1 #include <linux/interrupt.h>
2 #include <linux/dmar.h>
3 #include <linux/spinlock.h>
4 #include <linux/slab.h>
5 #include <linux/jiffies.h>
6 #include <linux/hpet.h>
9 #include <asm/io_apic.h>
12 #include <linux/intel-iommu.h>
13 #include "intr_remapping.h"
14 #include <acpi/acpi.h>
15 #include <asm/pci-direct.h>
17 static struct ioapic_scope ir_ioapic
[MAX_IO_APICS
];
18 static struct hpet_scope ir_hpet
[MAX_HPET_TBS
];
19 static int ir_ioapic_num
, ir_hpet_num
;
20 int intr_remapping_enabled
;
22 static int disable_intremap
;
23 static int disable_sourceid_checking
;
24 static int no_x2apic_optout
;
26 static __init
int setup_nointremap(char *str
)
31 early_param("nointremap", setup_nointremap
);
33 static __init
int setup_intremap(char *str
)
39 if (!strncmp(str
, "on", 2))
41 else if (!strncmp(str
, "off", 3))
43 else if (!strncmp(str
, "nosid", 5))
44 disable_sourceid_checking
= 1;
45 else if (!strncmp(str
, "no_x2apic_optout", 16))
48 str
+= strcspn(str
, ",");
55 early_param("intremap", setup_intremap
);
57 static DEFINE_RAW_SPINLOCK(irq_2_ir_lock
);
59 static struct irq_2_iommu
*irq_2_iommu(unsigned int irq
)
61 struct irq_cfg
*cfg
= irq_get_chip_data(irq
);
62 return cfg
? &cfg
->irq_2_iommu
: NULL
;
65 int get_irte(int irq
, struct irte
*entry
)
67 struct irq_2_iommu
*irq_iommu
= irq_2_iommu(irq
);
71 if (!entry
|| !irq_iommu
)
74 raw_spin_lock_irqsave(&irq_2_ir_lock
, flags
);
76 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
77 *entry
= *(irq_iommu
->iommu
->ir_table
->base
+ index
);
79 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
83 int alloc_irte(struct intel_iommu
*iommu
, int irq
, u16 count
)
85 struct ir_table
*table
= iommu
->ir_table
;
86 struct irq_2_iommu
*irq_iommu
= irq_2_iommu(irq
);
87 u16 index
, start_index
;
88 unsigned int mask
= 0;
92 if (!count
|| !irq_iommu
)
96 * start the IRTE search from index 0.
98 index
= start_index
= 0;
101 count
= __roundup_pow_of_two(count
);
105 if (mask
> ecap_max_handle_mask(iommu
->ecap
)) {
107 "Requested mask %x exceeds the max invalidation handle"
108 " mask value %Lx\n", mask
,
109 ecap_max_handle_mask(iommu
->ecap
));
113 raw_spin_lock_irqsave(&irq_2_ir_lock
, flags
);
115 for (i
= index
; i
< index
+ count
; i
++)
116 if (table
->base
[i
].present
)
118 /* empty index found */
119 if (i
== index
+ count
)
122 index
= (index
+ count
) % INTR_REMAP_TABLE_ENTRIES
;
124 if (index
== start_index
) {
125 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
126 printk(KERN_ERR
"can't allocate an IRTE\n");
131 for (i
= index
; i
< index
+ count
; i
++)
132 table
->base
[i
].present
= 1;
134 irq_iommu
->iommu
= iommu
;
135 irq_iommu
->irte_index
= index
;
136 irq_iommu
->sub_handle
= 0;
137 irq_iommu
->irte_mask
= mask
;
139 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
144 static int qi_flush_iec(struct intel_iommu
*iommu
, int index
, int mask
)
148 desc
.low
= QI_IEC_IIDEX(index
) | QI_IEC_TYPE
| QI_IEC_IM(mask
)
152 return qi_submit_sync(&desc
, iommu
);
155 int map_irq_to_irte_handle(int irq
, u16
*sub_handle
)
157 struct irq_2_iommu
*irq_iommu
= irq_2_iommu(irq
);
164 raw_spin_lock_irqsave(&irq_2_ir_lock
, flags
);
165 *sub_handle
= irq_iommu
->sub_handle
;
166 index
= irq_iommu
->irte_index
;
167 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
171 int set_irte_irq(int irq
, struct intel_iommu
*iommu
, u16 index
, u16 subhandle
)
173 struct irq_2_iommu
*irq_iommu
= irq_2_iommu(irq
);
179 raw_spin_lock_irqsave(&irq_2_ir_lock
, flags
);
181 irq_iommu
->iommu
= iommu
;
182 irq_iommu
