1 #include <linux/interrupt.h>
2 #include <linux/dmar.h>
3 #include <linux/spinlock.h>
4 #include <linux/jiffies.h>
7 #include <asm/io_apic.h>
10 #include <linux/intel-iommu.h>
11 #include "intr_remapping.h"
12 #include <acpi/acpi.h>
13 #include <asm/pci-direct.h>
16 static struct ioapic_scope ir_ioapic
[MAX_IO_APICS
];
17 static int ir_ioapic_num
;
18 int intr_remapping_enabled
;
20 static int disable_intremap
;
21 static __init
int setup_nointremap(char *str
)
26 early_param("nointremap", setup_nointremap
);
29 struct intel_iommu
*iommu
;
35 #ifdef CONFIG_GENERIC_HARDIRQS
36 static struct irq_2_iommu
*get_one_free_irq_2_iommu(int node
)
38 struct irq_2_iommu
*iommu
;
40 iommu
= kzalloc_node(sizeof(*iommu
), GFP_ATOMIC
, node
);
41 printk(KERN_DEBUG
"alloc irq_2_iommu on node %d\n", node
);
46 static struct irq_2_iommu
*irq_2_iommu(unsigned int irq
)
48 struct irq_desc
*desc
;
50 desc
= irq_to_desc(irq
);
52 if (WARN_ON_ONCE(!desc
))
55 return desc
->irq_2_iommu
;
58 static struct irq_2_iommu
*irq_2_iommu_alloc_node(unsigned int irq
, int node
)
60 struct irq_desc
*desc
;
61 struct irq_2_iommu
*irq_iommu
;
64 * alloc irq desc if not allocated already.
66 desc
= irq_to_desc_alloc_node(irq
, node
);
68 printk(KERN_INFO
"can not get irq_desc for %d\n", irq
);
72 irq_iommu
= desc
->irq_2_iommu
;
75 desc
->irq_2_iommu
= get_one_free_irq_2_iommu(node
);
77 return desc
->irq_2_iommu
;
80 static struct irq_2_iommu
*irq_2_iommu_alloc(unsigned int irq
)
82 return irq_2_iommu_alloc_node(irq
, cpu_to_node(boot_cpu_id
));
85 #else /* !CONFIG_SPARSE_IRQ */
87 static struct irq_2_iommu irq_2_iommuX
[NR_IRQS
];
89 static struct irq_2_iommu
*irq_2_iommu(unsigned int irq
)
92 return &irq_2_iommuX
[irq
];
96 static struct irq_2_iommu
*irq_2_iommu_alloc(unsigned int irq
)
98 return irq_2_iommu(irq
);
102 static DEFINE_SPINLOCK(irq_2_ir_lock
);
104 static struct irq_2_iommu
*valid_irq_2_iommu(unsigned int irq
)
106 struct irq_2_iommu
*irq_iommu
;
108 irq_iommu
= irq_2_iommu(irq
);
113 if (!irq_iommu
->iommu
)
119 int irq_remapped(int irq
)
121 return valid_irq_2_iommu(irq
) != NULL
;
124 int get_irte(int irq
, struct irte
*entry
)
127 struct irq_2_iommu
*irq_iommu
;
133 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
134 irq_iommu
= valid_irq_2_iommu(irq
);
136 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
140 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
141 *entry
= *(irq_iommu
->iommu
->ir_table
->base
+ index
);
143 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
147 int alloc_irte(struct intel_iommu
*iommu
, int irq
, u16 count
)
149 struct ir_table
*table
= iommu
->ir_table
;
150 struct irq_2_iommu
*irq_iommu
;
151 u16 index
, start_index
;
152 unsigned int mask
= 0;
159 #ifndef CONFIG_SPARSE_IRQ
160 /* protect irq_2_iommu_alloc later */
166 * start the IRTE search from index 0.
