2 #define pr_fmt(fmt) "DMAR-IR: " fmt
4 #include <linux/interrupt.h>
5 #include <linux/dmar.h>
6 #include <linux/spinlock.h>
7 #include <linux/slab.h>
8 #include <linux/jiffies.h>
9 #include <linux/hpet.h>
10 #include <linux/pci.h>
11 #include <linux/irq.h>
12 #include <linux/intel-iommu.h>
13 #include <linux/acpi.h>
14 #include <linux/irqdomain.h>
15 #include <linux/crash_dump.h>
16 #include <asm/io_apic.h>
19 #include <asm/irq_remapping.h>
20 #include <asm/pci-direct.h>
21 #include <asm/msidef.h>
23 #include "irq_remapping.h"
31 struct intel_iommu *iommu;
33 unsigned int bus; /* PCI bus number */
34 unsigned int devfn; /* PCI devfn number */
38 struct intel_iommu *iommu;
45 struct intel_iommu *iommu;
52 struct intel_ir_data {
53 struct irq_2_iommu irq_2_iommu;
54 struct irte irte_entry;
56 struct msi_msg msi_entry;
60 #define IR_X2APIC_MODE(mode) (mode ? (1 << 11) : 0)
61 #define IRTE_DEST(dest) ((eim_mode) ? dest : dest << 8)
63 static int __read_mostly eim_mode;
64 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
65 static struct hpet_scope ir_hpet[MAX_HPET_TBS];
72 * ->iommu->register_lock
74 * intel_irq_remap_ops.{supported,prepare,enable,disable,reenable} are called
75 * in single-threaded environment with interrupt disabled, so no need to tabke
76 * the dmar_global_lock.
78 static DEFINE_RAW_SPINLOCK(irq_2_ir_lock);
79 static struct irq_domain_ops intel_ir_domain_ops;
81 static void iommu_disable_irq_remapping(struct intel_iommu *iommu);
82 static int __init parse_ioapics_under_ir(void);
84 static bool ir_pre_enabled(struct intel_iommu *iommu)
86 return (iommu->flags & VTD_FLAG_IRQ_REMAP_PRE_ENABLED);
89 static void clear_ir_pre_enabled(struct intel_iommu *iommu)
91 iommu->flags &= ~VTD_FLAG_IRQ_REMAP_PRE_ENABLED;
94 static void init_ir_status(struct intel_iommu *iommu)
98 gsts = readl(iommu->reg + DMAR_GSTS_REG);
99 if (gsts & DMA_GSTS_IRES)
100 iommu->flags |= VTD_FLAG_IRQ_REMAP_PRE_ENABLED;
103 static int alloc_irte(struct intel_iommu *iommu, int irq,
104 struct irq_2_iommu *irq_iommu, u16 count)
106 struct ir_table *table = iommu->ir_table;
107 unsigned int mask = 0;
111 if (!count || !irq_iommu)
115 count = __roundup_pow_of_two(count);
119 if (mask > ecap_max_handle_mask(iommu->ecap)) {
120 pr_err("Requested mask %x exceeds the max invalidation handle"
121 " mask value %Lx\n", mask,
122 ecap_max_handle_mask(iommu->ecap));
126 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
127 index = bitmap_find_free_region(table->bitmap,
128 INTR_REMAP_TABLE_ENTRIES, mask);
130 pr_warn("IR%d: can't allocate an IRTE\n", iommu->seq_id);
132 irq_iommu->iommu = iommu;
133 irq_iommu->irte_index = index;
134 irq_iommu->sub_handle = 0;
135 irq_iommu->irte_mask = mask;
136 irq_iommu->mode = IRQ_REMAPPING;
138 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
143 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
147 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
151 return qi_submit_sync(&desc, iommu);
154 static int modify_irte(struct irq_2_iommu *irq_iommu,
155 struct irte *irte_modified)
157 struct intel_iommu *iommu;
165 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
167 iommu = irq_iommu->iommu;
169 index = irq_iommu->irte_index + irq_iommu->sub_handle;
170 irte = &iommu->ir_table->base[index];
172 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE)
173 if ((irte->pst == 1) || (irte_modified->pst == 1)) {
176 ret = cmpxchg_double(&irte->low, &irte->high,
177 irte->low, irte->high,
178 irte_modified->low, irte_modified->high);
180 * We use cmpxchg16 to atomically update the 128-bit IRTE,
181 * and it cannot be updated by the hardware or other processors
182 * behind us, so the return value of cmpxchg16 should be the
183 * same as the old value.