->irte_index
= index
;
183 irq_iommu
->sub_handle
= subhandle
;
184 irq_iommu
->irte_mask
= 0;
186 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
191 int modify_irte(int irq
, struct irte
*irte_modified
)
193 struct irq_2_iommu
*irq_iommu
= irq_2_iommu(irq
);
194 struct intel_iommu
*iommu
;
202 raw_spin_lock_irqsave(&irq_2_ir_lock
, flags
);
204 iommu
= irq_iommu
->iommu
;
206 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
207 irte
= &iommu
->ir_table
->base
[index
];
209 set_64bit(&irte
->low
, irte_modified
->low
);
210 set_64bit(&irte
->high
, irte_modified
->high
);
211 __iommu_flush_cache(iommu
, irte
, sizeof(*irte
));
213 rc
= qi_flush_iec(iommu
, index
, 0);
214 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
219 struct intel_iommu
*map_hpet_to_ir(u8 hpet_id
)
223 for (i
= 0; i
< MAX_HPET_TBS
; i
++)
224 if (ir_hpet
[i
].id
== hpet_id
)
225 return ir_hpet
[i
].iommu
;
229 struct intel_iommu
*map_ioapic_to_ir(int apic
)
233 for (i
= 0; i
< MAX_IO_APICS
; i
++)
234 if (ir_ioapic
[i
].id
== apic
)
235 return ir_ioapic
[i
].iommu
;
239 struct intel_iommu
*map_dev_to_ir(struct pci_dev
*dev
)
241 struct dmar_drhd_unit
*drhd
;
243 drhd
= dmar_find_matched_drhd_unit(dev
);
250 static int clear_entries(struct irq_2_iommu
*irq_iommu
)
252 struct irte
*start
, *entry
, *end
;
253 struct intel_iommu
*iommu
;
256 if (irq_iommu
->sub_handle
)
259 iommu
= irq_iommu
->iommu
;
260 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
262 start
= iommu
->ir_table
->base
+ index
;
263 end
= start
+ (1 << irq_iommu
->irte_mask
);
265 for (entry
= start
; entry
< end
; entry
++) {
266 set_64bit(&entry
->low
, 0);
267 set_64bit(&entry
->high
, 0);
270 return qi_flush_iec(iommu
, index
, irq_iommu
->irte_mask
);
273 int free_irte(int irq
)
275 struct irq_2_iommu
*irq_iommu
= irq_2_iommu(irq
);
282 raw_spin_lock_irqsave(&irq_2_ir_lock
, flags
);
284 rc
= clear_entries(irq_iommu
);
286 irq_iommu
->iommu
= NULL
;
287 irq_iommu
->irte_index
= 0;
288 irq_iommu
->sub_handle
= 0;
289 irq_iommu
->irte_mask
= 0;
291 raw_spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
297 * source validation type
299 #define SVT_NO_VERIFY 0x0 /* no verification is required */
300 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fields */
301 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
304 * source-id qualifier
306 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
307 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
308 * the third least significant bit
310 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
311 * the second and third least significant bits
313 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
314 * the least three significant bits
318 * set SVT, SQ and SID fields of irte to verify
319 * source ids of interrupt requests
321 static void set_irte_sid(struct irte
*irte
, unsigned int svt
,
322 unsigned int sq
, unsigned int sid
)
324 if (disable_sourceid_checking
)
331 int set_ioapic_sid(struct irte
*irte
, int apic
)
339 for (i
= 0; i
< MAX_IO_APICS
; i
++) {
340 if (ir_ioapic
[i
].id
== apic
) {
341 sid
= (ir_ioapic
[i
].bus
<< 8) | ir_ioapic
[i
].devfn
;
347 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic
);
351 set_irte_sid(irte
, 1, 0, sid
);
356 int set_hpet_sid(struct irte
*irte
, u8 id
)
364 for (i
= 0; i
< MAX_HPET_TBS
; i
++) {
365 if (ir_hpet
[i
].id
== id
) {
366 sid
= (ir_hpet
[i
].bus
<< 8) | ir_hpet
[i
].devfn
;
372 pr_warning("Failed to set source-id of HPET block (%d)\n", id
);
377 * Should really use SQ_ALL_16. Some platforms are broken.
