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>
36 #define PREFIX "DMAR:"
38 /* No locks are needed as DMA remapping hardware unit
39 * list is constructed at boot time and hotplug of
40 * these units are not supported by the architecture.
42 LIST_HEAD(dmar_drhd_units);
44 static struct acpi_table_header * __initdata dmar_tbl;
46 static void __init dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
49 * add INCLUDE_ALL at the tail, so scan the list will find it at
52 if (drhd->include_all)
53 list_add_tail(&drhd->list, &dmar_drhd_units);
55 list_add(&drhd->list, &dmar_drhd_units);
58 static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope,
59 struct pci_dev **dev, u16 segment)
62 struct pci_dev *pdev = NULL;
63 struct acpi_dmar_pci_path *path;
66 bus = pci_find_bus(segment, scope->bus);
67 path = (struct acpi_dmar_pci_path *)(scope + 1);
68 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
69 / sizeof(struct acpi_dmar_pci_path);
75 * Some BIOSes list non-exist devices in DMAR table, just
80 PREFIX "Device scope bus [%d] not found\n",
84 pdev = pci_get_slot(bus, PCI_DEVFN(path->dev, path->fn));
86 printk(KERN_WARNING PREFIX
87 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
88 segment, bus->number, path->dev, path->fn);
93 bus = pdev->subordinate;
96 printk(KERN_WARNING PREFIX
97 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
98 segment, scope->bus, path->dev, path->fn);
102 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT && \
103 pdev->subordinate) || (scope->entry_type == \
104 ACPI_DMAR_SCOPE_TYPE_BRIDGE && !pdev->subordinate)) {
106 printk(KERN_WARNING PREFIX
107 "Device scope type does not match for %s\n",
115 static int __init dmar_parse_dev_scope(void *start, void *end, int *cnt,
116 struct pci_dev ***devices, u16 segment)
118 struct acpi_dmar_device_scope *scope;
124 while (start < end) {
126 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
127 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
130 printk(KERN_WARNING PREFIX
131 "Unsupported device scope\n");
132 start += scope->length;
137 *devices = kcalloc(*cnt, sizeof(struct pci_dev *), GFP_KERNEL);
143 while (start < end) {
145 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
146 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) {
147 ret = dmar_parse_one_dev_scope(scope,
148 &(*devices)[index], segment);
155 start += scope->length;
162 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
163 * structure which uniquely represent one DMA remapping hardware unit
164 * present in the platform
167 dmar_parse_one_drhd(struct acpi_dmar_header *header)
169 struct acpi_dmar_hardware_unit *drhd;
170 struct dmar_drhd_unit *dmaru;
173 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
178 drhd = (struct acpi_dmar_hardware_unit *)header;
179 dmaru->reg_base_addr = drhd->address;
180 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
182 ret = alloc_iommu(dmaru);
187 dmar_register_drhd_unit(dmaru);
191 static int __init dmar_parse_dev(struct dmar_drhd_unit *dmaru)
193 struct acpi_dmar_hardware_unit *drhd;
196 drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
198 if (dmaru->include_all)
201 ret = dmar_parse_dev_scope((void *)(drhd + 1),
202 ((void *)drhd) + drhd->header.length,
203 &dmaru->devices_cnt, &dmaru->devices,
206 list_del(&dmaru->list);
213 LIST_HEAD(dmar_rmrr_units);
215 static void __init dmar_register_rmrr_unit(struct dmar_rmrr_unit *rmrr)
217 list_add(&rmrr->list, &dmar_rmrr_units);
222 dmar_parse_one_rmrr(struct acpi_dmar_header *header)
224 struct acpi_dmar_reserved_memory *rmrr;
