2 * Copyright (c) 2006, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
17 * Copyright (C) 2006-2008 Intel Corporation
18 * Author: Ashok Raj <ashok.raj@intel.com>
19 * Author: Shaohua Li <shaohua.li@intel.com>
20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
22 * This file implements early detection/parsing of Remapping Devices
23 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
26 * These routines are used by both DMA-remapping and Interrupt-remapping
29 #include <linux/pci.h>
30 #include <linux/dmar.h>
31 #include <linux/iova.h>
32 #include <linux/intel-iommu.h>
33 #include <linux/timer.h>
34 #include <linux/irq.h>
35 #include <linux/interrupt.h>
36 #include <linux/tboot.h>
37 #include <linux/dmi.h>
39 #define PREFIX "DMAR: "
41 /* No locks are needed as DMA remapping hardware unit
42 * list is constructed at boot time and hotplug of
43 * these units are not supported by the architecture.
45 LIST_HEAD(dmar_drhd_units);
47 static struct acpi_table_header * __initdata dmar_tbl;
48 static acpi_size dmar_tbl_size;
50 static void __init dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
53 * add INCLUDE_ALL at the tail, so scan the list will find it at
56 if (drhd->include_all)
57 list_add_tail(&drhd->list, &dmar_drhd_units);
59 list_add(&drhd->list, &dmar_drhd_units);
62 static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope,
63 struct pci_dev **dev, u16 segment)
66 struct pci_dev *pdev = NULL;
67 struct acpi_dmar_pci_path *path;
70 bus = pci_find_bus(segment, scope->bus);
71 path = (struct acpi_dmar_pci_path *)(scope + 1);
72 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
73 / sizeof(struct acpi_dmar_pci_path);
79 * Some BIOSes list non-exist devices in DMAR table, just
84 PREFIX "Device scope bus [%d] not found\n",
88 pdev = pci_get_slot(bus, PCI_DEVFN(path->dev, path->fn));
90 printk(KERN_WARNING PREFIX
91 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
92 segment, bus->number, path->dev, path->fn);
97 bus = pdev->subordinate;
100 printk(KERN_WARNING PREFIX
101 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
102 segment, scope->bus, path->dev, path->fn);
106 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT && \
107 pdev->subordinate) || (scope->entry_type == \
108 ACPI_DMAR_SCOPE_TYPE_BRIDGE && !pdev->subordinate)) {
110 printk(KERN_WARNING PREFIX
111 "Device scope type does not match for %s\n",
119 static int __init dmar_parse_dev_scope(void *start, void *end, int *cnt,
120 struct pci_dev ***devices, u16 segment)
122 struct acpi_dmar_device_scope *scope;
128 while (start < end) {
130 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
131 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
133 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
134 printk(KERN_WARNING PREFIX
135 "Unsupported device scope\n");
137 start += scope->length;
142 *devices = kcalloc(*cnt, sizeof(struct pci_dev *), GFP_KERNEL);
148 while (start < end) {
150 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
151 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) {
152 ret = dmar_parse_one_dev_scope(scope,
153 &(*devices)[index], segment);
160 start += scope->length;
167 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
168 * structure which uniquely represent one DMA remapping hardware unit
169 * present in the platform
172 dmar_parse_one_drhd(struct acpi_dmar_header *header)
174 struct acpi_dmar_hardware_unit *drhd;
175 struct dmar_drhd_unit *dmaru;
178 drhd = (struct acpi_dmar_hardware_unit *)header;
