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>
18 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
19 static struct hpet_scope ir_hpet[MAX_HPET_TBS];
20 static int ir_ioapic_num, ir_hpet_num;
21 int intr_remapping_enabled;
23 static int disable_intremap;
24 static __init int setup_nointremap(char *str)
29 early_param("nointremap", setup_nointremap);
32 struct intel_iommu *iommu;
38 #ifdef CONFIG_GENERIC_HARDIRQS
39 static struct irq_2_iommu *get_one_free_irq_2_iommu(int node)
41 struct irq_2_iommu *iommu;
43 iommu = kzalloc_node(sizeof(*iommu), GFP_ATOMIC, node);
44 printk(KERN_DEBUG "alloc irq_2_iommu on node %d\n", node);
49 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
51 struct irq_desc *desc;
53 desc = irq_to_desc(irq);
55 if (WARN_ON_ONCE(!desc))
58 return desc->irq_2_iommu;
61 static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
63 struct irq_desc *desc;
64 struct irq_2_iommu *irq_iommu;
66 desc = irq_to_desc(irq);
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(irq_node(irq));
77 return desc->irq_2_iommu;
80 #else /* !CONFIG_SPARSE_IRQ */
82 static struct irq_2_iommu irq_2_iommuX[NR_IRQS];
84 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
87 return &irq_2_iommuX[irq];
91 static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
93 return irq_2_iommu(irq);
97 static DEFINE_SPINLOCK(irq_2_ir_lock);
99 static struct irq_2_iommu *valid_irq_2_iommu(unsigned int irq)
101 struct irq_2_iommu *irq_iommu;
103 irq_iommu = irq_2_iommu(irq);
108 if (!irq_iommu->iommu)
114 int irq_remapped(int irq)
116 return valid_irq_2_iommu(irq) != NULL;
119 int get_irte(int irq, struct irte *entry)
122 struct irq_2_iommu *irq_iommu;
128 spin_lock_irqsave(&irq_2_ir_lock, flags);
129 irq_iommu = valid_irq_2_iommu(irq);
131 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
135 index = irq_iommu->irte_index + irq_iommu->sub_handle;
136 *entry = *(irq_iommu->iommu->ir_table->base + index);
138 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
142 int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
144 struct ir_table *table = iommu->ir_table;
145 struct irq_2_iommu *irq_iommu;
146 u16 index, start_index;
147 unsigned int mask = 0;
154 #ifndef CONFIG_SPARSE_IRQ
155 /* protect irq_2_iommu_alloc later */
161 * start the IRTE search from index 0.
163 index = start_index = 0;
166 count = __roundup_pow_of_two(count);
170 if (mask > ecap_max_handle_mask(iommu->ecap)) {
172 "Requested mask %x exceeds the max invalidation handle"
173 " mask value %Lx\n", mask,
174 ecap_max_handle_mask(iommu->ecap));
178 spin_lock_irqsave(&irq_2_ir_lock, flags);
180 for (i = index; i < index + count; i++)
181 if (table->base[i].present)
183 /* empty index found */
184 if (i == index + count)
187 index = (index + count) % INTR_REMAP_TABLE_ENTRIES;
189 if (index == start_index) {
190 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
191 printk(KERN_ERR "can't allocate an IRTE\n");
196 for (i = index; i < index + count; i++)
197 table->base[i].present = 1;
199 irq_iommu = irq_2_iommu_alloc(irq);
201 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
202 printk(KERN_ERR "can't allocate irq_2_iommu\n");
206 irq_iommu->iommu = iommu;
207 irq_iommu->irte_index = index;
208 irq_iommu->sub_handle = 0;
209 irq_iommu->irte_mask = mask;
211 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
216 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
220 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
224 return qi_submit_sync(&desc, iommu);
227 int map_irq_to_irte_handle(int irq, u16 *sub_handle)
230 struct irq_2_iommu *irq_iommu;
233 spin_lock_irqsave(&irq_2_ir_lock, flags);
234 irq_iommu = valid_irq_2_iommu(irq);
236 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
240 *sub_handle = irq_iommu->sub_handle;
241 index = irq_iommu->irte_index;
242 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
246 int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
248 struct irq_2_iommu *irq_iommu;
251 spin_lock_irqsave(&irq_2_ir_lock, flags);
253 irq_iommu = irq_2_iommu_alloc(irq);
256 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
257 printk(KERN_ERR "can't allocate irq_2_iommu\n");
261 irq_iommu->iommu = iommu;
262 irq_iommu->irte_index = index;
263 irq_iommu->sub_handle = subhandle;
264 irq_iommu->irte_mask = 0;
266 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
