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 <linux/intel-iommu.h>
10 #include <linux/acpi.h>
11 #include <asm/io_apic.h>
14 #include <asm/irq_remapping.h>
15 #include <asm/pci-direct.h>
16 #include <asm/msidef.h>
18 #include "irq_remapping.h"
21 struct intel_iommu *iommu;
23 unsigned int bus; /* PCI bus number */
24 unsigned int devfn; /* PCI devfn number */
28 struct intel_iommu *iommu;
34 #define IR_X2APIC_MODE(mode) (mode ? (1 << 11) : 0)
35 #define IRTE_DEST(dest) ((x2apic_mode) ? dest : dest << 8)
37 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
38 static struct hpet_scope ir_hpet[MAX_HPET_TBS];
45 * ->iommu->register_lock
47 * intel_irq_remap_ops.{supported,prepare,enable,disable,reenable} are called
48 * in single-threaded environment with interrupt disabled, so no need to tabke
49 * the dmar_global_lock.
51 static DEFINE_RAW_SPINLOCK(irq_2_ir_lock);
53 static int __init parse_ioapics_under_ir(void);
55 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
57 struct irq_cfg *cfg = irq_cfg(irq);
58 return cfg ? &cfg->irq_2_iommu : NULL;
61 static int get_irte(int irq, struct irte *entry)
63 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
67 if (!entry || !irq_iommu)
70 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
72 if (unlikely(!irq_iommu->iommu)) {
73 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
77 index = irq_iommu->irte_index + irq_iommu->sub_handle;
78 *entry = *(irq_iommu->iommu->ir_table->base + index);
80 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
84 static int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
86 struct ir_table *table = iommu->ir_table;
87 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
88 struct irq_cfg *cfg = irq_cfg(irq);
89 unsigned int mask = 0;
93 if (!count || !irq_iommu)
97 count = __roundup_pow_of_two(count);
101 if (mask > ecap_max_handle_mask(iommu->ecap)) {
103 "Requested mask %x exceeds the max invalidation handle"
104 " mask value %Lx\n", mask,
105 ecap_max_handle_mask(iommu->ecap));
109 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
110 index = bitmap_find_free_region(table->bitmap,
111 INTR_REMAP_TABLE_ENTRIES, mask);
113 pr_warn("IR%d: can't allocate an IRTE\n", iommu->seq_id);
116 irq_iommu->iommu = iommu;
117 irq_iommu->irte_index = index;
118 irq_iommu->sub_handle = 0;
119 irq_iommu->irte_mask = mask;
121 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
126 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
130 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
134 return qi_submit_sync(&desc, iommu);
137 static int map_irq_to_irte_handle(int irq, u16 *sub_handle)
139 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
146 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
147 *sub_handle = irq_iommu->sub_handle;
148 index = irq_iommu->irte_index;
149 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
153 static int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
155 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
156 struct irq_cfg *cfg = irq_cfg(irq);
162 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
165 irq_iommu->iommu = iommu;
166 irq_iommu->irte_index = index;
167 irq_iommu->sub_handle = subhandle;
168 irq_iommu->irte_mask = 0;
170 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
175 static int modify_irte(int irq, struct irte *irte_modified)
177 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
178 struct intel_iommu *iommu;
186 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
188 iommu = irq_iommu->iommu;
190 index = irq_iommu->irte_index + irq_iommu->sub_handle;
191 irte = &iommu->ir_table->base[index];
193 set_64bit(&irte->low, irte_modified->low);
194 set_64bit(&irte->high, irte_modified->high);
195 __iommu_flush_cache(iommu, irte, sizeof(*irte));
197 rc = qi_flush_iec(iommu, index, 0);
198 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
203 static struct intel_iommu *map_hpet_to_ir(u8 hpet_id)
207 for (i = 0; i < MAX_HPET_TBS; i++)
208 if (ir_hpet[i].id == hpet_id && ir_hpet[i].iommu)
209 return ir_hpet[i].iommu;
213 static struct intel_iommu *map_ioapic_to_ir(int apic)
217 for (i = 0; i < MAX_IO_APICS; i++)
218 if (ir_ioapic[i].id == apic && ir_ioapic[i].iommu)
219 return ir_ioapic[i].iommu;
223 static struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
225 struct dmar_drhd_unit *drhd;
227 drhd = dmar_find_matched_drhd_unit(dev);
234 static int clear_entries(struct irq_2_iommu *irq_iommu)
236 struct irte *start, *entry, *end;
237 struct intel_iommu *iommu;
