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)
134 printk(KERN_WARNING PREFIX
135 "Unsupported device scope\n");
136 start += scope->length;
141 *devices = kcalloc(*cnt, sizeof(struct pci_dev *), GFP_KERNEL);
147 while (start < end) {
149 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
150 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) {
151 ret = dmar_parse_one_dev_scope(scope,
152 &(*devices)[index], segment);
159 start += scope->length;
166 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
167 * structure which uniquely represent one DMA remapping hardware unit
168 * present in the platform
171 dmar_parse_one_drhd(struct acpi_dmar_header *header)
173 struct acpi_dmar_hardware_unit *drhd;
174 struct dmar_drhd_unit *dmaru;
177 drhd = (struct acpi_dmar_hardware_unit *)header;
178 if (!drhd->address) {
179 /* Promote an attitude of violence to a BIOS engineer today */
180 WARN(1, "Your BIOS is broken; DMAR reported at address zero!\n"
181 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
182 dmi_get_system_info(DMI_BIOS_VENDOR),
183 dmi_get_system_info(DMI_BIOS_VERSION),
184 dmi_get_system_info(DMI_PRODUCT_VERSION));
187 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
192 dmaru->reg_base_addr = drhd->address;
193 dmaru->segment = drhd->segment;
194 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
196 ret = alloc_iommu(dmaru);
201 dmar_register_drhd_unit(dmaru);
205 static int __init dmar_parse_dev(struct dmar_drhd_unit *dmaru)
207 struct acpi_dmar_hardware_unit *drhd;
210 drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
212 if (dmaru->include_all)
215 ret = dmar_parse_dev_scope((void *)(drhd + 1),
216 ((void *)drhd) + drhd->header.length,
217 &dmaru->devices_cnt, &dmaru->devices,
220 list_del(&dmaru->list);
227 LIST_HEAD(dmar_rmrr_units);
229 static void __init dmar_register_rmrr_unit(struct dmar_rmrr_unit *rmrr)
231 list_add(&rmrr->list, &dmar_rmrr_units);
236 dmar_parse_one_rmrr(struct acpi_dmar_header *header)
238 struct acpi_dmar_reserved_memory *rmrr;
239 struct dmar_rmrr_unit *rmrru;
241 rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL);
246 rmrr = (struct acpi_dmar_reserved_memory *)header;
247 rmrru->base_address = rmrr->base_address;
248 rmrru->end_address = rmrr->end_address;
250 dmar_register_rmrr_unit(rmrru);
255 rmrr_parse_dev(struct dmar_rmrr_unit *rmrru)
257 struct acpi_dmar_reserved_memory *rmrr;
260 rmrr = (struct acpi_dmar_reserved_memory *) rmrru->hdr;
261 ret = dmar_parse_dev_scope((void *)(rmrr + 1),
262 ((void *)rmrr) + rmrr->header.length,
263 &rmrru->devices_cnt, &rmrru->devices, rmrr->segment);
265 if (ret || (rmrru->devices_cnt == 0)) {
266 list_del(&rmrru->list);
272 static LIST_HEAD(dmar_atsr_units);
274 static int __init dmar_parse_one_atsr(struct acpi_dmar_header *hdr)
276 struct acpi_dmar_atsr *atsr;
277 struct dmar_atsr_unit *atsru;
279 atsr = container_of(hdr, struct acpi_dmar_atsr, header);
280 atsru = kzalloc(sizeof(*atsru), GFP_KERNEL);
285 atsru->include_all = atsr->flags & 0x1;
287 list_add(&atsru->list, &dmar_atsr_units);