168 index
= start_index
= 0;
171 count
= __roundup_pow_of_two(count
);
175 if (mask
> ecap_max_handle_mask(iommu
->ecap
)) {
177 "Requested mask %x exceeds the max invalidation handle"
178 " mask value %Lx\n", mask
,
179 ecap_max_handle_mask(iommu
->ecap
));
183 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
185 for (i
= index
; i
< index
+ count
; i
++)
186 if (table
->base
[i
].present
)
188 /* empty index found */
189 if (i
== index
+ count
)
192 index
= (index
+ count
) % INTR_REMAP_TABLE_ENTRIES
;
194 if (index
== start_index
) {
195 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
196 printk(KERN_ERR
"can't allocate an IRTE\n");
201 for (i
= index
; i
< index
+ count
; i
++)
202 table
->base
[i
].present
= 1;
204 irq_iommu
= irq_2_iommu_alloc(irq
);
206 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
207 printk(KERN_ERR
"can't allocate irq_2_iommu\n");
211 irq_iommu
->iommu
= iommu
;
212 irq_iommu
->irte_index
= index
;
213 irq_iommu
->sub_handle
= 0;
214 irq_iommu
->irte_mask
= mask
;
216 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
221 static int qi_flush_iec(struct intel_iommu
*iommu
, int index
, int mask
)
225 desc
.low
= QI_IEC_IIDEX(index
) | QI_IEC_TYPE
| QI_IEC_IM(mask
)
229 return qi_submit_sync(&desc
, iommu
);
232 int map_irq_to_irte_handle(int irq
, u16
*sub_handle
)
235 struct irq_2_iommu
*irq_iommu
;
238 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
239 irq_iommu
= valid_irq_2_iommu(irq
);
241 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
245 *sub_handle
= irq_iommu
->sub_handle
;
246 index
= irq_iommu
->irte_index
;
247 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
251 int set_irte_irq(int irq
, struct intel_iommu
*iommu
, u16 index
, u16 subhandle
)
253 struct irq_2_iommu
*irq_iommu
;
256 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
258 irq_iommu
= irq_2_iommu_alloc(irq
);
261 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
262 printk(KERN_ERR
"can't allocate irq_2_iommu\n");
266 irq_iommu
->iommu
= iommu
;
267 irq_iommu
->irte_index
= index
;
268 irq_iommu
->sub_handle
= subhandle
;
269 irq_iommu
->irte_mask
= 0;
271 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
276 int clear_irte_irq(int irq
, struct intel_iommu
*iommu
, u16 index
)
278 struct irq_2_iommu
*irq_iommu
;
281 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
282 irq_iommu
= valid_irq_2_iommu(irq
);
284 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
288 irq_iommu
->iommu
= NULL
;
289 irq_iommu
->irte_index
= 0;
290 irq_iommu
->sub_handle
= 0;
291 irq_2_iommu(irq
)->irte_mask
= 0;
293 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
298 int modify_irte(int irq
, struct irte
*irte_modified
)
303 struct intel_iommu
*iommu
;
304 struct irq_2_iommu
*irq_iommu
;
307 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
308 irq_iommu
= valid_irq_2_iommu(irq
);
310 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
314 iommu
= irq_iommu
->iommu
;
316 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
317 irte
= &iommu
->ir_table
->base
[index
];
319 set_64bit((unsigned long *)&irte
->low
, irte_modified
->low
);
320 set_64bit((unsigned long *)&irte
->high
, irte_modified
->high
);
321 __iommu_flush_cache(iommu
, irte
, sizeof(*irte
));
323 rc
= qi_flush_iec(iommu
, index
, 0);
324 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
329 int flush_irte(int irq
)
333 struct intel_iommu
*iommu
;
334 struct irq_2_iommu
*irq_iommu
;
337 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
338 irq_iommu
= valid_irq_2_iommu(irq
);
340 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
344 iommu
= irq_iommu
->iommu
;
346 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
348 rc
= qi_flush_iec(iommu
, index
, irq_iommu
->irte_mask
);
349 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
