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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt /* has to precede printk.h */
31 #include <linux/pci.h>
32 #include <linux/dmar.h>
33 #include <linux/iova.h>
34 #include <linux/intel-iommu.h>
35 #include <linux/timer.h>
36 #include <linux/irq.h>
37 #include <linux/interrupt.h>
38 #include <linux/tboot.h>
39 #include <linux/dmi.h>
40 #include <linux/slab.h>
41 #include <asm/irq_remapping.h>
42 #include <asm/iommu_table.h>
44 /* No locks are needed as DMA remapping hardware unit
45 * list is constructed at boot time and hotplug of
46 * these units are not supported by the architecture.
48 LIST_HEAD(dmar_drhd_units);
50 struct acpi_table_header * __initdata dmar_tbl;
51 static acpi_size dmar_tbl_size;
53 static void __init dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
56 * add INCLUDE_ALL at the tail, so scan the list will find it at
59 if (drhd->include_all)
60 list_add_tail(&drhd->list, &dmar_drhd_units);
62 list_add(&drhd->list, &dmar_drhd_units);
65 static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope,
66 struct pci_dev **dev, u16 segment)
69 struct pci_dev *pdev = NULL;
70 struct acpi_dmar_pci_path *path;
73 bus = pci_find_bus(segment, scope->bus);
74 path = (struct acpi_dmar_pci_path *)(scope + 1);
75 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
76 / sizeof(struct acpi_dmar_pci_path);
82 * Some BIOSes list non-exist devices in DMAR table, just
86 pr_warn("Device scope bus [%d] not found\n", scope->bus);
89 pdev = pci_get_slot(bus, PCI_DEVFN(path->dev, path->fn));
91 /* warning will be printed below */
96 bus = pdev->subordinate;
99 pr_warn("Device scope device [%04x:%02x:%02x.%02x] not found\n",
100 segment, scope->bus, path->dev, path->fn);
104 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT && \
105 pdev->subordinate) || (scope->entry_type == \
106 ACPI_DMAR_SCOPE_TYPE_BRIDGE && !pdev->subordinate)) {
108 pr_warn("Device scope type does not match for %s\n",
116 int __init dmar_parse_dev_scope(void *start, void *end, int *cnt,
117 struct pci_dev ***devices, u16 segment)
119 struct acpi_dmar_device_scope *scope;
125 while (start < end) {
127 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
128 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
130 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
131 pr_warn("Unsupported device scope\n");
133 start += scope->length;
138 *devices = kcalloc(*cnt, sizeof(struct pci_dev *), GFP_KERNEL);
144 while (start < end) {
146 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
147 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) {
148 ret = dmar_parse_one_dev_scope(scope,
149 &(*devices)[index], segment);
156 start += scope->length;
163 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
164 * structure which uniquely represent one DMA remapping hardware unit
165 * present in the platform
168 dmar_parse_one_drhd(struct acpi_dmar_header *header)
170 struct acpi_dmar_hardware_unit *drhd;
171 struct dmar_drhd_unit *dmaru;
174 drhd = (struct acpi_dmar_hardware_unit *)header;
175 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
180 dmaru->reg_base_addr = drhd->address;
181 dmaru->segment = drhd->segment;
182 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
184 ret = alloc_iommu(dmaru);
189 dmar_register_drhd_unit(dmaru);
193 static int __init dmar_parse_dev(struct dmar_drhd_unit *dmaru)
195 struct acpi_dmar_hardware_unit *drhd;
198 drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
200 if (dmaru->include_all)
203 ret = dmar_parse_dev_scope((void *)(drhd + 1),
204 ((void *)drhd) + drhd->header.length,
205 &dmaru->devices_cnt, &dmaru->devices,
208 list_del(&dmaru->list);
214 #ifdef CONFIG_ACPI_NUMA
