2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
4 * Leo Duran <leo.duran@amd.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/pci.h>
21 #include <linux/acpi.h>
22 #include <linux/gfp.h>
23 #include <linux/list.h>
24 #include <linux/sysdev.h>
25 #include <linux/interrupt.h>
26 #include <linux/msi.h>
27 #include <asm/pci-direct.h>
28 #include <asm/amd_iommu_types.h>
29 #include <asm/amd_iommu.h>
30 #include <asm/iommu.h>
33 * definitions for the ACPI scanning code
35 #define IVRS_HEADER_LENGTH 48
37 #define ACPI_IVHD_TYPE 0x10
38 #define ACPI_IVMD_TYPE_ALL 0x20
39 #define ACPI_IVMD_TYPE 0x21
40 #define ACPI_IVMD_TYPE_RANGE 0x22
42 #define IVHD_DEV_ALL 0x01
43 #define IVHD_DEV_SELECT 0x02
44 #define IVHD_DEV_SELECT_RANGE_START 0x03
45 #define IVHD_DEV_RANGE_END 0x04
46 #define IVHD_DEV_ALIAS 0x42
47 #define IVHD_DEV_ALIAS_RANGE 0x43
48 #define IVHD_DEV_EXT_SELECT 0x46
49 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
51 #define IVHD_FLAG_HT_TUN_EN 0x00
52 #define IVHD_FLAG_PASSPW_EN 0x01
53 #define IVHD_FLAG_RESPASSPW_EN 0x02
54 #define IVHD_FLAG_ISOC_EN 0x03
56 #define IVMD_FLAG_EXCL_RANGE 0x08
57 #define IVMD_FLAG_UNITY_MAP 0x01
59 #define ACPI_DEVFLAG_INITPASS 0x01
60 #define ACPI_DEVFLAG_EXTINT 0x02
61 #define ACPI_DEVFLAG_NMI 0x04
62 #define ACPI_DEVFLAG_SYSMGT1 0x10
63 #define ACPI_DEVFLAG_SYSMGT2 0x20
64 #define ACPI_DEVFLAG_LINT0 0x40
65 #define ACPI_DEVFLAG_LINT1 0x80
66 #define ACPI_DEVFLAG_ATSDIS 0x10000000
69 * ACPI table definitions
71 * These data structures are laid over the table to parse the important values
76 * structure describing one IOMMU in the ACPI table. Typically followed by one
77 * or more ivhd_entrys.
89 } __attribute__((packed));
92 * A device entry describing which devices a specific IOMMU translates and
93 * which requestor ids they use.
100 } __attribute__((packed));
103 * An AMD IOMMU memory definition structure. It defines things like exclusion
104 * ranges for devices and regions that should be unity mapped.
115 } __attribute__((packed));
117 static int __initdata amd_iommu_detected;
119 u16 amd_iommu_last_bdf; /* largest PCI device id we have
121 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
123 unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
124 int amd_iommu_isolate = 1; /* if 1, device isolation is enabled */
125 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
127 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
131 * Pointer to the device table which is shared by all AMD IOMMUs
132 * it is indexed by the PCI device id or the HT unit id and contains
133 * information about the domain the device belongs to as well as the
134 * page table root pointer.
136 struct dev_table_entry *amd_iommu_dev_table;
139 * The alias table is a driver specific data structure which contains the
140 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
141 * More than one device can share the same requestor id.
143 u16 *amd_iommu_alias_table;
146 * The rlookup table is used to find the IOMMU which is responsible
147 * for a specific device. It is also indexed by the PCI device id.
149 struct amd_iommu **amd_iommu_rlookup_table;
152 * The pd table (protection domain table) is used to find the protection domain
153 * data structure a device belongs to. Indexed with the PCI device id too.
155 struct protection_domain **amd_iommu_pd_table;
158 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
159 * to know which ones are already in use.
161 unsigned long *amd_iommu_pd_alloc_bitmap;
163 static u32 dev_table_size; /* size of the device table */
164 static u32 alias_table_size; /* size of the alias table */
165 static u32 rlookup_table_size; /* size if the rlookup table */
167 static inline void update_last_devid(u16 devid)
169 if (devid > amd_iommu_last_bdf)
170 amd_iommu_last_bdf = devid;
173 static inline unsigned long tbl_size(int entry_size)
175 unsigned shift = PAGE_SHIFT +
176 get_order(amd_iommu_last_bdf * entry_size);
181 /****************************************************************************
183 * AMD IOMMU MMIO register space handling functions
185 * These functions are used to program the IOMMU device registers in
186 * MMIO space required for that driver.