354 struct intel_iommu
*map_ioapic_to_ir(int apic
)
358 for (i
= 0; i
< MAX_IO_APICS
; i
++)
359 if (ir_ioapic
[i
].id
== apic
)
360 return ir_ioapic
[i
].iommu
;
364 struct intel_iommu
*map_dev_to_ir(struct pci_dev
*dev
)
366 struct dmar_drhd_unit
*drhd
;
368 drhd
= dmar_find_matched_drhd_unit(dev
);
375 static int clear_entries(struct irq_2_iommu
*irq_iommu
)
377 struct irte
*start
, *entry
, *end
;
378 struct intel_iommu
*iommu
;
381 if (irq_iommu
->sub_handle
)
384 iommu
= irq_iommu
->iommu
;
385 index
= irq_iommu
->irte_index
+ irq_iommu
->sub_handle
;
387 start
= iommu
->ir_table
->base
+ index
;
388 end
= start
+ (1 << irq_iommu
->irte_mask
);
390 for (entry
= start
; entry
< end
; entry
++) {
391 set_64bit((unsigned long *)&entry
->low
, 0);
392 set_64bit((unsigned long *)&entry
->high
, 0);
395 return qi_flush_iec(iommu
, index
, irq_iommu
->irte_mask
);
398 int free_irte(int irq
)
401 struct irq_2_iommu
*irq_iommu
;
404 spin_lock_irqsave(&irq_2_ir_lock
, flags
);
405 irq_iommu
= valid_irq_2_iommu(irq
);
407 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
411 rc
= clear_entries(irq_iommu
);
413 irq_iommu
->iommu
= NULL
;
414 irq_iommu
->irte_index
= 0;
415 irq_iommu
->sub_handle
= 0;
416 irq_iommu
->irte_mask
= 0;
418 spin_unlock_irqrestore(&irq_2_ir_lock
, flags
);
424 * source validation type
426 #define SVT_NO_VERIFY 0x0 /* no verification is required */
427 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fiels */
428 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
431 * source-id qualifier
433 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
434 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
435 * the third least significant bit
437 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
438 * the second and third least significant bits
440 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
441 * the least three significant bits
445 * set SVT, SQ and SID fields of irte to verify
446 * source ids of interrupt requests
448 static void set_irte_sid(struct irte
*irte
, unsigned int svt
,
449 unsigned int sq
, unsigned int sid
)
456 int set_ioapic_sid(struct irte
*irte
, int apic
)
464 for (i
= 0; i
< MAX_IO_APICS
; i
++) {
465 if (ir_ioapic
[i
].id
== apic
) {
466 sid
= (ir_ioapic
[i
].bus
<< 8) | ir_ioapic
[i
].devfn
;
472 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic
);
476 set_irte_sid(irte
, 1, 0, sid
);
481 int set_msi_sid(struct irte
*irte
, struct pci_dev
*dev
)
483 struct pci_dev
*bridge
;
488 /* PCIe device or Root Complex integrated PCI device */
489 if (dev
->is_pcie
|| !dev
->bus
->parent
) {
490 set_irte_sid(irte
, SVT_VERIFY_SID_SQ
, SQ_ALL_16
,
491 (dev
->bus
->number
<< 8) | dev
->devfn
);
495 bridge
= pci_find_upstream_pcie_bridge(dev
);
497 if (bridge
->is_pcie
) /* this is a PCIE-to-PCI/PCIX bridge */
498 set_irte_sid(irte
, SVT_VERIFY_BUS
, SQ_ALL_16
,
499 (bridge
->bus
->number
<< 8) | dev
->bus
->number
);
500 else /* this is a legacy PCI bridge */
501 set_irte_sid(irte
, SVT_VERIFY_SID_SQ
, SQ_ALL_16
,
502 (bridge
->bus
->number
<< 8) | bridge
->devfn
);
508 static void iommu_set_intr_remapping(struct intel_iommu
*iommu
, int mode
)
514 addr
= virt_to_phys((void *)iommu
->ir_table
->base
);
516 spin_lock_irqsave(&iommu
->register_lock
, flags
);
518 dmar_writeq(iommu
->reg
+ DMAR_IRTA_REG
,
519 (addr
) | IR_X2APIC_MODE(mode
) | INTR_REMAP_TABLE_REG_SIZE
);
521 /* Set interrupt-remapping table pointer */
522 iommu
->gcmd
|= DMA_GCMD_SIRTP
;
523 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
525 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
,
526 readl
, (sts
& DMA_GSTS_IRTPS
), sts
);
527 spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
530 * global invalidation of interrupt entry cache before enabling
531 * interrupt-remapping.