216 dmar_parse_one_rhsa(struct acpi_dmar_header *header)
218 struct acpi_dmar_rhsa *rhsa;
219 struct dmar_drhd_unit *drhd;
221 rhsa = (struct acpi_dmar_rhsa *)header;
222 for_each_drhd_unit(drhd) {
223 if (drhd->reg_base_addr == rhsa->base_address) {
224 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
226 if (!node_online(node))
228 drhd->iommu->node = node;
233 1, TAINT_FIRMWARE_WORKAROUND,
234 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
235 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
237 dmi_get_system_info(DMI_BIOS_VENDOR),
238 dmi_get_system_info(DMI_BIOS_VERSION),
239 dmi_get_system_info(DMI_PRODUCT_VERSION));
246 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
248 struct acpi_dmar_hardware_unit *drhd;
249 struct acpi_dmar_reserved_memory *rmrr;
250 struct acpi_dmar_atsr *atsr;
251 struct acpi_dmar_rhsa *rhsa;
253 switch (header->type) {
254 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
255 drhd = container_of(header, struct acpi_dmar_hardware_unit,
257 pr_info("DRHD base: %#016Lx flags: %#x\n",
258 (unsigned long long)drhd->address, drhd->flags);
260 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
261 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
263 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
264 (unsigned long long)rmrr->base_address,
265 (unsigned long long)rmrr->end_address);
267 case ACPI_DMAR_TYPE_ATSR:
268 atsr = container_of(header, struct acpi_dmar_atsr, header);
269 pr_info("ATSR flags: %#x\n", atsr->flags);
271 case ACPI_DMAR_HARDWARE_AFFINITY:
272 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
273 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
274 (unsigned long long)rhsa->base_address,
275 rhsa->proximity_domain);
281 * dmar_table_detect - checks to see if the platform supports DMAR devices
283 static int __init dmar_table_detect(void)
285 acpi_status status = AE_OK;
287 /* if we could find DMAR table, then there are DMAR devices */
288 status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
289 (struct acpi_table_header **)&dmar_tbl,
292 if (ACPI_SUCCESS(status) && !dmar_tbl) {
293 pr_warn("Unable to map DMAR\n");
294 status = AE_NOT_FOUND;
297 return (ACPI_SUCCESS(status) ? 1 : 0);
301 * parse_dmar_table - parses the DMA reporting table
304 parse_dmar_table(void)
306 struct acpi_table_dmar *dmar;
307 struct acpi_dmar_header *entry_header;
311 * Do it again, earlier dmar_tbl mapping could be mapped with
317 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
318 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
320 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
322 dmar = (struct acpi_table_dmar *)dmar_tbl;
326 if (dmar->width < PAGE_SHIFT - 1) {
327 pr_warn("Invalid DMAR haw\n");
331 pr_info("Host address width %d\n", dmar->width + 1);
333 entry_header = (struct acpi_dmar_header *)(dmar + 1);
334 while (((unsigned long)entry_header) <
335 (((unsigned long)dmar) + dmar_tbl->length)) {
336 /* Avoid looping forever on bad ACPI tables */
337 if (entry_header->length == 0) {
338 pr_warn("Invalid 0-length structure\n");
343 dmar_table_print_dmar_entry(entry_header);
345 switch (entry_header->type) {
346 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
347 ret = dmar_parse_one_drhd(entry_header);
349 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
350 ret = dmar_parse_one_rmrr(entry_header);
352 case ACPI_DMAR_TYPE_ATSR:
353 ret = dmar_parse_one_atsr(entry_header);
355 case ACPI_DMAR_HARDWARE_AFFINITY:
356 #ifdef CONFIG_ACPI_NUMA
357 ret = dmar_parse_one_rhsa(entry_header);
361 pr_warn("Unknown DMAR structure type %d\n",
363 ret = 0; /* for forward compatibility */
369 entry_header = ((void *)entry_header + entry_header->length);
