x86-64, NUMA: Unify use of memblk in all init methods
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / pci / intr_remapping.c
blobec87cd66f3eb21b792381b52c6d8f1656beadede
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>
7 #include <linux/pci.h>
8 #include <linux/irq.h>
9 #include <asm/io_apic.h>
10 #include <asm/smp.h>
11 #include <asm/cpu.h>
12 #include <linux/intel-iommu.h>
13 #include "intr_remapping.h"
14 #include <acpi/acpi.h>
15 #include <asm/pci-direct.h>
16 #include "pci.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 int disable_sourceid_checking;
26 static __init int setup_nointremap(char *str)
28 disable_intremap = 1;
29 return 0;
31 early_param("nointremap", setup_nointremap);
33 static __init int setup_intremap(char *str)
35 if (!str)
36 return -EINVAL;
38 if (!strncmp(str, "on", 2))
39 disable_intremap = 0;
40 else if (!strncmp(str, "off", 3))
41 disable_intremap = 1;
42 else if (!strncmp(str, "nosid", 5))
43 disable_sourceid_checking = 1;
45 return 0;
47 early_param("intremap", setup_intremap);
49 static DEFINE_SPINLOCK(irq_2_ir_lock);
51 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
53 struct irq_cfg *cfg = get_irq_chip_data(irq);
54 return cfg ? &cfg->irq_2_iommu : NULL;
57 int get_irte(int irq, struct irte *entry)
59 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
60 unsigned long flags;
61 int index;
63 if (!entry || !irq_iommu)
64 return -1;
66 spin_lock_irqsave(&irq_2_ir_lock, flags);
68 index = irq_iommu->irte_index + irq_iommu->sub_handle;
69 *entry = *(irq_iommu->iommu->ir_table->base + index);
71 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
72 return 0;
75 int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
77 struct ir_table *table = iommu->ir_table;
78 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
79 u16 index, start_index;
80 unsigned int mask = 0;
81 unsigned long flags;
82 int i;
84 if (!count || !irq_iommu)
85 return -1;
88 * start the IRTE search from index 0.
90 index = start_index = 0;
92 if (count > 1) {
93 count = __roundup_pow_of_two(count);
94 mask = ilog2(count);
97 if (mask > ecap_max_handle_mask(iommu->ecap)) {
98 printk(KERN_ERR
99 "Requested mask %x exceeds the max invalidation handle"
100 " mask value %Lx\n", mask,
101 ecap_max_handle_mask(iommu->ecap));
102 return -1;
105 spin_lock_irqsave(&irq_2_ir_lock, flags);
106 do {
107 for (i = index; i < index + count; i++)
108 if (table->base[i].present)
109 break;
110 /* empty index found */
111 if (i == index + count)
112 break;
114 index = (index + count) % INTR_REMAP_TABLE_ENTRIES;
116 if (index == start_index) {
117 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
118 printk(KERN_ERR "can't allocate an IRTE\n");
119 return -1;
121 } while (1);
123 for (i = index; i < index + count; i++)
124 table->base[i].present = 1;
126 irq_iommu->iommu = iommu;
127 irq_iommu->irte_index = index;
128 irq_iommu->sub_handle = 0;
129 irq_iommu->irte_mask = mask;
131 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
133 return index;
136 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
138 struct qi_desc desc;
140 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
141 | QI_IEC_SELECTIVE;
142 desc.high = 0;
144 return qi_submit_sync(&desc, iommu);
147 int map_irq_to_irte_handle(int irq, u16 *sub_handle)
149 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
150 unsigned long flags;
151 int index;
153 if (!irq_iommu)
154 return -1;
156 spin_lock_irqsave(&irq_2_ir_lock, flags);
157 *sub_handle = irq_iommu->sub_handle;
158 index = irq_iommu->irte_index;
159 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
160 return index;
163 int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
165 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
166 unsigned long flags;
168 if (!irq_iommu)
169 return -1;
171 spin_lock_irqsave(&irq_2_ir_lock, flags);
173 irq_iommu->iommu = iommu;
174 irq_iommu->irte_index = index;
175 irq_iommu->sub_handle = subhandle;
176 irq_iommu->irte_mask = 0;
178 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
180 return 0;
183 int modify_irte(int irq, struct irte *irte_modified)
185 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
186 struct intel_iommu *iommu;
187 unsigned long flags;
188 struct irte *irte;
189 int rc, index;
191 if (!irq_iommu)
192 return -1;
194 spin_lock_irqsave(&irq_2_ir_lock, flags);
196 iommu = irq_iommu->iommu;
198 index = irq_iommu->irte_index + irq_iommu->sub_handle;
199 irte = &iommu->ir_table->base[index];
201 set_64bit(&irte->low, irte_modified->low);
202 set_64bit(&irte->high, irte_modified->high);
203 __iommu_flush_cache(iommu, irte, sizeof(*irte));
205 rc = qi_flush_iec(iommu, index, 0);
206 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
208 return rc;
211 struct intel_iommu *map_hpet_to_ir(u8 hpet_id)
213 int i;
