommu_flush_cache(iommu
, &hw
[wait_index
], sizeof(struct qi_desc
));
822 qi
->free_head
= (qi
->free_head
+ 2) % QI_LENGTH
;
826 * update the HW tail register indicating the presence of
829 writel(qi
->free_head
<< DMAR_IQ_SHIFT
, iommu
->reg
+ DMAR_IQT_REG
);
831 while (qi
->desc_status
[wait_index
] != QI_DONE
) {
833 * We will leave the interrupts disabled, to prevent interrupt
834 * context to queue another cmd while a cmd is already submitted
835 * and waiting for completion on this cpu. This is to avoid
836 * a deadlock where the interrupt context can wait indefinitely
837 * for free slots in the queue.
839 rc
= qi_check_fault(iommu
, index
);
843 spin_unlock(&qi
->q_lock
);
845 spin_lock(&qi
->q_lock
);
848 qi
->desc_status
[index
] = QI_DONE
;
850 reclaim_free_desc(qi
);
851 spin_unlock_irqrestore(&qi
->q_lock
, flags
);
860 * Flush the global interrupt entry cache.
862 void qi_global_iec(struct intel_iommu
*iommu
)
866 desc
.low
= QI_IEC_TYPE
;
869 /* should never fail */
870 qi_submit_sync(&desc
, iommu
);
873 void qi_flush_context(struct intel_iommu
*iommu
, u16 did
, u16 sid
, u8 fm
,
878 desc
.low
= QI_CC_FM(fm
) | QI_CC_SID(sid
) | QI_CC_DID(did
)
879 | QI_CC_GRAN(type
) | QI_CC_TYPE
;
882 qi_submit_sync(&desc
, iommu
);
885 void qi_flush_iotlb(struct intel_iommu
*iommu
, u16 did
, u64 addr
,
886 unsigned int size_order
, u64 type
)
893 if (cap_write_drain(iommu
->cap
))
896 if (cap_read_drain(iommu
->cap
))
899 desc
.low
= QI_IOTLB_DID(did
) | QI_IOTLB_DR(dr
) | QI_IOTLB_DW(dw
)
900 | QI_IOTLB_GRAN(type
) | QI_IOTLB_TYPE
;
901 desc
.high
= QI_IOTLB_ADDR(addr
) | QI_IOTLB_IH(ih
)
902 | QI_IOTLB_AM(size_order
);
904 qi_submit_sync(&desc
, iommu
);
907 void qi_flush_dev_iotlb(struct intel_iommu
*iommu
, u16 sid
, u16 qdep
,
908 u64 addr
, unsigned mask
)
913 BUG_ON(addr
& ((1 << (VTD_PAGE_SHIFT
+ mask
)) - 1));
914 addr
|= (1 << (VTD_PAGE_SHIFT
+ mask
- 1)) - 1;
915 desc
.high
= QI_DEV_IOTLB_ADDR(addr
) | QI_DEV_IOTLB_SIZE
;
917 desc
.high
= QI_DEV_IOTLB_ADDR(addr
);
919 if (qdep
>= QI_DEV_IOTLB_MAX_INVS
)
922 desc
.low
= QI_DEV_IOTLB_SID(sid
) | QI_DEV_IOTLB_QDEP(qdep
) |
925 qi_submit_sync(&desc
, iommu
);
929 * Disable Queued Invalidation interface.
931 void dmar_disable_qi(struct intel_iommu
*iommu
)
935 cycles_t start_time
= get_cycles();
937 if (!ecap_qis(iommu
->ecap
))
940 spin_lock_irqsave(&iommu
->register_lock
, flags
);
942 sts
= dmar_readq(iommu
->reg
+ DMAR_GSTS_REG
);
943 if (!(sts
& DMA_GSTS_QIES
))
947 * Give a chance to HW to complete the pending invalidation requests.
949 while ((readl(iommu
->reg
+ DMAR_IQT_REG
) !=
950 readl(iommu
->reg
+ DMAR_IQH_REG
)) &&
951 (DMAR_OPERATION_TIMEOUT
> (get_cycles() - start_time
)))
954 iommu
->gcmd
&= ~DMA_GCMD_QIE
;
955 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
957 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
,
958 !(sts
& DMA_GSTS_QIES
), sts
);
960 spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
964 * Enable queued invalidation.
966 static void __dmar_enable_qi(struct intel_iommu
*iommu
)
970 struct q_inval
*qi
= iommu
->qi
;
972 qi
->free_head
= qi
->free_tail
= 0;
973 qi
->free_cnt
= QI_LENGTH
;
975 spin_lock_irqsave(&iommu
->register_lock
, flags
);
977 /* write zero to the tail reg */
978 writel(0, iommu
->reg
+ DMAR_IQT_REG
);
980 dmar_writeq(iommu
->reg
+ DMAR_IQA_REG
, virt_to_phys(qi
->desc
));
982 iommu
->gcmd
|= DMA_GCMD_QIE
;
983 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
985 /* Make sure hardware complete it */
986 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
, readl
, (sts
& DMA_GSTS_QIES
), sts
);
988 spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
992 * Enable Queued Invalidation interface. This is a must to support
993 * interrupt-remapping. Also used by DMA-remapping, which replaces
994 * register based IOTLB invalidation.
