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>
34 * definitions for the ACPI scanning code
36 #define IVRS_HEADER_LENGTH 48
38 #define ACPI_IVHD_TYPE 0x10
39 #define ACPI_IVMD_TYPE_ALL 0x20
40 #define ACPI_IVMD_TYPE 0x21
41 #define ACPI_IVMD_TYPE_RANGE 0x22
43 #define IVHD_DEV_ALL 0x01
44 #define IVHD_DEV_SELECT 0x02
45 #define IVHD_DEV_SELECT_RANGE_START 0x03
46 #define IVHD_DEV_RANGE_END 0x04
47 #define IVHD_DEV_ALIAS 0x42
48 #define IVHD_DEV_ALIAS_RANGE 0x43
49 #define IVHD_DEV_EXT_SELECT 0x46
50 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
52 #define IVHD_FLAG_HT_TUN_EN 0x00
53 #define IVHD_FLAG_PASSPW_EN 0x01
54 #define IVHD_FLAG_RESPASSPW_EN 0x02
55 #define IVHD_FLAG_ISOC_EN 0x03
57 #define IVMD_FLAG_EXCL_RANGE 0x08
58 #define IVMD_FLAG_UNITY_MAP 0x01
60 #define ACPI_DEVFLAG_INITPASS 0x01
61 #define ACPI_DEVFLAG_EXTINT 0x02
62 #define ACPI_DEVFLAG_NMI 0x04
63 #define ACPI_DEVFLAG_SYSMGT1 0x10
64 #define ACPI_DEVFLAG_SYSMGT2 0x20
65 #define ACPI_DEVFLAG_LINT0 0x40
66 #define ACPI_DEVFLAG_LINT1 0x80
67 #define ACPI_DEVFLAG_ATSDIS 0x10000000
70 * ACPI table definitions
72 * These data structures are laid over the table to parse the important values
77 * structure describing one IOMMU in the ACPI table. Typically followed by one
78 * or more ivhd_entrys.
90 } __attribute__((packed));
93 * A device entry describing which devices a specific IOMMU translates and
94 * which requestor ids they use.
101 } __attribute__((packed));
104 * An AMD IOMMU memory definition structure. It defines things like exclusion
105 * ranges for devices and regions that should be unity mapped.
116 } __attribute__((packed));
118 static int __initdata amd_iommu_detected;
120 u16 amd_iommu_last_bdf; /* largest PCI device id we have
122 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
124 unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
125 int amd_iommu_isolate = 1; /* if 1, device isolation is enabled */
126 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
128 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
132 * Pointer to the device table which is shared by all AMD IOMMUs
133 * it is indexed by the PCI device id or the HT unit id and contains
134 * information about the domain the device belongs to as well as the
135 * page table root pointer.
137 struct dev_table_entry *amd_iommu_dev_table;
140 * The alias table is a driver specific data structure which contains the
141 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
142 * More than one device can share the same requestor id.
144 u16 *amd_iommu_alias_table;
147 * The rlookup table is used to find the IOMMU which is responsible
148 * for a specific device. It is also indexed by the PCI device id.
150 struct amd_iommu **amd_iommu_rlookup_table;
153 * The pd table (protection domain table) is used to find the protection domain
154 * data structure a device belongs to. Indexed with the PCI device id too.
156 struct protection_domain **amd_iommu_pd_table;
159 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
160 * to know which ones are already in use.
162 unsigned long *amd_iommu_pd_alloc_bitmap;
164 static u32 dev_table_size; /* size of the device table */
165 static u32 alias_table_size; /* size of the alias table */
166 static u32 rlookup_table_size; /* size if the rlookup table */
168 static inline void update_last_devid(u16 devid)
170 if (devid > amd_iommu_last_bdf)
171 amd_iommu_last_bdf = devid;
174 static inline unsigned long tbl_size(int entry_size)
176 unsigned shift = PAGE_SHIFT +
177 get_order(amd_iommu_last_bdf * entry_size);
182 /****************************************************************************
184 * AMD IOMMU MMIO register space handling functions
186 * These functions are used to program the IOMMU device registers in
187 * MMIO space required for that driver.