271 int clear_irte_irq(int irq, struct intel_iommu *iommu, u16 index)
273 struct irq_2_iommu *irq_iommu;
276 spin_lock_irqsave(&irq_2_ir_lock, flags);
277 irq_iommu = valid_irq_2_iommu(irq);
279 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
283 irq_iommu->iommu = NULL;
284 irq_iommu->irte_index = 0;
285 irq_iommu->sub_handle = 0;
286 irq_2_iommu(irq)->irte_mask = 0;
288 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
293 int modify_irte(int irq, struct irte *irte_modified)
298 struct intel_iommu *iommu;
299 struct irq_2_iommu *irq_iommu;
302 spin_lock_irqsave(&irq_2_ir_lock, flags);
303 irq_iommu = valid_irq_2_iommu(irq);
305 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
309 iommu = irq_iommu->iommu;
311 index = irq_iommu->irte_index + irq_iommu->sub_handle;
312 irte = &iommu->ir_table->base[index];
314 set_64bit((unsigned long *)&irte->low, irte_modified->low);
315 set_64bit((unsigned long *)&irte->high, irte_modified->high);
316 __iommu_flush_cache(iommu, irte, sizeof(*irte));
318 rc = qi_flush_iec(iommu, index, 0);
319 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
324 int flush_irte(int irq)
328 struct intel_iommu *iommu;
329 struct irq_2_iommu *irq_iommu;
332 spin_lock_irqsave(&irq_2_ir_lock, flags);
333 irq_iommu = valid_irq_2_iommu(irq);
335 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
339 iommu = irq_iommu->iommu;
341 index = irq_iommu->irte_index + irq_iommu->sub_handle;
343 rc = qi_flush_iec(iommu, index, irq_iommu->irte_mask);
344 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
349 struct intel_iommu *map_hpet_to_ir(u8 hpet_id)
353 for (i = 0; i < MAX_HPET_TBS; i++)
354 if (ir_hpet[i].id == hpet_id)
355 return ir_hpet[i].iommu;
359 struct intel_iommu *map_ioapic_to_ir(int apic)
363 for (i = 0; i < MAX_IO_APICS; i++)
364 if (ir_ioapic[i].id == apic)
365 return ir_ioapic[i].iommu;
369 struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
371 struct dmar_drhd_unit *drhd;
373 drhd = dmar_find_matched_drhd_unit(dev);
380 static int clear_entries(struct irq_2_iommu *irq_iommu)
382 struct irte *start, *entry, *end;
383 struct intel_iommu *iommu;
386 if (irq_iommu->sub_handle)
389 iommu = irq_iommu->iommu;
390 index = irq_iommu->irte_index + irq_iommu->sub_handle;
392 start = iommu->ir_table->base + index;
393 end = start + (1 << irq_iommu->irte_mask);
395 for (entry = start; entry < end; entry++) {
396 set_64bit((unsigned long *)&entry->low, 0);
397 set_64bit((unsigned long *)&entry->high, 0);
400 return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
403 int free_irte(int irq)
406 struct irq_2_iommu *irq_iommu;
409 spin_lock_irqsave(&irq_2_ir_lock, flags);
410 irq_iommu = valid_irq_2_iommu(irq);
412 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
416 rc = clear_entries(irq_iommu);
418 irq_iommu->iommu = NULL;
419 irq_iommu->irte_index = 0;
420 irq_iommu->sub_handle = 0;
421 irq_iommu->irte_mask = 0;
423 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
429 * source validation type
431 #define SVT_NO_VERIFY 0x0 /* no verification is required */
432 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fiels */
433 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
436 * source-id qualifier
438 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
439 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
440 * the third least significant bit
442 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
443 * the second and third least significant bits
445 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
446 * the least three significant bits
450 * set SVT, SQ and SID fields of irte to verify
451 * source ids of interrupt requests
453 static void set_irte_sid(struct irte *irte, unsigned int svt,
454 unsigned int sq, unsigned int sid)
461 int set_ioapic_sid(struct irte *irte, int apic)
469 for (i = 0; i < MAX_IO_APICS; i++) {
470 if (ir_ioapic[i].id == apic) {
471 sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
477 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
481 set_irte_sid(irte, 1, 0, sid);
486 int set_hpet_sid(struct irte *irte, u8 id)
494 for (i = 0; i < MAX_HPET_TBS; i++) {
495 if (ir_hpet[i].id == id) {
496 sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
502 pr_warning("Failed to set source-id of HPET block (%d)\n", id);
507 * Should really use SQ_ALL_16. Some platforms are broken.