179 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
184 dmaru->reg_base_addr = drhd->address;
185 dmaru->segment = drhd->segment;
186 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
188 ret = alloc_iommu(dmaru);
193 dmar_register_drhd_unit(dmaru);
197 static int __init dmar_parse_dev(struct dmar_drhd_unit *dmaru)
199 struct acpi_dmar_hardware_unit *drhd;
202 drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
204 if (dmaru->include_all)
207 ret = dmar_parse_dev_scope((void *)(drhd + 1),
208 ((void *)drhd) + drhd->header.length,
209 &dmaru->devices_cnt, &dmaru->devices,
212 list_del(&dmaru->list);
219 LIST_HEAD(dmar_rmrr_units);
221 static void __init dmar_register_rmrr_unit(struct dmar_rmrr_unit *rmrr)
223 list_add(&rmrr->list, &dmar_rmrr_units);
228 dmar_parse_one_rmrr(struct acpi_dmar_header *header)
230 struct acpi_dmar_reserved_memory *rmrr;
231 struct dmar_rmrr_unit *rmrru;
233 rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL);
238 rmrr = (struct acpi_dmar_reserved_memory *)header;
239 rmrru->base_address = rmrr->base_address;
240 rmrru->end_address = rmrr->end_address;
242 dmar_register_rmrr_unit(rmrru);
247 rmrr_parse_dev(struct dmar_rmrr_unit *rmrru)
249 struct acpi_dmar_reserved_memory *rmrr;
252 rmrr = (struct acpi_dmar_reserved_memory *) rmrru->hdr;
253 ret = dmar_parse_dev_scope((void *)(rmrr + 1),
254 ((void *)rmrr) + rmrr->header.length,
255 &rmrru->devices_cnt, &rmrru->devices, rmrr->segment);
257 if (ret || (rmrru->devices_cnt == 0)) {
258 list_del(&rmrru->list);
264 static LIST_HEAD(dmar_atsr_units);
266 static int __init dmar_parse_one_atsr(struct acpi_dmar_header *hdr)
268 struct acpi_dmar_atsr *atsr;
269 struct dmar_atsr_unit *atsru;
271 atsr = container_of(hdr, struct acpi_dmar_atsr, header);
272 atsru = kzalloc(sizeof(*atsru), GFP_KERNEL);
277 atsru->include_all = atsr->flags & 0x1;
279 list_add(&atsru->list, &dmar_atsr_units);
284 static int __init atsr_parse_dev(struct dmar_atsr_unit *atsru)
287 struct acpi_dmar_atsr *atsr;
289 if (atsru->include_all)
292 atsr = container_of(atsru->hdr, struct acpi_dmar_atsr, header);
293 rc = dmar_parse_dev_scope((void *)(atsr + 1),
294 (void *)atsr + atsr->header.length,
295 &atsru->devices_cnt, &atsru->devices,
297 if (rc || !atsru->devices_cnt) {
298 list_del(&atsru->list);
305 int dmar_find_matched_atsr_unit(struct pci_dev *dev)
309 struct acpi_dmar_atsr *atsr;
310 struct dmar_atsr_unit *atsru;
312 list_for_each_entry(atsru, &dmar_atsr_units, list) {
313 atsr = container_of(atsru->hdr, struct acpi_dmar_atsr, header);
314 if (atsr->segment == pci_domain_nr(dev->bus))
321 for (bus = dev->bus; bus; bus = bus->parent) {
322 struct pci_dev *bridge = bus->self;
324 if (!bridge || !pci_is_pcie(bridge) ||
325 bridge->pcie_type == PCI_EXP_TYPE_PCI_BRIDGE)
328 if (bridge->pcie_type == PCI_EXP_TYPE_ROOT_PORT) {
329 for (i = 0; i < atsru->devices_cnt; i++)
330 if (atsru->devices[i] == bridge)
336 if (atsru->include_all)
343 #ifdef CONFIG_ACPI_NUMA
345 dmar_parse_one_rhsa(struct acpi_dmar_header *header)
347 struct acpi_dmar_rhsa *rhsa;
348 struct dmar_drhd_unit *drhd;
350 rhsa = (struct acpi_dmar_rhsa *)header;
351 for_each_drhd_unit(drhd) {
352 if (drhd->reg_base_addr == rhsa->base_address) {
353 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
355 if (!node_online(node))
357 drhd->iommu->node = node;
361 WARN(1, "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
362 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