189 ****************************************************************************/
192 * This function set the exclusion range in the IOMMU. DMA accesses to the
193 * exclusion range are passed through untranslated
195 static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
197 u64 start = iommu->exclusion_start & PAGE_MASK;
198 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
201 if (!iommu->exclusion_start)
204 entry = start | MMIO_EXCL_ENABLE_MASK;
205 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
206 &entry, sizeof(entry));
209 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
210 &entry, sizeof(entry));
213 /* Programs the physical address of the device table into the IOMMU hardware */
214 static void __init iommu_set_device_table(struct amd_iommu *iommu)
218 BUG_ON(iommu->mmio_base == NULL);
220 entry = virt_to_phys(amd_iommu_dev_table);
221 entry |= (dev_table_size >> 12) - 1;
222 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
223 &entry, sizeof(entry));
226 /* Generic functions to enable/disable certain features of the IOMMU. */
227 static void __init iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
231 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
233 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
236 static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
240 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
242 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
245 /* Function to enable the hardware */
246 void __init iommu_enable(struct amd_iommu *iommu)
248 printk(KERN_INFO "AMD IOMMU: Enabling IOMMU "
249 "at %02x:%02x.%x cap 0x%hx\n",
250 iommu->dev->bus->number,
251 PCI_SLOT(iommu->dev->devfn),
252 PCI_FUNC(iommu->dev->devfn),
255 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
258 /* Function to enable IOMMU event logging and event interrupts */
259 void __init iommu_enable_event_logging(struct amd_iommu *iommu)
261 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
262 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
266 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
267 * the system has one.
269 static u8 * __init iommu_map_mmio_space(u64 address)
273 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu"))
276 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
280 release_mem_region(address, MMIO_REGION_LENGTH);
285 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
287 if (iommu->mmio_base)
288 iounmap(iommu->mmio_base);
289 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
292 /****************************************************************************
294 * The functions below belong to the first pass of AMD IOMMU ACPI table
295 * parsing. In this pass we try to find out the highest device id this
296 * code has to handle. Upon this information the size of the shared data
297 * structures is determined later.
299 ****************************************************************************/
302 * This function calculates the length of a given IVHD entry
304 static inline int ivhd_entry_length(u8 *ivhd)
306 return 0x04 << (*ivhd >> 6);
310 * This function reads the last device id the IOMMU has to handle from the PCI
311 * capability header for this IOMMU
313 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
317 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
318 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
324 * After reading the highest device id from the IOMMU PCI capability header
325 * this function looks if there is a higher device id defined in the ACPI table
327 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
329 u8 *p = (void *)h, *end = (void *)h;
330 struct ivhd_entry *dev;
335 find_last_devid_on_pci(PCI_BUS(h->devid),
341 dev = (struct ivhd_entry *)p;
343 case IVHD_DEV_SELECT:
344 case IVHD_DEV_RANGE_END:
346 case IVHD_DEV_EXT_SELECT:
347 /* all the above subfield types refer to device ids */
348 update_last_devid(dev->devid);
353 p += ivhd_entry_length(p);
362 * Iterate over all IVHD entries in the ACPI table and find the highest device
363 * id which we need to handle. This is the first of three functions which parse
364 * the ACPI table. So we check the checksum here.
366 static int __init find_last_devid_acpi(struct acpi_table_header *table)
369 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
370 struct ivhd_header *h;
373 * Validate checksum here so we don't need to do it when
374 * we actually parse the table
376 for (i = 0; i < table->length; ++i)
379 /* ACPI table corrupt */
382 p += IVRS_HEADER_LENGTH;
384 end += table->length;
386 h = (struct ivhd_header *)p;
389 find_last_devid_from_ivhd(h);
401 /****************************************************************************
403 * The following functions belong the the code path which parses the ACPI table
404 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
405 * data structures, initialize the device/alias/rlookup table and also
406 * basically initialize the hardware.