188 ****************************************************************************/
191 * This function set the exclusion range in the IOMMU. DMA accesses to the
192 * exclusion range are passed through untranslated
194 static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
196 u64 start = iommu->exclusion_start & PAGE_MASK;
197 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
200 if (!iommu->exclusion_start)
203 entry = start | MMIO_EXCL_ENABLE_MASK;
204 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
205 &entry, sizeof(entry));
208 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
209 &entry, sizeof(entry));
212 /* Programs the physical address of the device table into the IOMMU hardware */
213 static void __init iommu_set_device_table(struct amd_iommu *iommu)
217 BUG_ON(iommu->mmio_base == NULL);
219 entry = virt_to_phys(amd_iommu_dev_table);
220 entry |= (dev_table_size >> 12) - 1;
221 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
222 &entry, sizeof(entry));
225 /* Generic functions to enable/disable certain features of the IOMMU. */
226 static void __init iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
230 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
232 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
235 static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
239 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
241 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
244 /* Function to enable the hardware */
245 void __init iommu_enable(struct amd_iommu *iommu)
247 printk(KERN_INFO "AMD IOMMU: Enabling IOMMU "
248 "at %02x:%02x.%x cap 0x%hx\n",
249 iommu->dev->bus->number,
250 PCI_SLOT(iommu->dev->devfn),
251 PCI_FUNC(iommu->dev->devfn),
254 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
257 /* Function to enable IOMMU event logging and event interrupts */
258 void __init iommu_enable_event_logging(struct amd_iommu *iommu)
260 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
261 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
265 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
266 * the system has one.
268 static u8 * __init iommu_map_mmio_space(u64 address)
272 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu"))
275 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
279 release_mem_region(address, MMIO_REGION_LENGTH);
284 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
286 if (iommu->mmio_base)
287 iounmap(iommu->mmio_base);
288 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
291 /****************************************************************************
293 * The functions below belong to the first pass of AMD IOMMU ACPI table
294 * parsing. In this pass we try to find out the highest device id this
295 * code has to handle. Upon this information the size of the shared data
296 * structures is determined later.
298 ****************************************************************************/
301 * This function calculates the length of a given IVHD entry
303 static inline int ivhd_entry_length(u8 *ivhd)
305 return 0x04 << (*ivhd >> 6);
309 * This function reads the last device id the IOMMU has to handle from the PCI
310 * capability header for this IOMMU
312 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
316 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
317 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
323 * After reading the highest device id from the IOMMU PCI capability header
324 * this function looks if there is a higher device id defined in the ACPI table
326 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
328 u8 *p = (void *)h, *end = (void *)h;
329 struct ivhd_entry *dev;
334 find_last_devid_on_pci(PCI_BUS(h->devid),
340 dev = (struct ivhd_entry *)p;
342 case IVHD_DEV_SELECT:
343 case IVHD_DEV_RANGE_END:
345 case IVHD_DEV_EXT_SELECT:
346 /* all the above subfield types refer to device ids */
347 update_last_devid(dev->devid);
352 p += ivhd_entry_length(p);
361 * Iterate over all IVHD entries in the ACPI table and find the highest device
362 * id which we need to handle. This is the first of three functions which parse
363 * the ACPI table. So we check the checksum here.
365 static int __init find_last_devid_acpi(struct acpi_table_header *table)
368 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
369 struct ivhd_header *h;
372 * Validate checksum here so we don't need to do it when
373 * we actually parse the table
375 for (i = 0; i < table->length; ++i)
378 /* ACPI table corrupt */
381 p += IVRS_HEADER_LENGTH;
383 end += table->length;
385 h = (struct ivhd_header *)p;
388 find_last_devid_from_ivhd(h);
400 /****************************************************************************
402 * The following functions belong the the code path which parses the ACPI table
403 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
404 * data structures, initialize the device/alias/rlookup table and also
405 * basically initialize the hardware.