364 dmi_get_system_info(DMI_BIOS_VENDOR),
365 dmi_get_system_info(DMI_BIOS_VERSION),
366 dmi_get_system_info(DMI_PRODUCT_VERSION));
373 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
375 struct acpi_dmar_hardware_unit *drhd;
376 struct acpi_dmar_reserved_memory *rmrr;
377 struct acpi_dmar_atsr *atsr;
378 struct acpi_dmar_rhsa *rhsa;
380 switch (header->type) {
381 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
382 drhd = container_of(header, struct acpi_dmar_hardware_unit,
384 printk (KERN_INFO PREFIX
385 "DRHD base: %#016Lx flags: %#x\n",
386 (unsigned long long)drhd->address, drhd->flags);
388 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
389 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
391 printk (KERN_INFO PREFIX
392 "RMRR base: %#016Lx end: %#016Lx\n",
393 (unsigned long long)rmrr->base_address,
394 (unsigned long long)rmrr->end_address);
396 case ACPI_DMAR_TYPE_ATSR:
397 atsr = container_of(header, struct acpi_dmar_atsr, header);
398 printk(KERN_INFO PREFIX "ATSR flags: %#x\n", atsr->flags);
400 case ACPI_DMAR_HARDWARE_AFFINITY:
401 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
402 printk(KERN_INFO PREFIX "RHSA base: %#016Lx proximity domain: %#x\n",
403 (unsigned long long)rhsa->base_address,
404 rhsa->proximity_domain);
410 * dmar_table_detect - checks to see if the platform supports DMAR devices
412 static int __init dmar_table_detect(void)
414 acpi_status status = AE_OK;
416 /* if we could find DMAR table, then there are DMAR devices */
417 status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
418 (struct acpi_table_header **)&dmar_tbl,
421 if (ACPI_SUCCESS(status) && !dmar_tbl) {
422 printk (KERN_WARNING PREFIX "Unable to map DMAR\n");
423 status = AE_NOT_FOUND;
426 return (ACPI_SUCCESS(status) ? 1 : 0);
430 * parse_dmar_table - parses the DMA reporting table
433 parse_dmar_table(void)
435 struct acpi_table_dmar *dmar;
436 struct acpi_dmar_header *entry_header;
440 * Do it again, earlier dmar_tbl mapping could be mapped with
446 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
447 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
449 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
451 dmar = (struct acpi_table_dmar *)dmar_tbl;
455 if (dmar->width < PAGE_SHIFT - 1) {
456 printk(KERN_WARNING PREFIX "Invalid DMAR haw\n");
460 printk (KERN_INFO PREFIX "Host address width %d\n",
463 entry_header = (struct acpi_dmar_header *)(dmar + 1);
464 while (((unsigned long)entry_header) <
465 (((unsigned long)dmar) + dmar_tbl->length)) {
466 /* Avoid looping forever on bad ACPI tables */
467 if (entry_header->length == 0) {
468 printk(KERN_WARNING PREFIX
469 "Invalid 0-length structure\n");
474 dmar_table_print_dmar_entry(entry_header);
476 switch (entry_header->type) {
477 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
478 ret = dmar_parse_one_drhd(entry_header);
480 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
482 ret = dmar_parse_one_rmrr(entry_header);
485 case ACPI_DMAR_TYPE_ATSR:
487 ret = dmar_parse_one_atsr(entry_header);
490 case ACPI_DMAR_HARDWARE_AFFINITY:
491 #ifdef CONFIG_ACPI_NUMA
492 ret = dmar_parse_one_rhsa(entry_header);
496 printk(KERN_WARNING PREFIX
497 "Unknown DMAR structure type %d\n",
499 ret = 0; /* for forward compatibility */
505 entry_header = ((void *)entry_header + entry_header->length);
510 int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
516 for (index = 0; index < cnt; index++)
517 if (dev == devices[index])
520 /* Check our parent */