292 static int __init atsr_parse_dev(struct dmar_atsr_unit *atsru)
295 struct acpi_dmar_atsr *atsr;
297 if (atsru->include_all)
300 atsr = container_of(atsru->hdr, struct acpi_dmar_atsr, header);
301 rc = dmar_parse_dev_scope((void *)(atsr + 1),
302 (void *)atsr + atsr->header.length,
303 &atsru->devices_cnt, &atsru->devices,
305 if (rc || !atsru->devices_cnt) {
306 list_del(&atsru->list);
313 int dmar_find_matched_atsr_unit(struct pci_dev *dev)
317 struct acpi_dmar_atsr *atsr;
318 struct dmar_atsr_unit *atsru;
320 list_for_each_entry(atsru, &dmar_atsr_units, list) {
321 atsr = container_of(atsru->hdr, struct acpi_dmar_atsr, header);
322 if (atsr->segment == pci_domain_nr(dev->bus))
329 for (bus = dev->bus; bus; bus = bus->parent) {
330 struct pci_dev *bridge = bus->self;
332 if (!bridge || !bridge->is_pcie ||
333 bridge->pcie_type == PCI_EXP_TYPE_PCI_BRIDGE)
336 if (bridge->pcie_type == PCI_EXP_TYPE_ROOT_PORT) {
337 for (i = 0; i < atsru->devices_cnt; i++)
338 if (atsru->devices[i] == bridge)
344 if (atsru->include_all)
352 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
354 struct acpi_dmar_hardware_unit *drhd;
355 struct acpi_dmar_reserved_memory *rmrr;
356 struct acpi_dmar_atsr *atsr;
357 struct acpi_dmar_rhsa *rhsa;
359 switch (header->type) {
360 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
361 drhd = container_of(header, struct acpi_dmar_hardware_unit,
363 printk (KERN_INFO PREFIX
364 "DRHD base: %#016Lx flags: %#x\n",
365 (unsigned long long)drhd->address, drhd->flags);
367 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
368 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
370 printk (KERN_INFO PREFIX
371 "RMRR base: %#016Lx end: %#016Lx\n",
372 (unsigned long long)rmrr->base_address,
373 (unsigned long long)rmrr->end_address);
375 case ACPI_DMAR_TYPE_ATSR:
376 atsr = container_of(header, struct acpi_dmar_atsr, header);
377 printk(KERN_INFO PREFIX "ATSR flags: %#x\n", atsr->flags);
379 case ACPI_DMAR_HARDWARE_AFFINITY:
380 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
381 printk(KERN_INFO PREFIX "RHSA base: %#016Lx proximity domain: %#x\n",
382 (unsigned long long)rhsa->base_address,
383 rhsa->proximity_domain);
389 * dmar_table_detect - checks to see if the platform supports DMAR devices
391 static int __init dmar_table_detect(void)
393 acpi_status status = AE_OK;
395 /* if we could find DMAR table, then there are DMAR devices */
396 status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
397 (struct acpi_table_header **)&dmar_tbl,
400 if (ACPI_SUCCESS(status) && !dmar_tbl) {
401 printk (KERN_WARNING PREFIX "Unable to map DMAR\n");
402 status = AE_NOT_FOUND;
405 return (ACPI_SUCCESS(status) ? 1 : 0);
409 * parse_dmar_table - parses the DMA reporting table
412 parse_dmar_table(void)
414 struct acpi_table_dmar *dmar;
415 struct acpi_dmar_header *entry_header;
419 * Do it again, earlier dmar_tbl mapping could be mapped with
425 * ACPI tables may not be DMA protected by tboot, so use DMAR copy
426 * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
428 dmar_tbl = tboot_get_dmar_table(dmar_tbl);