240 if (irq_iommu->sub_handle)
243 iommu = irq_iommu->iommu;
244 index = irq_iommu->irte_index + irq_iommu->sub_handle;
246 start = iommu->ir_table->base + index;
247 end = start + (1 << irq_iommu->irte_mask);
249 for (entry = start; entry < end; entry++) {
250 set_64bit(&entry->low, 0);
251 set_64bit(&entry->high, 0);
253 bitmap_release_region(iommu->ir_table->bitmap, index,
254 irq_iommu->irte_mask);
256 return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
259 static int free_irte(int irq)
261 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
268 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
270 rc = clear_entries(irq_iommu);
272 irq_iommu->iommu = NULL;
273 irq_iommu->irte_index = 0;
274 irq_iommu->sub_handle = 0;
275 irq_iommu->irte_mask = 0;
277 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
283 * source validation type
285 #define SVT_NO_VERIFY 0x0 /* no verification is required */
286 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fields */
287 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
290 * source-id qualifier
292 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
293 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
294 * the third least significant bit
296 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
297 * the second and third least significant bits
299 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
300 * the least three significant bits
304 * set SVT, SQ and SID fields of irte to verify
305 * source ids of interrupt requests
307 static void set_irte_sid(struct irte *irte, unsigned int svt,
308 unsigned int sq, unsigned int sid)
310 if (disable_sourceid_checking)
317 static int set_ioapic_sid(struct irte *irte, int apic)
325 down_read(&dmar_global_lock);
326 for (i = 0; i < MAX_IO_APICS; i++) {
327 if (ir_ioapic[i].iommu && ir_ioapic[i].id == apic) {
328 sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
332 up_read(&dmar_global_lock);
335 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
339 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, sid);
344 static int set_hpet_sid(struct irte *irte, u8 id)
352 down_read(&dmar_global_lock);
353 for (i = 0; i < MAX_HPET_TBS; i++) {
354 if (ir_hpet[i].iommu && ir_hpet[i].id == id) {
355 sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
359 up_read(&dmar_global_lock);
362 pr_warning("Failed to set source-id of HPET block (%d)\n", id);
367 * Should really use SQ_ALL_16. Some platforms are broken.
368 * While we figure out the right quirks for these broken platforms, use
369 * SQ_13_IGNORE_3 for now.
371 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
376 struct set_msi_sid_data {
377 struct pci_dev *pdev;
381 static int set_msi_sid_cb(struct pci_dev *pdev, u16 alias, void *opaque)
383 struct set_msi_sid_data *data = opaque;
391 static int set_msi_sid(struct irte *irte, struct pci_dev *dev)
393 struct set_msi_sid_data data;
398 pci_for_each_dma_alias(dev, set_msi_sid_cb, &data);
401 * DMA alias provides us with a PCI device and alias. The only case
402 * where the it will return an alias on a different bus than the
403 * device is the case of a PCIe-to-PCI bridge, where the alias is for
404 * the subordinate bus. In this case we can only verify the bus.
406 * If the alias device is on a different bus than our source device
407 * then we have a topology based alias, use it.
409 * Otherwise, the alias is for a device DMA quirk and we cannot
410 * assume that MSI uses the same requester ID. Therefore use the
413 if (PCI_BUS_NUM(data.alias) != data.pdev->bus->number)
414 set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
415 PCI_DEVID(PCI_BUS_NUM(data.alias),
417 else if (data.pdev->bus->number != dev->bus->number)
418 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, data.alias);
420 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
421 PCI_DEVID(dev->bus->number, dev->devfn));
426 static void iommu_set_irq_remapping(struct intel_iommu *iommu, int mode)
432 addr = virt_to_phys((void *)iommu->ir_table->base);
434 raw_spin_lock_irqsave(&iommu->register_lock, flags);
436 dmar_writeq(iommu->reg + DMAR_IRTA_REG,
437 (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
439 /* Set interrupt-remapping table pointer */
440 writel(iommu->gcmd | DMA_GCMD_SIRTP, iommu->reg + DMAR_GCMD_REG);
442 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
443 readl, (sts & DMA_GSTS_IRTPS), sts);
444 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
447 * global invalidation of interrupt entry cache before enabling
448 * interrupt-remapping.