225 struct dmar_rmrr_unit *rmrru;
227 rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL);
232 rmrr = (struct acpi_dmar_reserved_memory *)header;
233 rmrru->base_address = rmrr->base_address;
234 rmrru->end_address = rmrr->end_address;
236 dmar_register_rmrr_unit(rmrru);
241 rmrr_parse_dev(struct dmar_rmrr_unit *rmrru)
243 struct acpi_dmar_reserved_memory *rmrr;
246 rmrr = (struct acpi_dmar_reserved_memory *) rmrru->hdr;
247 ret = dmar_parse_dev_scope((void *)(rmrr + 1),
248 ((void *)rmrr) + rmrr->header.length,
249 &rmrru->devices_cnt, &rmrru->devices, rmrr->segment);
251 if (ret || (rmrru->devices_cnt == 0)) {
252 list_del(&rmrru->list);
260 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
262 struct acpi_dmar_hardware_unit *drhd;
263 struct acpi_dmar_reserved_memory *rmrr;
265 switch (header->type) {
266 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
267 drhd = (struct acpi_dmar_hardware_unit *)header;
268 printk (KERN_INFO PREFIX
269 "DRHD (flags: 0x%08x)base: 0x%016Lx\n",
270 drhd->flags, (unsigned long long)drhd->address);
272 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
273 rmrr = (struct acpi_dmar_reserved_memory *)header;
275 printk (KERN_INFO PREFIX
276 "RMRR base: 0x%016Lx end: 0x%016Lx\n",
277 (unsigned long long)rmrr->base_address,
278 (unsigned long long)rmrr->end_address);
284 * dmar_table_detect - checks to see if the platform supports DMAR devices
286 static int __init dmar_table_detect(void)
288 acpi_status status = AE_OK;
290 /* if we could find DMAR table, then there are DMAR devices */
291 status = acpi_get_table(ACPI_SIG_DMAR, 0,
292 (struct acpi_table_header **)&dmar_tbl);
294 if (ACPI_SUCCESS(status) && !dmar_tbl) {
295 printk (KERN_WARNING PREFIX "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
318 dmar = (struct acpi_table_dmar *)dmar_tbl;
322 if (dmar->width < PAGE_SHIFT - 1) {
323 printk(KERN_WARNING PREFIX "Invalid DMAR haw\n");
327 printk (KERN_INFO PREFIX "Host address width %d\n",
330 entry_header = (struct acpi_dmar_header *)(dmar + 1);
331 while (((unsigned long)entry_header) <
332 (((unsigned long)dmar) + dmar_tbl->length)) {
333 /* Avoid looping forever on bad ACPI tables */
334 if (entry_header->length == 0) {
335 printk(KERN_WARNING PREFIX
336 "Invalid 0-length structure\n");
341 dmar_table_print_dmar_entry(entry_header);
343 switch (entry_header->type) {
344 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
345 ret = dmar_parse_one_drhd(entry_header);
347 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
349 ret = dmar_parse_one_rmrr(entry_header);
353 printk(KERN_WARNING PREFIX
354 "Unknown DMAR structure type\n");
355 ret = 0; /* for forward compatibility */
361 entry_header = ((void *)entry_header + entry_header->length);
366 int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
372 for (index = 0; index < cnt; index++)
373 if (dev == devices[index])
376 /* Check our parent */
377 dev = dev->bus->self;
383 struct dmar_drhd_unit *
384 dmar_find_matched_drhd_unit(struct pci_dev *dev)
386 struct dmar_drhd_unit *dmaru = NULL;
387 struct acpi_dmar_hardware_unit *drhd;
389 list_for_each_entry(dmaru, &dmar_drhd_units, list) {
390 drhd = container_of(dmaru->hdr,
391 struct acpi_dmar_hardware_unit,
394 if (dmaru->include_all &&
395 drhd->segment == pci_domain_nr(dev->bus))
398 if (dmar_pci_device_match(dmaru->devices,
399 dmaru->devices_cnt, dev))
406 int __init dmar_dev_scope_init(void)
408 struct dmar_drhd_unit *drhd, *drhd_n;
411 list_for_each_entry_safe(drhd, drhd_n, &dmar_drhd_units, list) {
