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 #include "irq_remapping.h"
46 /* No locks are needed as DMA remapping hardware unit
47 * list is constructed at boot time and hotplug of
48 * these units are not supported by the architecture.
50 LIST_HEAD(dmar_drhd_units);
52 struct acpi_table_header * __initdata dmar_tbl;
53 static acpi_size dmar_tbl_size;
55 static void __init dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
58 * add INCLUDE_ALL at the tail, so scan the list will find it at
61 if (drhd->include_all)
62 list_add_tail(&drhd->list, &dmar_drhd_units);
64 list_add(&drhd->list, &dmar_drhd_units);
67 static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope,
68 struct pci_dev **dev, u16 segment)
71 struct pci_dev *pdev = NULL;
72 struct acpi_dmar_pci_path *path;
75 bus = pci_find_bus(segment, scope->bus);
76 path = (struct acpi_dmar_pci_path *)(scope + 1);
77 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
78 / sizeof(struct acpi_dmar_pci_path);
84 * Some BIOSes list non-exist devices in DMAR table, just
88 pr_warn("Device scope bus [%d] not found\n", scope->bus);
91 pdev = pci_get_slot(bus, PCI_DEVFN(path->dev, path->fn));
93 /* warning will be printed below */
98 bus = pdev->subordinate;
101 pr_warn("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 pr_warn("Device scope type does not match for %s\n",
118 int __init dmar_parse_dev_scope(void *start, void *end, int *cnt,
119 struct pci_dev ***devices, u16 segment)
121 struct acpi_dmar_device_scope *scope;
127 while (start < end) {
129 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
130 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
132 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
133 pr_warn("Unsupported device scope\n");
135 start += scope->length;
140 *devices = kcalloc(*cnt, sizeof(struct pci_dev *), GFP_KERNEL);
146 while (start < end) {
148 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
149 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) {
150 ret = dmar_parse_one_dev_scope(scope,
151 &(*devices)[index], segment);
158 start += scope->length;
165 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
166 * structure which uniquely represent one DMA remapping hardware unit
167 * present in the platform
170 dmar_parse_one_drhd(struct acpi_dmar_header *header)
172 struct acpi_dmar_hardware_unit *drhd;
173 struct dmar_drhd_unit *dmaru;
176 drhd = (struct acpi_dmar_hardware_unit *)header;
177 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
182 dmaru->reg_base_addr = drhd->address;
183 dmaru->segment = drhd->segment;
184 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
186 ret = alloc_iommu(dmaru);
191 dmar_register_drhd_unit(dmaru);
195 static int __init dmar_parse_dev(struct dmar_drhd_unit *dmaru)
197 struct acpi_dmar_hardware_unit *drhd;
200 drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
202 if (dmaru->include_all)
205 ret = dmar_parse_dev_scope((void *)(drhd + 1),
206 ((void *)drhd) + drhd->header.length,
207 &dmaru->devices_cnt, &dmaru->devices,
210 list_del(&dmaru->list);
216 #ifdef CONFIG_ACPI_NUMA
218 dmar_parse_one_rhsa(struct acpi_dmar_header *header)
220 struct acpi_dmar_rhsa *rhsa;
221 struct dmar_drhd_unit *drhd;
223 rhsa = (struct acpi_dmar_rhsa *)header;
224 for_each_drhd_unit(drhd) {
225 if (drhd->reg_base_addr == rhsa->base_address) {
226 int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
228 if (!node_online(node))
230 drhd->iommu->node = node;
235 1, TAINT_FIRMWARE_WORKAROUND,
236 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
237 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
239 dmi_get_system_info(DMI_BIOS_VENDOR),
240 dmi_get_system_info(DMI_BIOS_VERSION),
241 dmi_get_system_info(DMI_PRODUCT_VERSION));
248 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
250 struct acpi_dmar_hardware_unit *drhd;
251 struct acpi_dmar_reserved_memory *rmrr;
252 struct acpi_dmar_atsr *atsr;
253 struct acpi_dmar_rhsa *rhsa;
255 switch (header->type) {
256 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
257 drhd = container_of(header, struct acpi_dmar_hardware_unit,
259 pr_info("DRHD base: %#016Lx flags: %#x\n",
260 (unsigned long long)drhd->address, drhd->flags);
262 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