189 set_64bit(&irte->low, irte_modified->low);
190 set_64bit(&irte->high, irte_modified->high);
192 __iommu_flush_cache(iommu, irte, sizeof(*irte));
194 rc = qi_flush_iec(iommu, index, 0);
196 /* Update iommu mode according to the IRTE mode */
197 irq_iommu->mode = irte->pst ? IRQ_POSTING : IRQ_REMAPPING;
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;
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);
260 * source validation type
262 #define SVT_NO_VERIFY 0x0 /* no verification is required */
263 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fields */
264 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
267 * source-id qualifier
269 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
270 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
271 * the third least significant bit
273 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
274 * the second and third least significant bits
276 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
277 * the least three significant bits
281 * set SVT, SQ and SID fields of irte to verify
282 * source ids of interrupt requests
284 static void set_irte_sid(struct irte *irte, unsigned int svt,
285 unsigned int sq, unsigned int sid)
287 if (disable_sourceid_checking)
294 static int set_ioapic_sid(struct irte *irte, int apic)
302 down_read(&dmar_global_lock);
303 for (i = 0; i < MAX_IO_APICS; i++) {
304 if (ir_ioapic[i].iommu && ir_ioapic[i].id == apic) {
305 sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
309 up_read(&dmar_global_lock);
312 pr_warn("Failed to set source-id of IOAPIC (%d)\n", apic);
316 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, sid);
321 static int set_hpet_sid(struct irte *irte, u8 id)
329 down_read(&dmar_global_lock);
330 for (i = 0; i < MAX_HPET_TBS; i++) {
331 if (ir_hpet[i].iommu && ir_hpet[i].id == id) {
332 sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
336 up_read(&dmar_global_lock);
339 pr_warn("Failed to set source-id of HPET block (%d)\n", id);
344 * Should really use SQ_ALL_16. Some platforms are broken.
345 * While we figure out the right quirks for these broken platforms, use
346 * SQ_13_IGNORE_3 for now.
348 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
353 struct set_msi_sid_data {
354 struct pci_dev *pdev;
358 static int set_msi_sid_cb(struct pci_dev *pdev, u16 alias, void *opaque)
360 struct set_msi_sid_data *data = opaque;
368 static int set_msi_sid(struct irte *irte, struct pci_dev *dev)
370 struct set_msi_sid_data data;
375 pci_for_each_dma_alias(dev, set_msi_sid_cb, &data);
378 * DMA alias provides us with a PCI device and alias. The only case
379 * where the it will return an alias on a different bus than the
380 * device is the case of a PCIe-to-PCI bridge, where the alias is for
381 * the subordinate bus. In this case we can only verify the bus.
383 * If the alias device is on a different bus than our source device
384 * then we have a topology based alias, use it.
386 * Otherwise, the alias is for a device DMA quirk and we cannot
387 * assume that MSI uses the same requester ID. Therefore use the
390 if (PCI_BUS_NUM(data.alias) != data.pdev->bus->number)
391 set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
392 PCI_DEVID(PCI_BUS_NUM(data.alias),
394 else if (data.pdev->bus->number != dev->bus->number)
395 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, data.alias);
397 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
398 PCI_DEVID(dev->bus->number, dev->devfn));
403 static int iommu_load_old_irte(struct intel_iommu *iommu)
405 struct irte *old_ir_table;
406 phys_addr_t irt_phys;
411 if (!is_kdump_kernel()) {
412 pr_warn("IRQ remapping was enabled on %s but we are not in kdump mode\n",
414 clear_ir_pre_enabled(iommu);
415 iommu_disable_irq_remapping(iommu);
419 /* Check whether the old ir-table has the same size as ours */
420 irta = dmar_readq(iommu->reg + DMAR_IRTA_REG);
421 if ((irta & INTR_REMAP_TABLE_REG_SIZE_MASK)
422 != INTR_REMAP_TABLE_REG_SIZE)
425 irt_phys = irta & VTD_PAGE_MASK;
426 size = INTR_REMAP_TABLE_ENTRIES*sizeof(struct irte);
428 /* Map the old IR table */
429 old_ir_table = memremap(irt_phys, size, MEMREMAP_WB);
434 memcpy(iommu->ir_table->base, old_ir_table, size);
436 __iommu_flush_cache(iommu, iommu->ir_table->base, size);
439 * Now check the table for used entries and mark those as
440 * allocated in the bitmap
442 for (i = 0; i < INTR_REMAP_TABLE_ENTRIES; i++) {
443 if (iommu->ir_table->base[i].present)
444 bitmap_set(iommu->ir_table->bitmap, i, 1);
447 memunmap(old_ir_table);
453 static void iommu_set_irq_remapping(struct intel_iommu *iommu, int mode)
459 addr = virt_to_phys((void *)iommu->ir_table->base);
461 raw_spin_lock_irqsave(&iommu->register_lock, flags);
463 dmar_writeq(iommu->reg + DMAR_IRTA_REG,
464 (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
466 /* Set interrupt-remapping table pointer */
467 writel(iommu->gcmd | DMA_GCMD_SIRTP, iommu->reg + DMAR_GCMD_REG);
469 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
470 readl, (sts & DMA_GSTS_IRTPS), sts);
471 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
474 * Global invalidation of interrupt entry cache to make sure the
475 * hardware uses the new irq remapping table.