374 static int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
380 for (index = 0; index < cnt; index++)
381 if (dev == devices[index])
384 /* Check our parent */
385 dev = dev->bus->self;
391 struct dmar_drhd_unit *
392 dmar_find_matched_drhd_unit(struct pci_dev *dev)
394 struct dmar_drhd_unit *dmaru = NULL;
395 struct acpi_dmar_hardware_unit *drhd;
397 dev = pci_physfn(dev);
399 list_for_each_entry(dmaru, &dmar_drhd_units, list) {
400 drhd = container_of(dmaru->hdr,
401 struct acpi_dmar_hardware_unit,
404 if (dmaru->include_all &&
405 drhd->segment == pci_domain_nr(dev->bus))
408 if (dmar_pci_device_match(dmaru->devices,
409 dmaru->devices_cnt, dev))
416 int __init dmar_dev_scope_init(void)
418 static int dmar_dev_scope_initialized;
419 struct dmar_drhd_unit *drhd, *drhd_n;
422 if (dmar_dev_scope_initialized)
423 return dmar_dev_scope_initialized;
425 if (list_empty(&dmar_drhd_units))
428 list_for_each_entry_safe(drhd, drhd_n, &dmar_drhd_units, list) {
429 ret = dmar_parse_dev(drhd);
434 ret = dmar_parse_rmrr_atsr_dev();
438 dmar_dev_scope_initialized = 1;
442 dmar_dev_scope_initialized = ret;
447 int __init dmar_table_init(void)
449 static int dmar_table_initialized;
452 if (dmar_table_initialized)
455 dmar_table_initialized = 1;
457 ret = parse_dmar_table();
460 pr_info("parse DMAR table failure.\n");
464 if (list_empty(&dmar_drhd_units)) {
465 pr_info("No DMAR devices found\n");
472 static void warn_invalid_dmar(u64 addr, const char *message)
475 1, TAINT_FIRMWARE_WORKAROUND,
476 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
477 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
479 dmi_get_system_info(DMI_BIOS_VENDOR),
480 dmi_get_system_info(DMI_BIOS_VERSION),
481 dmi_get_system_info(DMI_PRODUCT_VERSION));
484 int __init check_zero_address(void)
486 struct acpi_table_dmar *dmar;
487 struct acpi_dmar_header *entry_header;
488 struct acpi_dmar_hardware_unit *drhd;
490 dmar = (struct acpi_table_dmar *)dmar_tbl;
491 entry_header = (struct acpi_dmar_header *)(dmar + 1);
493 while (((unsigned long)entry_header) <
494 (((unsigned long)dmar) + dmar_tbl->length)) {
495 /* Avoid looping forever on bad ACPI tables */
496 if (entry_header->length == 0) {
497 pr_warn("Invalid 0-length structure\n");
501 if (entry_header->type == ACPI_DMAR_TYPE_HARDWARE_UNIT) {
505 drhd = (void *)entry_header;
506 if (!drhd->address) {
507 warn_invalid_dmar(0, "");
511 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
513 printk("IOMMU: can't validate: %llx\n", drhd->address);
516 cap = dmar_readq(addr + DMAR_CAP_REG);
517 ecap = dmar_readq(addr + DMAR_ECAP_REG);
518 early_iounmap(addr, VTD_PAGE_SIZE);
519 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
520 warn_invalid_dmar(drhd->address,
521 " returns all ones");
526 entry_header = ((void *)entry_header + entry_header->length);
534 int __init detect_intel_iommu(void)
538 ret = dmar_table_detect();
540 ret = check_zero_address();
542 struct acpi_table_dmar *dmar;
544 dmar = (struct acpi_table_dmar *) dmar_tbl;
546 if (ret && irq_remapping_enabled && cpu_has_x2apic &&
548 pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
550 if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
552 /* Make sure ACS will be enabled */
558 x86_init.iommu.iommu_init = intel_iommu_init;
561 early_acpi_os_unmap_memory(dmar_tbl, dmar_tbl_size);
564 return ret ? 1 : -ENODEV;
568 static void unmap_iommu(struct intel_iommu *iommu)
571 release_mem_region(iommu->reg_phys, iommu->reg_size);
575 * map_iommu: map the iommu's registers
576 * @iommu: the iommu to map
577 * @phys_addr: the physical address of the base resgister
579 * Memory map the iommu's registers. Start w/ a single page, and
580 * possibly expand if that turns out to be insufficent.