430 dmar = (struct acpi_table_dmar *)dmar_tbl;
434 if (dmar->width < PAGE_SHIFT - 1) {
435 printk(KERN_WARNING PREFIX "Invalid DMAR haw\n");
439 printk (KERN_INFO PREFIX "Host address width %d\n",
442 entry_header = (struct acpi_dmar_header *)(dmar + 1);
443 while (((unsigned long)entry_header) <
444 (((unsigned long)dmar) + dmar_tbl->length)) {
445 /* Avoid looping forever on bad ACPI tables */
446 if (entry_header->length == 0) {
447 printk(KERN_WARNING PREFIX
448 "Invalid 0-length structure\n");
453 dmar_table_print_dmar_entry(entry_header);
455 switch (entry_header->type) {
456 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
457 ret = dmar_parse_one_drhd(entry_header);
459 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
461 ret = dmar_parse_one_rmrr(entry_header);
464 case ACPI_DMAR_TYPE_ATSR:
466 ret = dmar_parse_one_atsr(entry_header);
469 case ACPI_DMAR_HARDWARE_AFFINITY:
470 /* We don't do anything with RHSA (yet?) */
473 printk(KERN_WARNING PREFIX
474 "Unknown DMAR structure type %d\n",
476 ret = 0; /* for forward compatibility */
482 entry_header = ((void *)entry_header + entry_header->length);
487 int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
493 for (index = 0; index < cnt; index++)
494 if (dev == devices[index])
497 /* Check our parent */
498 dev = dev->bus->self;
504 struct dmar_drhd_unit *
505 dmar_find_matched_drhd_unit(struct pci_dev *dev)
507 struct dmar_drhd_unit *dmaru = NULL;
508 struct acpi_dmar_hardware_unit *drhd;
510 list_for_each_entry(dmaru, &dmar_drhd_units, list) {
511 drhd = container_of(dmaru->hdr,
512 struct acpi_dmar_hardware_unit,
515 if (dmaru->include_all &&
516 drhd->segment == pci_domain_nr(dev->bus))
519 if (dmar_pci_device_match(dmaru->devices,
520 dmaru->devices_cnt, dev))
527 int __init dmar_dev_scope_init(void)
529 struct dmar_drhd_unit *drhd, *drhd_n;
532 list_for_each_entry_safe(drhd, drhd_n, &dmar_drhd_units, list) {
533 ret = dmar_parse_dev(drhd);
540 struct dmar_rmrr_unit *rmrr, *rmrr_n;
541 struct dmar_atsr_unit *atsr, *atsr_n;
543 list_for_each_entry_safe(rmrr, rmrr_n, &dmar_rmrr_units, list) {
544 ret = rmrr_parse_dev(rmrr);
549 list_for_each_entry_safe(atsr, atsr_n, &dmar_atsr_units, list) {
550 ret = atsr_parse_dev(atsr);
561 int __init dmar_table_init(void)
563 static int dmar_table_initialized;
566 if (dmar_table_initialized)
569 dmar_table_initialized = 1;
571 ret = parse_dmar_table();
574 printk(KERN_INFO PREFIX "parse DMAR table failure.\n");
578 if (list_empty(&dmar_drhd_units)) {
579 printk(KERN_INFO PREFIX "No DMAR devices found\n");
584 if (list_empty(&dmar_rmrr_units))
585 printk(KERN_INFO PREFIX "No RMRR found\n");
587 if (list_empty(&dmar_atsr_units))
588 printk(KERN_INFO PREFIX "No ATSR found\n");
594 void __init detect_intel_iommu(void)
598 ret = dmar_table_detect();
601 #ifdef CONFIG_INTR_REMAP
602 struct acpi_table_dmar *dmar;
604 * for now we will disable dma-remapping when interrupt
605 * remapping is enabled.
606 * When support for queued invalidation for IOTLB invalidation
607 * is added, we will not need this any more.