521 dev = dev->bus->self;
527 struct dmar_drhd_unit *
528 dmar_find_matched_drhd_unit(struct pci_dev *dev)
530 struct dmar_drhd_unit *dmaru = NULL;
531 struct acpi_dmar_hardware_unit *drhd;
533 list_for_each_entry(dmaru, &dmar_drhd_units, list) {
534 drhd = container_of(dmaru->hdr,
535 struct acpi_dmar_hardware_unit,
538 if (dmaru->include_all &&
539 drhd->segment == pci_domain_nr(dev->bus))
542 if (dmar_pci_device_match(dmaru->devices,
543 dmaru->devices_cnt, dev))
550 int __init dmar_dev_scope_init(void)
552 struct dmar_drhd_unit *drhd, *drhd_n;
555 list_for_each_entry_safe(drhd, drhd_n, &dmar_drhd_units, list) {
556 ret = dmar_parse_dev(drhd);
563 struct dmar_rmrr_unit *rmrr, *rmrr_n;
564 struct dmar_atsr_unit *atsr, *atsr_n;
566 list_for_each_entry_safe(rmrr, rmrr_n, &dmar_rmrr_units, list) {
567 ret = rmrr_parse_dev(rmrr);
572 list_for_each_entry_safe(atsr, atsr_n, &dmar_atsr_units, list) {
573 ret = atsr_parse_dev(atsr);
584 int __init dmar_table_init(void)
586 static int dmar_table_initialized;
589 if (dmar_table_initialized)
592 dmar_table_initialized = 1;
594 ret = parse_dmar_table();
597 printk(KERN_INFO PREFIX "parse DMAR table failure.\n");
601 if (list_empty(&dmar_drhd_units)) {
602 printk(KERN_INFO PREFIX "No DMAR devices found\n");
607 if (list_empty(&dmar_rmrr_units))
608 printk(KERN_INFO PREFIX "No RMRR found\n");
610 if (list_empty(&dmar_atsr_units))
611 printk(KERN_INFO PREFIX "No ATSR found\n");
617 static int bios_warned;
619 int __init check_zero_address(void)
621 struct acpi_table_dmar *dmar;
622 struct acpi_dmar_header *entry_header;
623 struct acpi_dmar_hardware_unit *drhd;
625 dmar = (struct acpi_table_dmar *)dmar_tbl;
626 entry_header = (struct acpi_dmar_header *)(dmar + 1);
628 while (((unsigned long)entry_header) <
629 (((unsigned long)dmar) + dmar_tbl->length)) {
630 /* Avoid looping forever on bad ACPI tables */
631 if (entry_header->length == 0) {
632 printk(KERN_WARNING PREFIX
633 "Invalid 0-length structure\n");
637 if (entry_header->type == ACPI_DMAR_TYPE_HARDWARE_UNIT) {
641 drhd = (void *)entry_header;
642 if (!drhd->address) {
643 /* Promote an attitude of violence to a BIOS engineer today */
644 WARN(1, "Your BIOS is broken; DMAR reported at address zero!\n"
645 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
646 dmi_get_system_info(DMI_BIOS_VENDOR),
647 dmi_get_system_info(DMI_BIOS_VERSION),
648 dmi_get_system_info(DMI_PRODUCT_VERSION));
653 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
655 printk("IOMMU: can't validate: %llx\n", drhd->address);
658 cap = dmar_readq(addr + DMAR_CAP_REG);
659 ecap = dmar_readq(addr + DMAR_ECAP_REG);
660 early_iounmap(addr, VTD_PAGE_SIZE);
661 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
662 /* Promote an attitude of violence to a BIOS engineer today */
663 WARN(1, "Your BIOS is broken; DMAR reported at address %llx returns all ones!\n"
664 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
666 dmi_get_system_info(DMI_BIOS_VENDOR),
667 dmi_get_system_info(DMI_BIOS_VERSION),
668 dmi_get_system_info(DMI_PRODUCT_VERSION));
674 entry_header = ((void *)entry_header + entry_header->length);
685 void __init detect_intel_iommu(void)
689 ret = dmar_table_detect();
691 ret = check_zero_address();
693 #ifdef CONFIG_INTR_REMAP
694 struct acpi_table_dmar *dmar;
696 * for now we will disable dma-remapping when interrupt
697 * remapping is enabled.
698 * When support for queued invalidation for IOTLB invalidation
699 * is added, we will not need this any more.