407 ****************************************************************************/
410 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
411 * write commands to that buffer later and the IOMMU will execute them
414 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
416 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
417 get_order(CMD_BUFFER_SIZE));
423 iommu->cmd_buf_size = CMD_BUFFER_SIZE;
425 entry = (u64)virt_to_phys(cmd_buf);
426 entry |= MMIO_CMD_SIZE_512;
427 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
428 &entry, sizeof(entry));
430 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
435 static void __init free_command_buffer(struct amd_iommu *iommu)
437 free_pages((unsigned long)iommu->cmd_buf,
438 get_order(iommu->cmd_buf_size));
441 /* allocates the memory where the IOMMU will log its events to */
442 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
445 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
446 get_order(EVT_BUFFER_SIZE));
448 if (iommu->evt_buf == NULL)
451 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
452 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
453 &entry, sizeof(entry));
455 iommu->evt_buf_size = EVT_BUFFER_SIZE;
457 return iommu->evt_buf;
460 static void __init free_event_buffer(struct amd_iommu *iommu)
462 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
465 /* sets a specific bit in the device table entry. */
466 static void set_dev_entry_bit(u16 devid, u8 bit)
468 int i = (bit >> 5) & 0x07;
469 int _bit = bit & 0x1f;
471 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
474 /* Writes the specific IOMMU for a device into the rlookup table */
475 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
477 amd_iommu_rlookup_table[devid] = iommu;
481 * This function takes the device specific flags read from the ACPI
482 * table and sets up the device table entry with that information
484 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
485 u16 devid, u32 flags, u32 ext_flags)
487 if (flags & ACPI_DEVFLAG_INITPASS)
488 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
489 if (flags & ACPI_DEVFLAG_EXTINT)
490 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
491 if (flags & ACPI_DEVFLAG_NMI)
492 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
493 if (flags & ACPI_DEVFLAG_SYSMGT1)
494 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
495 if (flags & ACPI_DEVFLAG_SYSMGT2)
496 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
497 if (flags & ACPI_DEVFLAG_LINT0)
498 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
499 if (flags & ACPI_DEVFLAG_LINT1)
500 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
502 set_iommu_for_device(iommu, devid);
506 * Reads the device exclusion range from ACPI and initialize IOMMU with
509 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
511 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
513 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
518 * We only can configure exclusion ranges per IOMMU, not
519 * per device. But we can enable the exclusion range per
520 * device. This is done here
522 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
523 iommu->exclusion_start = m->range_start;
524 iommu->exclusion_length = m->range_length;
529 * This function reads some important data from the IOMMU PCI space and
530 * initializes the driver data structure with it. It reads the hardware
531 * capabilities and the first/last device entries
533 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
535 int cap_ptr = iommu->cap_ptr;
538 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
540 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
542 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
545 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
547 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
549 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
553 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
554 * initializes the hardware and our data structures with it.
556 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
557 struct ivhd_header *h)
560 u8 *end = p, flags = 0;
561 u16 dev_i, devid = 0, devid_start = 0, devid_to = 0;
564 struct ivhd_entry *e;
567 * First set the recommended feature enable bits from ACPI
568 * into the IOMMU control registers
570 h->flags & IVHD_FLAG_HT_TUN_EN ?
571 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
572 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
574 h->flags & IVHD_FLAG_PASSPW_EN ?
575 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
576 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
578 h->flags & IVHD_FLAG_RESPASSPW_EN ?
579 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
580 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
582 h->flags & IVHD_FLAG_ISOC_EN ?