609 dmar = (struct acpi_table_dmar *) dmar_tbl;
610 if (ret && cpu_has_x2apic && dmar->flags & 0x1)
612 "Queued invalidation will be enabled to support "
613 "x2apic and Intr-remapping.\n");
616 if (ret && !no_iommu && !iommu_detected && !swiotlb &&
622 x86_init.iommu.iommu_init = intel_iommu_init;
625 early_acpi_os_unmap_memory(dmar_tbl, dmar_tbl_size);
630 int alloc_iommu(struct dmar_drhd_unit *drhd)
632 struct intel_iommu *iommu;
635 static int iommu_allocated = 0;
639 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
643 iommu->seq_id = iommu_allocated++;
644 sprintf (iommu->name, "dmar%d", iommu->seq_id);
646 iommu->reg = ioremap(drhd->reg_base_addr, VTD_PAGE_SIZE);
648 printk(KERN_ERR "IOMMU: can't map the region\n");
651 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
652 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
654 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
655 /* Promote an attitude of violence to a BIOS engineer today */
656 WARN(1, "Your BIOS is broken; DMAR reported at address %llx returns all ones!\n"
657 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
659 dmi_get_system_info(DMI_BIOS_VENDOR),
660 dmi_get_system_info(DMI_BIOS_VERSION),
661 dmi_get_system_info(DMI_PRODUCT_VERSION));
666 agaw = iommu_calculate_agaw(iommu);
669 "Cannot get a valid agaw for iommu (seq_id = %d)\n",
673 msagaw = iommu_calculate_max_sagaw(iommu);
676 "Cannot get a valid max agaw for iommu (seq_id = %d)\n",
682 iommu->msagaw = msagaw;
684 /* the registers might be more than one page */
685 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
686 cap_max_fault_reg_offset(iommu->cap));
687 map_size = VTD_PAGE_ALIGN(map_size);
688 if (map_size > VTD_PAGE_SIZE) {
690 iommu->reg = ioremap(drhd->reg_base_addr, map_size);
692 printk(KERN_ERR "IOMMU: can't map the region\n");
697 ver = readl(iommu->reg + DMAR_VER_REG);
698 pr_info("IOMMU %llx: ver %d:%d cap %llx ecap %llx\n",
699 (unsigned long long)drhd->reg_base_addr,
700 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
701 (unsigned long long)iommu->cap,
702 (unsigned long long)iommu->ecap);
704 spin_lock_init(&iommu->register_lock);
716 void free_iommu(struct intel_iommu *iommu)
722 free_dmar_iommu(iommu);
731 * Reclaim all the submitted descriptors which have completed its work.
733 static inline void reclaim_free_desc(struct q_inval *qi)
735 while (qi->desc_status[qi->free_tail] == QI_DONE ||
736 qi->desc_status[qi->free_tail] == QI_ABORT) {
737 qi->desc_status[qi->free_tail] = QI_FREE;
738 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
743 static int qi_check_fault(struct intel_iommu *iommu, int index)
747 struct q_inval *qi = iommu->qi;
748 int wait_index = (index + 1) % QI_LENGTH;
750 if (qi->desc_status[wait_index] == QI_ABORT)
753 fault = readl(iommu->reg + DMAR_FSTS_REG);
756 * If IQE happens, the head points to the descriptor associated
757 * with the error. No new descriptors are fetched until the IQE
760 if (fault & DMA_FSTS_IQE) {
761 head = readl(iommu->reg + DMAR_IQH_REG);
762 if ((head >> DMAR_IQ_SHIFT) == index) {
763 printk(KERN_ERR "VT-d detected invalid descriptor: "
764 "low=%llx, high=%llx\n",
765 (unsigned long long)qi->desc[index].low,
766 (unsigned long long)qi->desc[index].high);
767 memcpy(&qi->desc[index], &qi->desc[wait_index],
768 sizeof(struct qi_desc));
769 __iommu_flush_cache(iommu, &qi->desc[index],
770 sizeof(struct qi_desc));
771 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
777 * If ITE happens, all pending wait_desc commands are aborted.
778 * No new descriptors are fetched until the ITE is cleared.
780 if (fault & DMA_FSTS_ITE) {
781 head = readl(iommu->reg + DMAR_IQH_REG);
782 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
784 tail = readl(iommu->reg + DMAR_IQT_REG);
785 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
787 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
790 if (qi->desc_status[head] == QI_IN_USE)
791 qi->desc_status[head] = QI_ABORT;
792 head = (head - 2 + QI_LENGTH) % QI_LENGTH;
793 } while (head != tail);
795 if (qi->desc_status[wait_index] == QI_ABORT)
799 if (fault & DMA_FSTS_ICE)
800 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
806 * Submit the queued invalidation descriptor to the remapping
807 * hardware unit and wait for its completion.