450 qi_global_iec(iommu);
452 raw_spin_lock_irqsave(&iommu->register_lock, flags);
454 /* Enable interrupt-remapping */
455 iommu->gcmd |= DMA_GCMD_IRE;
456 iommu->gcmd &= ~DMA_GCMD_CFI; /* Block compatibility-format MSIs */
457 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
459 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
460 readl, (sts & DMA_GSTS_IRES), sts);
463 * With CFI clear in the Global Command register, we should be
464 * protected from dangerous (i.e. compatibility) interrupts
465 * regardless of x2apic status. Check just to be sure.
467 if (sts & DMA_GSTS_CFIS)
469 "Compatibility-format IRQs enabled despite intr remapping;\n"
470 "you are vulnerable to IRQ injection.\n");
472 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
475 static int intel_setup_irq_remapping(struct intel_iommu *iommu)
477 struct ir_table *ir_table;
479 unsigned long *bitmap;
484 ir_table = kzalloc(sizeof(struct ir_table), GFP_ATOMIC);
488 pages = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
489 INTR_REMAP_PAGE_ORDER);
492 pr_err("IR%d: failed to allocate pages of order %d\n",
493 iommu->seq_id, INTR_REMAP_PAGE_ORDER);
497 bitmap = kcalloc(BITS_TO_LONGS(INTR_REMAP_TABLE_ENTRIES),
498 sizeof(long), GFP_ATOMIC);
499 if (bitmap == NULL) {
500 pr_err("IR%d: failed to allocate bitmap\n", iommu->seq_id);
504 ir_table->base = page_address(pages);
505 ir_table->bitmap = bitmap;
506 iommu->ir_table = ir_table;
510 __free_pages(pages, INTR_REMAP_PAGE_ORDER);
516 static void intel_teardown_irq_remapping(struct intel_iommu *iommu)
518 if (iommu && iommu->ir_table) {
519 free_pages((unsigned long)iommu->ir_table->base,
520 INTR_REMAP_PAGE_ORDER);
521 kfree(iommu->ir_table->bitmap);
522 kfree(iommu->ir_table);
523 iommu->ir_table = NULL;
528 * Disable Interrupt Remapping.
530 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
535 if (!ecap_ir_support(iommu->ecap))
539 * global invalidation of interrupt entry cache before disabling
540 * interrupt-remapping.
542 qi_global_iec(iommu);
544 raw_spin_lock_irqsave(&iommu->register_lock, flags);
546 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
547 if (!(sts & DMA_GSTS_IRES))
550 iommu->gcmd &= ~DMA_GCMD_IRE;
551 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
553 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
554 readl, !(sts & DMA_GSTS_IRES), sts);
557 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
560 static int __init dmar_x2apic_optout(void)
562 struct acpi_table_dmar *dmar;
563 dmar = (struct acpi_table_dmar *)dmar_tbl;
564 if (!dmar || no_x2apic_optout)
566 return dmar->flags & DMAR_X2APIC_OPT_OUT;
569 static int __init intel_irq_remapping_supported(void)
571 struct dmar_drhd_unit *drhd;
572 struct intel_iommu *iommu;
574 if (disable_irq_remap)
576 if (irq_remap_broken) {
578 "This system BIOS has enabled interrupt remapping\n"
579 "on a chipset that contains an erratum making that\n"
580 "feature unstable. To maintain system stability\n"
581 "interrupt remapping is being disabled. Please\n"
582 "contact your BIOS vendor for an update\n");
583 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
584 disable_irq_remap = 1;
588 if (!dmar_ir_support())
591 for_each_iommu(iommu, drhd)
592 if (!ecap_ir_support(iommu->ecap))
598 static int __init intel_enable_irq_remapping(void)
600 struct dmar_drhd_unit *drhd;
601 struct intel_iommu *iommu;
606 x2apic_present = x2apic_supported();
608 if (parse_ioapics_under_ir() != 1) {
609 printk(KERN_INFO "Not enable interrupt remapping\n");
613 if (x2apic_present) {
614 pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
616 eim = !dmar_x2apic_optout();
618 pr_info("x2apic is disabled because BIOS sets x2apic opt out bit. You can use 'intremap=no_x2apic_optout' to override the BIOS setting.\n");
621 for_each_iommu(iommu, drhd) {
623 * If the queued invalidation is already initialized,
624 * shouldn't disable it.