583 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
584 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
587 * make IOMMU memory accesses cache coherent
589 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
592 * Done. Now parse the device entries
594 p += sizeof(struct ivhd_header);
598 e = (struct ivhd_entry *)p;
601 for (dev_i = iommu->first_device;
602 dev_i <= iommu->last_device; ++dev_i)
603 set_dev_entry_from_acpi(iommu, dev_i,
606 case IVHD_DEV_SELECT:
608 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
610 case IVHD_DEV_SELECT_RANGE_START:
611 devid_start = e->devid;
618 devid_to = e->ext >> 8;
619 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
620 amd_iommu_alias_table[devid] = devid_to;
622 case IVHD_DEV_ALIAS_RANGE:
623 devid_start = e->devid;
625 devid_to = e->ext >> 8;
629 case IVHD_DEV_EXT_SELECT:
631 set_dev_entry_from_acpi(iommu, devid, e->flags,
634 case IVHD_DEV_EXT_SELECT_RANGE:
635 devid_start = e->devid;
640 case IVHD_DEV_RANGE_END:
642 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
644 amd_iommu_alias_table[dev_i] = devid_to;
645 set_dev_entry_from_acpi(iommu,
646 amd_iommu_alias_table[dev_i],
654 p += ivhd_entry_length(p);
658 /* Initializes the device->iommu mapping for the driver */
659 static int __init init_iommu_devices(struct amd_iommu *iommu)
663 for (i = iommu->first_device; i <= iommu->last_device; ++i)
664 set_iommu_for_device(iommu, i);
669 static void __init free_iommu_one(struct amd_iommu *iommu)
671 free_command_buffer(iommu);
672 free_event_buffer(iommu);
673 iommu_unmap_mmio_space(iommu);
676 static void __init free_iommu_all(void)
678 struct amd_iommu *iommu, *next;
680 list_for_each_entry_safe(iommu, next, &amd_iommu_list, list) {
681 list_del(&iommu->list);
682 free_iommu_one(iommu);
688 * This function clues the initialization function for one IOMMU
689 * together and also allocates the command buffer and programs the
690 * hardware. It does NOT enable the IOMMU. This is done afterwards.
692 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
694 spin_lock_init(&iommu->lock);
695 list_add_tail(&iommu->list, &amd_iommu_list);
698 * Copy data from ACPI table entry to the iommu struct
700 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
704 iommu->cap_ptr = h->cap_ptr;
705 iommu->pci_seg = h->pci_seg;
706 iommu->mmio_phys = h->mmio_phys;
707 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
708 if (!iommu->mmio_base)
711 iommu_set_device_table(iommu);
712 iommu->cmd_buf = alloc_command_buffer(iommu);
716 iommu->evt_buf = alloc_event_buffer(iommu);
720 iommu->int_enabled = false;
722 init_iommu_from_pci(iommu);
723 init_iommu_from_acpi(iommu, h);
724 init_iommu_devices(iommu);
726 return pci_enable_device(iommu->dev);
730 * Iterates over all IOMMU entries in the ACPI table, allocates the
731 * IOMMU structure and initializes it with init_iommu_one()
733 static int __init init_iommu_all(struct acpi_table_header *table)
735 u8 *p = (u8 *)table, *end = (u8 *)table;
736 struct ivhd_header *h;
737 struct amd_iommu *iommu;
740 end += table->length;
741 p += IVRS_HEADER_LENGTH;
744 h = (struct ivhd_header *)p;
747 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
750 ret = init_iommu_one(iommu, h);
765 /****************************************************************************
767 * The following functions initialize the MSI interrupts for all IOMMUs
768 * in the system. Its a bit challenging because there could be multiple
769 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
772 ****************************************************************************/
774 static int __init iommu_setup_msix(struct amd_iommu *iommu)
776 struct amd_iommu *curr;
777 struct msix_entry entries[32]; /* only 32 supported by AMD IOMMU */
780 list_for_each_entry(curr, &amd_iommu_list, list) {
781 if (curr->dev == iommu->dev) {
782 entries[nvec].entry = curr->evt_msi_num;
783 entries[nvec].vector = 0;
784 curr->int_enabled = true;
789 if (pci_enable_msix(iommu->dev, entries, nvec)) {
790 pci_disable_msix(iommu->dev);
794 for (i = 0; i < nvec; ++i) {
795 int r = request_irq(entries->vector, amd_iommu_int_handler,
806 for (i -= 1; i >= 0; --i)
807 free_irq(entries->vector, NULL);
809 pci_disable_msix(iommu->dev);
814 static int __init iommu_setup_msi(struct amd_iommu *iommu)
817 struct amd_iommu *curr;
819 list_for_each_entry(curr, &amd_iommu_list, list) {
820 if (curr->dev == iommu->dev)
821 curr->int_enabled = true;
825 if (pci_enable_msi(iommu->dev))
828 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
834 pci_disable_msi(iommu->dev);
841 static int __init iommu_init_msi(struct amd_iommu *iommu)
843 if (iommu->int_enabled)
846 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSIX))
847 return iommu_setup_msix(iommu);
848 else if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
849 return iommu_setup_msi(iommu);
854 /****************************************************************************
856 * The next functions belong to the third pass of parsing the ACPI
857 * table. In this last pass the memory mapping requirements are
858 * gathered (like exclusion and unity mapping reanges).