263 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
265 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
266 (unsigned long long)rmrr->base_address,
267 (unsigned long long)rmrr->end_address);
269 case ACPI_DMAR_TYPE_ATSR:
270 atsr = container_of(header, struct acpi_dmar_atsr, header);
271 pr_info("ATSR flags: %#x\n", atsr->flags);
273 case ACPI_DMAR_HARDWARE_AFFINITY:
274 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
275 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
276 (unsigned long long)rhsa->base_address,
277 rhsa->proximity_domain);
283 * dmar_table_detect - checks to see if the platform supports DMAR devices
285 static int __init dmar_table_detect(void)
287 acpi_status status = AE_OK;
289 /* if we could find DMAR table, then there are DMAR devices */
290 status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
291 (struct acpi_table_header **)&dmar_tbl,
294 if (ACPI_SUCCESS(status) && !dmar_tbl) {
295 pr_warn("Unable to map DMAR\n");
296 status = AE_NOT_FOUND;
299 return (ACPI_SUCCESS(status) ? 1 : 0);
303 * parse_dmar_table - parses the DMA reporting table
306 parse_dmar_table(void)
308 struct acpi_table_dmar *dmar;
309 struct acpi_dmar_header *entry_header;
313 * Do it again, earlier dmar_tbl mapping could be mapped with
319 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
320 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
322 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
324 dmar = (struct acpi_table_dmar *)dmar_tbl;
328 if (dmar->width < PAGE_SHIFT - 1) {
329 pr_warn("Invalid DMAR haw\n");
333 pr_info("Host address width %d\n", dmar->width + 1);
335 entry_header = (struct acpi_dmar_header *)(dmar + 1);
336 while (((unsigned long)entry_header) <
337 (((unsigned long)dmar) + dmar_tbl->length)) {
338 /* Avoid looping forever on bad ACPI tables */
339 if (entry_header->length == 0) {
340 pr_warn("Invalid 0-length structure\n");
345 dmar_table_print_dmar_entry(entry_header);
347 switch (entry_header->type) {
348 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
349 ret = dmar_parse_one_drhd(entry_header);
351 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
352 ret = dmar_parse_one_rmrr(entry_header);
354 case ACPI_DMAR_TYPE_ATSR:
355 ret = dmar_parse_one_atsr(entry_header);
357 case ACPI_DMAR_HARDWARE_AFFINITY:
358 #ifdef CONFIG_ACPI_NUMA
359 ret = dmar_parse_one_rhsa(entry_header);
363 pr_warn("Unknown DMAR structure type %d\n",
365 ret = 0; /* for forward compatibility */
371 entry_header = ((void *)entry_header + entry_header->length);
376 static int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
382 for (index = 0; index < cnt; index++)
383 if (dev == devices[index])
386 /* Check our parent */
387 dev = dev->bus->self;
393 struct dmar_drhd_unit *
394 dmar_find_matched_drhd_unit(struct pci_dev *dev)
396 struct dmar_drhd_unit *dmaru = NULL;
397 struct acpi_dmar_hardware_unit *drhd;
399 dev = pci_physfn(dev);
401 list_for_each_entry(dmaru, &dmar_drhd_units, list) {
402 drhd = container_of(dmaru->hdr,
403 struct acpi_dmar_hardware_unit,
406 if (dmaru->include_all &&
407 drhd->segment == pci_domain_nr(dev->bus))
410 if (dmar_pci_device_match(dmaru->devices,
411 dmaru->devices_cnt, dev))
418 int __init dmar_dev_scope_init(void)
420 static int dmar_dev_scope_initialized;
421 struct dmar_drhd_unit *drhd, *drhd_n;
424 if (dmar_dev_scope_initialized)
425 return dmar_dev_scope_initialized;
427 if (list_empty(&dmar_drhd_units))
430 list_for_each_entry_safe(drhd, drhd_n, &dmar_drhd_units, list) {
431 ret = dmar_parse_dev(drhd);
436 ret = dmar_parse_rmrr_atsr_dev();
440 dmar_dev_scope_initialized = 1;
444 dmar_dev_scope_initialized = ret;
449 int __init dmar_table_init(void)
451 static int dmar_table_initialized;
454 if (dmar_table_initialized)
457 dmar_table_initialized = 1;
459 ret = parse_dmar_table();
462 pr_info("parse DMAR table failure.\n");
466 if (list_empty(&dmar_drhd_units)) {
467 pr_info("No DMAR devices found\n");
474 static void warn_invalid_dmar(u64 addr, const char *message)
477 1, TAINT_FIRMWARE_WORKAROUND,
478 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
479 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
481 dmi_get_system_info(DMI_BIOS_VENDOR),
482 dmi_get_system_info(DMI_BIOS_VERSION),
483 dmi_get_system_info(DMI_PRODUCT_VERSION));