630 * Clear previous faults.
632 dmar_fault(-1, iommu);
635 * Disable intr remapping and queued invalidation, if already
636 * enabled prior to OS handover.
638 iommu_disable_irq_remapping(iommu);
640 dmar_disable_qi(iommu);
644 * check for the Interrupt-remapping support
646 for_each_iommu(iommu, drhd) {
647 if (!ecap_ir_support(iommu->ecap))
650 if (eim && !ecap_eim_support(iommu->ecap)) {
651 printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
652 " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
658 * Enable queued invalidation for all the DRHD's.
660 for_each_iommu(iommu, drhd) {
661 int ret = dmar_enable_qi(iommu);
664 printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
665 " invalidation, ecap %Lx, ret %d\n",
666 drhd->reg_base_addr, iommu->ecap, ret);
672 * Setup Interrupt-remapping for all the DRHD's now.
674 for_each_iommu(iommu, drhd) {
675 if (!ecap_ir_support(iommu->ecap))
678 if (intel_setup_irq_remapping(iommu))
681 iommu_set_irq_remapping(iommu, eim);
688 irq_remapping_enabled = 1;
691 * VT-d has a different layout for IO-APIC entries when
692 * interrupt remapping is enabled. So it needs a special routine
693 * to print IO-APIC entries for debugging purposes too.
695 x86_io_apic_ops.print_entries = intel_ir_io_apic_print_entries;
697 pr_info("Enabled IRQ remapping in %s mode\n", eim ? "x2apic" : "xapic");
699 return eim ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
702 for_each_iommu(iommu, drhd)
703 if (ecap_ir_support(iommu->ecap)) {
704 iommu_disable_irq_remapping(iommu);
705 intel_teardown_irq_remapping(iommu);
709 pr_warn("Failed to enable irq remapping. You are vulnerable to irq-injection attacks.\n");
714 static int ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
715 struct intel_iommu *iommu,
716 struct acpi_dmar_hardware_unit *drhd)
718 struct acpi_dmar_pci_path *path;
720 int count, free = -1;
723 path = (struct acpi_dmar_pci_path *)(scope + 1);
724 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
725 / sizeof(struct acpi_dmar_pci_path);
727 while (--count > 0) {
729 * Access PCI directly due to the PCI
730 * subsystem isn't initialized yet.
732 bus = read_pci_config_byte(bus, path->device, path->function,
737 for (count = 0; count < MAX_HPET_TBS; count++) {
738 if (ir_hpet[count].iommu == iommu &&
739 ir_hpet[count].id == scope->enumeration_id)
741 else if (ir_hpet[count].iommu == NULL && free == -1)
745 pr_warn("Exceeded Max HPET blocks\n");
749 ir_hpet[free].iommu = iommu;
750 ir_hpet[free].id = scope->enumeration_id;
751 ir_hpet[free].bus = bus;
752 ir_hpet[free].devfn = PCI_DEVFN(path->device, path->function);
753 pr_info("HPET id %d under DRHD base 0x%Lx\n",
754 scope->enumeration_id, drhd->address);
759 static int ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
760 struct intel_iommu *iommu,
761 struct acpi_dmar_hardware_unit *drhd)
763 struct acpi_dmar_pci_path *path;
765 int count, free = -1;
768 path = (struct acpi_dmar_pci_path *)(scope + 1);
769 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
770 / sizeof(struct acpi_dmar_pci_path);
772 while (--count > 0) {
774 * Access PCI directly due to the PCI
775 * subsystem isn't initialized yet.