477 qi_global_iec(iommu);
480 static void iommu_enable_irq_remapping(struct intel_iommu *iommu)
485 raw_spin_lock_irqsave(&iommu->register_lock, flags);
487 /* Enable interrupt-remapping */
488 iommu->gcmd |= DMA_GCMD_IRE;
489 iommu->gcmd &= ~DMA_GCMD_CFI; /* Block compatibility-format MSIs */
490 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
492 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
493 readl, (sts & DMA_GSTS_IRES), sts);
496 * With CFI clear in the Global Command register, we should be
497 * protected from dangerous (i.e. compatibility) interrupts
498 * regardless of x2apic status. Check just to be sure.
500 if (sts & DMA_GSTS_CFIS)
502 "Compatibility-format IRQs enabled despite intr remapping;\n"
503 "you are vulnerable to IRQ injection.\n");
505 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
508 static int intel_setup_irq_remapping(struct intel_iommu *iommu)
510 struct ir_table *ir_table;
512 unsigned long *bitmap;
517 ir_table = kzalloc(sizeof(struct ir_table), GFP_KERNEL);
521 pages = alloc_pages_node(iommu->node, GFP_KERNEL | __GFP_ZERO,
522 INTR_REMAP_PAGE_ORDER);
524 pr_err("IR%d: failed to allocate pages of order %d\n",
525 iommu->seq_id, INTR_REMAP_PAGE_ORDER);
529 bitmap = kcalloc(BITS_TO_LONGS(INTR_REMAP_TABLE_ENTRIES),
530 sizeof(long), GFP_ATOMIC);
531 if (bitmap == NULL) {
532 pr_err("IR%d: failed to allocate bitmap\n", iommu->seq_id);
536 iommu->ir_domain = irq_domain_add_hierarchy(arch_get_ir_parent_domain(),
537 0, INTR_REMAP_TABLE_ENTRIES,
538 NULL, &intel_ir_domain_ops,
540 if (!iommu->ir_domain) {
541 pr_err("IR%d: failed to allocate irqdomain\n", iommu->seq_id);
542 goto out_free_bitmap;
544 iommu->ir_msi_domain = arch_create_msi_irq_domain(iommu->ir_domain);
546 ir_table->base = page_address(pages);
547 ir_table->bitmap = bitmap;
548 iommu->ir_table = ir_table;
551 * If the queued invalidation is already initialized,
552 * shouldn't disable it.
556 * Clear previous faults.
558 dmar_fault(-1, iommu);
559 dmar_disable_qi(iommu);
561 if (dmar_enable_qi(iommu)) {
562 pr_err("Failed to enable queued invalidation\n");
563 goto out_free_bitmap;
567 init_ir_status(iommu);
569 if (ir_pre_enabled(iommu)) {
570 if (iommu_load_old_irte(iommu))
571 pr_err("Failed to copy IR table for %s from previous kernel\n",
574 pr_info("Copied IR table for %s from previous kernel\n",
578 iommu_set_irq_remapping(iommu, eim_mode);
585 __free_pages(pages, INTR_REMAP_PAGE_ORDER);
589 iommu->ir_table = NULL;
594 static void intel_teardown_irq_remapping(struct intel_iommu *iommu)
596 if (iommu && iommu->ir_table) {
597 if (iommu->ir_msi_domain) {
598 irq_domain_remove(iommu->ir_msi_domain);
599 iommu->ir_msi_domain = NULL;
601 if (iommu->ir_domain) {
602 irq_domain_remove(iommu->ir_domain);
603 iommu->ir_domain = NULL;
605 free_pages((unsigned long)iommu->ir_table->base,
606 INTR_REMAP_PAGE_ORDER);
607 kfree(iommu->ir_table->bitmap);
608 kfree(iommu->ir_table);
609 iommu->ir_table = NULL;
614 * Disable Interrupt Remapping.
616 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
621 if (!ecap_ir_support(iommu->ecap))
625 * global invalidation of interrupt entry cache before disabling
626 * interrupt-remapping.