408 ****************************************************************************/
411 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
412 * write commands to that buffer later and the IOMMU will execute them
415 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
417 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
418 get_order(CMD_BUFFER_SIZE));
424 iommu->cmd_buf_size = CMD_BUFFER_SIZE;
426 entry = (u64)virt_to_phys(cmd_buf);
427 entry |= MMIO_CMD_SIZE_512;
428 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
429 &entry, sizeof(entry));
431 /* set head and tail to zero manually */
432 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
433 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
435 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
440 static void __init free_command_buffer(struct amd_iommu *iommu)
442 free_pages((unsigned long)iommu->cmd_buf,
443 get_order(iommu->cmd_buf_size));
446 /* allocates the memory where the IOMMU will log its events to */
447 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
450 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
451 get_order(EVT_BUFFER_SIZE));
453 if (iommu->evt_buf == NULL)
456 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
457 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
458 &entry, sizeof(entry));
460 iommu->evt_buf_size = EVT_BUFFER_SIZE;
462 return iommu->evt_buf;
465 static void __init free_event_buffer(struct amd_iommu *iommu)
467 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
470 /* sets a specific bit in the device table entry. */
471 static void set_dev_entry_bit(u16 devid, u8 bit)
473 int i = (bit >> 5) & 0x07;
474 int _bit = bit & 0x1f;
476 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
479 /* Writes the specific IOMMU for a device into the rlookup table */
480 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
482 amd_iommu_rlookup_table[devid] = iommu;
486 * This function takes the device specific flags read from the ACPI
487 * table and sets up the device table entry with that information
489 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
490 u16 devid, u32 flags, u32 ext_flags)
492 if (flags & ACPI_DEVFLAG_INITPASS)
493 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
494 if (flags & ACPI_DEVFLAG_EXTINT)
495 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
496 if (flags & ACPI_DEVFLAG_NMI)
497 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
498 if (flags & ACPI_DEVFLAG_SYSMGT1)
499 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
500 if (flags & ACPI_DEVFLAG_SYSMGT2)
501 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
502 if (flags & ACPI_DEVFLAG_LINT0)
503 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
504 if (flags & ACPI_DEVFLAG_LINT1)
505 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
507 set_iommu_for_device(iommu, devid);
511 * Reads the device exclusion range from ACPI and initialize IOMMU with
514 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
516 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
518 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
523 * We only can configure exclusion ranges per IOMMU, not
524 * per device. But we can enable the exclusion range per
525 * device. This is done here
527 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
528 iommu->exclusion_start = m->range_start;
529 iommu->exclusion_length = m->range_length;
534 * This function reads some important data from the IOMMU PCI space and
535 * initializes the driver data structure with it. It reads the hardware
536 * capabilities and the first/last device entries
538 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
540 int cap_ptr = iommu->cap_ptr;
543 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
545 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
547 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
550 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
552 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
554 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
558 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
559 * initializes the hardware and our data structures with it.
561 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
562 struct ivhd_header *h)
565 u8 *end = p, flags = 0;
566 u16 dev_i, devid = 0, devid_start = 0, devid_to = 0;
569 struct ivhd_entry *e;
572 * First set the recommended feature enable bits from ACPI
573 * into the IOMMU control registers
575 h->flags & IVHD_FLAG_HT_TUN_EN ?
576 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
577 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
579 h->flags & IVHD_FLAG_PASSPW_EN ?
580 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
581 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
583 h->flags & IVHD_FLAG_RESPASSPW_EN ?
584 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
585 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
587 h->flags & IVHD_FLAG_ISOC_EN ?