777 bus = read_pci_config_byte(bus, path->device, path->function,
782 for (count = 0; count < MAX_IO_APICS; count++) {
783 if (ir_ioapic[count].iommu == iommu &&
784 ir_ioapic[count].id == scope->enumeration_id)
786 else if (ir_ioapic[count].iommu == NULL && free == -1)
790 pr_warn("Exceeded Max IO APICS\n");
794 ir_ioapic[free].bus = bus;
795 ir_ioapic[free].devfn = PCI_DEVFN(path->device, path->function);
796 ir_ioapic[free].iommu = iommu;
797 ir_ioapic[free].id = scope->enumeration_id;
798 pr_info("IOAPIC id %d under DRHD base 0x%Lx IOMMU %d\n",
799 scope->enumeration_id, drhd->address, iommu->seq_id);
804 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
805 struct intel_iommu *iommu)
808 struct acpi_dmar_hardware_unit *drhd;
809 struct acpi_dmar_device_scope *scope;
812 drhd = (struct acpi_dmar_hardware_unit *)header;
813 start = (void *)(drhd + 1);
814 end = ((void *)drhd) + header->length;
816 while (start < end && ret == 0) {
818 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC)
819 ret = ir_parse_one_ioapic_scope(scope, iommu, drhd);
820 else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET)
821 ret = ir_parse_one_hpet_scope(scope, iommu, drhd);
822 start += scope->length;
828 static void ir_remove_ioapic_hpet_scope(struct intel_iommu *iommu)
832 for (i = 0; i < MAX_HPET_TBS; i++)
833 if (ir_hpet[i].iommu == iommu)
834 ir_hpet[i].iommu = NULL;
836 for (i = 0; i < MAX_IO_APICS; i++)
837 if (ir_ioapic[i].iommu == iommu)
838 ir_ioapic[i].iommu = NULL;
842 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
845 static int __init parse_ioapics_under_ir(void)
847 struct dmar_drhd_unit *drhd;
848 struct intel_iommu *iommu;
849 int ir_supported = 0;
852 for_each_iommu(iommu, drhd)
853 if (ecap_ir_support(iommu->ecap)) {
854 if (ir_parse_ioapic_hpet_scope(drhd->hdr, iommu))
863 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
864 int ioapic_id = mpc_ioapic_id(ioapic_idx);
865 if (!map_ioapic_to_ir(ioapic_id)) {
866 pr_err(FW_BUG "ioapic %d has no mapping iommu, "
867 "interrupt remapping will be disabled\n",
876 static int __init ir_dev_scope_init(void)
880 if (!irq_remapping_enabled)
883 down_write(&dmar_global_lock);
884 ret = dmar_dev_scope_init();
885 up_write(&dmar_global_lock);
889 rootfs_initcall(ir_dev_scope_init);
891 static void disable_irq_remapping(void)
893 struct dmar_drhd_unit *drhd;
894 struct intel_iommu *iommu = NULL;
897 * Disable Interrupt-remapping for all the DRHD's now.
899 for_each_iommu(iommu, drhd) {
900 if (!ecap_ir_support(iommu->ecap))
903 iommu_disable_irq_remapping(iommu);
907 static int reenable_irq_remapping(int eim)
909 struct dmar_drhd_unit *drhd;
911 struct intel_iommu *iommu = NULL;
913 for_each_iommu(iommu, drhd)
915 dmar_reenable_qi(iommu);
918 * Setup Interrupt-remapping for all the DRHD's now.
920 for_each_iommu(iommu, drhd) {
921 if (!ecap_ir_support(iommu->ecap))
924 /* Set up interrupt remapping for iommu.*/
925 iommu_set_irq_remapping(iommu, eim);
936 * handle error condition gracefully here!
941 static void prepare_irte(struct irte *irte, int vector,
944 memset(irte, 0, sizeof(*irte));
947 irte->dst_mode = apic->irq_dest_mode;
949 * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
950 * actual level or edge trigger will be setup in the IO-APIC
951 * RTE. This will help simplify level triggered irq migration.
952 * For more details, see the comments (in io_apic.c) explainig IO-APIC
953 * irq migration in the presence of interrupt-remapping.