628 qi_global_iec(iommu);
630 raw_spin_lock_irqsave(&iommu->register_lock, flags);
632 sts = readl(iommu->reg + DMAR_GSTS_REG);
633 if (!(sts & DMA_GSTS_IRES))
636 iommu->gcmd &= ~DMA_GCMD_IRE;
637 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
639 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
640 readl, !(sts & DMA_GSTS_IRES), sts);
643 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
646 static int __init dmar_x2apic_optout(void)
648 struct acpi_table_dmar *dmar;
649 dmar = (struct acpi_table_dmar *)dmar_tbl;
650 if (!dmar || no_x2apic_optout)
652 return dmar->flags & DMAR_X2APIC_OPT_OUT;
655 static void __init intel_cleanup_irq_remapping(void)
657 struct dmar_drhd_unit *drhd;
658 struct intel_iommu *iommu;
660 for_each_iommu(iommu, drhd) {
661 if (ecap_ir_support(iommu->ecap)) {
662 iommu_disable_irq_remapping(iommu);
663 intel_teardown_irq_remapping(iommu);
667 if (x2apic_supported())
668 pr_warn("Failed to enable irq remapping. You are vulnerable to irq-injection attacks.\n");
671 static int __init intel_prepare_irq_remapping(void)
673 struct dmar_drhd_unit *drhd;
674 struct intel_iommu *iommu;
677 if (irq_remap_broken) {
678 pr_warn("This system BIOS has enabled interrupt remapping\n"
679 "on a chipset that contains an erratum making that\n"
680 "feature unstable. To maintain system stability\n"
681 "interrupt remapping is being disabled. Please\n"
682 "contact your BIOS vendor for an update\n");
683 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
687 if (dmar_table_init() < 0)
690 if (!dmar_ir_support())
693 if (parse_ioapics_under_ir()) {
694 pr_info("Not enabling interrupt remapping\n");
698 /* First make sure all IOMMUs support IRQ remapping */
699 for_each_iommu(iommu, drhd)
700 if (!ecap_ir_support(iommu->ecap))
703 /* Detect remapping mode: lapic or x2apic */
704 if (x2apic_supported()) {
705 eim = !dmar_x2apic_optout();
707 pr_info("x2apic is disabled because BIOS sets x2apic opt out bit.");
708 pr_info("Use 'intremap=no_x2apic_optout' to override the BIOS setting.\n");
712 for_each_iommu(iommu, drhd) {
713 if (eim && !ecap_eim_support(iommu->ecap)) {
714 pr_info("%s does not support EIM\n", iommu->name);
721 pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
723 /* Do the initializations early */
724 for_each_iommu(iommu, drhd) {
725 if (intel_setup_irq_remapping(iommu)) {
726 pr_err("Failed to setup irq remapping for %s\n",
735 intel_cleanup_irq_remapping();
740 * Set Posted-Interrupts capability.
742 static inline void set_irq_posting_cap(void)
744 struct dmar_drhd_unit *drhd;
745 struct intel_iommu *iommu;
747 if (!disable_irq_post) {
749 * If IRTE is in posted format, the 'pda' field goes across the
750 * 64-bit boundary, we need use cmpxchg16b to atomically update
751 * it. We only expose posted-interrupt when X86_FEATURE_CX16
752 * is supported. Actually, hardware platforms supporting PI
753 * should have X86_FEATURE_CX16 support, this has been confirmed
754 * with Intel hardware guys.
756 if (boot_cpu_has(X86_FEATURE_CX16))
757 intel_irq_remap_ops.capability |= 1 << IRQ_POSTING_CAP;
759 for_each_iommu(iommu, drhd)
760 if (!cap_pi_support(iommu->cap)) {
761 intel_irq_remap_ops.capability &=
762 ~(1 << IRQ_POSTING_CAP);
768 static int __init intel_enable_irq_remapping(void)
770 struct dmar_drhd_unit *drhd;
771 struct intel_iommu *iommu;
775 * Setup Interrupt-remapping for all the DRHD's now.
777 for_each_iommu(iommu, drhd) {
778 if (!ir_pre_enabled(iommu))
779 iommu_enable_irq_remapping(iommu);
786 irq_remapping_enabled = 1;
788 set_irq_posting_cap();
790 pr_info("Enabled IRQ remapping in %s mode\n", eim_mode ? "x2apic" : "xapic");
792 return eim_mode ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
795 intel_cleanup_irq_remapping();
799 static int ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
800 struct intel_iommu *iommu,
801 struct acpi_dmar_hardware_unit *drhd)
803 struct acpi_dmar_pci_path *path;
805 int count, free = -1;
808 path = (struct acpi_dmar_pci_path *)(scope + 1);
809 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
810 / sizeof(struct acpi_dmar_pci_path);
812 while (--count > 0) {
814 * Access PCI directly due to the PCI
815 * subsystem isn't initialized yet.