486 int __init check_zero_address(void)
488 struct acpi_table_dmar *dmar;
489 struct acpi_dmar_header *entry_header;
490 struct acpi_dmar_hardware_unit *drhd;
492 dmar = (struct acpi_table_dmar *)dmar_tbl;
493 entry_header = (struct acpi_dmar_header *)(dmar + 1);
495 while (((unsigned long)entry_header) <
496 (((unsigned long)dmar) + dmar_tbl->length)) {
497 /* Avoid looping forever on bad ACPI tables */
498 if (entry_header->length == 0) {
499 pr_warn("Invalid 0-length structure\n");
503 if (entry_header->type == ACPI_DMAR_TYPE_HARDWARE_UNIT) {
507 drhd = (void *)entry_header;
508 if (!drhd->address) {
509 warn_invalid_dmar(0, "");
513 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
515 printk("IOMMU: can't validate: %llx\n", drhd->address);
518 cap = dmar_readq(addr + DMAR_CAP_REG);
519 ecap = dmar_readq(addr + DMAR_ECAP_REG);
520 early_iounmap(addr, VTD_PAGE_SIZE);
521 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
522 warn_invalid_dmar(drhd->address,
523 " returns all ones");
528 entry_header = ((void *)entry_header + entry_header->length);
536 int __init detect_intel_iommu(void)
540 ret = dmar_table_detect();
542 ret = check_zero_address();
544 struct acpi_table_dmar *dmar;
546 dmar = (struct acpi_table_dmar *) dmar_tbl;
548 if (ret && irq_remapping_enabled && cpu_has_x2apic &&
550 pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
552 if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
554 /* Make sure ACS will be enabled */
560 x86_init.iommu.iommu_init = intel_iommu_init;
563 early_acpi_os_unmap_memory(dmar_tbl, dmar_tbl_size);
566 return ret ? 1 : -ENODEV;
570 static void unmap_iommu(struct intel_iommu *iommu)
573 release_mem_region(iommu->reg_phys, iommu->reg_size);
577 * map_iommu: map the iommu's registers
578 * @iommu: the iommu to map
579 * @phys_addr: the physical address of the base resgister
581 * Memory map the iommu's registers. Start w/ a single page, and
582 * possibly expand if that turns out to be insufficent.
584 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
588 iommu->reg_phys = phys_addr;
589 iommu->reg_size = VTD_PAGE_SIZE;
591 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
592 pr_err("IOMMU: can't reserve memory\n");
597 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
599 pr_err("IOMMU: can't map the region\n");
604 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
605 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
607 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
609 warn_invalid_dmar(phys_addr, " returns all ones");
613 /* the registers might be more than one page */
614 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
615 cap_max_fault_reg_offset(iommu->cap));
616 map_size = VTD_PAGE_ALIGN(map_size);
617 if (map_size > iommu->reg_size) {
619 release_mem_region(iommu->reg_phys, iommu->reg_size);
620 iommu->reg_size = map_size;
621 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
623 pr_err("IOMMU: can't reserve memory\n");
627 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
629 pr_err("IOMMU: can't map the region\n");
640 release_mem_region(iommu->reg_phys, iommu->reg_size);
645 int alloc_iommu(struct dmar_drhd_unit *drhd)
647 struct intel_iommu *iommu;
649 static int iommu_allocated = 0;
654 if (!drhd->reg_base_addr) {
655 warn_invalid_dmar(0, "");
659 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
663 iommu->seq_id = iommu_allocated++;
664 sprintf (iommu->name, "dmar%d", iommu->seq_id);
666 err = map_iommu(iommu, drhd->reg_base_addr);
668 pr_err("IOMMU: failed to map %s\n", iommu->name);
673 agaw = iommu_calculate_agaw(iommu);
675 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
679 msagaw = iommu_calculate_max_sagaw(iommu);
681 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
686 iommu->msagaw = msagaw;
690 ver = readl(iommu->reg + DMAR_VER_REG);
691 pr_info("IOMMU %d: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
693 (unsigned long long)drhd->reg_base_addr,
694 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
695 (unsigned long long)iommu->cap,
696 (unsigned long long)iommu->ecap);
698 raw_spin_lock_init(&iommu->register_lock);
710 void free_iommu(struct intel_iommu *iommu)
715 free_dmar_iommu(iommu);
724 * Reclaim all the submitted descriptors which have completed its work.