582 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
586 iommu->reg_phys = phys_addr;
587 iommu->reg_size = VTD_PAGE_SIZE;
589 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
590 pr_err("IOMMU: can't reserve memory\n");
595 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
597 pr_err("IOMMU: can't map the region\n");
602 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
603 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
605 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
607 warn_invalid_dmar(phys_addr, " returns all ones");
611 /* the registers might be more than one page */
612 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
613 cap_max_fault_reg_offset(iommu->cap));
614 map_size = VTD_PAGE_ALIGN(map_size);
615 if (map_size > iommu->reg_size) {
617 release_mem_region(iommu->reg_phys, iommu->reg_size);
618 iommu->reg_size = map_size;
619 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
621 pr_err("IOMMU: can't reserve memory\n");
625 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
627 pr_err("IOMMU: can't map the region\n");
638 release_mem_region(iommu->reg_phys, iommu->reg_size);
643 int alloc_iommu(struct dmar_drhd_unit *drhd)
645 struct intel_iommu *iommu;
647 static int iommu_allocated = 0;
652 if (!drhd->reg_base_addr) {
653 warn_invalid_dmar(0, "");
657 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
661 iommu->seq_id = iommu_allocated++;
662 sprintf (iommu->name, "dmar%d", iommu->seq_id);
664 err = map_iommu(iommu, drhd->reg_base_addr);
666 pr_err("IOMMU: failed to map %s\n", iommu->name);
671 agaw = iommu_calculate_agaw(iommu);
673 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
677 msagaw = iommu_calculate_max_sagaw(iommu);
679 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
684 iommu->msagaw = msagaw;
688 ver = readl(iommu->reg + DMAR_VER_REG);
689 pr_info("IOMMU %d: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
691 (unsigned long long)drhd->reg_base_addr,
692 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
693 (unsigned long long)iommu->cap,
694 (unsigned long long)iommu->ecap);
696 raw_spin_lock_init(&iommu->register_lock);
708 void free_iommu(struct intel_iommu *iommu)
713 free_dmar_iommu(iommu);
722 * Reclaim all the submitted descriptors which have completed its work.
724 static inline void reclaim_free_desc(struct q_inval *qi)
726 while (qi->desc_status[qi->free_tail] == QI_DONE ||
727 qi->desc_status[qi->free_tail] == QI_ABORT) {
728 qi->desc_status[qi->free_tail] = QI_FREE;
729 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
734 static int qi_check_fault(struct intel_iommu *iommu, int index)
738 struct q_inval *qi = iommu->qi;
739 int wait_index = (index + 1) % QI_LENGTH;
741 if (qi->desc_status[wait_index] == QI_ABORT)
744 fault = readl(iommu->reg + DMAR_FSTS_REG);
747 * If IQE happens, the head points to the descriptor associated
748 * with the error. No new descriptors are fetched until the IQE
751 if (fault & DMA_FSTS_IQE) {
752 head = readl(iommu->reg + DMAR_IQH_REG);
753 if ((head >> DMAR_IQ_SHIFT) == index) {
754 pr_err("VT-d detected invalid descriptor: "
755 "low=%llx, high=%llx\n",
756 (unsigned long long)qi->desc[index].low,
757 (unsigned long long)qi->desc[index].high);
758 memcpy(&qi->desc[index], &qi->desc[wait_index],
759 sizeof(struct qi_desc));
760 __iommu_flush_cache(iommu, &qi->desc[index],
761 sizeof(struct qi_desc));
762 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
768 * If ITE happens, all pending wait_desc commands are aborted.
