2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
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_MASK 0x01
53 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
54 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
55 #define IVHD_FLAG_ISOC_EN_MASK 0x08
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));
120 static int __initdata amd_iommu_detected;
122 u16 amd_iommu_last_bdf; /* largest PCI device id we have
124 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
126 unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
127 bool amd_iommu_isolate = true; /* if true, device isolation is
129 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
131 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
135 * Pointer to the device table which is shared by all AMD IOMMUs
136 * it is indexed by the PCI device id or the HT unit id and contains
137 * information about the domain the device belongs to as well as the
138 * page table root pointer.
140 struct dev_table_entry *amd_iommu_dev_table;
143 * The alias table is a driver specific data structure which contains the
144 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
145 * More than one device can share the same requestor id.
147 u16 *amd_iommu_alias_table;
150 * The rlookup table is used to find the IOMMU which is responsible
151 * for a specific device. It is also indexed by the PCI device id.
153 struct amd_iommu **amd_iommu_rlookup_table;
156 * The pd table (protection domain table) is used to find the protection domain
157 * data structure a device belongs to. Indexed with the PCI device id too.
159 struct protection_domain **amd_iommu_pd_table;
162 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
163 * to know which ones are already in use.
165 unsigned long *amd_iommu_pd_alloc_bitmap;
167 static u32 dev_table_size; /* size of the device table */
168 static u32 alias_table_size; /* size of the alias table */
169 static u32 rlookup_table_size; /* size if the rlookup table */
171 static inline void update_last_devid(u16 devid)
173 if (devid > amd_iommu_last_bdf)
174 amd_iommu_last_bdf = devid;
177 static inline unsigned long tbl_size(int entry_size)
179 unsigned shift = PAGE_SHIFT +
180 get_order(amd_iommu_last_bdf * entry_size);
185 /****************************************************************************
187 * AMD IOMMU MMIO register space handling functions
189 * These functions are used to program the IOMMU device registers in
190 * MMIO space required for that driver.
192 ****************************************************************************/
195 * This function set the exclusion range in the IOMMU. DMA accesses to the
196 * exclusion range are passed through untranslated
198 static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
200 u64 start = iommu->exclusion_start & PAGE_MASK;
201 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
204 if (!iommu->exclusion_start)
207 entry = start | MMIO_EXCL_ENABLE_MASK;
208 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
209 &entry, sizeof(entry));
212 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
213 &entry, sizeof(entry));
216 /* Programs the physical address of the device table into the IOMMU hardware */
217 static void __init iommu_set_device_table(struct amd_iommu *iommu)
221 BUG_ON(iommu->mmio_base == NULL);
223 entry = virt_to_phys(amd_iommu_dev_table);
224 entry |= (dev_table_size >> 12) - 1;
225 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
226 &entry, sizeof(entry));
229 /* Generic functions to enable/disable certain features of the IOMMU. */
230 static void __init iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
234 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
236 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
239 static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
243 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
245 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
248 /* Function to enable the hardware */
249 static void __init iommu_enable(struct amd_iommu *iommu)
251 printk(KERN_INFO "AMD IOMMU: Enabling IOMMU at %s cap 0x%hx\n",
252 dev_name(&iommu->dev->dev), iommu->cap_ptr);
254 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
257 /* Function to enable IOMMU event logging and event interrupts */
258 static 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 /* set head and tail to zero manually */
431 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
432 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
434 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
439 static void __init free_command_buffer(struct amd_iommu *iommu)
441 free_pages((unsigned long)iommu->cmd_buf,
442 get_order(iommu->cmd_buf_size));
445 /* allocates the memory where the IOMMU will log its events to */
446 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
449 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
450 get_order(EVT_BUFFER_SIZE));
452 if (iommu->evt_buf == NULL)
455 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
456 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