726 static inline void reclaim_free_desc(struct q_inval *qi)
728 while (qi->desc_status[qi->free_tail] == QI_DONE ||
729 qi->desc_status[qi->free_tail] == QI_ABORT) {
730 qi->desc_status[qi->free_tail] = QI_FREE;
731 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
736 static int qi_check_fault(struct intel_iommu *iommu, int index)
740 struct q_inval *qi = iommu->qi;
741 int wait_index = (index + 1) % QI_LENGTH;
743 if (qi->desc_status[wait_index] == QI_ABORT)
746 fault = readl(iommu->reg + DMAR_FSTS_REG);
749 * If IQE happens, the head points to the descriptor associated
750 * with the error. No new descriptors are fetched until the IQE
753 if (fault & DMA_FSTS_IQE) {
754 head = readl(iommu->reg + DMAR_IQH_REG);
755 if ((head >> DMAR_IQ_SHIFT) == index) {
756 pr_err("VT-d detected invalid descriptor: "
757 "low=%llx, high=%llx\n",
758 (unsigned long long)qi->desc[index].low,
759 (unsigned long long)qi->desc[index].high);
760 memcpy(&qi->desc[index], &qi->desc[wait_index],
761 sizeof(struct qi_desc));
762 __iommu_flush_cache(iommu, &qi->desc[index],
763 sizeof(struct qi_desc));
764 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
770 * If ITE happens, all pending wait_desc commands are aborted.
771 * No new descriptors are fetched until the ITE is cleared.
773 if (fault & DMA_FSTS_ITE) {
774 head = readl(iommu->reg + DMAR_IQH_REG);
775 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
777 tail = readl(iommu->reg + DMAR_IQT_REG);
778 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
780 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
783 if (qi->desc_status[head] == QI_IN_USE)
784 qi->desc_status[head] = QI_ABORT;
785 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
786 } while (head != tail);
788 if (qi->desc_status[wait_index] == QI_ABORT)
792 if (fault & DMA_FSTS_ICE)
793 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
799 * Submit the queued invalidation descriptor to the remapping
800 * hardware unit and wait for its completion.
802 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
805 struct q_inval *qi = iommu->qi;
806 struct qi_desc *hw, wait_desc;
807 int wait_index, index;
818 raw_spin_lock_irqsave(&qi->q_lock, flags);
819 while (qi->free_cnt < 3) {
820 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
822 raw_spin_lock_irqsave(&qi->q_lock, flags);
825 index = qi->free_head;
826 wait_index = (index + 1) % QI_LENGTH;
828 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
832 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
833 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
834 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
836 hw[wait_index] = wait_desc;
838 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
839 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
841 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
845 * update the HW tail register indicating the presence of
848 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
850 while (qi->desc_status[wait_index] != QI_DONE) {
852 * We will leave the interrupts disabled, to prevent interrupt
853 * context to queue another cmd while a cmd is already submitted
854 * and waiting for completion on this cpu. This is to avoid
855 * a deadlock where the interrupt context can wait indefinitely
856 * for free slots in the queue.
858 rc = qi_check_fault(iommu, index);
862 raw_spin_unlock(&qi->q_lock);
864 raw_spin_lock(&qi->q_lock);
867 qi->desc_status[index] = QI_DONE;
869 reclaim_free_desc(qi);
870 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
879 * Flush the global interrupt entry cache.
881 void qi_global_iec(struct intel_iommu *iommu)
885 desc.low = QI_IEC_TYPE;
888 /* should never fail */
889 qi_submit_sync(&desc, iommu);
892 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
897 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
898 | QI_CC_GRAN(type) | QI_CC_TYPE;
901 qi_submit_sync(&desc, iommu);
904 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
905 unsigned int size_order, u64 type)
912 if (cap_write_drain(iommu->cap))
915 if (cap_read_drain(iommu->cap))
918 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
919 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
920 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
921 | QI_IOTLB_AM(size_order);
923 qi_submit_sync(&desc, iommu);
926 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
927 u64 addr, unsigned mask)
932 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
933 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
934 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
936 desc.high = QI_DEV_IOTLB_ADDR(addr);
938 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
941 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
944 qi_submit_sync(&desc, iommu);
948 * Disable Queued Invalidation interface.