701 dmar = (struct acpi_table_dmar *) dmar_tbl;
702 if (ret && cpu_has_x2apic && dmar->flags & 0x1)
704 "Queued invalidation will be enabled to support "
705 "x2apic and Intr-remapping.\n");
708 if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
710 /* Make sure ACS will be enabled */
716 x86_init.iommu.iommu_init = intel_iommu_init;
719 early_acpi_os_unmap_memory(dmar_tbl, dmar_tbl_size);
724 int alloc_iommu(struct dmar_drhd_unit *drhd)
726 struct intel_iommu *iommu;
729 static int iommu_allocated = 0;
733 if (!drhd->reg_base_addr) {
735 WARN(1, "Your BIOS is broken; DMAR reported at address zero!\n"
736 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
737 dmi_get_system_info(DMI_BIOS_VENDOR),
738 dmi_get_system_info(DMI_BIOS_VERSION),
739 dmi_get_system_info(DMI_PRODUCT_VERSION));
745 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
749 iommu->seq_id = iommu_allocated++;
750 sprintf (iommu->name, "dmar%d", iommu->seq_id);
752 iommu->reg = ioremap(drhd->reg_base_addr, VTD_PAGE_SIZE);
754 printk(KERN_ERR "IOMMU: can't map the region\n");
757 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
758 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
760 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
762 /* Promote an attitude of violence to a BIOS engineer today */
763 WARN(1, "Your BIOS is broken; DMAR reported at address %llx returns all ones!\n"
764 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
766 dmi_get_system_info(DMI_BIOS_VENDOR),
767 dmi_get_system_info(DMI_BIOS_VERSION),
768 dmi_get_system_info(DMI_PRODUCT_VERSION));
775 agaw = iommu_calculate_agaw(iommu);
778 "Cannot get a valid agaw for iommu (seq_id = %d)\n",
782 msagaw = iommu_calculate_max_sagaw(iommu);
785 "Cannot get a valid max agaw for iommu (seq_id = %d)\n",
791 iommu->msagaw = msagaw;
795 /* the registers might be more than one page */
796 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
797 cap_max_fault_reg_offset(iommu->cap));
798 map_size = VTD_PAGE_ALIGN(map_size);
799 if (map_size > VTD_PAGE_SIZE) {
801 iommu->reg = ioremap(drhd->reg_base_addr, map_size);
803 printk(KERN_ERR "IOMMU: can't map the region\n");
808 ver = readl(iommu->reg + DMAR_VER_REG);
809 pr_info("IOMMU %d: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
811 (unsigned long long)drhd->reg_base_addr,
812 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
813 (unsigned long long)iommu->cap,
814 (unsigned long long)iommu->ecap);
816 spin_lock_init(&iommu->register_lock);
828 void free_iommu(struct intel_iommu *iommu)
834 free_dmar_iommu(iommu);
843 * Reclaim all the submitted descriptors which have completed its work.
845 static inline void reclaim_free_desc(struct q_inval *qi)
847 while (qi->desc_status[qi->free_tail] == QI_DONE ||
848 qi->desc_status[qi->free_tail] == QI_ABORT) {
849 qi->desc_status[qi->free_tail] = QI_FREE;
850 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
855 static int qi_check_fault(struct intel_iommu *iommu, int index)
859 struct q_inval *qi = iommu->qi;
860 int wait_index = (index + 1) % QI_LENGTH;
862 if (qi->desc_status[wait_index] == QI_ABORT)
865 fault = readl(iommu->reg + DMAR_FSTS_REG);
868 * If IQE happens, the head points to the descriptor associated
869 * with the error. No new descriptors are fetched until the IQE
872 if (fault & DMA_FSTS_IQE) {
873 head = readl(iommu->reg + DMAR_IQH_REG);
874 if ((head >> DMAR_IQ_SHIFT) == index) {
875 printk(KERN_ERR "VT-d detected invalid descriptor: "
876 "low=%llx, high=%llx\n",
877 (unsigned long long)qi->desc[index].low,
878 (unsigned long long)qi->desc[index].high);
879 memcpy(&qi->desc[index], &qi->desc[wait_index],
880 sizeof(struct qi_desc));
881 __iommu_flush_cache(iommu, &qi->desc[index],
882 sizeof(struct qi_desc));
883 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
889 * If ITE happens, all pending wait_desc commands are aborted.
890 * No new descriptors are fetched until the ITE is cleared.
892 if (fault & DMA_FSTS_ITE) {
893 head = readl(iommu->reg + DMAR_IQH_REG);
894 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
896 tail = readl(iommu->reg + DMAR_IQT_REG);
897 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
899 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
902 if (qi->desc_status[head] == QI_IN_USE)
903 qi->desc_status[head] = QI_ABORT;
904 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
905 } while (head != tail);
907 if (qi->desc_status[wait_index] == QI_ABORT)
911 if (fault & DMA_FSTS_ICE)
912 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
918 * Submit the queued invalidation descriptor to the remapping
919 * hardware unit and wait for its completion.