996 int dmar_enable_qi(struct intel_iommu
*iommu
)
1000 if (!ecap_qis(iommu
->ecap
))
1004 * queued invalidation is already setup and enabled.
1009 iommu
->qi
= kmalloc(sizeof(*qi
), GFP_ATOMIC
);
1015 qi
->desc
= (void *)(get_zeroed_page(GFP_ATOMIC
));
1022 qi
->desc_status
= kmalloc(QI_LENGTH
* sizeof(int), GFP_ATOMIC
);
1023 if (!qi
->desc_status
) {
1024 free_page((unsigned long) qi
->desc
);
1030 qi
->free_head
= qi
->free_tail
= 0;
1031 qi
->free_cnt
= QI_LENGTH
;
1033 spin_lock_init(&qi
->q_lock
);
1035 __dmar_enable_qi(iommu
);
1040 /* iommu interrupt handling. Most stuff are MSI-like. */
1048 static const char *dma_remap_fault_reasons
[] =
1051 "Present bit in root entry is clear",
1052 "Present bit in context entry is clear",
1053 "Invalid context entry",
1054 "Access beyond MGAW",
1055 "PTE Write access is not set",
1056 "PTE Read access is not set",
1057 "Next page table ptr is invalid",
1058 "Root table address invalid",
1059 "Context table ptr is invalid",
1060 "non-zero reserved fields in RTP",
1061 "non-zero reserved fields in CTP",
1062 "non-zero reserved fields in PTE",
1065 static const char *intr_remap_fault_reasons
[] =
1067 "Detected reserved fields in the decoded interrupt-remapped request",
1068 "Interrupt index exceeded the interrupt-remapping table size",
1069 "Present field in the IRTE entry is clear",
1070 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1071 "Detected reserved fields in the IRTE entry",
1072 "Blocked a compatibility format interrupt request",
1073 "Blocked an interrupt request due to source-id verification failure",
1076 #define MAX_FAULT_REASON_IDX (ARRAY_SIZE(fault_reason_strings) - 1)
1078 const char *dmar_get_fault_reason(u8 fault_reason
, int *fault_type
)
1080 if (fault_reason
>= 0x20 && (fault_reason
<= 0x20 +
1081 ARRAY_SIZE(intr_remap_fault_reasons
))) {
1082 *fault_type
= INTR_REMAP
;
1083 return intr_remap_fault_reasons
[fault_reason
- 0x20];
1084 } else if (fault_reason
< ARRAY_SIZE(dma_remap_fault_reasons
)) {
1085 *fault_type
= DMA_REMAP
;
1086 return dma_remap_fault_reasons
[fault_reason
];
1088 *fault_type
= UNKNOWN
;
1093 void dmar_msi_unmask(unsigned int irq
)
1095 struct intel_iommu
*iommu
= get_irq_data(irq
);
1099 spin_lock_irqsave(&iommu
->register_lock
, flag
);
1100 writel(0, iommu
->reg
+ DMAR_FECTL_REG
);
1101 /* Read a reg to force flush the post write */
1102 readl(iommu
->reg
+ DMAR_FECTL_REG
);
1103 spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1106 void dmar_msi_mask(unsigned int irq
)
1109 struct intel_iommu
*iommu
= get_irq_data(irq
);
1112 spin_lock_irqsave(&iommu
->register_lock
, flag
);
1113 writel(DMA_FECTL_IM
, iommu
->reg
+ DMAR_FECTL_REG
);
1114 /* Read a reg to force flush the post write */
1115 readl(iommu
->reg
+ DMAR_FECTL_REG
);
1116 spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1119 void dmar_msi_write(int irq
, struct msi_msg
*msg
)
1121 struct intel_iommu
*iommu
= get_irq_data(irq
);
1124 spin_lock_irqsave(&iommu
->register_lock
, flag
);
1125 writel(msg
->data
, iommu
->reg
+ DMAR_FEDATA_REG
);
1126 writel(msg
->address_lo
, iommu
->reg
+ DMAR_FEADDR_REG
);
1127 writel(msg
->address_hi
, iommu
->reg
+ DMAR_FEUADDR_REG
);
1128 spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1131 void dmar_msi_read(int irq
, struct msi_msg
*msg
)
1133 struct intel_iommu
*iommu
= get_irq_data(irq
);
1136 spin_lock_irqsave(&iommu
->register_lock
, flag
);
1137 msg
->data
= readl(iommu
->reg
+ DMAR_FEDATA_REG
);
1138 msg
->address_lo
= readl(iommu
->reg
+ DMAR_FEADDR_REG
);
1139 msg
->address_hi
= readl(iommu
->reg
+ DMAR_FEUADDR_REG
);