457 &entry, sizeof(entry));
459 iommu->evt_buf_size = EVT_BUFFER_SIZE;
461 return iommu->evt_buf;
464 static void __init free_event_buffer(struct amd_iommu *iommu)
466 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
469 /* sets a specific bit in the device table entry. */
470 static void set_dev_entry_bit(u16 devid, u8 bit)
472 int i = (bit >> 5) & 0x07;
473 int _bit = bit & 0x1f;
475 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
478 /* Writes the specific IOMMU for a device into the rlookup table */
479 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
481 amd_iommu_rlookup_table[devid] = iommu;
485 * This function takes the device specific flags read from the ACPI
486 * table and sets up the device table entry with that information
488 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
489 u16 devid, u32 flags, u32 ext_flags)
491 if (flags & ACPI_DEVFLAG_INITPASS)
492 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
493 if (flags & ACPI_DEVFLAG_EXTINT)
494 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
495 if (flags & ACPI_DEVFLAG_NMI)
496 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
497 if (flags & ACPI_DEVFLAG_SYSMGT1)
498 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
499 if (flags & ACPI_DEVFLAG_SYSMGT2)
500 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
501 if (flags & ACPI_DEVFLAG_LINT0)
502 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
503 if (flags & ACPI_DEVFLAG_LINT1)
504 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
506 set_iommu_for_device(iommu, devid);
510 * Reads the device exclusion range from ACPI and initialize IOMMU with
513 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
515 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
517 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
522 * We only can configure exclusion ranges per IOMMU, not
523 * per device. But we can enable the exclusion range per
524 * device. This is done here
526 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
527 iommu->exclusion_start = m->range_start;
528 iommu->exclusion_length = m->range_length;
533 * This function reads some important data from the IOMMU PCI space and
534 * initializes the driver data structure with it. It reads the hardware
535 * capabilities and the first/last device entries
537 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
539 int cap_ptr = iommu->cap_ptr;
542 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
544 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
546 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
549 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
551 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
553 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
557 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
558 * initializes the hardware and our data structures with it.
560 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
561 struct ivhd_header *h)
564 u8 *end = p, flags = 0;
565 u16 dev_i, devid = 0, devid_start = 0, devid_to = 0;
568 struct ivhd_entry *e;
571 * First set the recommended feature enable bits from ACPI
572 * into the IOMMU control registers
574 h->flags & IVHD_FLAG_HT_TUN_EN_MASK ?
575 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
576 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
578 h->flags & IVHD_FLAG_PASSPW_EN_MASK ?
579 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
580 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
582 h->flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
583 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
584 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
586 h->flags & IVHD_FLAG_ISOC_EN_MASK ?
587 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
588 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
591 * make IOMMU memory accesses cache coherent
593 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
596 * Done. Now parse the device entries
598 p += sizeof(struct ivhd_header);
603 e = (struct ivhd_entry *)p;
607 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
608 " last device %02x:%02x.%x flags: %02x\n",
609 PCI_BUS(iommu->first_device),
610 PCI_SLOT(iommu->first_device),
611 PCI_FUNC(iommu->first_device),
612 PCI_BUS(iommu->last_device),
613 PCI_SLOT(iommu->last_device),
614 PCI_FUNC(iommu->last_device),
617 for (dev_i = iommu->first_device;
618 dev_i <= iommu->last_device; ++dev_i)
619 set_dev_entry_from_acpi(iommu, dev_i,
622 case IVHD_DEV_SELECT:
624 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
632 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
634 case IVHD_DEV_SELECT_RANGE_START:
636 DUMP_printk(" DEV_SELECT_RANGE_START\t "
637 "devid: %02x:%02x.%x flags: %02x\n",
643 devid_start = e->devid;
650 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
651 "flags: %02x devid_to: %02x:%02x.%x\n",
656 PCI_BUS(e->ext >> 8),
657 PCI_SLOT(e->ext >> 8),
658 PCI_FUNC(e->ext >> 8));
661 devid_to = e->ext >> 8;
662 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
663 amd_iommu_alias_table[devid] = devid_to;
665 case IVHD_DEV_ALIAS_RANGE:
