x86, pat: Fix set_memory_wc related corruption
[linux-2.6/mini2440.git] / drivers / pci / intr_remapping.c
blob4f5b8712931f0d34c1702440a80ff0bb71f46964
1 #include <linux/interrupt.h>
2 #include <linux/dmar.h>
3 #include <linux/spinlock.h>
4 #include <linux/jiffies.h>
5 #include <linux/pci.h>
6 #include <linux/irq.h>
7 #include <asm/io_apic.h>
8 #include <asm/smp.h>
9 #include <asm/cpu.h>
10 #include <linux/intel-iommu.h>
11 #include "intr_remapping.h"
12 #include <acpi/acpi.h>
13 #include <asm/pci-direct.h>
14 #include "pci.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)
23 disable_intremap = 1;
24 return 0;
26 early_param("nointremap", setup_nointremap);
28 struct irq_2_iommu {
29 struct intel_iommu *iommu;
30 u16 irte_index;
31 u16 sub_handle;
32 u8 irte_mask;
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);
43 return iommu;
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))
53 return NULL;
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);
67 if (!desc) {
68 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
69 return NULL;
72 irq_iommu = desc->irq_2_iommu;
74 if (!irq_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)
91 if (irq < nr_irqs)
92 return &irq_2_iommuX[irq];
94 return NULL;
96 static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
98 return irq_2_iommu(irq);
100 #endif
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);
110 if (!irq_iommu)
111 return NULL;
113 if (!irq_iommu->iommu)
114 return NULL;
116 return irq_iommu;
119 int irq_remapped(int irq)
121 return valid_irq_2_iommu(irq) != NULL;
124 int get_irte(int irq, struct irte *entry)
126 int index;
127 struct irq_2_iommu *irq_iommu;
128 unsigned long flags;
130 if (!entry)
131 return -1;
133 spin_lock_irqsave(&irq_2_ir_lock, flags);
134 irq_iommu = valid_irq_2_iommu(irq);
135 if (!irq_iommu) {
136 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
137 return -1;
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);
144 return 0;
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;
153 unsigned long flags;
154 int i;
156 if (!count)
157 return -1;
159 #ifndef CONFIG_SPARSE_IRQ
160 /* protect irq_2_iommu_alloc later */
161 if (irq >= nr_irqs)
162 return -1;
163 #endif
166 * start the IRTE search from index 0.
168 index = start_index = 0;
170 if (count > 1) {
171 count = __roundup_pow_of_two(count);
172 mask = ilog2(count);
175 if (mask > ecap_max_handle_mask(iommu->ecap)) {
176 printk(KERN_ERR
177 "Requested mask %x exceeds the max invalidation handle"
178 " mask value %Lx\n", mask,
179 ecap_max_handle_mask(iommu->ecap));
180 return -1;
183 spin_lock_irqsave(&irq_2_ir_lock, flags);
184 do {
185 for (i = index; i < index + count; i++)
186 if (table->base[i].present)
187 break;
188 /* empty index found */
189 if (i == index + count)
190 break;
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");
197 return -1;
199 } while (1);
201 for (i = index; i < index + count; i++)
202 table->base[i].present = 1;
204 irq_iommu = irq_2_iommu_alloc(irq);
205 if (!irq_iommu) {
206 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
207 printk(KERN_ERR "can't allocate irq_2_iommu\n");
208 return -1;
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);
218 return index;
221 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
223 struct qi_desc desc;
225 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
226 | QI_IEC_SELECTIVE;
227 desc.high = 0;
229 return qi_submit_sync(&desc, iommu);
232 int map_irq_to_irte_handle(int irq, u16 *sub_handle)
234 int index;
235 struct irq_2_iommu *irq_iommu;
236 unsigned long flags;
238 spin_lock_irqsave(&irq_2_ir_lock, flags);
239 irq_iommu = valid_irq_2_iommu(irq);
240 if (!irq_iommu) {
241 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
242 return -1;
245 *sub_handle = irq_iommu->sub_handle;
246 index = irq_iommu->irte_index;
247 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
248 return index;
251 int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
253 struct irq_2_iommu *irq_iommu;
254 unsigned long flags;
256 spin_lock_irqsave(&irq_2_ir_lock, flags);
258 irq_iommu = irq_2_iommu_alloc(irq);
260 if (!irq_iommu) {
261 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
262 printk(KERN_ERR "can't allocate irq_2_iommu\n");
263 return -1;
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);
273 return 0;
276 int clear_irte_irq(int irq, struct intel_iommu *iommu, u16 index)
278 struct irq_2_iommu *irq_iommu;
279 unsigned long flags;
281 spin_lock_irqsave(&irq_2_ir_lock, flags);
282 irq_iommu = valid_irq_2_iommu(irq);
283 if (!irq_iommu) {
284 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
285 return -1;
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);
295 return 0;
298 int modify_irte(int irq, struct irte *irte_modified)
300 int rc;
301 int index;
302 struct irte *irte;
303 struct intel_iommu *iommu;
304 struct irq_2_iommu *irq_iommu;
305 unsigned long flags;
307 spin_lock_irqsave(&irq_2_ir_lock, flags);
308 irq_iommu = valid_irq_2_iommu(irq);
309 if (!irq_iommu) {
310 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
311 return -1;
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);
326 return rc;
329 int flush_irte(int irq)
331 int rc;
332 int index;
333 struct intel_iommu *iommu;
334 struct irq_2_iommu *irq_iommu;
335 unsigned long flags;
337 spin_lock_irqsave(&irq_2_ir_lock, flags);
338 irq_iommu = valid_irq_2_iommu(irq);
339 if (!irq_iommu) {
340 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
341 return -1;