1140 spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1143 static int dmar_fault_do_one(struct intel_iommu
*iommu
, int type
,
1144 u8 fault_reason
, u16 source_id
, unsigned long long addr
)
1149 reason
= dmar_get_fault_reason(fault_reason
, &fault_type
);
1151 if (fault_type
== INTR_REMAP
)
1152 printk(KERN_ERR
"INTR-REMAP: Request device [[%02x:%02x.%d] "
1153 "fault index %llx\n"
1154 "INTR-REMAP:[fault reason %02d] %s\n",
1155 (source_id
>> 8), PCI_SLOT(source_id
& 0xFF),
1156 PCI_FUNC(source_id
& 0xFF), addr
>> 48,
1157 fault_reason
, reason
);
1160 "DMAR:[%s] Request device [%02x:%02x.%d] "
1161 "fault addr %llx \n"
1162 "DMAR:[fault reason %02d] %s\n",
1163 (type
? "DMA Read" : "DMA Write"),
1164 (source_id
>> 8), PCI_SLOT(source_id
& 0xFF),
1165 PCI_FUNC(source_id
& 0xFF), addr
, fault_reason
, reason
);
1169 #define PRIMARY_FAULT_REG_LEN (16)
1170 irqreturn_t
dmar_fault(int irq
, void *dev_id
)
1172 struct intel_iommu
*iommu
= dev_id
;
1173 int reg
, fault_index
;
1177 spin_lock_irqsave(&iommu
->register_lock
, flag
);
1178 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1180 printk(KERN_ERR
"DRHD: handling fault status reg %x\n",
1183 /* TBD: ignore advanced fault log currently */
1184 if (!(fault_status
& DMA_FSTS_PPF
))
1187 fault_index
= dma_fsts_fault_record_index(fault_status
);
1188 reg
= cap_fault_reg_offset(iommu
->cap
);
1196 /* highest 32 bits */
1197 data
= readl(iommu
->reg
+ reg
+
1198 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1199 if (!(data
& DMA_FRCD_F
))
1202 fault_reason
= dma_frcd_fault_reason(data
);
1203 type
= dma_frcd_type(data
);
1205 data
= readl(iommu
->reg
+ reg
+
1206 fault_index
* PRIMARY_FAULT_REG_LEN
+ 8);
1207 source_id
= dma_frcd_source_id(data
);
1209 guest_addr
= dmar_readq(iommu
->reg
+ reg
+
1210 fault_index
* PRIMARY_FAULT_REG_LEN
);
1211 guest_addr
= dma_frcd_page_addr(guest_addr
);
1212 /* clear the fault */
1213 writel(DMA_FRCD_F
, iommu
->reg
+ reg
+
1214 fault_index
* PRIMARY_FAULT_REG_LEN
+ 12);
1216 spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1218 dmar_fault_do_one(iommu
, type
, fault_reason
,
1219 source_id
, guest_addr
);
1222 if (fault_index
> cap_num_fault_regs(iommu
->cap
))
1224 spin_lock_irqsave(&iommu
->register_lock
, flag
);
1227 /* clear all the other faults */
1228 fault_status
= readl(iommu
->reg
+ DMAR_FSTS_REG
);
1229 writel(fault_status
, iommu
->reg
+ DMAR_FSTS_REG
);
1231 spin_unlock_irqrestore(&iommu
->register_lock
, flag
);
1235 int dmar_set_interrupt(struct intel_iommu
*iommu
)
1240 * Check if the fault interrupt is already initialized.
1247 printk(KERN_ERR
"IOMMU: no free vectors\n");
1251 set_irq_data(irq
, iommu
);
1254 ret
= arch_setup_dmar_msi(irq
);
1256 set_irq_data(irq
, NULL
);
1262 ret
= request_irq(irq
, dmar_fault
, 0, iommu
->name
, iommu
);
1264 printk(KERN_ERR
"IOMMU: can't request irq\n");
1268 int __init
enable_drhd_fault_handling(void)
1270 struct dmar_drhd_unit
*drhd
;
1273 * Enable fault control interrupt.
1275 for_each_drhd_unit(drhd
) {
1277 struct intel_iommu
*iommu
= drhd
->iommu
;
1278 ret
= dmar_set_interrupt(iommu
);
1281 printk(KERN_ERR
"DRHD %Lx: failed to enable fault, "
1282 " interrupt, ret %d\n",
1283 (unsigned long long)drhd
->reg_base_addr
, ret
);
1292 * Re-enable Queued Invalidation interface.
1294 int dmar_reenable_qi(struct intel_iommu
*iommu
)
1296 if (!ecap_qis(iommu
->ecap
))
1303 * First disable queued invalidation.
1305 dmar_disable_qi(iommu
);
1307 * Then enable queued invalidation again. Since there is no pending
1308 * invalidation requests now, it's safe to re-enable queued
1311 __dmar_enable_qi(iommu
);