508 * While we figure out the right quirks for these broken platforms, use
509 * SQ_13_IGNORE_3 for now.
511 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
516 int set_msi_sid(struct irte *irte, struct pci_dev *dev)
518 struct pci_dev *bridge;
523 /* PCIe device or Root Complex integrated PCI device */
524 if (pci_is_pcie(dev) || !dev->bus->parent) {
525 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
526 (dev->bus->number << 8) | dev->devfn);
530 bridge = pci_find_upstream_pcie_bridge(dev);
532 if (pci_is_pcie(bridge))/* this is a PCIe-to-PCI/PCIX bridge */
533 set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
534 (bridge->bus->number << 8) | dev->bus->number);
535 else /* this is a legacy PCI bridge */
536 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
537 (bridge->bus->number << 8) | bridge->devfn);
543 static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
549 addr = virt_to_phys((void *)iommu->ir_table->base);
551 spin_lock_irqsave(&iommu->register_lock, flags);
553 dmar_writeq(iommu->reg + DMAR_IRTA_REG,
554 (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
556 /* Set interrupt-remapping table pointer */
557 iommu->gcmd |= DMA_GCMD_SIRTP;
558 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
560 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
561 readl, (sts & DMA_GSTS_IRTPS), sts);
562 spin_unlock_irqrestore(&iommu->register_lock, flags);
565 * global invalidation of interrupt entry cache before enabling
566 * interrupt-remapping.
568 qi_global_iec(iommu);
570 spin_lock_irqsave(&iommu->register_lock, flags);
572 /* Enable interrupt-remapping */
573 iommu->gcmd |= DMA_GCMD_IRE;
574 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
576 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
577 readl, (sts & DMA_GSTS_IRES), sts);
579 spin_unlock_irqrestore(&iommu->register_lock, flags);
583 static int setup_intr_remapping(struct intel_iommu *iommu, int mode)
585 struct ir_table *ir_table;
588 ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
591 if (!iommu->ir_table)
594 pages = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
595 INTR_REMAP_PAGE_ORDER);
598 printk(KERN_ERR "failed to allocate pages of order %d\n",
599 INTR_REMAP_PAGE_ORDER);
600 kfree(iommu->ir_table);
604 ir_table->base = page_address(pages);
606 iommu_set_intr_remapping(iommu, mode);
611 * Disable Interrupt Remapping.
613 static void iommu_disable_intr_remapping(struct intel_iommu *iommu)
618 if (!ecap_ir_support(iommu->ecap))
622 * global invalidation of interrupt entry cache before disabling
623 * interrupt-remapping.
625 qi_global_iec(iommu);
627 spin_lock_irqsave(&iommu->register_lock, flags);
629 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
630 if (!(sts & DMA_GSTS_IRES))
633 iommu->gcmd &= ~DMA_GCMD_IRE;
634 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
636 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
637 readl, !(sts & DMA_GSTS_IRES), sts);
640 spin_unlock_irqrestore(&iommu->register_lock, flags);
643 int __init intr_remapping_supported(void)
645 struct dmar_drhd_unit *drhd;
647 if (disable_intremap)
650 if (!dmar_ir_support())
653 for_each_drhd_unit(drhd) {
654 struct intel_iommu *iommu = drhd->iommu;
656 if (!ecap_ir_support(iommu->ecap))
663 int __init enable_intr_remapping(int eim)
665 struct dmar_drhd_unit *drhd;
668 if (parse_ioapics_under_ir() != 1) {
669 printk(KERN_INFO "Not enable interrupt remapping\n");
673 for_each_drhd_unit(drhd) {
674 struct intel_iommu *iommu = drhd->iommu;
677 * If the queued invalidation is already initialized,
678 * shouldn't disable it.
684 * Clear previous faults.
686 dmar_fault(-1, iommu);
689 * Disable intr remapping and queued invalidation, if already
690 * enabled prior to OS handover.
692 iommu_disable_intr_remapping(iommu);
694 dmar_disable_qi(iommu);
698 * check for the Interrupt-remapping support
700 for_each_drhd_unit(drhd) {
701 struct intel_iommu *iommu = drhd->iommu;
703 if (!ecap_ir_support(iommu->ecap))
706 if (eim && !ecap_eim_support(iommu->ecap)) {
707 printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
708 " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
714 * Enable queued invalidation for all the DRHD's.
716 for_each_drhd_unit(drhd) {
718 struct intel_iommu *iommu = drhd->iommu;
719 ret = dmar_enable_qi(iommu);
722 printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
723 " invalidation, ecap %Lx, ret %d\n",
724 drhd->reg_base_addr, iommu->ecap, ret);
730 * Setup Interrupt-remapping for all the DRHD's now.