412 ret = dmar_parse_dev(drhd);
419 struct dmar_rmrr_unit *rmrr, *rmrr_n;
420 list_for_each_entry_safe(rmrr, rmrr_n, &dmar_rmrr_units, list) {
421 ret = rmrr_parse_dev(rmrr);
432 int __init dmar_table_init(void)
434 static int dmar_table_initialized;
437 if (dmar_table_initialized)
440 dmar_table_initialized = 1;
442 ret = parse_dmar_table();
445 printk(KERN_INFO PREFIX "parse DMAR table failure.\n");
449 if (list_empty(&dmar_drhd_units)) {
450 printk(KERN_INFO PREFIX "No DMAR devices found\n");
455 if (list_empty(&dmar_rmrr_units))
456 printk(KERN_INFO PREFIX "No RMRR found\n");
459 #ifdef CONFIG_INTR_REMAP
460 parse_ioapics_under_ir();
465 void __init detect_intel_iommu(void)
469 ret = dmar_table_detect();
472 #ifdef CONFIG_INTR_REMAP
473 struct acpi_table_dmar *dmar;
475 * for now we will disable dma-remapping when interrupt
476 * remapping is enabled.
477 * When support for queued invalidation for IOTLB invalidation
478 * is added, we will not need this any more.
480 dmar = (struct acpi_table_dmar *) dmar_tbl;
481 if (ret && cpu_has_x2apic && dmar->flags & 0x1)
483 "Queued invalidation will be enabled to support "
484 "x2apic and Intr-remapping.\n");
487 if (ret && !no_iommu && !iommu_detected && !swiotlb &&
496 int alloc_iommu(struct dmar_drhd_unit *drhd)
498 struct intel_iommu *iommu;
501 static int iommu_allocated = 0;
504 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
508 iommu->seq_id = iommu_allocated++;
510 iommu->reg = ioremap(drhd->reg_base_addr, VTD_PAGE_SIZE);
512 printk(KERN_ERR "IOMMU: can't map the region\n");
515 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
516 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
519 agaw = iommu_calculate_agaw(iommu);
522 "Cannot get a valid agaw for iommu (seq_id = %d)\n",
529 /* the registers might be more than one page */
530 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
531 cap_max_fault_reg_offset(iommu->cap));
532 map_size = VTD_PAGE_ALIGN(map_size);
533 if (map_size > VTD_PAGE_SIZE) {
535 iommu->reg = ioremap(drhd->reg_base_addr, map_size);
537 printk(KERN_ERR "IOMMU: can't map the region\n");
542 ver = readl(iommu->reg + DMAR_VER_REG);
543 pr_debug("IOMMU %llx: ver %d:%d cap %llx ecap %llx\n",
544 (unsigned long long)drhd->reg_base_addr,
545 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
546 (unsigned long long)iommu->cap,
547 (unsigned long long)iommu->ecap);
549 spin_lock_init(&iommu->register_lock);
558 void free_iommu(struct intel_iommu *iommu)
564 free_dmar_iommu(iommu);
573 * Reclaim all the submitted descriptors which have completed its work.
575 static inline void reclaim_free_desc(struct q_inval *qi)
577 while (qi->desc_status[qi->free_tail] == QI_DONE) {
578 qi->desc_status[qi->free_tail] = QI_FREE;
579 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
584 static int qi_check_fault(struct intel_iommu *iommu, int index)
588 struct q_inval *qi = iommu->qi;
589 int wait_index = (index + 1) % QI_LENGTH;
591 fault = readl(iommu->reg + DMAR_FSTS_REG);
594 * If IQE happens, the head points to the descriptor associated
595 * with the error. No new descriptors are fetched until the IQE
598 if (fault & DMA_FSTS_IQE) {
599 head = readl(iommu->reg + DMAR_IQH_REG);
600 if ((head >> 4) == index) {
601 memcpy(&qi->desc[index], &qi->desc[wait_index],
602 sizeof(struct qi_desc));
603 __iommu_flush_cache(iommu, &qi->desc[index],
604 sizeof(struct qi_desc));
605 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
614 * Submit the queued invalidation descriptor to the remapping
615 * hardware unit and wait for its completion.