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);
351 return rc;
354 struct intel_iommu *map_ioapic_to_ir(int apic)
356 int i;
358 for (i = 0; i < MAX_IO_APICS; i++)
359 if (ir_ioapic[i].id == apic)
360 return ir_ioapic[i].iommu;
361 return NULL;
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);
369 if (!drhd)
370 return NULL;
372 return drhd->iommu;
375 static int clear_entries(struct irq_2_iommu *irq_iommu)
377 struct irte *start, *entry, *end;
378 struct intel_iommu *iommu;
379 int index;
381 if (irq_iommu->sub_handle)
382 return 0;
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)
400 int rc = 0;
401 struct irq_2_iommu *irq_iommu;
402 unsigned long flags;
404 spin_lock_irqsave(&irq_2_ir_lock, flags);
405 irq_iommu = valid_irq_2_iommu(irq);
406 if (!irq_iommu) {
407 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
408 return -1;
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);
420 return rc;
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)
451 irte->svt = svt;
452 irte->sq = sq;
453 irte->sid = sid;
456 int set_ioapic_sid(struct irte *irte, int apic)
458 int i;
459 u16 sid = 0;
461 if (!irte)
462 return -1;
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;
467 break;
471 if (sid == 0) {
472 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
473 return -1;
476 set_irte_sid(irte, 1, 0, sid);
478 return 0;
481 int set_msi_sid(struct irte *irte, struct pci_dev *dev)
483 struct pci_dev *bridge;
485 if (!irte || !dev)
486 return -1;
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);
492 return 0;
495 bridge = pci_find_upstream_pcie_bridge(dev);
496 if (bridge) {
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);
505 return 0;
508 static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
510 u64 addr;
511 u32 sts;
512 unsigned long flags;
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;
551 struct page *pages;
553 ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
554 GFP_ATOMIC);
556 if (!iommu->ir_table)
557 return -ENOMEM;
559 pages = alloc_pages(GFP_ATOMIC | __GFP_ZERO, INTR_REMAP_PAGE_ORDER);
561 if (!pages) {
562 printk(KERN_ERR "failed to allocate pages of order %d\n",
563 INTR_REMAP_PAGE_ORDER);
564 kfree(iommu->ir_table);
565 return -ENOMEM;
568 ir_table->base = page_address(pages);
570 iommu_set_intr_remapping(iommu, mode);
571 return 0;
575 * Disable Interrupt Remapping.
577 static void iommu_disable_intr_remapping(struct intel_iommu *iommu)
579 unsigned long flags;
580 u32 sts;
582 if (!ecap_ir_support(iommu->ecap))
583 return;
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))
595 goto end;
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);
603 end:
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)
612 return 0;
614 for_each_drhd_unit(drhd) {
615 struct intel_iommu *iommu = drhd->iommu;
617 if (!ecap_ir_support(iommu->ecap))
618 return 0;
621 return 1;
624 int __init enable_intr_remapping(int eim)
626 struct dmar_drhd_unit *drhd;
627 int setup = 0;
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.
636 if (iommu->qi)
637 continue;
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))
660 continue;
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);
665 return -1;
670 * Enable queued invalidation for all the DRHD's.
672 for_each_drhd_unit(drhd) {
673 int ret;
674 struct intel_iommu *iommu = drhd->iommu;
675 ret = dmar_enable_qi(iommu);
677 if (ret) {
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);
681 return -1;
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))
692 continue;
694 if (setup_intr_remapping(iommu, eim))
695 goto error;
697 setup = 1;
700 if (!setup)
701 goto error;
703 intr_remapping_enabled = 1;
705 return 0;
707 error:
709 * handle error condition gracefully here!
711 return -1;
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;
718 u8 bus;
719 int count;
721 bus = scope->bus;
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,
732 PCI_SECONDARY_BUS);
733 path++;
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;
740 ir_ioapic_num++;
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;
748 void *start, *end;
750 drhd = (struct acpi_dmar_hardware_unit *)header;
752 start = (void *)(drhd + 1);
753 end = ((void *)drhd) + header->length;
755 while (start < end) {
756 scope = start;
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");
760 return -1;
763 printk(KERN_INFO "IOAPIC id %d under DRHD base"
764 " 0x%Lx\n", scope->enumeration_id,
765 drhd->address);
767 ir_parse_one_ioapic_scope(scope, iommu);
769 start += scope->length;
772 return 0;
776 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
777 * hardware unit.
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))
789 return -1;
791 ir_supported = 1;
795 if (ir_supported && ir_ioapic_num != nr_ioapics) {
796 printk(KERN_WARNING
797 "Not all IO-APIC's listed under remapping hardware\n");
798 return -1;
801 return ir_supported;
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))
814 continue;
816 iommu_disable_intr_remapping(iommu);
820 int reenable_intr_remapping(int eim)
822 struct dmar_drhd_unit *drhd;
823 int setup = 0;
824 struct intel_iommu *iommu = NULL;
826 for_each_iommu(iommu, drhd)
827 if (iommu->qi)
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))
835 continue;
837 /* Set up interrupt remapping for iommu.*/
838 iommu_set_intr_remapping(iommu, eim);
839 setup = 1;
842 if (!setup)
843 goto error;
845 return 0;
847 error:
849 * handle error condition gracefully here!
851 return -1;