588 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
589 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
592 * make IOMMU memory accesses cache coherent
594 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
597 * Done. Now parse the device entries
599 p += sizeof(struct ivhd_header);
603 e = (struct ivhd_entry *)p;
606 for (dev_i = iommu->first_device;
607 dev_i <= iommu->last_device; ++dev_i)
608 set_dev_entry_from_acpi(iommu, dev_i,
611 case IVHD_DEV_SELECT:
613 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
615 case IVHD_DEV_SELECT_RANGE_START:
616 devid_start = e->devid;
623 devid_to = e->ext >> 8;
624 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
625 amd_iommu_alias_table[devid] = devid_to;
627 case IVHD_DEV_ALIAS_RANGE:
628 devid_start = e->devid;
630 devid_to = e->ext >> 8;
634 case IVHD_DEV_EXT_SELECT:
636 set_dev_entry_from_acpi(iommu, devid, e->flags,
639 case IVHD_DEV_EXT_SELECT_RANGE:
640 devid_start = e->devid;
645 case IVHD_DEV_RANGE_END:
647 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
649 amd_iommu_alias_table[dev_i] = devid_to;
650 set_dev_entry_from_acpi(iommu,
651 amd_iommu_alias_table[dev_i],
659 p += ivhd_entry_length(p);
663 /* Initializes the device->iommu mapping for the driver */
664 static int __init init_iommu_devices(struct amd_iommu *iommu)
668 for (i = iommu->first_device; i <= iommu->last_device; ++i)
669 set_iommu_for_device(iommu, i);
674 static void __init free_iommu_one(struct amd_iommu *iommu)
676 free_command_buffer(iommu);
677 free_event_buffer(iommu);
678 iommu_unmap_mmio_space(iommu);
681 static void __init free_iommu_all(void)
683 struct amd_iommu *iommu, *next;
685 list_for_each_entry_safe(iommu, next, &amd_iommu_list, list) {
686 list_del(&iommu->list);
687 free_iommu_one(iommu);
693 * This function clues the initialization function for one IOMMU
694 * together and also allocates the command buffer and programs the
695 * hardware. It does NOT enable the IOMMU. This is done afterwards.
697 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
699 spin_lock_init(&iommu->lock);
700 list_add_tail(&iommu->list, &amd_iommu_list);
703 * Copy data from ACPI table entry to the iommu struct
705 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
709 iommu->cap_ptr = h->cap_ptr;
710 iommu->pci_seg = h->pci_seg;
711 iommu->mmio_phys = h->mmio_phys;
712 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
713 if (!iommu->mmio_base)
716 iommu_set_device_table(iommu);
717 iommu->cmd_buf = alloc_command_buffer(iommu);
721 iommu->evt_buf = alloc_event_buffer(iommu);
725 iommu->int_enabled = false;
727 init_iommu_from_pci(iommu);
728 init_iommu_from_acpi(iommu, h);
729 init_iommu_devices(iommu);
731 return pci_enable_device(iommu->dev);
735 * Iterates over all IOMMU entries in the ACPI table, allocates the
736 * IOMMU structure and initializes it with init_iommu_one()
738 static int __init init_iommu_all(struct acpi_table_header *table)
740 u8 *p = (u8 *)table, *end = (u8 *)table;
741 struct ivhd_header *h;
742 struct amd_iommu *iommu;
745 end += table->length;
746 p += IVRS_HEADER_LENGTH;
749 h = (struct ivhd_header *)p;
752 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
755 ret = init_iommu_one(iommu, h);
770 /****************************************************************************
772 * The following functions initialize the MSI interrupts for all IOMMUs
773 * in the system. Its a bit challenging because there could be multiple
774 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
777 ****************************************************************************/
779 static int __init iommu_setup_msix(struct amd_iommu *iommu)
781 struct amd_iommu *curr;
782 struct msix_entry entries[32]; /* only 32 supported by AMD IOMMU */
785 list_for_each_entry(curr, &amd_iommu_list, list) {
786 if (curr->dev == iommu->dev) {
787 entries[nvec].entry = curr->evt_msi_num;
788 entries[nvec].vector = 0;
789 curr->int_enabled = true;
794 if (pci_enable_msix(iommu->dev, entries, nvec)) {
795 pci_disable_msix(iommu->dev);
799 for (i = 0; i < nvec; ++i) {
800 int r = request_irq(entries->vector, amd_iommu_int_handler,
811 for (i -= 1; i >= 0; --i)
812 free_irq(entries->vector, NULL);
814 pci_disable_msix(iommu->dev);
819 static int __init iommu_setup_msi(struct amd_iommu *iommu)
822 struct amd_iommu *curr;
824 list_for_each_entry(curr, &amd_iommu_list, list) {
825 if (curr->dev == iommu->dev)
826 curr->int_enabled = true;
830 if (pci_enable_msi(iommu->dev))
833 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
839 pci_disable_msi(iommu->dev);
846 static int __init iommu_init_msi(struct amd_iommu *iommu)
848 if (iommu->int_enabled)
851 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSIX))
852 return iommu_setup_msix(iommu);
853 else if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
854 return iommu_setup_msi(iommu);
859 /****************************************************************************
861 * The next functions belong to the third pass of parsing the ACPI
862 * table. In this last pass the memory mapping requirements are
863 * gathered (like exclusion and unity mapping reanges).