769 * No new descriptors are fetched until the ITE is cleared.
771 if (fault & DMA_FSTS_ITE) {
772 head = readl(iommu->reg + DMAR_IQH_REG);
773 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
775 tail = readl(iommu->reg + DMAR_IQT_REG);
776 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
778 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
781 if (qi->desc_status[head] == QI_IN_USE)
782 qi->desc_status[head] = QI_ABORT;
783 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
784 } while (head != tail);
786 if (qi->desc_status[wait_index] == QI_ABORT)
790 if (fault & DMA_FSTS_ICE)
791 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
797 * Submit the queued invalidation descriptor to the remapping
798 * hardware unit and wait for its completion.
800 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
803 struct q_inval *qi = iommu->qi;
804 struct qi_desc *hw, wait_desc;
805 int wait_index, index;
816 raw_spin_lock_irqsave(&qi->q_lock, flags);
817 while (qi->free_cnt < 3) {
818 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
820 raw_spin_lock_irqsave(&qi->q_lock, flags);
823 index = qi->free_head;
824 wait_index = (index + 1) % QI_LENGTH;
826 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
830 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
831 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
832 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
834 hw[wait_index] = wait_desc;
836 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
837 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
839 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
843 * update the HW tail register indicating the presence of
846 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
848 while (qi->desc_status[wait_index] != QI_DONE) {
850 * We will leave the interrupts disabled, to prevent interrupt
851 * context to queue another cmd while a cmd is already submitted
852 * and waiting for completion on this cpu. This is to avoid
853 * a deadlock where the interrupt context can wait indefinitely
854 * for free slots in the queue.
856 rc = qi_check_fault(iommu, index);
860 raw_spin_unlock(&qi->q_lock);
862 raw_spin_lock(&qi->q_lock);
865 qi->desc_status[index] = QI_DONE;
867 reclaim_free_desc(qi);
868 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
877 * Flush the global interrupt entry cache.
879 void qi_global_iec(struct intel_iommu *iommu)
883 desc.low = QI_IEC_TYPE;
886 /* should never fail */
887 qi_submit_sync(&desc, iommu);
890 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
895 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
896 | QI_CC_GRAN(type) | QI_CC_TYPE;
899 qi_submit_sync(&desc, iommu);
902 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
903 unsigned int size_order, u64 type)
910 if (cap_write_drain(iommu->cap))
913 if (cap_read_drain(iommu->cap))
916 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
917 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
918 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
919 | QI_IOTLB_AM(size_order);
921 qi_submit_sync(&desc, iommu);
924 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
925 u64 addr, unsigned mask)
930 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
931 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
932 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
934 desc.high = QI_DEV_IOTLB_ADDR(addr);
936 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
939 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
942 qi_submit_sync(&desc, iommu);
946 * Disable Queued Invalidation interface.
948 void dmar_disable_qi(struct intel_iommu *iommu)
952 cycles_t start_time = get_cycles();
954 if (!ecap_qis(iommu->ecap))
957 raw_spin_lock_irqsave(&iommu->register_lock, flags);
959 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
960 if (!(sts & DMA_GSTS_QIES))
964 * Give a chance to HW to complete the pending invalidation requests.
966 while ((readl(iommu->reg + DMAR_IQT_REG) !=
967 readl(iommu->reg + DMAR_IQH_REG)) &&
968 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
971 iommu->gcmd &= ~DMA_GCMD_QIE;
972 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
974 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
975 !(sts & DMA_GSTS_QIES), sts);
977 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
981 * Enable queued invalidation.