215 for (i = 0; i < MAX_HPET_TBS; i++)
216 if (ir_hpet[i].id == hpet_id)
217 return ir_hpet[i].iommu;
218 return NULL;
221 struct intel_iommu *map_ioapic_to_ir(int apic)
223 int i;
225 for (i = 0; i < MAX_IO_APICS; i++)
226 if (ir_ioapic[i].id == apic)
227 return ir_ioapic[i].iommu;
228 return NULL;
231 struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
233 struct dmar_drhd_unit *drhd;
235 drhd = dmar_find_matched_drhd_unit(dev);
236 if (!drhd)
237 return NULL;
239 return drhd->iommu;
242 static int clear_entries(struct irq_2_iommu *irq_iommu)
244 struct irte *start, *entry, *end;
245 struct intel_iommu *iommu;
246 int index;
248 if (irq_iommu->sub_handle)
249 return 0;
251 iommu = irq_iommu->iommu;
252 index = irq_iommu->irte_index + irq_iommu->sub_handle;
254 start = iommu->ir_table->base + index;
255 end = start + (1 << irq_iommu->irte_mask);
257 for (entry = start; entry < end; entry++) {
258 set_64bit(&entry->low, 0);
259 set_64bit(&entry->high, 0);
262 return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
265 int free_irte(int irq)
267 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
268 unsigned long flags;
269 int rc;
271 if (!irq_iommu)
272 return -1;
274 spin_lock_irqsave(&irq_2_ir_lock, flags);
276 rc = clear_entries(irq_iommu);
278 irq_iommu->iommu = NULL;
279 irq_iommu->irte_index = 0;
280 irq_iommu->sub_handle = 0;
281 irq_iommu->irte_mask = 0;
283 spin_unlock_irqrestore(&irq_2_ir_lock, flags);
285 return rc;
289 * source validation type
291 #define SVT_NO_VERIFY 0x0 /* no verification is required */
292 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fiels */
293 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
296 * source-id qualifier
298 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
299 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
300 * the third least significant bit
302 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
303 * the second and third least significant bits
305 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
306 * the least three significant bits
310 * set SVT, SQ and SID fields of irte to verify
311 * source ids of interrupt requests
313 static void set_irte_sid(struct irte *irte, unsigned int svt,
314 unsigned int sq, unsigned int sid)
316 if (disable_sourceid_checking)
317 svt = SVT_NO_VERIFY;
318 irte->svt = svt;
319 irte->sq = sq;
320 irte->sid = sid;
323 int set_ioapic_sid(struct irte *irte, int apic)
325 int i;
326 u16 sid = 0;
328 if (!irte)
329 return -1;
331 for (i = 0; i < MAX_IO_APICS; i++) {
332 if (ir_ioapic[i].id == apic) {
333 sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
334 break;
338 if (sid == 0) {
339 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
340 return -1;
343 set_irte_sid(irte, 1, 0, sid);
345 return 0;
348 int set_hpet_sid(struct irte *irte, u8 id)
350 int i;
351 u16 sid = 0;
353 if (!irte)
354 return -1;
356 for (i = 0; i < MAX_HPET_TBS; i++) {
357 if (ir_hpet[i].id == id) {
358 sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
359 break;
363 if (sid == 0) {
364 pr_warning("Failed to set source-id of HPET block (%d)\n", id);
365 return -1;
369 * Should really use SQ_ALL_16. Some platforms are broken.
370 * While we figure out the right quirks for these broken platforms, use
371 * SQ_13_IGNORE_3 for now.
373 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
375 return 0;
378 int set_msi_sid(struct irte *irte, struct pci_dev *dev)
380 struct pci_dev *bridge;
382 if (!irte || !dev)
383 return -1;
385 /* PCIe device or Root Complex integrated PCI device */
386 if (pci_is_pcie(dev) || !dev->bus->parent) {
387 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
388 (dev->bus->number << 8) | dev->devfn);
389 return 0;
392 bridge = pci_find_upstream_pcie_bridge(dev);
393 if (bridge) {
394 if (pci_is_pcie(bridge))/* this is a PCIe-to-PCI/PCIX bridge */
395 set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
396 (bridge->bus->number << 8) | dev->bus->number);
397 else /* this is a legacy PCI bridge */
398 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
399 (bridge->bus->number << 8) | bridge->devfn);
402 return 0;
405 static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
407 u64 addr;
408 u32 sts;
409 unsigned long flags;
411 addr = virt_to_phys((void *)iommu->ir_table->base);
413 spin_lock_irqsave(&iommu->register_lock, flags);
415 dmar_writeq(iommu->reg + DMAR_IRTA_REG,
416 (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
418 /* Set interrupt-remapping table pointer */
419 iommu->gcmd |= DMA_GCMD_SIRTP;
420 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
422 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
423 readl, (sts & DMA_GSTS_IRTPS), sts);
424 spin_unlock_irqrestore(&iommu->register_lock, flags);
427 * global invalidation of interrupt entry cache before enabling
428 * interrupt-remapping.