378 * While we figure out the right quirks for these broken platforms, use
379 * SQ_13_IGNORE_3 for now.
381 set_irte_sid(irte
, SVT_VERIFY_SID_SQ
, SQ_13_IGNORE_3
, sid
);
386 int set_msi_sid(struct irte
*irte
, struct pci_dev
*dev
)
388 struct pci_dev
*bridge
;
393 /* PCIe device or Root Complex integrated PCI device */
394 if (pci_is_pcie(dev
) || !dev
->bus
->parent
) {
395 set_irte_sid(irte
, SVT_VERIFY_SID_SQ
, SQ_ALL_16
,
396 (dev
->bus
->number
<< 8) | dev
->devfn
);
400 bridge
= pci_find_upstream_pcie_bridge(dev
);
402 if (pci_is_pcie(bridge
))/* this is a PCIe-to-PCI/PCIX bridge */
403 set_irte_sid(irte
, SVT_VERIFY_BUS
, SQ_ALL_16
,
404 (bridge
->bus
->number
<< 8) | dev
->bus
->number
);
405 else /* this is a legacy PCI bridge */
406 set_irte_sid(irte
, SVT_VERIFY_SID_SQ
, SQ_ALL_16
,
407 (bridge
->bus
->number
<< 8) | bridge
->devfn
);
413 static void iommu_set_intr_remapping(struct intel_iommu
*iommu
, int mode
)
419 addr
= virt_to_phys((void *)iommu
->ir_table
->base
);
421 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
423 dmar_writeq(iommu
->reg
+ DMAR_IRTA_REG
,
424 (addr
) | IR_X2APIC_MODE(mode
) | INTR_REMAP_TABLE_REG_SIZE
);
426 /* Set interrupt-remapping table pointer */
427 iommu
->gcmd
|= DMA_GCMD_SIRTP
;
428 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
430 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
,
431 readl
, (sts
& DMA_GSTS_IRTPS
), sts
);
432 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
435 * global invalidation of interrupt entry cache before enabling
436 * interrupt-remapping.
438 qi_global_iec(iommu
);
440 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
442 /* Enable interrupt-remapping */
443 iommu
->gcmd
|= DMA_GCMD_IRE
;
444 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
446 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
,
447 readl
, (sts
& DMA_GSTS_IRES
), sts
);
449 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
453 static int setup_intr_remapping(struct intel_iommu
*iommu
, int mode
)
455 struct ir_table
*ir_table
;
458 ir_table
= iommu
->ir_table
= kzalloc(sizeof(struct ir_table
),
461 if (!iommu
->ir_table
)
464 pages
= alloc_pages_node(iommu
->node
, GFP_ATOMIC
| __GFP_ZERO
,
465 INTR_REMAP_PAGE_ORDER
);
468 printk(KERN_ERR
"failed to allocate pages of order %d\n",
469 INTR_REMAP_PAGE_ORDER
);
470 kfree(iommu
->ir_table
);
474 ir_table
->base
= page_address(pages
);
476 iommu_set_intr_remapping(iommu
, mode
);
481 * Disable Interrupt Remapping.
483 static void iommu_disable_intr_remapping(struct intel_iommu
*iommu
)
488 if (!ecap_ir_support(iommu
->ecap
))
492 * global invalidation of interrupt entry cache before disabling
493 * interrupt-remapping.