667 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
668 "devid: %02x:%02x.%x flags: %02x "
669 "devid_to: %02x:%02x.%x\n",
674 PCI_BUS(e->ext >> 8),
675 PCI_SLOT(e->ext >> 8),
676 PCI_FUNC(e->ext >> 8));
678 devid_start = e->devid;
680 devid_to = e->ext >> 8;
684 case IVHD_DEV_EXT_SELECT:
686 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
687 "flags: %02x ext: %08x\n",
694 set_dev_entry_from_acpi(iommu, devid, e->flags,
697 case IVHD_DEV_EXT_SELECT_RANGE:
699 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
700 "%02x:%02x.%x flags: %02x ext: %08x\n",
706 devid_start = e->devid;
711 case IVHD_DEV_RANGE_END:
713 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
719 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
721 amd_iommu_alias_table[dev_i] = devid_to;
722 set_dev_entry_from_acpi(iommu,
723 amd_iommu_alias_table[dev_i],
731 p += ivhd_entry_length(p);
735 /* Initializes the device->iommu mapping for the driver */
736 static int __init init_iommu_devices(struct amd_iommu *iommu)
740 for (i = iommu->first_device; i <= iommu->last_device; ++i)
741 set_iommu_for_device(iommu, i);
746 static void __init free_iommu_one(struct amd_iommu *iommu)
748 free_command_buffer(iommu);
749 free_event_buffer(iommu);
750 iommu_unmap_mmio_space(iommu);
753 static void __init free_iommu_all(void)
755 struct amd_iommu *iommu, *next;
757 list_for_each_entry_safe(iommu, next, &amd_iommu_list, list) {
758 list_del(&iommu->list);
759 free_iommu_one(iommu);
765 * This function clues the initialization function for one IOMMU
766 * together and also allocates the command buffer and programs the
767 * hardware. It does NOT enable the IOMMU. This is done afterwards.
769 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
771 spin_lock_init(&iommu->lock);
772 list_add_tail(&iommu->list, &amd_iommu_list);
775 * Copy data from ACPI table entry to the iommu struct
777 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
781 iommu->cap_ptr = h->cap_ptr;
782 iommu->pci_seg = h->pci_seg;
783 iommu->mmio_phys = h->mmio_phys;
784 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
785 if (!iommu->mmio_base)
788 iommu_set_device_table(iommu);
789 iommu->cmd_buf = alloc_command_buffer(iommu);
793 iommu->evt_buf = alloc_event_buffer(iommu);
797 iommu->int_enabled = false;
799 init_iommu_from_pci(iommu);
800 init_iommu_from_acpi(iommu, h);
801 init_iommu_devices(iommu);
803 return pci_enable_device(iommu->dev);
807 * Iterates over all IOMMU entries in the ACPI table, allocates the
808 * IOMMU structure and initializes it with init_iommu_one()
810 static int __init init_iommu_all(struct acpi_table_header *table)
812 u8 *p = (u8 *)table, *end = (u8 *)table;
813 struct ivhd_header *h;
814 struct amd_iommu *iommu;
817 end += table->length;
818 p += IVRS_HEADER_LENGTH;
821 h = (struct ivhd_header *)p;
825 DUMP_printk("IOMMU: device: %02x:%02x.%01x cap: %04x "
826 "seg: %d flags: %01x info %04x\n",
827 PCI_BUS(h->devid), PCI_SLOT(h->devid),
828 PCI_FUNC(h->devid), h->cap_ptr,
829 h->pci_seg, h->flags, h->info);
830 DUMP_printk(" mmio-addr: %016llx\n",
833 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
836 ret = init_iommu_one(iommu, h);
851 /****************************************************************************
853 * The following functions initialize the MSI interrupts for all IOMMUs
854 * in the system. Its a bit challenging because there could be multiple
855 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
858 ****************************************************************************/
860 static int __init iommu_setup_msix(struct amd_iommu *iommu)
862 struct amd_iommu *curr;
863 struct msix_entry entries[32]; /* only 32 supported by AMD IOMMU */
866 list_for_each_entry(curr, &amd_iommu_list, list) {
867 if (curr->dev == iommu->dev) {
868 entries[nvec].entry = curr->evt_msi_num;
869 entries[nvec].vector = 0;
870 curr->int_enabled = true;
875 if (pci_enable_msix(iommu->dev, entries, nvec)) {
876 pci_disable_msix(iommu->dev);
880 for (i = 0; i < nvec; ++i) {
881 int r = request_irq(entries->vector, amd_iommu_int_handler,
892 for (i -= 1; i >= 0; --i)
893 free_irq(entries->vector, NULL);
895 pci_disable_msix(iommu->dev);
900 static int __init iommu_setup_msi(struct amd_iommu *iommu)
903 struct amd_iommu *curr;
905 list_for_each_entry(curr, &amd_iommu_list, list) {
906 if (curr->dev == iommu->dev)
907 curr->int_enabled = true;
911 if (pci_enable_msi(iommu->dev))
914 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
920 pci_disable_msi(iommu->dev);
927 static int __init iommu_init_msi(struct amd_iommu *iommu)
929 if (iommu->int_enabled)
932 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSIX))
933 return iommu_setup_msix(iommu);
934 else if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
935 return iommu_setup_msi(iommu);
940 /****************************************************************************
942 * The next functions belong to the third pass of parsing the ACPI
943 * table. In this last pass the memory mapping requirements are
944 * gathered (like exclusion and unity mapping reanges).