533 qi_global_iec(iommu
);
535 spin_lock_irqsave(&iommu
->register_lock
, flags
);
537 /* Enable interrupt-remapping */
538 iommu
->gcmd
|= DMA_GCMD_IRE
;
539 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
541 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
,
542 readl
, (sts
& DMA_GSTS_IRES
), sts
);
544 spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
548 static int setup_intr_remapping(struct intel_iommu
*iommu
, int mode
)
550 struct ir_table
*ir_table
;
553 ir_table
= iommu
->ir_table
= kzalloc(sizeof(struct ir_table
),
556 if (!iommu
->ir_table
)
559 pages
= alloc_pages(GFP_ATOMIC
| __GFP_ZERO
, INTR_REMAP_PAGE_ORDER
);
562 printk(KERN_ERR
"failed to allocate pages of order %d\n",
563 INTR_REMAP_PAGE_ORDER
);
564 kfree(iommu
->ir_table
);
568 ir_table
->base
= page_address(pages
);
570 iommu_set_intr_remapping(iommu
, mode
);
575 * Disable Interrupt Remapping.
577 static void iommu_disable_intr_remapping(struct intel_iommu
*iommu
)
582 if (!ecap_ir_support(iommu
->ecap
))
586 * global invalidation of interrupt entry cache before disabling
587 * interrupt-remapping.
589 qi_global_iec(iommu
);
591 spin_lock_irqsave(&iommu
->register_lock
, flags
);
593 sts
= dmar_readq(iommu
->reg
+ DMAR_GSTS_REG
);
594 if (!(sts
& DMA_GSTS_IRES
))
597 iommu
->gcmd
&= ~DMA_GCMD_IRE
;
598 writel(iommu
->gcmd
, iommu
->reg
+ DMAR_GCMD_REG
);
600 IOMMU_WAIT_OP(iommu
, DMAR_GSTS_REG
,
601 readl
, !(sts
& DMA_GSTS_IRES
), sts
);
604 spin_unlock_irqrestore(&iommu
->register_lock
, flags
);
607 int __init
intr_remapping_supported(void)
609 struct dmar_drhd_unit
*drhd
;
611 if (disable_intremap
)
614 for_each_drhd_unit(drhd
) {
615 struct intel_iommu
*iommu
= drhd
->iommu
;
617 if (!ecap_ir_support(iommu
->ecap
))
624 int __init
enable_intr_remapping(int eim
)
626 struct dmar_drhd_unit
*drhd
;
629 for_each_drhd_unit(drhd
) {
630 struct intel_iommu
*iommu
= drhd
->iommu
;
633 * If the queued invalidation is already initialized,
634 * shouldn't disable it.
640 * Clear previous faults.
642 dmar_fault(-1, iommu
);
645 * Disable intr remapping and queued invalidation, if already
646 * enabled prior to OS handover.
648 iommu_disable_intr_remapping(iommu
);
650 dmar_disable_qi(iommu
);
654 * check for the Interrupt-remapping support
656 for_each_drhd_unit(drhd
) {
657 struct intel_iommu
*iommu
= drhd
->iommu
;
659 if (!ecap_ir_support(iommu
->ecap
))
662 if (eim
&& !ecap_eim_support(iommu
->ecap
)) {
663 printk(KERN_INFO
"DRHD %Lx: EIM not supported by DRHD, "
664 " ecap %Lx\n", drhd
->reg_base_addr
, iommu
->ecap
);
670 * Enable queued invalidation for all the DRHD's.
672 for_each_drhd_unit(drhd
) {
674 struct intel_iommu
*iommu
= drhd
->iommu
;
675 ret
= dmar_enable_qi(iommu
);
678 printk(KERN_ERR
"DRHD %Lx: failed to enable queued, "
679 " invalidation, ecap %Lx, ret %d\n",
680 drhd
->reg_base_addr
, iommu
->ecap
, ret
);
686 * Setup Interrupt-remapping for all the DRHD's now.