495 qi_global_iec(iommu
);
497 raw_spin_lock_irqsave(&iommu
->register_lock
, flags
);
499 sts
= dmar_readq(iommu
->reg
+ DMAR_GSTS_REG
);
500 if (!(sts
& DMA_GSTS_IRES
))
503 iommu
->gcmd
&= ~DMA_GCMD_IRE
;
504 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
506 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
,
507 readl
, !(sts
& DMA_GSTS_IRES
), sts
);
510 raw_spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
513 static int __init
dmar_x2apic_optout(void)
515 struct acpi_table_dmar
*dmar
;
516 dmar
= (struct acpi_table_dmar
*)dmar_tbl
;
517 if (!dmar
|| no_x2apic_optout
)
519 return dmar
->flags
& DMAR_X2APIC_OPT_OUT
;
522 int __init
intr_remapping_supported(void)
524 struct dmar_drhd_unit
*drhd
;
526 if (disable_intremap
)
529 if (!dmar_ir_support())
532 for_each_drhd_unit(drhd
) {
533 struct intel_iommu
*iommu
= drhd
->iommu
;
535 if (!ecap_ir_support(iommu
->ecap
))
542 int __init
enable_intr_remapping(void)
544 struct dmar_drhd_unit
*drhd
;
548 if (parse_ioapics_under_ir() != 1) {
549 printk(KERN_INFO
"Not enable interrupt remapping\n");
553 if (x2apic_supported()) {
554 eim
= !dmar_x2apic_optout();
555 WARN(!eim
, KERN_WARNING
556 "Your BIOS is broken and requested that x2apic be disabled\n"
557 "This will leave your machine vulnerable to irq-injection attacks\n"
558 "Use 'intremap=no_x2apic_optout' to override BIOS request\n");
561 for_each_drhd_unit(drhd
) {
562 struct intel_iommu
*iommu
= drhd
->iommu
;
565 * If the queued invalidation is already initialized,
566 * shouldn't disable it.
572 * Clear previous faults.
574 dmar_fault(-1, iommu
);
577 * Disable intr remapping and queued invalidation, if already
578 * enabled prior to OS handover.
580 iommu_disable_intr_remapping(iommu
);
582 dmar_disable_qi(iommu
);
586 * check for the Interrupt-remapping support
588 for_each_drhd_unit(drhd
) {
589 struct intel_iommu
*iommu
= drhd
->iommu
;
591 if (!ecap_ir_support(iommu
->ecap
))
594 if (eim
&& !ecap_eim_support(iommu
->ecap
)) {
595 printk(KERN_INFO
"DRHD %Lx: EIM not supported by DRHD, "
596 " ecap %Lx\n", drhd
->reg_base_addr
, iommu
->ecap
);
602 * Enable queued invalidation for all the DRHD's.
604 for_each_drhd_unit(drhd
) {
606 struct intel_iommu
*iommu
= drhd
->iommu
;
607 ret
= dmar_enable_qi(iommu
);
610 printk(KERN_ERR
"DRHD %Lx: failed to enable queued, "
611 " invalidation, ecap %Lx, ret %d\n",
612 drhd
->reg_base_addr
, iommu
->ecap
, ret
);
618 * Setup Interrupt-remapping for all the DRHD's now.
620 for_each_drhd_unit(drhd
) {
621 struct intel_iommu
*iommu
= drhd
->iommu
;
623 if (!ecap_ir_support(iommu
->ecap
))
626 if (setup_intr_remapping(iommu
, eim
))
635 intr_remapping_enabled
= 1;
636 pr_info("Enabled IRQ remapping in %s mode\n", eim
? "x2apic" : "xapic");
638 return eim
? IRQ_REMAP_X2APIC_MODE
: IRQ_REMAP_XAPIC_MODE
;
642 * handle error condition gracefully here!
647 static void ir_parse_one_hpet_scope(struct acpi_dmar_device_scope
*scope
,
648 struct intel_iommu
*iommu
)
650 struct acpi_dmar_pci_path
*path
;
655 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
656 count
= (scope
->length
- sizeof(struct acpi_dmar_device_scope
))
657 / sizeof(struct acpi_dmar_pci_path
);
659 while (--count
> 0) {
661 * Access PCI directly due to the PCI
662 * subsystem isn't initialized yet.