921 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
924 struct q_inval *qi = iommu->qi;
925 struct qi_desc *hw, wait_desc;
926 int wait_index, index;
937 spin_lock_irqsave(&qi->q_lock, flags);
938 while (qi->free_cnt < 3) {
939 spin_unlock_irqrestore(&qi->q_lock, flags);
941 spin_lock_irqsave(&qi->q_lock, flags);
944 index = qi->free_head;
945 wait_index = (index + 1) % QI_LENGTH;
947 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
951 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
952 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
953 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
955 hw[wait_index] = wait_desc;
957 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
958 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
960 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
964 * update the HW tail register indicating the presence of
967 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
969 while (qi->desc_status[wait_index] != QI_DONE) {
971 * We will leave the interrupts disabled, to prevent interrupt
972 * context to queue another cmd while a cmd is already submitted
973 * and waiting for completion on this cpu. This is to avoid
974 * a deadlock where the interrupt context can wait indefinitely
975 * for free slots in the queue.
977 rc = qi_check_fault(iommu, index);
981 spin_unlock(&qi->q_lock);
983 spin_lock(&qi->q_lock);
986 qi->desc_status[index] = QI_DONE;
988 reclaim_free_desc(qi);
989 spin_unlock_irqrestore(&qi->q_lock, flags);
998 * Flush the global interrupt entry cache.
1000 void qi_global_iec(struct intel_iommu *iommu)
1002 struct qi_desc desc;
1004 desc.low = QI_IEC_TYPE;
1007 /* should never fail */
1008 qi_submit_sync(&desc, iommu);
1011 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1014 struct qi_desc desc;
1016 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1017 | QI_CC_GRAN(type) | QI_CC_TYPE;
1020 qi_submit_sync(&desc, iommu);
1023 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1024 unsigned int size_order, u64 type)
1028 struct qi_desc desc;
1031 if (cap_write_drain(iommu->cap))
1034 if (cap_read_drain(iommu->cap))
1037 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1038 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1039 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1040 | QI_IOTLB_AM(size_order);
1042 qi_submit_sync(&desc, iommu);
1045 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
1046 u64 addr, unsigned mask)
1048 struct qi_desc desc;
1051 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
1052 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1053 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1055 desc.high = QI_DEV_IOTLB_ADDR(addr);
1057 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1060 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1063 qi_submit_sync(&desc, iommu);
1067 * Disable Queued Invalidation interface.
1069 void dmar_disable_qi(struct intel_iommu *iommu)
1071 unsigned long flags;
1073 cycles_t start_time = get_cycles();
1075 if (!ecap_qis(iommu->ecap))
1078 spin_lock_irqsave(&iommu->register_lock, flags);
1080 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
1081 if (!(sts & DMA_GSTS_QIES))
1085 * Give a chance to HW to complete the pending invalidation requests.
1087 while ((readl(iommu->reg + DMAR_IQT_REG) !=
1088 readl(iommu->reg + DMAR_IQH_REG)) &&
1089 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1092 iommu->gcmd &= ~DMA_GCMD_QIE;
1093 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1095 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1096 !(sts & DMA_GSTS_QIES), sts);
1098 spin_unlock_irqrestore(&iommu->register_lock, flags);
1102 * Enable queued invalidation.
1104 static void __dmar_enable_qi(struct intel_iommu *iommu)
1107 unsigned long flags;
1108 struct q_inval *qi = iommu->qi;
1110 qi->free_head = qi->free_tail = 0;
1111 qi->free_cnt = QI_LENGTH;
1113 spin_lock_irqsave(&iommu->register_lock, flags);
1115 /* write zero to the tail reg */
1116 writel(0, iommu->reg + DMAR_IQT_REG);
1118 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1120 iommu->gcmd |= DMA_GCMD_QIE;
1121 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1123 /* Make sure hardware complete it */
1124 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1126 spin_unlock_irqrestore(&iommu->register_lock, flags);
1130 * Enable Queued Invalidation interface. This is a must to support
1131 * interrupt-remapping. Also used by DMA-remapping, which replaces
1132 * register based IOTLB invalidation.