983 static void __dmar_enable_qi(struct intel_iommu *iommu)
987 struct q_inval *qi = iommu->qi;
989 qi->free_head = qi->free_tail = 0;
990 qi->free_cnt = QI_LENGTH;
992 raw_spin_lock_irqsave(&iommu->register_lock, flags);
994 /* write zero to the tail reg */
995 writel(0, iommu->reg + DMAR_IQT_REG);
997 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
999 iommu->gcmd |= DMA_GCMD_QIE;
1000 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1002 /* Make sure hardware complete it */
1003 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1005 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1009 * Enable Queued Invalidation interface. This is a must to support
1010 * interrupt-remapping. Also used by DMA-remapping, which replaces
1011 * register based IOTLB invalidation.
1013 int dmar_enable_qi(struct intel_iommu *iommu)
1016 struct page *desc_page;
1018 if (!ecap_qis(iommu->ecap))
1022 * queued invalidation is already setup and enabled.
1027 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1034 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1041 qi->desc = page_address(desc_page);
1043 qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1044 if (!qi->desc_status) {
1045 free_page((unsigned long) qi->desc);
1051 qi->free_head = qi->free_tail = 0;
1052 qi->free_cnt = QI_LENGTH;
1054 raw_spin_lock_init(&qi->q_lock);
1056 __dmar_enable_qi(iommu);
1061 /* iommu interrupt handling. Most stuff are MSI-like. */
1069 static const char *dma_remap_fault_reasons[] =
1072 "Present bit in root entry is clear",
1073 "Present bit in context entry is clear",
1074 "Invalid context entry",
1075 "Access beyond MGAW",
1076 "PTE Write access is not set",
1077 "PTE Read access is not set",
1078 "Next page table ptr is invalid",
1079 "Root table address invalid",
1080 "Context table ptr is invalid",
1081 "non-zero reserved fields in RTP",
1082 "non-zero reserved fields in CTP",
1083 "non-zero reserved fields in PTE",
1086 static const char *irq_remap_fault_reasons[] =
1088 "Detected reserved fields in the decoded interrupt-remapped request",
1089 "Interrupt index exceeded the interrupt-remapping table size",
1090 "Present field in the IRTE entry is clear",
1091 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1092 "Detected reserved fields in the IRTE entry",
1093 "Blocked a compatibility format interrupt request",
1094 "Blocked an interrupt request due to source-id verification failure",
1097 #define MAX_FAULT_REASON_IDX (ARRAY_SIZE(fault_reason_strings) - 1)
1099 const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1101 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1102 ARRAY_SIZE(irq_remap_fault_reasons))) {
1103 *fault_type = INTR_REMAP;
1104 return irq_remap_fault_reasons[fault_reason - 0x20];
1105 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1106 *fault_type = DMA_REMAP;
1107 return dma_remap_fault_reasons[fault_reason];
1109 *fault_type = UNKNOWN;
1114 void dmar_msi_unmask(struct irq_data *data)
1116 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1120 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1121 writel(0, iommu->reg + DMAR_FECTL_REG);
1122 /* Read a reg to force flush the post write */
1123 readl(iommu->reg + DMAR_FECTL_REG);
1124 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1127 void dmar_msi_mask(struct irq_data *data)
1130 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1133 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1134 writel(DMA_FECTL_IM, iommu->reg + DMAR_FECTL_REG);
1135 /* Read a reg to force flush the post write */
1136 readl(iommu->reg + DMAR_FECTL_REG);
1137 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1140 void dmar_msi_write(int irq, struct msi_msg *msg)
1142 struct intel_iommu *iommu = irq_get_handler_data(irq);
1145 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1146 writel(msg->data, iommu->reg + DMAR_FEDATA_REG);
1147 writel(msg->address_lo, iommu->reg + DMAR_FEADDR_REG);
1148 writel(msg->address_hi, iommu->reg + DMAR_FEUADDR_REG);
1149 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1152 void dmar_msi_read(int irq, struct msi_msg *msg)