865 ****************************************************************************/
867 static void __init free_unity_maps(void)
869 struct unity_map_entry *entry, *next;
871 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
872 list_del(&entry->list);
877 /* called when we find an exclusion range definition in ACPI */
878 static int __init init_exclusion_range(struct ivmd_header *m)
884 set_device_exclusion_range(m->devid, m);
886 case ACPI_IVMD_TYPE_ALL:
887 for (i = 0; i <= amd_iommu_last_bdf; ++i)
888 set_device_exclusion_range(i, m);
890 case ACPI_IVMD_TYPE_RANGE:
891 for (i = m->devid; i <= m->aux; ++i)
892 set_device_exclusion_range(i, m);
901 /* called for unity map ACPI definition */
902 static int __init init_unity_map_range(struct ivmd_header *m)
904 struct unity_map_entry *e = 0;
906 e = kzalloc(sizeof(*e), GFP_KERNEL);
913 e->devid_start = e->devid_end = m->devid;
915 case ACPI_IVMD_TYPE_ALL:
917 e->devid_end = amd_iommu_last_bdf;
919 case ACPI_IVMD_TYPE_RANGE:
920 e->devid_start = m->devid;
921 e->devid_end = m->aux;
924 e->address_start = PAGE_ALIGN(m->range_start);
925 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
926 e->prot = m->flags >> 1;
928 list_add_tail(&e->list, &amd_iommu_unity_map);
933 /* iterates over all memory definitions we find in the ACPI table */
934 static int __init init_memory_definitions(struct acpi_table_header *table)
936 u8 *p = (u8 *)table, *end = (u8 *)table;
937 struct ivmd_header *m;
939 end += table->length;
940 p += IVRS_HEADER_LENGTH;
943 m = (struct ivmd_header *)p;
944 if (m->flags & IVMD_FLAG_EXCL_RANGE)
945 init_exclusion_range(m);
946 else if (m->flags & IVMD_FLAG_UNITY_MAP)
947 init_unity_map_range(m);
956 * Init the device table to not allow DMA access for devices and
957 * suppress all page faults
959 static void init_device_table(void)
963 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
964 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
965 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
970 * This function finally enables all IOMMUs found in the system after
971 * they have been initialized
973 static void __init enable_iommus(void)
975 struct amd_iommu *iommu;
977 list_for_each_entry(iommu, &amd_iommu_list, list) {
978 iommu_set_exclusion_range(iommu);
979 iommu_init_msi(iommu);
980 iommu_enable_event_logging(iommu);
986 * Suspend/Resume support
987 * disable suspend until real resume implemented
990 static int amd_iommu_resume(struct sys_device *dev)
995 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
1000 static struct sysdev_class amd_iommu_sysdev_class = {
1001 .name = "amd_iommu",
1002 .suspend = amd_iommu_suspend,
1003 .resume = amd_iommu_resume,
1006 static struct sys_device device_amd_iommu = {
1008 .cls = &amd_iommu_sysdev_class,
1012 * This is the core init function for AMD IOMMU hardware in the system.
1013 * This function is called from the generic x86 DMA layer initialization
1016 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1019 * 1 pass) Find the highest PCI device id the driver has to handle.
1020 * Upon this information the size of the data structures is
1021 * determined that needs to be allocated.
1023 * 2 pass) Initialize the data structures just allocated with the
1024 * information in the ACPI table about available AMD IOMMUs
1025 * in the system. It also maps the PCI devices in the
1026 * system to specific IOMMUs
1028 * 3 pass) After the basic data structures are allocated and
1029 * initialized we update them with information about memory
1030 * remapping requirements parsed out of the ACPI table in
1033 * After that the hardware is initialized and ready to go. In the last
1034 * step we do some Linux specific things like registering the driver in
1035 * the dma_ops interface and initializing the suspend/resume support
1036 * functions. Finally it prints some information about AMD IOMMUs and
1037 * the driver state and enables the hardware.