809 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
812 struct q_inval *qi = iommu->qi;
813 struct qi_desc *hw, wait_desc;
814 int wait_index, index;
825 spin_lock_irqsave(&qi->q_lock, flags);
826 while (qi->free_cnt < 3) {
827 spin_unlock_irqrestore(&qi->q_lock, flags);
829 spin_lock_irqsave(&qi->q_lock, flags);
832 index = qi->free_head;
833 wait_index = (index + 1) % QI_LENGTH;
835 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
839 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
840 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
841 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
843 hw[wait_index] = wait_desc;
845 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
846 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
848 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
852 * update the HW tail register indicating the presence of
855 writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
857 while (qi->desc_status[wait_index] != QI_DONE) {
859 * We will leave the interrupts disabled, to prevent interrupt
860 * context to queue another cmd while a cmd is already submitted
861 * and waiting for completion on this cpu. This is to avoid
862 * a deadlock where the interrupt context can wait indefinitely
863 * for free slots in the queue.
865 rc = qi_check_fault(iommu, index);
869 spin_unlock(&qi->q_lock);
871 spin_lock(&qi->q_lock);
874 qi->desc_status[index] = QI_DONE;
876 reclaim_free_desc(qi);
877 spin_unlock_irqrestore(&qi->q_lock, flags);
886 * Flush the global interrupt entry cache.
888 void qi_global_iec(struct intel_iommu *iommu)
892 desc.low = QI_IEC_TYPE;
895 /* should never fail */
896 qi_submit_sync(&desc, iommu);
899 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
904 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
905 | QI_CC_GRAN(type) | QI_CC_TYPE;
908 qi_submit_sync(&desc, iommu);
911 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
912 unsigned int size_order, u64 type)
919 if (cap_write_drain(iommu->cap))
922 if (cap_read_drain(iommu->cap))
925 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
926 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
927 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
928 | QI_IOTLB_AM(size_order);
930 qi_submit_sync(&desc, iommu);
933 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
934 u64 addr, unsigned mask)
939 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
940 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
941 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
943 desc.high = QI_DEV_IOTLB_ADDR(addr);
945 if (qdep >= QI_DEV_IOTLB_MAX_INVS)
948 desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
951 qi_submit_sync(&desc, iommu);
955 * Disable Queued Invalidation interface.
957 void dmar_disable_qi(struct intel_iommu *iommu)
961 cycles_t start_time = get_cycles();
963 if (!ecap_qis(iommu->ecap))
966 spin_lock_irqsave(&iommu->register_lock, flags);
968 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
969 if (!(sts & DMA_GSTS_QIES))
973 * Give a chance to HW to complete the pending invalidation requests.
975 while ((readl(iommu->reg + DMAR_IQT_REG) !=
976 readl(iommu->reg + DMAR_IQH_REG)) &&
977 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
980 iommu->gcmd &= ~DMA_GCMD_QIE;
981 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
983 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
984 !(sts & DMA_GSTS_QIES), sts);
986 spin_unlock_irqrestore(&iommu->register_lock, flags);
990 * Enable queued invalidation.
992 static void __dmar_enable_qi(struct intel_iommu *iommu)
996 struct q_inval *qi = iommu->qi;
998 qi->free_head = qi->free_tail = 0;
999 qi->free_cnt = QI_LENGTH;
1001 spin_lock_irqsave(&iommu->register_lock, flags);
1003 /* write zero to the tail reg */
1004 writel(0, iommu->reg + DMAR_IQT_REG);
1006 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1008 iommu->gcmd |= DMA_GCMD_QIE;
1009 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1011 /* Make sure hardware complete it */
1012 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1014 spin_unlock_irqrestore(&iommu->register_lock, flags);
1018 * Enable Queued Invalidation interface. This is a must to support
1019 * interrupt-remapping. Also used by DMA-remapping, which replaces
1020 * register based IOTLB invalidation.
1022 int dmar_enable_qi(struct intel_iommu *iommu)
1026 if (!ecap_qis(iommu->ecap))
1030 * queued invalidation is already setup and enabled.