817 bus = read_pci_config_byte(bus, path->device, path->function,
822 for (count = 0; count < MAX_HPET_TBS; count++) {
823 if (ir_hpet[count].iommu == iommu &&
824 ir_hpet[count].id == scope->enumeration_id)
826 else if (ir_hpet[count].iommu == NULL && free == -1)
830 pr_warn("Exceeded Max HPET blocks\n");
834 ir_hpet[free].iommu = iommu;
835 ir_hpet[free].id = scope->enumeration_id;
836 ir_hpet[free].bus = bus;
837 ir_hpet[free].devfn = PCI_DEVFN(path->device, path->function);
838 pr_info("HPET id %d under DRHD base 0x%Lx\n",
839 scope->enumeration_id, drhd->address);
844 static int ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
845 struct intel_iommu *iommu,
846 struct acpi_dmar_hardware_unit *drhd)
848 struct acpi_dmar_pci_path *path;
850 int count, free = -1;
853 path = (struct acpi_dmar_pci_path *)(scope + 1);
854 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
855 / sizeof(struct acpi_dmar_pci_path);
857 while (--count > 0) {
859 * Access PCI directly due to the PCI
860 * subsystem isn't initialized yet.
862 bus = read_pci_config_byte(bus, path->device, path->function,
867 for (count = 0; count < MAX_IO_APICS; count++) {
868 if (ir_ioapic[count].iommu == iommu &&
869 ir_ioapic[count].id == scope->enumeration_id)
871 else if (ir_ioapic[count].iommu == NULL && free == -1)
875 pr_warn("Exceeded Max IO APICS\n");
879 ir_ioapic[free].bus = bus;
880 ir_ioapic[free].devfn = PCI_DEVFN(path->device, path->function);
881 ir_ioapic[free].iommu = iommu;
882 ir_ioapic[free].id = scope->enumeration_id;
883 pr_info("IOAPIC id %d under DRHD base 0x%Lx IOMMU %d\n",
884 scope->enumeration_id, drhd->address, iommu->seq_id);
889 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
890 struct intel_iommu *iommu)
893 struct acpi_dmar_hardware_unit *drhd;
894 struct acpi_dmar_device_scope *scope;
897 drhd = (struct acpi_dmar_hardware_unit *)header;
898 start = (void *)(drhd + 1);
899 end = ((void *)drhd) + header->length;
901 while (start < end && ret == 0) {
903 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC)
904 ret = ir_parse_one_ioapic_scope(scope, iommu, drhd);
905 else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET)
906 ret = ir_parse_one_hpet_scope(scope, iommu, drhd);
907 start += scope->length;
913 static void ir_remove_ioapic_hpet_scope(struct intel_iommu *iommu)
917 for (i = 0; i < MAX_HPET_TBS; i++)
918 if (ir_hpet[i].iommu == iommu)
919 ir_hpet[i].iommu = NULL;
921 for (i = 0; i < MAX_IO_APICS; i++)
922 if (ir_ioapic[i].iommu == iommu)
923 ir_ioapic[i].iommu = NULL;
927 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
930 static int __init parse_ioapics_under_ir(void)
932 struct dmar_drhd_unit *drhd;
933 struct intel_iommu *iommu;
934 bool ir_supported = false;
937 for_each_iommu(iommu, drhd) {
940 if (!ecap_ir_support(iommu->ecap))
943 ret = ir_parse_ioapic_hpet_scope(drhd->hdr, iommu);
953 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
954 int ioapic_id = mpc_ioapic_id(ioapic_idx);
955 if (!map_ioapic_to_ir(ioapic_id)) {
956 pr_err(FW_BUG "ioapic %d has no mapping iommu, "
957 "interrupt remapping will be disabled\n",
966 static int __init ir_dev_scope_init(void)
970 if (!irq_remapping_enabled)
973 down_write(&dmar_global_lock);
974 ret = dmar_dev_scope_init();
975 up_write(&dmar_global_lock);
979 rootfs_initcall(ir_dev_scope_init);
981 static void disable_irq_remapping(void)
983 struct dmar_drhd_unit *drhd;
984 struct intel_iommu *iommu = NULL;
987 * Disable Interrupt-remapping for all the DRHD's now.
989 for_each_iommu(iommu, drhd) {
990 if (!ecap_ir_support(iommu->ecap))
993 iommu_disable_irq_remapping(iommu);
997 * Clear Posted-Interrupts capability.
999 if (!disable_irq_post)
1000 intel_irq_remap_ops.capability &= ~(1 << IRQ_POSTING_CAP);
1003 static int reenable_irq_remapping(int eim)
1005 struct dmar_drhd_unit *drhd;
1007 struct intel_iommu *iommu = NULL;
1009 for_each_iommu(iommu, drhd)
1011 dmar_reenable_qi(iommu);
1014 * Setup Interrupt-remapping for all the DRHD's now.
1016 for_each_iommu(iommu, drhd) {
1017 if (!ecap_ir_support(iommu->ecap))
1020 /* Set up interrupt remapping for iommu.*/
1021 iommu_set_irq_remapping(iommu, eim);
1022 iommu_enable_irq_remapping(iommu);
1029 set_irq_posting_cap();
1035 * handle error condition gracefully here!