1039 int __init amd_iommu_init(void)
1045 printk(KERN_INFO "AMD IOMMU disabled by kernel command line\n");
1049 if (!amd_iommu_detected)
1053 * First parse ACPI tables to find the largest Bus/Dev/Func
1054 * we need to handle. Upon this information the shared data
1055 * structures for the IOMMUs in the system will be allocated
1057 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1060 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1061 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1062 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1066 /* Device table - directly used by all IOMMUs */
1067 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1068 get_order(dev_table_size));
1069 if (amd_iommu_dev_table == NULL)
1073 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1074 * IOMMU see for that device
1076 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1077 get_order(alias_table_size));
1078 if (amd_iommu_alias_table == NULL)
1081 /* IOMMU rlookup table - find the IOMMU for a specific device */
1082 amd_iommu_rlookup_table = (void *)__get_free_pages(
1083 GFP_KERNEL | __GFP_ZERO,
1084 get_order(rlookup_table_size));
1085 if (amd_iommu_rlookup_table == NULL)
1089 * Protection Domain table - maps devices to protection domains
1090 * This table has the same size as the rlookup_table
1092 amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1093 get_order(rlookup_table_size));
1094 if (amd_iommu_pd_table == NULL)
1097 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1098 GFP_KERNEL | __GFP_ZERO,
1099 get_order(MAX_DOMAIN_ID/8));
1100 if (amd_iommu_pd_alloc_bitmap == NULL)
1103 /* init the device table */
1104 init_device_table();
1107 * let all alias entries point to itself
1109 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1110 amd_iommu_alias_table[i] = i;
1113 * never allocate domain 0 because its used as the non-allocated and
1114 * error value placeholder
1116 amd_iommu_pd_alloc_bitmap[0] = 1;
1119 * now the data structures are allocated and basically initialized
1120 * start the real acpi table scan
1123 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1126 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1129 ret = sysdev_class_register(&amd_iommu_sysdev_class);
1133 ret = sysdev_register(&device_amd_iommu);
1137 ret = amd_iommu_init_dma_ops();
1143 printk(KERN_INFO "AMD IOMMU: aperture size is %d MB\n",
1144 (1 << (amd_iommu_aperture_order-20)));
1146 printk(KERN_INFO "AMD IOMMU: device isolation ");
1147 if (amd_iommu_isolate)
1148 printk("enabled\n");
1150 printk("disabled\n");
1152 if (amd_iommu_unmap_flush)
1153 printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n");
1155 printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n");
1161 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1162 get_order(MAX_DOMAIN_ID/8));
1164 free_pages((unsigned long)amd_iommu_pd_table,
1165 get_order(rlookup_table_size));
1167 free_pages((unsigned long)amd_iommu_rlookup_table,
1168 get_order(rlookup_table_size));
1170 free_pages((unsigned long)amd_iommu_alias_table,
1171 get_order(alias_table_size));
1173 free_pages((unsigned long)amd_iommu_dev_table,
1174 get_order(dev_table_size));
1183 /****************************************************************************
1185 * Early detect code. This code runs at IOMMU detection time in the DMA
1186 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1189 ****************************************************************************/
1190 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1195 void __init amd_iommu_detect(void)
1197 if (swiotlb || no_iommu || (iommu_detected && !gart_iommu_aperture))
1200 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1202 amd_iommu_detected = 1;
1203 #ifdef CONFIG_GART_IOMMU
1204 gart_iommu_aperture_disabled = 1;
1205 gart_iommu_aperture = 0;
1210 /****************************************************************************
1212 * Parsing functions for the AMD IOMMU specific kernel command line
1215 ****************************************************************************/
1217 static int __init parse_amd_iommu_options(char *str)
1219 for (; *str; ++str) {
1220 if (strncmp(str, "isolate", 7) == 0)
1221 amd_iommu_isolate = 1;
1222 if (strncmp(str, "share", 5) == 0)
1223 amd_iommu_isolate = 0;
1224 if (strncmp(str, "fullflush", 9) == 0)
1225 amd_iommu_unmap_flush = true;
1231 static int __init parse_amd_iommu_size_options(char *str)
1233 unsigned order = PAGE_SHIFT + get_order(memparse(str, &str));
1235 if ((order > 24) && (order < 31))
1236 amd_iommu_aperture_order = order;
1241 __setup("amd_iommu=", parse_amd_iommu_options);
1242 __setup("amd_iommu_size=", parse_amd_iommu_size_options);