430 qi_global_iec(iommu);
432 spin_lock_irqsave(&iommu->register_lock, flags);
434 /* Enable interrupt-remapping */
435 iommu->gcmd |= DMA_GCMD_IRE;
436 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
438 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
439 readl, (sts & DMA_GSTS_IRES), sts);
441 spin_unlock_irqrestore(&iommu->register_lock, flags);
445 static int setup_intr_remapping(struct intel_iommu *iommu, int mode)
447 struct ir_table *ir_table;
448 struct page *pages;
450 ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
451 GFP_ATOMIC);
453 if (!iommu->ir_table)
454 return -ENOMEM;
456 pages = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
457 INTR_REMAP_PAGE_ORDER);
459 if (!pages) {
460 printk(KERN_ERR "failed to allocate pages of order %d\n",
461 INTR_REMAP_PAGE_ORDER);
462 kfree(iommu->ir_table);
463 return -ENOMEM;
466 ir_table->base = page_address(pages);
468 iommu_set_intr_remapping(iommu, mode);
469 return 0;
473 * Disable Interrupt Remapping.
475 static void iommu_disable_intr_remapping(struct intel_iommu *iommu)
477 unsigned long flags;
478 u32 sts;
480 if (!ecap_ir_support(iommu->ecap))
481 return;
484 * global invalidation of interrupt entry cache before disabling
485 * interrupt-remapping.
487 qi_global_iec(iommu);
489 spin_lock_irqsave(&iommu->register_lock, flags);
491 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
492 if (!(sts & DMA_GSTS_IRES))
493 goto end;
495 iommu->gcmd &= ~DMA_GCMD_IRE;
496 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
498 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
499 readl, !(sts & DMA_GSTS_IRES), sts);
501 end:
502 spin_unlock_irqrestore(&iommu->register_lock, flags);
505 int __init intr_remapping_supported(void)
507 struct dmar_drhd_unit *drhd;
509 if (disable_intremap)
510 return 0;
512 if (!dmar_ir_support())
513 return 0;
515 for_each_drhd_unit(drhd) {
516 struct intel_iommu *iommu = drhd->iommu;
518 if (!ecap_ir_support(iommu->ecap))
519 return 0;
522 return 1;
525 int __init enable_intr_remapping(int eim)
527 struct dmar_drhd_unit *drhd;
528 int setup = 0;
530 if (parse_ioapics_under_ir() != 1) {
531 printk(KERN_INFO "Not enable interrupt remapping\n");
532 return -1;
535 for_each_drhd_unit(drhd) {
536 struct intel_iommu *iommu = drhd->iommu;
539 * If the queued invalidation is already initialized,
540 * shouldn't disable it.
542 if (iommu->qi)
543 continue;
546 * Clear previous faults.
548 dmar_fault(-1, iommu);
551 * Disable intr remapping and queued invalidation, if already
552 * enabled prior to OS handover.
554 iommu_disable_intr_remapping(iommu);
556 dmar_disable_qi(iommu);
560 * check for the Interrupt-remapping support
562 for_each_drhd_unit(drhd) {
563 struct intel_iommu *iommu = drhd->iommu;
565 if (!ecap_ir_support(iommu->ecap))
566 continue;
568 if (eim && !ecap_eim_support(iommu->ecap)) {
569 printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
570 " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
571 return -1;
576 * Enable queued invalidation for all the DRHD's.
578 for_each_drhd_unit(drhd) {
579 int ret;
580 struct intel_iommu *iommu = drhd->iommu;
581 ret = dmar_enable_qi(iommu);
583 if (ret) {
584 printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
585 " invalidation, ecap %Lx, ret %d\n",
586 drhd->reg_base_addr, iommu->ecap, ret);
587 return -1;
592 * Setup Interrupt-remapping for all the DRHD's now.
594 for_each_drhd_unit(drhd) {
595 struct intel_iommu *iommu = drhd->iommu;
597 if (!ecap_ir_support(iommu->ecap))
598 continue;
600 if (setup_intr_remapping(iommu, eim))
601 goto error;
603 setup = 1;
606 if (!setup)
607 goto error;
609 intr_remapping_enabled = 1;
611 return 0;
613 error:
615 * handle error condition gracefully here!