1040 static void prepare_irte(struct irte *irte, int vector, unsigned int dest)
1042 memset(irte, 0, sizeof(*irte));
1045 irte->dst_mode = apic->irq_dest_mode;
1047 * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
1048 * actual level or edge trigger will be setup in the IO-APIC
1049 * RTE. This will help simplify level triggered irq migration.
1050 * For more details, see the comments (in io_apic.c) explainig IO-APIC
1051 * irq migration in the presence of interrupt-remapping.
1053 irte->trigger_mode = 0;
1054 irte->dlvry_mode = apic->irq_delivery_mode;
1055 irte->vector = vector;
1056 irte->dest_id = IRTE_DEST(dest);
1057 irte->redir_hint = 1;
1060 static struct irq_domain *intel_get_ir_irq_domain(struct irq_alloc_info *info)
1062 struct intel_iommu *iommu = NULL;
1067 switch (info->type) {
1068 case X86_IRQ_ALLOC_TYPE_IOAPIC:
1069 iommu = map_ioapic_to_ir(info->ioapic_id);
1071 case X86_IRQ_ALLOC_TYPE_HPET:
1072 iommu = map_hpet_to_ir(info->hpet_id);
1074 case X86_IRQ_ALLOC_TYPE_MSI:
1075 case X86_IRQ_ALLOC_TYPE_MSIX:
1076 iommu = map_dev_to_ir(info->msi_dev);
1083 return iommu ? iommu->ir_domain : NULL;
1086 static struct irq_domain *intel_get_irq_domain(struct irq_alloc_info *info)
1088 struct intel_iommu *iommu;
1093 switch (info->type) {
1094 case X86_IRQ_ALLOC_TYPE_MSI:
1095 case X86_IRQ_ALLOC_TYPE_MSIX:
1096 iommu = map_dev_to_ir(info->msi_dev);
1098 return iommu->ir_msi_domain;
1107 struct irq_remap_ops intel_irq_remap_ops = {
1108 .prepare = intel_prepare_irq_remapping,
1109 .enable = intel_enable_irq_remapping,
1110 .disable = disable_irq_remapping,
1111 .reenable = reenable_irq_remapping,
1112 .enable_faulting = enable_drhd_fault_handling,
1113 .get_ir_irq_domain = intel_get_ir_irq_domain,
1114 .get_irq_domain = intel_get_irq_domain,
1118 * Migrate the IO-APIC irq in the presence of intr-remapping.
1120 * For both level and edge triggered, irq migration is a simple atomic
1121 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
1123 * For level triggered, we eliminate the io-apic RTE modification (with the
1124 * updated vector information), by using a virtual vector (io-apic pin number).
1125 * Real vector that is used for interrupting cpu will be coming from
1126 * the interrupt-remapping table entry.
1128 * As the migration is a simple atomic update of IRTE, the same mechanism
1129 * is used to migrate MSI irq's in the presence of interrupt-remapping.
1132 intel_ir_set_affinity(struct irq_data *data, const struct cpumask *mask,
1135 struct intel_ir_data *ir_data = data->chip_data;
1136 struct irte *irte = &ir_data->irte_entry;
1137 struct irq_cfg *cfg = irqd_cfg(data);
1138 struct irq_data *parent = data->parent_data;
1141 ret = parent->chip->irq_set_affinity(parent, mask, force);
1142 if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
1146 * Atomically updates the IRTE with the new destination, vector
1147 * and flushes the interrupt entry cache.
1149 irte->vector = cfg->vector;
1150 irte->dest_id = IRTE_DEST(cfg->dest_apicid);
1152 /* Update the hardware only if the interrupt is in remapped mode. */
1153 if (ir_data->irq_2_iommu.mode == IRQ_REMAPPING)
1154 modify_irte(&ir_data->irq_2_iommu, irte);
1157 * After this point, all the interrupts will start arriving
1158 * at the new destination. So, time to cleanup the previous
1159 * vector allocation.