688 for_each_drhd_unit(drhd
) {
689 struct intel_iommu
*iommu
= drhd
->iommu
;
691 if (!ecap_ir_support(iommu
->ecap
))
694 if (setup_intr_remapping(iommu
, eim
))
703 intr_remapping_enabled
= 1;
709 * handle error condition gracefully here!
714 static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope
*scope
,
715 struct intel_iommu
*iommu
)
717 struct acpi_dmar_pci_path
*path
;
722 path
= (struct acpi_dmar_pci_path
*)(scope
+ 1);
723 count
= (scope
->length
- sizeof(struct acpi_dmar_device_scope
))
724 / sizeof(struct acpi_dmar_pci_path
);
726 while (--count
> 0) {
728 * Access PCI directly due to the PCI
729 * subsystem isn't initialized yet.
731 bus
= read_pci_config_byte(bus
, path
->dev
, path
->fn
,
736 ir_ioapic
[ir_ioapic_num
].bus
= bus
;
737 ir_ioapic
[ir_ioapic_num
].devfn
= PCI_DEVFN(path
->dev
, path
->fn
);
738 ir_ioapic
[ir_ioapic_num
].iommu
= iommu
;
739 ir_ioapic
[ir_ioapic_num
].id
= scope
->enumeration_id
;
743 static int ir_parse_ioapic_scope(struct acpi_dmar_header
*header
,
744 struct intel_iommu
*iommu
)
746 struct acpi_dmar_hardware_unit
*drhd
;
747 struct acpi_dmar_device_scope
*scope
;
750 drhd
= (struct acpi_dmar_hardware_unit
*)header
;
752 start
= (void *)(drhd
+ 1);
753 end
= ((void *)drhd
) + header
->length
;
755 while (start
< end
) {
757 if (scope
->entry_type
== ACPI_DMAR_SCOPE_TYPE_IOAPIC
) {
758 if (ir_ioapic_num
== MAX_IO_APICS
) {
759 printk(KERN_WARNING
"Exceeded Max IO APICS\n");
763 printk(KERN_INFO
"IOAPIC id %d under DRHD base"
764 " 0x%Lx\n", scope
->enumeration_id
,
767 ir_parse_one_ioapic_scope(scope
, iommu
);
769 start
+= scope
->length
;
776 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
779 int __init
parse_ioapics_under_ir(void)
781 struct dmar_drhd_unit
*drhd
;
782 int ir_supported
= 0;
784 for_each_drhd_unit(drhd
) {
785 struct intel_iommu
*iommu
= drhd
->iommu
;
787 if (ecap_ir_support(iommu
->ecap
)) {
788 if (ir_parse_ioapic_scope(drhd
->hdr
, iommu
))
795 if (ir_supported
&& ir_ioapic_num
!= nr_ioapics
) {
797 "Not all IO-APIC's listed under remapping hardware\n");
804 void disable_intr_remapping(void)
806 struct dmar_drhd_unit
*drhd
;
807 struct intel_iommu
*iommu
= NULL
;
810 * Disable Interrupt-remapping for all the DRHD's now.
812 for_each_iommu(iommu
, drhd
) {
813 if (!ecap_ir_support(iommu
->ecap
))
816 iommu_disable_intr_remapping(iommu
);
820 int reenable_intr_remapping(int eim
)
822 struct dmar_drhd_unit
*drhd
;
824 struct intel_iommu
*iommu
= NULL
;
826 for_each_iommu(iommu
, drhd
)
828 dmar_reenable_qi(iommu
);
831 * Setup Interrupt-remapping for all the DRHD's now.
833 for_each_iommu(iommu
, drhd
) {
834 if (!ecap_ir_support(iommu
->ecap
))
837 /* Set up interrupt remapping for iommu.*/
838 iommu_set_intr_remapping(iommu
, eim
);
849 * handle error condition gracefully here!