617 return -1;
620 static void ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
621 struct intel_iommu *iommu)
623 struct acpi_dmar_pci_path *path;
624 u8 bus;
625 int count;
627 bus = scope->bus;
628 path = (struct acpi_dmar_pci_path *)(scope + 1);
629 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
630 / sizeof(struct acpi_dmar_pci_path);
632 while (--count > 0) {
634 * Access PCI directly due to the PCI
635 * subsystem isn't initialized yet.
637 bus = read_pci_config_byte(bus, path->dev, path->fn,
638 PCI_SECONDARY_BUS);
639 path++;
641 ir_hpet[ir_hpet_num].bus = bus;
642 ir_hpet[ir_hpet_num].devfn = PCI_DEVFN(path->dev, path->fn);
643 ir_hpet[ir_hpet_num].iommu = iommu;
644 ir_hpet[ir_hpet_num].id = scope->enumeration_id;
645 ir_hpet_num++;
648 static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
649 struct intel_iommu *iommu)
651 struct acpi_dmar_pci_path *path;
652 u8 bus;
653 int count;
655 bus = scope->bus;
656 path = (struct acpi_dmar_pci_path *)(scope + 1);
657 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
658 / sizeof(struct acpi_dmar_pci_path);
660 while (--count > 0) {
662 * Access PCI directly due to the PCI
663 * subsystem isn't initialized yet.
665 bus = read_pci_config_byte(bus, path->dev, path->fn,
666 PCI_SECONDARY_BUS);
667 path++;
670 ir_ioapic[ir_ioapic_num].bus = bus;
671 ir_ioapic[ir_ioapic_num].devfn = PCI_DEVFN(path->dev, path->fn);
672 ir_ioapic[ir_ioapic_num].iommu = iommu;
673 ir_ioapic[ir_ioapic_num].id = scope->enumeration_id;
674 ir_ioapic_num++;
677 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
678 struct intel_iommu *iommu)
680 struct acpi_dmar_hardware_unit *drhd;
681 struct acpi_dmar_device_scope *scope;
682 void *start, *end;
684 drhd = (struct acpi_dmar_hardware_unit *)header;
686 start = (void *)(drhd + 1);
687 end = ((void *)drhd) + header->length;
689 while (start < end) {
690 scope = start;
691 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
692 if (ir_ioapic_num == MAX_IO_APICS) {
693 printk(KERN_WARNING "Exceeded Max IO APICS\n");
694 return -1;
697 printk(KERN_INFO "IOAPIC id %d under DRHD base "
698 " 0x%Lx IOMMU %d\n", scope->enumeration_id,
699 drhd->address, iommu->seq_id);
701 ir_parse_one_ioapic_scope(scope, iommu);
702 } else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET) {
703 if (ir_hpet_num == MAX_HPET_TBS) {
704 printk(KERN_WARNING "Exceeded Max HPET blocks\n");
705 return -1;
708 printk(KERN_INFO "HPET id %d under DRHD base"
709 " 0x%Lx\n", scope->enumeration_id,
710 drhd->address);
712 ir_parse_one_hpet_scope(scope, iommu);
714 start += scope->length;
717 return 0;
721 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
722 * hardware unit.
724 int __init parse_ioapics_under_ir(void)
726 struct dmar_drhd_unit *drhd;
727 int ir_supported = 0;
729 for_each_drhd_unit(drhd) {
730 struct intel_iommu *iommu = drhd->iommu;
732 if (ecap_ir_support(iommu->ecap)) {
733 if (ir_parse_ioapic_hpet_scope(drhd->hdr, iommu))
734 return -1;
736 ir_supported = 1;
740 if (ir_supported && ir_ioapic_num != nr_ioapics) {
741 printk(KERN_WARNING
742 "Not all IO-APIC's listed under remapping hardware\n");
743 return -1;
746 return ir_supported;
749 void disable_intr_remapping(void)
751 struct dmar_drhd_unit *drhd;
752 struct intel_iommu *iommu = NULL;
755 * Disable Interrupt-remapping for all the DRHD's now.
757 for_each_iommu(iommu, drhd) {
758 if (!ecap_ir_support(iommu->ecap))
759 continue;
761 iommu_disable_intr_remapping(iommu);
765 int reenable_intr_remapping(int eim)
767 struct dmar_drhd_unit *drhd;
768 int setup = 0;
769 struct intel_iommu *iommu = NULL;
771 for_each_iommu(iommu, drhd)
772 if (iommu->qi)
773 dmar_reenable_qi(iommu);
776 * Setup Interrupt-remapping for all the DRHD's now.
778 for_each_iommu(iommu, drhd) {
779 if (!ecap_ir_support(iommu->ecap))
780 continue;
782 /* Set up interrupt remapping for iommu.*/
783 iommu_set_intr_remapping(iommu, eim);
784 setup = 1;
787 if (!setup)
788 goto error;
790 return 0;
792 error:
794 * handle error condition gracefully here!
796 return -1;