1161 send_cleanup_vector(cfg);
1163 return IRQ_SET_MASK_OK_DONE;
1166 static void intel_ir_compose_msi_msg(struct irq_data *irq_data,
1167 struct msi_msg *msg)
1169 struct intel_ir_data *ir_data = irq_data->chip_data;
1171 *msg = ir_data->msi_entry;
1174 static int intel_ir_set_vcpu_affinity(struct irq_data *data, void *info)
1176 struct intel_ir_data *ir_data = data->chip_data;
1177 struct vcpu_data *vcpu_pi_info = info;
1179 /* stop posting interrupts, back to remapping mode */
1180 if (!vcpu_pi_info) {
1181 modify_irte(&ir_data->irq_2_iommu, &ir_data->irte_entry);
1183 struct irte irte_pi;
1186 * We are not caching the posted interrupt entry. We
1187 * copy the data from the remapped entry and modify
1188 * the fields which are relevant for posted mode. The
1189 * cached remapped entry is used for switching back to
1192 memset(&irte_pi, 0, sizeof(irte_pi));
1193 dmar_copy_shared_irte(&irte_pi, &ir_data->irte_entry);
1195 /* Update the posted mode fields */
1197 irte_pi.p_urgent = 0;
1198 irte_pi.p_vector = vcpu_pi_info->vector;
1199 irte_pi.pda_l = (vcpu_pi_info->pi_desc_addr >>
1200 (32 - PDA_LOW_BIT)) & ~(-1UL << PDA_LOW_BIT);
1201 irte_pi.pda_h = (vcpu_pi_info->pi_desc_addr >> 32) &
1202 ~(-1UL << PDA_HIGH_BIT);
1204 modify_irte(&ir_data->irq_2_iommu, &irte_pi);
1210 static struct irq_chip intel_ir_chip = {
1211 .irq_ack = ir_ack_apic_edge,
1212 .irq_set_affinity = intel_ir_set_affinity,
1213 .irq_compose_msi_msg = intel_ir_compose_msi_msg,
1214 .irq_set_vcpu_affinity = intel_ir_set_vcpu_affinity,
1217 static void intel_irq_remapping_prepare_irte(struct intel_ir_data *data,
1218 struct irq_cfg *irq_cfg,
1219 struct irq_alloc_info *info,
1220 int index, int sub_handle)
1222 struct IR_IO_APIC_route_entry *entry;
1223 struct irte *irte = &data->irte_entry;
1224 struct msi_msg *msg = &data->msi_entry;
1226 prepare_irte(irte, irq_cfg->vector, irq_cfg->dest_apicid);
1227 switch (info->type) {
1228 case X86_IRQ_ALLOC_TYPE_IOAPIC:
1229 /* Set source-id of interrupt request */
1230 set_ioapic_sid(irte, info->ioapic_id);
1231 apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: Set IRTE entry (P:%d FPD:%d Dst_Mode:%d Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X Avail:%X Vector:%02X Dest:%08X SID:%04X SQ:%X SVT:%X)\n",
1232 info->ioapic_id, irte->present, irte->fpd,
1233 irte->dst_mode, irte->redir_hint,
1234 irte->trigger_mode, irte->dlvry_mode,
1235 irte->avail, irte->vector, irte->dest_id,
1236 irte->sid, irte->sq, irte->svt);
1238 entry = (struct IR_IO_APIC_route_entry *)info->ioapic_entry;
1239 info->ioapic_entry = NULL;
1240 memset(entry, 0, sizeof(*entry));
1241 entry->index2 = (index >> 15) & 0x1;
1244 entry->index = (index & 0x7fff);
1246 * IO-APIC RTE will be configured with virtual vector.
1247 * irq handler will do the explicit EOI to the io-apic.
1249 entry->vector = info->ioapic_pin;
1250 entry->mask = 0; /* enable IRQ */
1251 entry->trigger = info->ioapic_trigger;
1252 entry->polarity = info->ioapic_polarity;
1253 if (info->ioapic_trigger)
1254 entry->mask = 1; /* Mask level triggered irqs. */
1257 case X86_IRQ_ALLOC_TYPE_HPET:
1258 case X86_IRQ_ALLOC_TYPE_MSI:
1259 case X86_IRQ_ALLOC_TYPE_MSIX:
1260 if (info->type == X86_IRQ_ALLOC_TYPE_HPET)
1261 set_hpet_sid(irte, info->hpet_id);
1263 set_msi_sid(irte, info->msi_dev);
1265 msg->address_hi = MSI_ADDR_BASE_HI;
1266 msg->data = sub_handle;
1267 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
1269 MSI_ADDR_IR_INDEX1(index) |
1270 MSI_ADDR_IR_INDEX2(index);
1279 static void intel_free_irq_resources(struct irq_domain *domain,
1280 unsigned int virq, unsigned int nr_irqs)
1282 struct irq_data *irq_data;
1283 struct intel_ir_data *data;
1284 struct irq_2_iommu *irq_iommu;
1285 unsigned long flags;
1287 for (i = 0; i < nr_irqs; i++) {
1288 irq_data = irq_domain_get_irq_data(domain, virq + i);
1289 if (irq_data && irq_data->chip_data) {
1290 data = irq_data->chip_data;
1291 irq_iommu = &data->irq_2_iommu;
1292 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
1293 clear_entries(irq_iommu);
1294 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
1295 irq_domain_reset_irq_data(irq_data);
1301 static int intel_irq_remapping_alloc(struct irq_domain *domain,
1302 unsigned int virq, unsigned int nr_irqs,
1305 struct intel_iommu *iommu = domain->host_data;
1306 struct irq_alloc_info *info = arg;
1307 struct intel_ir_data *data, *ird;
1308 struct irq_data *irq_data;
1309 struct irq_cfg *irq_cfg;
1312 if (!info || !iommu)
1314 if (nr_irqs > 1 && info->type != X86_IRQ_ALLOC_TYPE_MSI &&
1315 info->type != X86_IRQ_ALLOC_TYPE_MSIX)
1319 * With IRQ remapping enabled, don't need contiguous CPU vectors
1320 * to support multiple MSI interrupts.