950 void dmar_disable_qi(struct intel_iommu *iommu)
954 cycles_t start_time = get_cycles();
956 if (!ecap_qis(iommu->ecap))
959 raw_spin_lock_irqsave(&iommu->register_lock, flags);
961 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
962 if (!(sts & DMA_GSTS_QIES))
966 * Give a chance to HW to complete the pending invalidation requests.
968 while ((readl(iommu->reg + DMAR_IQT_REG) !=
969 readl(iommu->reg + DMAR_IQH_REG)) &&
970 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
973 iommu->gcmd &= ~DMA_GCMD_QIE;
974 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
976 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
977 !(sts & DMA_GSTS_QIES), sts);
979 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
983 * Enable queued invalidation.
985 static void __dmar_enable_qi(struct intel_iommu *iommu)
989 struct q_inval *qi = iommu->qi;
991 qi->free_head = qi->free_tail = 0;
992 qi->free_cnt = QI_LENGTH;
994 raw_spin_lock_irqsave(&iommu->register_lock, flags);
996 /* write zero to the tail reg */
997 writel(0, iommu->reg + DMAR_IQT_REG);
999 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1001 iommu->gcmd |= DMA_GCMD_QIE;
1002 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1004 /* Make sure hardware complete it */
1005 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1007 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1011 * Enable Queued Invalidation interface. This is a must to support
1012 * interrupt-remapping. Also used by DMA-remapping, which replaces
1013 * register based IOTLB invalidation.
1015 int dmar_enable_qi(struct intel_iommu *iommu)
1018 struct page *desc_page;
1020 if (!ecap_qis(iommu->ecap))
1024 * queued invalidation is already setup and enabled.
1029 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1036 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1043 qi->desc = page_address(desc_page);
1045 qi->desc_status = kzalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1046 if (!qi->desc_status) {
1047 free_page((unsigned long) qi->desc);
1053 qi->free_head = qi->free_tail = 0;
1054 qi->free_cnt = QI_LENGTH;
1056 raw_spin_lock_init(&qi->q_lock);
1058 __dmar_enable_qi(iommu);
1063 /* iommu interrupt handling. Most stuff are MSI-like. */
1071 static const char *dma_remap_fault_reasons[] =
1074 "Present bit in root entry is clear",
1075 "Present bit in context entry is clear",
1076 "Invalid context entry",
1077 "Access beyond MGAW",
1078 "PTE Write access is not set",
1079 "PTE Read access is not set",
1080 "Next page table ptr is invalid",
1081 "Root table address invalid",
1082 "Context table ptr is invalid",
1083 "non-zero reserved fields in RTP",
1084 "non-zero reserved fields in CTP",
1085 "non-zero reserved fields in PTE",
1086 "PCE for translation request specifies blocking",
1089 static const char *irq_remap_fault_reasons[] =
1091 "Detected reserved fields in the decoded interrupt-remapped request",
1092 "Interrupt index exceeded the interrupt-remapping table size",
1093 "Present field in the IRTE entry is clear",
1094 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1095 "Detected reserved fields in the IRTE entry",
1096 "Blocked a compatibility format interrupt request",
1097 "Blocked an interrupt request due to source-id verification failure",
1100 #define MAX_FAULT_REASON_IDX (ARRAY_SIZE(fault_reason_strings) - 1)
1102 const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1104 if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1105 ARRAY_SIZE(irq_remap_fault_reasons))) {
1106 *fault_type = INTR_REMAP;
1107 return irq_remap_fault_reasons[fault_reason - 0x20];
1108 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1109 *fault_type = DMA_REMAP;
1110 return dma_remap_fault_reasons[fault_reason];
1112 *fault_type = UNKNOWN;
1117 void dmar_msi_unmask(struct irq_data *data)
1119 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1123 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1124 writel(0, iommu->reg + DMAR_FECTL_REG);
1125 /* Read a reg to force flush the post write */
1126 readl(iommu->reg + DMAR_FECTL_REG);
1127 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1130 void dmar_msi_mask(struct irq_data *data)
1133 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1136 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1137 writel(DMA_FECTL_IM, iommu->reg + DMAR_FECTL_REG);
1138 /* Read a reg to force flush the post write */
1139 readl(iommu->reg + DMAR_FECTL_REG);
1140 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1143 void dmar_msi_write(int irq, struct msi_msg *msg)