617 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
620 struct q_inval *qi = iommu->qi;
621 struct qi_desc *hw, wait_desc;
622 int wait_index, index;
630 spin_lock_irqsave(&qi->q_lock, flags);
631 while (qi->free_cnt < 3) {
632 spin_unlock_irqrestore(&qi->q_lock, flags);
634 spin_lock_irqsave(&qi->q_lock, flags);
637 index = qi->free_head;
638 wait_index = (index + 1) % QI_LENGTH;
640 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
644 wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
645 QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
646 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
648 hw[wait_index] = wait_desc;
650 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
651 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
653 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
657 * update the HW tail register indicating the presence of
660 writel(qi->free_head << 4, iommu->reg + DMAR_IQT_REG);
662 while (qi->desc_status[wait_index] != QI_DONE) {
664 * We will leave the interrupts disabled, to prevent interrupt
665 * context to queue another cmd while a cmd is already submitted
666 * and waiting for completion on this cpu. This is to avoid
667 * a deadlock where the interrupt context can wait indefinitely
668 * for free slots in the queue.
670 rc = qi_check_fault(iommu, index);
674 spin_unlock(&qi->q_lock);
676 spin_lock(&qi->q_lock);
679 qi->desc_status[index] = qi->desc_status[wait_index] = QI_DONE;
681 reclaim_free_desc(qi);
682 spin_unlock_irqrestore(&qi->q_lock, flags);
688 * Flush the global interrupt entry cache.
690 void qi_global_iec(struct intel_iommu *iommu)
694 desc.low = QI_IEC_TYPE;
697 /* should never fail */
698 qi_submit_sync(&desc, iommu);
701 int qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
702 u64 type, int non_present_entry_flush)
706 if (non_present_entry_flush) {
707 if (!cap_caching_mode(iommu->cap))
713 desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
714 | QI_CC_GRAN(type) | QI_CC_TYPE;
717 return qi_submit_sync(&desc, iommu);
720 int qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
721 unsigned int size_order, u64 type,
722 int non_present_entry_flush)
729 if (non_present_entry_flush) {
730 if (!cap_caching_mode(iommu->cap))
736 if (cap_write_drain(iommu->cap))
739 if (cap_read_drain(iommu->cap))
742 desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
743 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
744 desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
745 | QI_IOTLB_AM(size_order);
747 return qi_submit_sync(&desc, iommu);
751 * Enable Queued Invalidation interface. This is a must to support
752 * interrupt-remapping. Also used by DMA-remapping, which replaces
753 * register based IOTLB invalidation.
755 int dmar_enable_qi(struct intel_iommu *iommu)
761 if (!ecap_qis(iommu->ecap))
765 * queued invalidation is already setup and enabled.
770 iommu->qi = kmalloc(sizeof(*qi), GFP_KERNEL);
776 qi->desc = (void *)(get_zeroed_page(GFP_KERNEL));
783 qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_KERNEL);
784 if (!qi->desc_status) {
785 free_page((unsigned long) qi->desc);
791 qi->free_head = qi->free_tail = 0;
792 qi->free_cnt = QI_LENGTH;
794 spin_lock_init(&qi->q_lock);
796 spin_lock_irqsave(&iommu->register_lock, flags);
797 /* write zero to the tail reg */
798 writel(0, iommu->reg + DMAR_IQT_REG);
800 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
802 cmd = iommu->gcmd | DMA_GCMD_QIE;
803 iommu->gcmd |= DMA_GCMD_QIE;
804 writel(cmd, iommu->reg + DMAR_GCMD_REG);
806 /* Make sure hardware complete it */
807 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
808 spin_unlock_irqrestore(&iommu->register_lock, flags);