1322 if (info->type == X86_IRQ_ALLOC_TYPE_MSI)
1323 info->flags &= ~X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
1325 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
1330 data = kzalloc(sizeof(*data), GFP_KERNEL);
1332 goto out_free_parent;
1334 down_read(&dmar_global_lock);
1335 index = alloc_irte(iommu, virq, &data->irq_2_iommu, nr_irqs);
1336 up_read(&dmar_global_lock);
1338 pr_warn("Failed to allocate IRTE\n");
1340 goto out_free_parent;
1343 for (i = 0; i < nr_irqs; i++) {
1344 irq_data = irq_domain_get_irq_data(domain, virq + i);
1345 irq_cfg = irqd_cfg(irq_data);
1346 if (!irq_data || !irq_cfg) {
1352 ird = kzalloc(sizeof(*ird), GFP_KERNEL);
1355 /* Initialize the common data */
1356 ird->irq_2_iommu = data->irq_2_iommu;
1357 ird->irq_2_iommu.sub_handle = i;
1362 irq_data->hwirq = (index << 16) + i;
1363 irq_data->chip_data = ird;
1364 irq_data->chip = &intel_ir_chip;
1365 intel_irq_remapping_prepare_irte(ird, irq_cfg, info, index, i);
1366 irq_set_status_flags(virq + i, IRQ_MOVE_PCNTXT);
1371 intel_free_irq_resources(domain, virq, i);
1373 irq_domain_free_irqs_common(domain, virq, nr_irqs);
1377 static void intel_irq_remapping_free(struct irq_domain *domain,
1378 unsigned int virq, unsigned int nr_irqs)
1380 intel_free_irq_resources(domain, virq, nr_irqs);
1381 irq_domain_free_irqs_common(domain, virq, nr_irqs);
1384 static void intel_irq_remapping_activate(struct irq_domain *domain,
1385 struct irq_data *irq_data)
1387 struct intel_ir_data *data = irq_data->chip_data;
1389 modify_irte(&data->irq_2_iommu, &data->irte_entry);
1392 static void intel_irq_remapping_deactivate(struct irq_domain *domain,
1393 struct irq_data *irq_data)
1395 struct intel_ir_data *data = irq_data->chip_data;
1398 memset(&entry, 0, sizeof(entry));
1399 modify_irte(&data->irq_2_iommu, &entry);
1402 static struct irq_domain_ops intel_ir_domain_ops = {
1403 .alloc = intel_irq_remapping_alloc,
1404 .free = intel_irq_remapping_free,
1405 .activate = intel_irq_remapping_activate,
1406 .deactivate = intel_irq_remapping_deactivate,
1410 * Support of Interrupt Remapping Unit Hotplug
1412 static int dmar_ir_add(struct dmar_drhd_unit *dmaru, struct intel_iommu *iommu)
1415 int eim = x2apic_enabled();
1417 if (eim && !ecap_eim_support(iommu->ecap)) {
1418 pr_info("DRHD %Lx: EIM not supported by DRHD, ecap %Lx\n",
1419 iommu->reg_phys, iommu->ecap);
1423 if (ir_parse_ioapic_hpet_scope(dmaru->hdr, iommu)) {
1424 pr_warn("DRHD %Lx: failed to parse managed IOAPIC/HPET\n",
1429 /* TODO: check all IOAPICs are covered by IOMMU */
1431 /* Setup Interrupt-remapping now. */
1432 ret = intel_setup_irq_remapping(iommu);
1434 pr_err("Failed to setup irq remapping for %s\n",
1436 intel_teardown_irq_remapping(iommu);
1437 ir_remove_ioapic_hpet_scope(iommu);
1439 iommu_enable_irq_remapping(iommu);
1445 int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert)
1448 struct intel_iommu *iommu = dmaru->iommu;
1450 if (!irq_remapping_enabled)
1454 if (!ecap_ir_support(iommu->ecap))
1456 if (irq_remapping_cap(IRQ_POSTING_CAP) &&
1457 !cap_pi_support(iommu->cap))
1461 if (!iommu->ir_table)
1462 ret = dmar_ir_add(dmaru, iommu);
1464 if (iommu->ir_table) {
1465 if (!bitmap_empty(iommu->ir_table->bitmap,
1466 INTR_REMAP_TABLE_ENTRIES)) {
1469 iommu_disable_irq_remapping(iommu);
1470 intel_teardown_irq_remapping(iommu);
1471 ir_remove_ioapic_hpet_scope(iommu);