955 irte->trigger_mode = 0;
956 irte->dlvry_mode = apic->irq_delivery_mode;
957 irte->vector = vector;
958 irte->dest_id = IRTE_DEST(dest);
959 irte->redir_hint = 1;
962 static int intel_setup_ioapic_entry(int irq,
963 struct IO_APIC_route_entry *route_entry,
964 unsigned int destination, int vector,
965 struct io_apic_irq_attr *attr)
967 int ioapic_id = mpc_ioapic_id(attr->ioapic);
968 struct intel_iommu *iommu;
969 struct IR_IO_APIC_route_entry *entry;
973 down_read(&dmar_global_lock);
974 iommu = map_ioapic_to_ir(ioapic_id);
976 pr_warn("No mapping iommu for ioapic %d\n", ioapic_id);
979 index = alloc_irte(iommu, irq, 1);
981 pr_warn("Failed to allocate IRTE for ioapic %d\n",
986 up_read(&dmar_global_lock);
990 prepare_irte(&irte, vector, destination);
992 /* Set source-id of interrupt request */
993 set_ioapic_sid(&irte, ioapic_id);
995 modify_irte(irq, &irte);
997 apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: "
998 "Set IRTE entry (P:%d FPD:%d Dst_Mode:%d "
999 "Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X "
1000 "Avail:%X Vector:%02X Dest:%08X "
1001 "SID:%04X SQ:%X SVT:%X)\n",
1002 attr->ioapic, irte.present, irte.fpd, irte.dst_mode,
1003 irte.redir_hint, irte.trigger_mode, irte.dlvry_mode,
1004 irte.avail, irte.vector, irte.dest_id,
1005 irte.sid, irte.sq, irte.svt);
1007 entry = (struct IR_IO_APIC_route_entry *)route_entry;
1008 memset(entry, 0, sizeof(*entry));
1010 entry->index2 = (index >> 15) & 0x1;
1013 entry->index = (index & 0x7fff);
1015 * IO-APIC RTE will be configured with virtual vector.
1016 * irq handler will do the explicit EOI to the io-apic.
1018 entry->vector = attr->ioapic_pin;
1019 entry->mask = 0; /* enable IRQ */
1020 entry->trigger = attr->trigger;
1021 entry->polarity = attr->polarity;
1023 /* Mask level triggered irqs.
1024 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1033 * Migrate the IO-APIC irq in the presence of intr-remapping.
1035 * For both level and edge triggered, irq migration is a simple atomic
1036 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
1038 * For level triggered, we eliminate the io-apic RTE modification (with the
1039 * updated vector information), by using a virtual vector (io-apic pin number).
1040 * Real vector that is used for interrupting cpu will be coming from
1041 * the interrupt-remapping table entry.
1043 * As the migration is a simple atomic update of IRTE, the same mechanism
1044 * is used to migrate MSI irq's in the presence of interrupt-remapping.
1047 intel_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
1050 struct irq_cfg *cfg = irqd_cfg(data);
1051 unsigned int dest, irq = data->irq;
1055 if (!config_enabled(CONFIG_SMP))
1058 if (!cpumask_intersects(mask, cpu_online_mask))
1061 if (get_irte(irq, &irte))
1064 err = assign_irq_vector(irq, cfg, mask);
1068 err = apic->cpu_mask_to_apicid_and(cfg->domain, mask, &dest);
1070 if (assign_irq_vector(irq, cfg, data->affinity))
1071 pr_err("Failed to recover vector for irq %d\n", irq);
1075 irte.vector = cfg->vector;
1076 irte.dest_id = IRTE_DEST(dest);
1079 * Atomically updates the IRTE with the new destination, vector
1080 * and flushes the interrupt entry cache.
1082 modify_irte(irq, &irte);
1085 * After this point, all the interrupts will start arriving
1086 * at the new destination. So, time to cleanup the previous
1087 * vector allocation.
1089 if (cfg->move_in_progress)
1090 send_cleanup_vector(cfg);
1092 cpumask_copy(data->affinity, mask);
1096 static void intel_compose_msi_msg(struct pci_dev *pdev,
1097 unsigned int irq, unsigned int dest,
1098 struct msi_msg *msg, u8 hpet_id)
1100 struct irq_cfg *cfg;
1107 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
1108 BUG_ON(ir_index == -1);
1110 prepare_irte(&irte, cfg->vector, dest);
1112 /* Set source-id of interrupt request */
1114 set_msi_sid(&irte, pdev);
1116 set_hpet_sid(&irte, hpet_id);
1118 modify_irte(irq, &irte);
1120 msg->address_hi = MSI_ADDR_BASE_HI;
1121 msg->data = sub_handle;
1122 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
1124 MSI_ADDR_IR_INDEX1(ir_index) |
1125 MSI_ADDR_IR_INDEX2(ir_index);
1129 * Map the PCI dev to the corresponding remapping hardware unit
1130 * and allocate 'nvec' consecutive interrupt-remapping table entries
1133 static int intel_msi_alloc_irq(struct pci_dev *dev, int irq, int nvec)
1135 struct intel_iommu *iommu;
1138 down_read(&dmar_global_lock);
1139 iommu = map_dev_to_ir(dev);
1142 "Unable to map PCI %s to iommu\n", pci_name(dev));
1145 index = alloc_irte(iommu, irq, nvec);
1148 "Unable to allocate %d IRTE for PCI %s\n",
1149 nvec, pci_name(dev));
1153 up_read(&dmar_global_lock);
1158 static int intel_msi_setup_irq(struct pci_dev *pdev, unsigned int irq,
1159 int index, int sub_handle)
1161 struct intel_iommu *iommu;
1164 down_read(&dmar_global_lock);
1165 iommu = map_dev_to_ir(pdev);
1168 * setup the mapping between the irq and the IRTE
1169 * base index, the sub_handle pointing to the
1170 * appropriate interrupt remap table entry.