860 ****************************************************************************/
862 static void __init free_unity_maps(void)
864 struct unity_map_entry *entry, *next;
866 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
867 list_del(&entry->list);
872 /* called when we find an exclusion range definition in ACPI */
873 static int __init init_exclusion_range(struct ivmd_header *m)
879 set_device_exclusion_range(m->devid, m);
881 case ACPI_IVMD_TYPE_ALL:
882 for (i = 0; i <= amd_iommu_last_bdf; ++i)
883 set_device_exclusion_range(i, m);
885 case ACPI_IVMD_TYPE_RANGE:
886 for (i = m->devid; i <= m->aux; ++i)
887 set_device_exclusion_range(i, m);
896 /* called for unity map ACPI definition */
897 static int __init init_unity_map_range(struct ivmd_header *m)
899 struct unity_map_entry *e = 0;
901 e = kzalloc(sizeof(*e), GFP_KERNEL);
908 e->devid_start = e->devid_end = m->devid;
910 case ACPI_IVMD_TYPE_ALL:
912 e->devid_end = amd_iommu_last_bdf;
914 case ACPI_IVMD_TYPE_RANGE:
915 e->devid_start = m->devid;
916 e->devid_end = m->aux;
919 e->address_start = PAGE_ALIGN(m->range_start);
920 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
921 e->prot = m->flags >> 1;
923 list_add_tail(&e->list, &amd_iommu_unity_map);
928 /* iterates over all memory definitions we find in the ACPI table */
929 static int __init init_memory_definitions(struct acpi_table_header *table)
931 u8 *p = (u8 *)table, *end = (u8 *)table;
932 struct ivmd_header *m;
934 end += table->length;
935 p += IVRS_HEADER_LENGTH;
938 m = (struct ivmd_header *)p;
939 if (m->flags & IVMD_FLAG_EXCL_RANGE)
940 init_exclusion_range(m);
941 else if (m->flags & IVMD_FLAG_UNITY_MAP)
942 init_unity_map_range(m);
951 * Init the device table to not allow DMA access for devices and
952 * suppress all page faults
954 static void init_device_table(void)
958 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
959 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
960 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
965 * This function finally enables all IOMMUs found in the system after
966 * they have been initialized
968 static void __init enable_iommus(void)
970 struct amd_iommu *iommu;
972 list_for_each_entry(iommu, &amd_iommu_list, list) {
973 iommu_set_exclusion_range(iommu);
974 iommu_init_msi(iommu);
975 iommu_enable_event_logging(iommu);
981 * Suspend/Resume support
982 * disable suspend until real resume implemented
985 static int amd_iommu_resume(struct sys_device *dev)
990 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
995 static struct sysdev_class amd_iommu_sysdev_class = {
997 .suspend = amd_iommu_suspend,
998 .resume = amd_iommu_resume,
1001 static struct sys_device device_amd_iommu = {
1003 .cls = &amd_iommu_sysdev_class,
1007 * This is the core init function for AMD IOMMU hardware in the system.
1008 * This function is called from the generic x86 DMA layer initialization
1011 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1014 * 1 pass) Find the highest PCI device id the driver has to handle.
1015 * Upon this information the size of the data structures is
1016 * determined that needs to be allocated.
1018 * 2 pass) Initialize the data structures just allocated with the
1019 * information in the ACPI table about available AMD IOMMUs
1020 * in the system. It also maps the PCI devices in the
1021 * system to specific IOMMUs
1023 * 3 pass) After the basic data structures are allocated and
1024 * initialized we update them with information about memory
1025 * remapping requirements parsed out of the ACPI table in
1028 * After that the hardware is initialized and ready to go. In the last
1029 * step we do some Linux specific things like registering the driver in
1030 * the dma_ops interface and initializing the suspend/resume support
1031 * functions. Finally it prints some information about AMD IOMMUs and
1032 * the driver state and enables the hardware.