1035 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1041 qi->desc = (void *)(get_zeroed_page(GFP_ATOMIC));
1048 qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1049 if (!qi->desc_status) {
1050 free_page((unsigned long) qi->desc);
1056 qi->free_head = qi->free_tail = 0;
1057 qi->free_cnt = QI_LENGTH;
1059 spin_lock_init(&qi->q_lock);
1061 __dmar_enable_qi(iommu);
1066 /* iommu interrupt handling. Most stuff are MSI-like. */
1074 static const char *dma_remap_fault_reasons[] =
1077 "Present bit in root entry is clear",
1078 "Present bit in context entry is clear",
1079 "Invalid context entry",
1080 "Access beyond MGAW",
1081 "PTE Write access is not set",
1082 "PTE Read access is not set",
1083 "Next page table ptr is invalid",
1084 "Root table address invalid",
1085 "Context table ptr is invalid",
1086 "non-zero reserved fields in RTP",
1087 "non-zero reserved fields in CTP",
1088 "non-zero reserved fields in PTE",
1091 static const char *intr_remap_fault_reasons[] =
1093 "Detected reserved fields in the decoded interrupt-remapped request",
1094 "Interrupt index exceeded the interrupt-remapping table size",
1095 "Present field in the IRTE entry is clear",
1096 "Error accessing interrupt-remapping table pointed by IRTA_REG",
1097 "Detected reserved fields in the IRTE entry",
1098 "Blocked a compatibility format interrupt request",
1099 "Blocked an interrupt request due to source-id verification failure",
1102 #define MAX_FAULT_REASON_IDX (ARRAY_SIZE(fault_reason_strings) - 1)
1104 const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1106 if (fault_reason >= 0x20 && (fault_reason <= 0x20 +
1107 ARRAY_SIZE(intr_remap_fault_reasons))) {
1108 *fault_type = INTR_REMAP;
1109 return intr_remap_fault_reasons[fault_reason - 0x20];
1110 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1111 *fault_type = DMA_REMAP;
1112 return dma_remap_fault_reasons[fault_reason];
1114 *fault_type = UNKNOWN;
1119 void dmar_msi_unmask(unsigned int irq)
1121 struct intel_iommu *iommu = get_irq_data(irq);
1125 spin_lock_irqsave(&iommu->register_lock, flag);
1126 writel(0, iommu->reg + DMAR_FECTL_REG);
1127 /* Read a reg to force flush the post write */
1128 readl(iommu->reg + DMAR_FECTL_REG);
1129 spin_unlock_irqrestore(&iommu->register_lock, flag);
1132 void dmar_msi_mask(unsigned int irq)
1135 struct intel_iommu *iommu = get_irq_data(irq);
1138 spin_lock_irqsave(&iommu->register_lock, flag);
1139 writel(DMA_FECTL_IM, iommu->reg + DMAR_FECTL_REG);
1140 /* Read a reg to force flush the post write */
1141 readl(iommu->reg + DMAR_FECTL_REG);
1142 spin_unlock_irqrestore(&iommu->register_lock, flag);
1145 void dmar_msi_write(int irq, struct msi_msg *msg)
1147 struct intel_iommu *iommu = get_irq_data(irq);
1150 spin_lock_irqsave(&iommu->register_lock, flag);
1151 writel(msg->data, iommu->reg + DMAR_FEDATA_REG);
1152 writel(msg->address_lo, iommu->reg + DMAR_FEADDR_REG);
1153 writel(msg->address_hi, iommu->reg + DMAR_FEUADDR_REG);
1154 spin_unlock_irqrestore(&iommu->register_lock, flag);
1157 void dmar_msi_read(int irq, struct msi_msg *msg)
1159 struct intel_iommu *iommu = get_irq_data(irq);
1162 spin_lock_irqsave(&iommu->register_lock, flag);
1163 msg->data = readl(iommu->reg + DMAR_FEDATA_REG);
1164 msg->address_lo = readl(iommu->reg + DMAR_FEADDR_REG);
1165 msg->address_hi = readl(iommu->reg + DMAR_FEUADDR_REG);