1145 struct intel_iommu *iommu = irq_get_handler_data(irq);
1148 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1149 writel(msg->data, iommu->reg + DMAR_FEDATA_REG);
1150 writel(msg->address_lo, iommu->reg + DMAR_FEADDR_REG);
1151 writel(msg->address_hi, iommu->reg + DMAR_FEUADDR_REG);
1152 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1155 void dmar_msi_read(int irq, struct msi_msg *msg)
1157 struct intel_iommu *iommu = irq_get_handler_data(irq);
1160 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1161 msg->data = readl(iommu->reg + DMAR_FEDATA_REG);
1162 msg->address_lo = readl(iommu->reg + DMAR_FEADDR_REG);
1163 msg->address_hi = readl(iommu->reg + DMAR_FEUADDR_REG);
1164 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1167 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1168 u8 fault_reason, u16 source_id, unsigned long long addr)
1173 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1175 if (fault_type == INTR_REMAP)
1176 pr_err("INTR-REMAP: Request device [[%02x:%02x.%d] "
1177 "fault index %llx\n"
1178 "INTR-REMAP:[fault reason %02d] %s\n",
1179 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1180 PCI_FUNC(source_id & 0xFF), addr >> 48,
1181 fault_reason, reason);
1183 pr_err("DMAR:[%s] Request device [%02x:%02x.%d] "
1184 "fault addr %llx \n"
1185 "DMAR:[fault reason %02d] %s\n",
1186 (type ? "DMA Read" : "DMA Write"),
1187 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1188 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1192 #define PRIMARY_FAULT_REG_LEN (16)
1193 irqreturn_t dmar_fault(int irq, void *dev_id)
1195 struct intel_iommu *iommu = dev_id;
1196 int reg, fault_index;
1200 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1201 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1203 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1205 /* TBD: ignore advanced fault log currently */
1206 if (!(fault_status & DMA_FSTS_PPF))
1209 fault_index = dma_fsts_fault_record_index(fault_status);
1210 reg = cap_fault_reg_offset(iommu->cap);
1218 /* highest 32 bits */
1219 data = readl(iommu->reg + reg +
1220 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1221 if (!(data & DMA_FRCD_F))
1224 fault_reason = dma_frcd_fault_reason(data);
1225 type = dma_frcd_type(data);
1227 data = readl(iommu->reg + reg +
1228 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1229 source_id = dma_frcd_source_id(data);
1231 guest_addr = dmar_readq(iommu->reg + reg +
1232 fault_index * PRIMARY_FAULT_REG_LEN);
1233 guest_addr = dma_frcd_page_addr(guest_addr);
1234 /* clear the fault */
1235 writel(DMA_FRCD_F, iommu->reg + reg +
1236 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1238 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1240 dmar_fault_do_one(iommu, type, fault_reason,
1241 source_id, guest_addr);
1244 if (fault_index >= cap_num_fault_regs(iommu->cap))
1246 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1249 /* clear all the other faults */
1250 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1251 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1253 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1257 int dmar_set_interrupt(struct intel_iommu *iommu)
1262 * Check if the fault interrupt is already initialized.
1269 pr_err("IOMMU: no free vectors\n");
1273 irq_set_handler_data(irq, iommu);
1276 ret = arch_setup_dmar_msi(irq);
1278 irq_set_handler_data(irq, NULL);
1284 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1286 pr_err("IOMMU: can't request irq\n");
1290 int __init enable_drhd_fault_handling(void)
1292 struct dmar_drhd_unit *drhd;
1295 * Enable fault control interrupt.
1297 for_each_drhd_unit(drhd) {
1299 struct intel_iommu *iommu = drhd->iommu;
1300 ret = dmar_set_interrupt(iommu);
1303 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1304 (unsigned long long)drhd->reg_base_addr, ret);
1309 * Clear any previous faults.
1311 dmar_fault(iommu->irq, iommu);
1318 * Re-enable Queued Invalidation interface.
1320 int dmar_reenable_qi(struct intel_iommu *iommu)
1322 if (!ecap_qis(iommu->ecap))
1329 * First disable queued invalidation.
1331 dmar_disable_qi(iommu);
1333 * Then enable queued invalidation again. Since there is no pending
1334 * invalidation requests now, it's safe to re-enable queued
1337 __dmar_enable_qi(iommu);
1343 * Check interrupt remapping support in DMAR table description.
1345 int __init dmar_ir_support(void)
1347 struct acpi_table_dmar *dmar;
1348 dmar = (struct acpi_table_dmar *)dmar_tbl;
1351 return dmar->flags & 0x1;
1353 IOMMU_INIT_POST(detect_intel_iommu);