664 bus
= read_pci_config_byte(bus
, path
->dev
, path
->fn
,
668 ir_hpet
[ir_hpet_num
].bus
= bus
;
669 ir_hpet
[ir_hpet_num
].devfn
= PCI_DEVFN(path
->dev
, path
->fn
);
670 ir_hpet
[ir_hpet_num
].iommu
= iommu
;
671 ir_hpet
[ir_hpet_num
].id
= scope
->enumeration_id
;
675 static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope
*scope
,
676 struct intel_iommu
*iommu
)
678 struct acpi_dmar_pci_path
*path
;
683 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
684 count
= (scope
->length
- sizeof(struct acpi_dmar_device_scope
))
685 / sizeof(struct acpi_dmar_pci_path
);
687 while (--count
> 0) {
689 * Access PCI directly due to the PCI
690 * subsystem isn't initialized yet.
692 bus
= read_pci_config_byte(bus
, path
->dev
, path
->fn
,
697 ir_ioapic
[ir_ioapic_num
].bus
= bus
;
698 ir_ioapic
[ir_ioapic_num
].devfn
= PCI_DEVFN(path
->dev
, path
->fn
);
699 ir_ioapic
[ir_ioapic_num
].iommu
= iommu
;
700 ir_ioapic
[ir_ioapic_num
].id
= scope
->enumeration_id
;
704 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header
*header
,
705 struct intel_iommu
*iommu
)
707 struct acpi_dmar_hardware_unit
*drhd
;
708 struct acpi_dmar_device_scope
*scope
;
711 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
713 start
= (void *)(drhd
+ 1);
714 end
= ((void *)drhd
) + header
->length
;
716 while (start
< end
) {
718 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_IOAPIC
) {
719 if (ir_ioapic_num
== MAX_IO_APICS
) {
720 printk(KERN_WARNING
"Exceeded Max IO APICS\n");
724 printk(KERN_INFO
"IOAPIC id %d under DRHD base "
725 " 0x%Lx IOMMU %d\n", scope
->enumeration_id
,
726 drhd
->address
, iommu
->seq_id
);
728 ir_parse_one_ioapic_scope(scope
, iommu
);
729 } else if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_HPET
) {
730 if (ir_hpet_num
== MAX_HPET_TBS
) {
731 printk(KERN_WARNING
"Exceeded Max HPET blocks\n");
735 printk(KERN_INFO
"HPET id %d under DRHD base"
736 " 0x%Lx\n", scope
->enumeration_id
,
739 ir_parse_one_hpet_scope(scope
, iommu
);
741 start
+= scope
->length
;
748 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
751 int __init
parse_ioapics_under_ir(void)
753 struct dmar_drhd_unit
*drhd
;
754 int ir_supported
= 0;
756 for_each_drhd_unit(drhd
) {
757 struct intel_iommu
*iommu
= drhd
->iommu
;
759 if (ecap_ir_support(iommu
->ecap
)) {
760 if (ir_parse_ioapic_hpet_scope(drhd
->hdr
, iommu
))
767 if (ir_supported
&& ir_ioapic_num
!= nr_ioapics
) {
769 "Not all IO-APIC's listed under remapping hardware\n");
776 int ir_dev_scope_init(void)
778 if (!intr_remapping_enabled
)
781 return dmar_dev_scope_init();
783 rootfs_initcall(ir_dev_scope_init
);
785 void disable_intr_remapping(void)
787 struct dmar_drhd_unit
*drhd
;
788 struct intel_iommu
*iommu
= NULL
;
791 * Disable Interrupt-remapping for all the DRHD's now.
793 for_each_iommu(iommu
, drhd
) {
794 if (!ecap_ir_support(iommu
->ecap
))
797 iommu_disable_intr_remapping(iommu
);
801 int reenable_intr_remapping(int eim
)
803 struct dmar_drhd_unit
*drhd
;
805 struct intel_iommu
*iommu
= NULL
;
807 for_each_iommu(iommu
, drhd
)
809 dmar_reenable_qi(iommu
);
812 * Setup Interrupt-remapping for all the DRHD's now.
814 for_each_iommu(iommu
, drhd
) {
815 if (!ecap_ir_support(iommu
->ecap
))
818 /* Set up interrupt remapping for iommu.*/
819 iommu_set_intr_remapping(iommu
, eim
);
830 * handle error condition gracefully here!