1172 set_irte_irq(irq, iommu, index, sub_handle);
1175 up_read(&dmar_global_lock);
1180 static int intel_alloc_hpet_msi(unsigned int irq, unsigned int id)
1183 struct intel_iommu *iommu;
1186 down_read(&dmar_global_lock);
1187 iommu = map_hpet_to_ir(id);
1189 index = alloc_irte(iommu, irq, 1);
1193 up_read(&dmar_global_lock);
1198 struct irq_remap_ops intel_irq_remap_ops = {
1199 .supported = intel_irq_remapping_supported,
1200 .prepare = dmar_table_init,
1201 .enable = intel_enable_irq_remapping,
1202 .disable = disable_irq_remapping,
1203 .reenable = reenable_irq_remapping,
1204 .enable_faulting = enable_drhd_fault_handling,
1205 .setup_ioapic_entry = intel_setup_ioapic_entry,
1206 .set_affinity = intel_ioapic_set_affinity,
1207 .free_irq = free_irte,
1208 .compose_msi_msg = intel_compose_msi_msg,
1209 .msi_alloc_irq = intel_msi_alloc_irq,
1210 .msi_setup_irq = intel_msi_setup_irq,
1211 .alloc_hpet_msi = intel_alloc_hpet_msi,
1215 * Support of Interrupt Remapping Unit Hotplug
1217 static int dmar_ir_add(struct dmar_drhd_unit *dmaru, struct intel_iommu *iommu)
1220 int eim = x2apic_enabled();
1222 if (eim && !ecap_eim_support(iommu->ecap)) {
1223 pr_info("DRHD %Lx: EIM not supported by DRHD, ecap %Lx\n",
1224 iommu->reg_phys, iommu->ecap);
1228 if (ir_parse_ioapic_hpet_scope(dmaru->hdr, iommu)) {
1229 pr_warn("DRHD %Lx: failed to parse managed IOAPIC/HPET\n",
1234 /* TODO: check all IOAPICs are covered by IOMMU */
1236 /* Setup Interrupt-remapping now. */
1237 ret = intel_setup_irq_remapping(iommu);
1239 pr_err("DRHD %Lx: failed to allocate resource\n",
1241 ir_remove_ioapic_hpet_scope(iommu);
1246 /* Clear previous faults. */
1247 dmar_fault(-1, iommu);
1248 iommu_disable_irq_remapping(iommu);
1249 dmar_disable_qi(iommu);
1252 /* Enable queued invalidation */
1253 ret = dmar_enable_qi(iommu);
1255 iommu_set_irq_remapping(iommu, eim);
1257 pr_err("DRHD %Lx: failed to enable queued invalidation, ecap %Lx, ret %d\n",
1258 iommu->reg_phys, iommu->ecap, ret);
1259 intel_teardown_irq_remapping(iommu);
1260 ir_remove_ioapic_hpet_scope(iommu);
1266 int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert)
1269 struct intel_iommu *iommu = dmaru->iommu;
1271 if (!irq_remapping_enabled)
1275 if (!ecap_ir_support(iommu->ecap))
1279 if (!iommu->ir_table)
1280 ret = dmar_ir_add(dmaru, iommu);
1282 if (iommu->ir_table) {
1283 if (!bitmap_empty(iommu->ir_table->bitmap,
1284 INTR_REMAP_TABLE_ENTRIES)) {
1287 iommu_disable_irq_remapping(iommu);
1288 intel_teardown_irq_remapping(iommu);
1289 ir_remove_ioapic_hpet_scope(iommu);