1154 struct intel_iommu *iommu = irq_get_handler_data(irq);
1157 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1158 msg->data = readl(iommu->reg + DMAR_FEDATA_REG);
1159 msg->address_lo = readl(iommu->reg + DMAR_FEADDR_REG);
1160 msg->address_hi = readl(iommu->reg + DMAR_FEUADDR_REG);
1161 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1164 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1165 u8 fault_reason, u16 source_id, unsigned long long addr)
1170 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1172 if (fault_type == INTR_REMAP)
1173 pr_err("INTR-REMAP: Request device [[%02x:%02x.%d] "
1174 "fault index %llx\n"
1175 "INTR-REMAP:[fault reason %02d] %s\n",
1176 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1177 PCI_FUNC(source_id & 0xFF), addr >> 48,
1178 fault_reason, reason);
1180 pr_err("DMAR:[%s] Request device [%02x:%02x.%d] "
1181 "fault addr %llx \n"
1182 "DMAR:[fault reason %02d] %s\n",
1183 (type ? "DMA Read" : "DMA Write"),
1184 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1185 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1189 #define PRIMARY_FAULT_REG_LEN (16)
1190 irqreturn_t dmar_fault(int irq, void *dev_id)
1192 struct intel_iommu *iommu = dev_id;
1193 int reg, fault_index;
1197 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1198 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1200 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1202 /* TBD: ignore advanced fault log currently */
1203 if (!(fault_status & DMA_FSTS_PPF))
1206 fault_index = dma_fsts_fault_record_index(fault_status);
1207 reg = cap_fault_reg_offset(iommu->cap);
1215 /* highest 32 bits */
1216 data = readl(iommu->reg + reg +
1217 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1218 if (!(data & DMA_FRCD_F))
1221 fault_reason = dma_frcd_fault_reason(data);
1222 type = dma_frcd_type(data);
1224 data = readl(iommu->reg + reg +
1225 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1226 source_id = dma_frcd_source_id(data);
1228 guest_addr = dmar_readq(iommu->reg + reg +
1229 fault_index * PRIMARY_FAULT_REG_LEN);
1230 guest_addr = dma_frcd_page_addr(guest_addr);
1231 /* clear the fault */
1232 writel(DMA_FRCD_F, iommu->reg + reg +
1233 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1235 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1237 dmar_fault_do_one(iommu, type, fault_reason,
1238 source_id, guest_addr);
1241 if (fault_index >= cap_num_fault_regs(iommu->cap))
1243 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1246 /* clear all the other faults */
1247 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1248 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1250 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1254 int dmar_set_interrupt(struct intel_iommu *iommu)
1259 * Check if the fault interrupt is already initialized.
1266 pr_err("IOMMU: no free vectors\n");
1270 irq_set_handler_data(irq, iommu);
1273 ret = arch_setup_dmar_msi(irq);
1275 irq_set_handler_data(irq, NULL);
1281 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1283 pr_err("IOMMU: can't request irq\n");
1287 int __init enable_drhd_fault_handling(void)
1289 struct dmar_drhd_unit *drhd;
1292 * Enable fault control interrupt.
1294 for_each_drhd_unit(drhd) {
1296 struct intel_iommu *iommu = drhd->iommu;
1297 ret = dmar_set_interrupt(iommu);
1300 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1301 (unsigned long long)drhd->reg_base_addr, ret);
1306 * Clear any previous faults.
1308 dmar_fault(iommu->irq, iommu);
1315 * Re-enable Queued Invalidation interface.
1317 int dmar_reenable_qi(struct intel_iommu *iommu)
1319 if (!ecap_qis(iommu->ecap))
1326 * First disable queued invalidation.
1328 dmar_disable_qi(iommu);
1330 * Then enable queued invalidation again. Since there is no pending
1331 * invalidation requests now, it's safe to re-enable queued
1334 __dmar_enable_qi(iommu);
1340 * Check interrupt remapping support in DMAR table description.
1342 int __init dmar_ir_support(void)
1344 struct acpi_table_dmar *dmar;
1345 dmar = (struct acpi_table_dmar *)dmar_tbl;
1348 return dmar->flags & 0x1;
1350 IOMMU_INIT_POST(detect_intel_iommu);