946 ****************************************************************************/
948 static void __init free_unity_maps(void)
950 struct unity_map_entry *entry, *next;
952 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
953 list_del(&entry->list);
958 /* called when we find an exclusion range definition in ACPI */
959 static int __init init_exclusion_range(struct ivmd_header *m)
965 set_device_exclusion_range(m->devid, m);
967 case ACPI_IVMD_TYPE_ALL:
968 for (i = 0; i <= amd_iommu_last_bdf; ++i)
969 set_device_exclusion_range(i, m);
971 case ACPI_IVMD_TYPE_RANGE:
972 for (i = m->devid; i <= m->aux; ++i)
973 set_device_exclusion_range(i, m);
982 /* called for unity map ACPI definition */
983 static int __init init_unity_map_range(struct ivmd_header *m)
985 struct unity_map_entry *e = 0;
987 e = kzalloc(sizeof(*e), GFP_KERNEL);
994 e->devid_start = e->devid_end = m->devid;
996 case ACPI_IVMD_TYPE_ALL:
998 e->devid_end = amd_iommu_last_bdf;
1000 case ACPI_IVMD_TYPE_RANGE:
1001 e->devid_start = m->devid;
1002 e->devid_end = m->aux;
1005 e->address_start = PAGE_ALIGN(m->range_start);
1006 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1007 e->prot = m->flags >> 1;
1009 list_add_tail(&e->list, &amd_iommu_unity_map);
1014 /* iterates over all memory definitions we find in the ACPI table */
1015 static int __init init_memory_definitions(struct acpi_table_header *table)
1017 u8 *p = (u8 *)table, *end = (u8 *)table;
1018 struct ivmd_header *m;
1020 end += table->length;
1021 p += IVRS_HEADER_LENGTH;
1024 m = (struct ivmd_header *)p;
1025 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1026 init_exclusion_range(m);
1027 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1028 init_unity_map_range(m);
1037 * Init the device table to not allow DMA access for devices and
1038 * suppress all page faults
1040 static void init_device_table(void)
1044 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1045 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1046 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1051 * This function finally enables all IOMMUs found in the system after
1052 * they have been initialized
1054 static void __init enable_iommus(void)
1056 struct amd_iommu *iommu;
1058 list_for_each_entry(iommu, &amd_iommu_list, list) {
1059 iommu_set_exclusion_range(iommu);
1060 iommu_init_msi(iommu);
1061 iommu_enable_event_logging(iommu);
1062 iommu_enable(iommu);
1067 * Suspend/Resume support
1068 * disable suspend until real resume implemented
1071 static int amd_iommu_resume(struct sys_device *dev)
1076 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
1081 static struct sysdev_class amd_iommu_sysdev_class = {
1082 .name = "amd_iommu",
1083 .suspend = amd_iommu_suspend,
1084 .resume = amd_iommu_resume,
1087 static struct sys_device device_amd_iommu = {
1089 .cls = &amd_iommu_sysdev_class,
1093 * This is the core init function for AMD IOMMU hardware in the system.
1094 * This function is called from the generic x86 DMA layer initialization
1097 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1100 * 1 pass) Find the highest PCI device id the driver has to handle.
1101 * Upon this information the size of the data structures is
1102 * determined that needs to be allocated.