732 for_each_drhd_unit(drhd) {
733 struct intel_iommu *iommu = drhd->iommu;
735 if (!ecap_ir_support(iommu->ecap))
738 if (setup_intr_remapping(iommu, eim))
747 intr_remapping_enabled = 1;
753 * handle error condition gracefully here!
758 static void ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
759 struct intel_iommu *iommu)
761 struct acpi_dmar_pci_path *path;
766 path = (struct acpi_dmar_pci_path *)(scope + 1);
767 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
768 / sizeof(struct acpi_dmar_pci_path);
770 while (--count > 0) {
772 * Access PCI directly due to the PCI
773 * subsystem isn't initialized yet.
775 bus = read_pci_config_byte(bus, path->dev, path->fn,
779 ir_hpet[ir_hpet_num].bus = bus;
780 ir_hpet[ir_hpet_num].devfn = PCI_DEVFN(path->dev, path->fn);
781 ir_hpet[ir_hpet_num].iommu = iommu;
782 ir_hpet[ir_hpet_num].id = scope->enumeration_id;
786 static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
787 struct intel_iommu *iommu)
789 struct acpi_dmar_pci_path *path;
794 path = (struct acpi_dmar_pci_path *)(scope + 1);
795 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
796 / sizeof(struct acpi_dmar_pci_path);
798 while (--count > 0) {
800 * Access PCI directly due to the PCI
801 * subsystem isn't initialized yet.
803 bus = read_pci_config_byte(bus, path->dev, path->fn,
808 ir_ioapic[ir_ioapic_num].bus = bus;
809 ir_ioapic[ir_ioapic_num].devfn = PCI_DEVFN(path->dev, path->fn);
810 ir_ioapic[ir_ioapic_num].iommu = iommu;
811 ir_ioapic[ir_ioapic_num].id = scope->enumeration_id;
815 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
816 struct intel_iommu *iommu)
818 struct acpi_dmar_hardware_unit *drhd;
819 struct acpi_dmar_device_scope *scope;
822 drhd = (struct acpi_dmar_hardware_unit *)header;
824 start = (void *)(drhd + 1);
825 end = ((void *)drhd) + header->length;
827 while (start < end) {
829 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
830 if (ir_ioapic_num == MAX_IO_APICS) {
831 printk(KERN_WARNING "Exceeded Max IO APICS\n");
835 printk(KERN_INFO "IOAPIC id %d under DRHD base"
836 " 0x%Lx\n", scope->enumeration_id,
839 ir_parse_one_ioapic_scope(scope, iommu);
840 } else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET) {
841 if (ir_hpet_num == MAX_HPET_TBS) {
842 printk(KERN_WARNING "Exceeded Max HPET blocks\n");
846 printk(KERN_INFO "HPET id %d under DRHD base"
847 " 0x%Lx\n", scope->enumeration_id,
850 ir_parse_one_hpet_scope(scope, iommu);
852 start += scope->length;
859 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
862 int __init parse_ioapics_under_ir(void)
864 struct dmar_drhd_unit *drhd;
865 int ir_supported = 0;
867 for_each_drhd_unit(drhd) {
868 struct intel_iommu *iommu = drhd->iommu;
870 if (ecap_ir_support(iommu->ecap)) {
871 if (ir_parse_ioapic_hpet_scope(drhd->hdr, iommu))
878 if (ir_supported && ir_ioapic_num != nr_ioapics) {
880 "Not all IO-APIC's listed under remapping hardware\n");
887 void disable_intr_remapping(void)
889 struct dmar_drhd_unit *drhd;
890 struct intel_iommu *iommu = NULL;
893 * Disable Interrupt-remapping for all the DRHD's now.
895 for_each_iommu(iommu, drhd) {
896 if (!ecap_ir_support(iommu->ecap))
899 iommu_disable_intr_remapping(iommu);
903 int reenable_intr_remapping(int eim)
905 struct dmar_drhd_unit *drhd;
907 struct intel_iommu *iommu = NULL;
909 for_each_iommu(iommu, drhd)
911 dmar_reenable_qi(iommu);
914 * Setup Interrupt-remapping for all the DRHD's now.
916 for_each_iommu(iommu, drhd) {
917 if (!ecap_ir_support(iommu->ecap))
920 /* Set up interrupt remapping for iommu.*/
921 iommu_set_intr_remapping(iommu, eim);
932 * handle error condition gracefully here!