1166 spin_unlock_irqrestore(&iommu->register_lock, flag);
1169 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1170 u8 fault_reason, u16 source_id, unsigned long long addr)
1175 reason = dmar_get_fault_reason(fault_reason, &fault_type);
1177 if (fault_type == INTR_REMAP)
1178 printk(KERN_ERR "INTR-REMAP: Request device [[%02x:%02x.%d] "
1179 "fault index %llx\n"
1180 "INTR-REMAP:[fault reason %02d] %s\n",
1181 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1182 PCI_FUNC(source_id & 0xFF), addr >> 48,
1183 fault_reason, reason);
1186 "DMAR:[%s] Request device [%02x:%02x.%d] "
1187 "fault addr %llx \n"
1188 "DMAR:[fault reason %02d] %s\n",
1189 (type ? "DMA Read" : "DMA Write"),
1190 (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1191 PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1195 #define PRIMARY_FAULT_REG_LEN (16)
1196 irqreturn_t dmar_fault(int irq, void *dev_id)
1198 struct intel_iommu *iommu = dev_id;
1199 int reg, fault_index;
1203 spin_lock_irqsave(&iommu->register_lock, flag);
1204 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1206 printk(KERN_ERR "DRHD: handling fault status reg %x\n",
1209 /* TBD: ignore advanced fault log currently */
1210 if (!(fault_status & DMA_FSTS_PPF))
1213 fault_index = dma_fsts_fault_record_index(fault_status);
1214 reg = cap_fault_reg_offset(iommu->cap);
1222 /* highest 32 bits */
1223 data = readl(iommu->reg + reg +
1224 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1225 if (!(data & DMA_FRCD_F))
1228 fault_reason = dma_frcd_fault_reason(data);
1229 type = dma_frcd_type(data);
1231 data = readl(iommu->reg + reg +
1232 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1233 source_id = dma_frcd_source_id(data);
1235 guest_addr = dmar_readq(iommu->reg + reg +
1236 fault_index * PRIMARY_FAULT_REG_LEN);
1237 guest_addr = dma_frcd_page_addr(guest_addr);
1238 /* clear the fault */
1239 writel(DMA_FRCD_F, iommu->reg + reg +
1240 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1242 spin_unlock_irqrestore(&iommu->register_lock, flag);
1244 dmar_fault_do_one(iommu, type, fault_reason,
1245 source_id, guest_addr);
1248 if (fault_index >= cap_num_fault_regs(iommu->cap))
1250 spin_lock_irqsave(&iommu->register_lock, flag);
1253 /* clear all the other faults */
1254 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1255 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1257 spin_unlock_irqrestore(&iommu->register_lock, flag);
1261 int dmar_set_interrupt(struct intel_iommu *iommu)
1266 * Check if the fault interrupt is already initialized.
1273 printk(KERN_ERR "IOMMU: no free vectors\n");
1277 set_irq_data(irq, iommu);
1280 ret = arch_setup_dmar_msi(irq);
1282 set_irq_data(irq, NULL);
1288 ret = request_irq(irq, dmar_fault, 0, iommu->name, iommu);
1290 printk(KERN_ERR "IOMMU: can't request irq\n");
1294 int __init enable_drhd_fault_handling(void)
1296 struct dmar_drhd_unit *drhd;
1299 * Enable fault control interrupt.
1301 for_each_drhd_unit(drhd) {
1303 struct intel_iommu *iommu = drhd->iommu;
1304 ret = dmar_set_interrupt(iommu);
1307 printk(KERN_ERR "DRHD %Lx: failed to enable fault, "
1308 " interrupt, ret %d\n",
1309 (unsigned long long)drhd->reg_base_addr, ret);
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 dmar_ir_support(void)
1347 struct acpi_table_dmar *dmar;
1348 dmar = (struct acpi_table_dmar *)dmar_tbl;
1349 return dmar->flags & 0x1;