1134 int dmar_enable_qi(struct intel_iommu *iommu)
1137 struct page *desc_page;
1139 if (!ecap_qis(iommu->ecap))
1143 * queued invalidation is already setup and enabled.
1148 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1155 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1162 qi->desc = page_address(desc_page);
1164 qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1165 if (!qi->desc_status) {
1166 free_page((unsigned long) qi->desc);
1172 qi->free_head = qi->free_tail = 0;
1173 qi->free_cnt = QI_LENGTH;
1175 spin_lock_init(&qi->q_lock);
1177 __dmar_enable_qi(iommu);
1182 /* iommu interrupt handling. Most stuff are MSI-like. */
1190 static const char *dma_remap_fault_reasons[] =
1193 "Present bit in root entry is clear",
1194 "Present bit in context entry is clear",
1195 "Invalid context entry",
1196 "Access beyond MGAW",
1197 "PTE Write access is not set",
1198 "PTE Read access is not set",
1199 "Next page table ptr is invalid",
1200 "Root table address invalid",
1201 "Context table ptr is invalid",
1202 "non-zero reserved fields in RTP",
1203 "non-zero reserved fields in CTP",
1204 "non-zero reserved fields in PTE",
1207 static const char *intr_remap_fault_reasons[] =
1209 "Detected reserved fields in the decoded interrupt-remapped request",
1210 "Interrupt index exceeded the interrupt-remapping table size",
1211 "Present field in the IRTE entry is clear",
1212 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1213 "Detected reserved fields in the IRTE entry",
1214 "Blocked a compatibility format interrupt request",
1215 "Blocked an interrupt request due to source-id verification failure",
1218 #define MAX_FAULT_REASON_IDX (ARRAY_SIZE(fault_reason_strings) - 1)
1220 const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1222 if (fault_reason >= 0x20 && (fault_reason <= 0x20 +
1223 ARRAY_SIZE(intr_remap_fault_reasons))) {
1224 *fault_type = INTR_REMAP;
1225 return intr_remap_fault_reasons[fault_reason - 0x20];
1226 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1227 *fault_type = DMA_REMAP;
1228 return dma_remap_fault_reasons[fault_reason];
1230 *fault_type = UNKNOWN;
1235 void dmar_msi_unmask(unsigned int irq)
1237 struct intel_iommu *iommu = get_irq_data(irq);
1241 spin_lock_irqsave(&iommu->register_lock, flag);
1242 writel(0, iommu->reg + DMAR_FECTL_REG);
1243 /* Read a reg to force flush the post write */
1244 readl(iommu->reg + DMAR_FECTL_REG);
1245 spin_unlock_irqrestore(&iommu->register_lock, flag);
1248 void dmar_msi_mask(unsigned int irq)
1251 struct intel_iommu *iommu = get_irq_data(irq);
1254 spin_lock_irqsave(&iommu->register_lock, flag);
1255 writel(DMA_FECTL_IM, iommu->reg + DMAR_FECTL_REG);
1256 /* Read a reg to force flush the post write */
1257 readl(iommu->reg + DMAR_FECTL_REG);
1258 spin_unlock_irqrestore(&iommu->register_lock, flag);
1261 void dmar_msi_write(int irq, struct msi_msg *msg)
1263 struct intel_iommu *iommu = get_irq_data(irq);
1266 spin_lock_irqsave(&iommu->register_lock, flag);
1267 writel(msg->data, iommu->reg + DMAR_FEDATA_REG);
1268 writel(msg->address_lo, iommu->reg + DMAR_FEADDR_REG);
1269 writel(msg->address_hi, iommu->reg + DMAR_FEUADDR_REG);
1270 spin_unlock_irqrestore(&iommu->register_lock, flag);
1273 void dmar_msi_read(int irq, struct msi_msg *msg)
1275 struct intel_iommu *iommu = get_irq_data(irq);
1278 spin_lock_irqsave(&iommu->register_lock, flag);
1279 msg->data = readl(iommu->reg + DMAR_FEDATA_REG);
1280 msg->address_lo = readl(iommu->reg + DMAR_FEADDR_REG);
1281 msg->address_hi = readl(iommu->reg + DMAR_FEUADDR_REG);
1282 spin_unlock_irqrestore(&iommu->register_lock, flag);