1034 int __init amd_iommu_init(void)
1040 printk(KERN_INFO "AMD IOMMU disabled by kernel command line\n");
1044 if (!amd_iommu_detected)
1048 * First parse ACPI tables to find the largest Bus/Dev/Func
1049 * we need to handle. Upon this information the shared data
1050 * structures for the IOMMUs in the system will be allocated
1052 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1055 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1056 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1057 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1061 /* Device table - directly used by all IOMMUs */
1062 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1063 get_order(dev_table_size));
1064 if (amd_iommu_dev_table == NULL)
1068 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1069 * IOMMU see for that device
1071 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1072 get_order(alias_table_size));
1073 if (amd_iommu_alias_table == NULL)
1076 /* IOMMU rlookup table - find the IOMMU for a specific device */
1077 amd_iommu_rlookup_table = (void *)__get_free_pages(GFP_KERNEL,
1078 get_order(rlookup_table_size));
1079 if (amd_iommu_rlookup_table == NULL)
1083 * Protection Domain table - maps devices to protection domains
1084 * This table has the same size as the rlookup_table
1086 amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1087 get_order(rlookup_table_size));
1088 if (amd_iommu_pd_table == NULL)
1091 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1092 GFP_KERNEL | __GFP_ZERO,
1093 get_order(MAX_DOMAIN_ID/8));
1094 if (amd_iommu_pd_alloc_bitmap == NULL)
1097 /* init the device table */
1098 init_device_table();
1101 * let all alias entries point to itself
1103 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1104 amd_iommu_alias_table[i] = i;
1107 * never allocate domain 0 because its used as the non-allocated and
1108 * error value placeholder
1110 amd_iommu_pd_alloc_bitmap[0] = 1;
1113 * now the data structures are allocated and basically initialized
1114 * start the real acpi table scan
1117 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1120 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1123 ret = sysdev_class_register(&amd_iommu_sysdev_class);
1127 ret = sysdev_register(&device_amd_iommu);
1131 ret = amd_iommu_init_dma_ops();
1137 printk(KERN_INFO "AMD IOMMU: aperture size is %d MB\n",
1138 (1 << (amd_iommu_aperture_order-20)));
1140 printk(KERN_INFO "AMD IOMMU: device isolation ");
1141 if (amd_iommu_isolate)
1142 printk("enabled\n");
1144 printk("disabled\n");
1146 if (amd_iommu_unmap_flush)
1147 printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n");
1149 printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n");
1155 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1156 get_order(MAX_DOMAIN_ID/8));
1158 free_pages((unsigned long)amd_iommu_pd_table,
1159 get_order(rlookup_table_size));
1161 free_pages((unsigned long)amd_iommu_rlookup_table,
1162 get_order(rlookup_table_size));
1164 free_pages((unsigned long)amd_iommu_alias_table,
1165 get_order(alias_table_size));
1167 free_pages((unsigned long)amd_iommu_dev_table,
1168 get_order(dev_table_size));
1177 /****************************************************************************
1179 * Early detect code. This code runs at IOMMU detection time in the DMA
1180 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1183 ****************************************************************************/
1184 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1189 void __init amd_iommu_detect(void)
1191 if (swiotlb || no_iommu || (iommu_detected && !gart_iommu_aperture))
1194 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1196 amd_iommu_detected = 1;
1197 #ifdef CONFIG_GART_IOMMU
1198 gart_iommu_aperture_disabled = 1;
1199 gart_iommu_aperture = 0;
1204 /****************************************************************************
1206 * Parsing functions for the AMD IOMMU specific kernel command line
1209 ****************************************************************************/
1211 static int __init parse_amd_iommu_options(char *str)
1213 for (; *str; ++str) {
1214 if (strncmp(str, "isolate", 7) == 0)
1215 amd_iommu_isolate = 1;
1216 if (strncmp(str, "share", 5) == 0)
1217 amd_iommu_isolate = 0;
1218 if (strncmp(str, "fullflush", 9) == 0)
1219 amd_iommu_unmap_flush = true;
1225 static int __init parse_amd_iommu_size_options(char *str)
1227 unsigned order = PAGE_SHIFT + get_order(memparse(str, &str));
1229 if ((order > 24) && (order < 31))
1230 amd_iommu_aperture_order = order;
1235 __setup("amd_iommu=", parse_amd_iommu_options);
1236 __setup("amd_iommu_size=", parse_amd_iommu_size_options);