1285 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1286 u8 fault_reason, u16 source_id, unsigned long long addr)
1291 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1293 if (fault_type == INTR_REMAP)
1294 printk(KERN_ERR "INTR-REMAP: Request device [[%02x:%02x.%d] "
1295 "fault index %llx\n"
1296 "INTR-REMAP:[fault reason %02d] %s\n",
1297 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1298 PCI_FUNC(source_id & 0xFF), addr >> 48,
1299 fault_reason, reason);
1302 "DMAR:[%s] Request device [%02x:%02x.%d] "
1303 "fault addr %llx \n"
1304 "DMAR:[fault reason %02d] %s\n",
1305 (type ? "DMA Read" : "DMA Write"),
1306 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1307 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1311 #define PRIMARY_FAULT_REG_LEN (16)
1312 irqreturn_t dmar_fault(int irq, void *dev_id)
1314 struct intel_iommu *iommu = dev_id;
1315 int reg, fault_index;
1319 spin_lock_irqsave(&iommu->register_lock, flag);
1320 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1322 printk(KERN_ERR "DRHD: handling fault status reg %x\n",
1325 /* TBD: ignore advanced fault log currently */
1326 if (!(fault_status & DMA_FSTS_PPF))
1329 fault_index = dma_fsts_fault_record_index(fault_status);
1330 reg = cap_fault_reg_offset(iommu->cap);
1338 /* highest 32 bits */
1339 data = readl(iommu->reg + reg +
1340 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1341 if (!(data & DMA_FRCD_F))
1344 fault_reason = dma_frcd_fault_reason(data);
1345 type = dma_frcd_type(data);
1347 data = readl(iommu->reg + reg +
1348 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1349 source_id = dma_frcd_source_id(data);
1351 guest_addr = dmar_readq(iommu->reg + reg +
1352 fault_index * PRIMARY_FAULT_REG_LEN);
1353 guest_addr = dma_frcd_page_addr(guest_addr);
1354 /* clear the fault */
1355 writel(DMA_FRCD_F, iommu->reg + reg +
1356 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1358 spin_unlock_irqrestore(&iommu->register_lock, flag);
1360 dmar_fault_do_one(iommu, type, fault_reason,
1361 source_id, guest_addr);
1364 if (fault_index >= cap_num_fault_regs(iommu->cap))
1366 spin_lock_irqsave(&iommu->register_lock, flag);
1369 /* clear all the other faults */
1370 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1371 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1373 spin_unlock_irqrestore(&iommu->register_lock, flag);
1377 int dmar_set_interrupt(struct intel_iommu *iommu)
1382 * Check if the fault interrupt is already initialized.
1389 printk(KERN_ERR "IOMMU: no free vectors\n");
1393 set_irq_data(irq, iommu);
1396 ret = arch_setup_dmar_msi(irq);
1398 set_irq_data(irq, NULL);
1404 ret = request_irq(irq, dmar_fault, 0, iommu->name, iommu);
1406 printk(KERN_ERR "IOMMU: can't request irq\n");
1410 int __init enable_drhd_fault_handling(void)
1412 struct dmar_drhd_unit *drhd;
1415 * Enable fault control interrupt.
1417 for_each_drhd_unit(drhd) {
1419 struct intel_iommu *iommu = drhd->iommu;
1420 ret = dmar_set_interrupt(iommu);
1423 printk(KERN_ERR "DRHD %Lx: failed to enable fault, "
1424 " interrupt, ret %d\n",
1425 (unsigned long long)drhd->reg_base_addr, ret);
1434 * Re-enable Queued Invalidation interface.
1436 int dmar_reenable_qi(struct intel_iommu *iommu)
1438 if (!ecap_qis(iommu->ecap))
1445 * First disable queued invalidation.
1447 dmar_disable_qi(iommu);
1449 * Then enable queued invalidation again. Since there is no pending
1450 * invalidation requests now, it's safe to re-enable queued
1453 __dmar_enable_qi(iommu);
1459 * Check interrupt remapping support in DMAR table description.
1461 int __init dmar_ir_support(void)
1463 struct acpi_table_dmar *dmar;
1464 dmar = (struct acpi_table_dmar *)dmar_tbl;
1465 return dmar->flags & 0x1;