1104 * 2 pass) Initialize the data structures just allocated with the
1105 * information in the ACPI table about available AMD IOMMUs
1106 * in the system. It also maps the PCI devices in the
1107 * system to specific IOMMUs
1109 * 3 pass) After the basic data structures are allocated and
1110 * initialized we update them with information about memory
1111 * remapping requirements parsed out of the ACPI table in
1114 * After that the hardware is initialized and ready to go. In the last
1115 * step we do some Linux specific things like registering the driver in
1116 * the dma_ops interface and initializing the suspend/resume support
1117 * functions. Finally it prints some information about AMD IOMMUs and
1118 * the driver state and enables the hardware.
1120 int __init amd_iommu_init(void)
1126 printk(KERN_INFO "AMD IOMMU disabled by kernel command line\n");
1130 if (!amd_iommu_detected)
1134 * First parse ACPI tables to find the largest Bus/Dev/Func
1135 * we need to handle. Upon this information the shared data
1136 * structures for the IOMMUs in the system will be allocated
1138 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1141 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1142 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1143 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1147 /* Device table - directly used by all IOMMUs */
1148 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1149 get_order(dev_table_size));
1150 if (amd_iommu_dev_table == NULL)
1154 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1155 * IOMMU see for that device
1157 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1158 get_order(alias_table_size));
1159 if (amd_iommu_alias_table == NULL)
1162 /* IOMMU rlookup table - find the IOMMU for a specific device */
1163 amd_iommu_rlookup_table = (void *)__get_free_pages(
1164 GFP_KERNEL | __GFP_ZERO,
1165 get_order(rlookup_table_size));
1166 if (amd_iommu_rlookup_table == NULL)
1170 * Protection Domain table - maps devices to protection domains
1171 * This table has the same size as the rlookup_table
1173 amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1174 get_order(rlookup_table_size));
1175 if (amd_iommu_pd_table == NULL)
1178 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1179 GFP_KERNEL | __GFP_ZERO,
1180 get_order(MAX_DOMAIN_ID/8));
1181 if (amd_iommu_pd_alloc_bitmap == NULL)
1184 /* init the device table */
1185 init_device_table();
1188 * let all alias entries point to itself
1190 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1191 amd_iommu_alias_table[i] = i;
1194 * never allocate domain 0 because its used as the non-allocated and
1195 * error value placeholder
1197 amd_iommu_pd_alloc_bitmap[0] = 1;
1200 * now the data structures are allocated and basically initialized
1201 * start the real acpi table scan
1204 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1207 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1210 ret = sysdev_class_register(&amd_iommu_sysdev_class);
1214 ret = sysdev_register(&device_amd_iommu);
1218 ret = amd_iommu_init_dma_ops();
1224 printk(KERN_INFO "AMD IOMMU: aperture size is %d MB\n",
1225 (1 << (amd_iommu_aperture_order-20)));
1227 printk(KERN_INFO "AMD IOMMU: device isolation ");
1228 if (amd_iommu_isolate)
1229 printk("enabled\n");
1231 printk("disabled\n");
1233 if (amd_iommu_unmap_flush)
1234 printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n");
1236 printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n");
1242 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1243 get_order(MAX_DOMAIN_ID/8));
1245 free_pages((unsigned long)amd_iommu_pd_table,
1246 get_order(rlookup_table_size));
1248 free_pages((unsigned long)amd_iommu_rlookup_table,
1249 get_order(rlookup_table_size));
1251 free_pages((unsigned long)amd_iommu_alias_table,
1252 get_order(alias_table_size));
1254 free_pages((unsigned long)amd_iommu_dev_table,
1255 get_order(dev_table_size));
1264 /****************************************************************************
1266 * Early detect code. This code runs at IOMMU detection time in the DMA
1267 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1270 ****************************************************************************/
1271 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1276 void __init amd_iommu_detect(void)
1278 if (swiotlb || no_iommu || (iommu_detected && !gart_iommu_aperture))
1281 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1283 amd_iommu_detected = 1;
1284 #ifdef CONFIG_GART_IOMMU
1285 gart_iommu_aperture_disabled = 1;
1286 gart_iommu_aperture = 0;
1291 /****************************************************************************
1293 * Parsing functions for the AMD IOMMU specific kernel command line
1296 ****************************************************************************/
1298 static int __init parse_amd_iommu_dump(char *str)
1300 amd_iommu_dump = true;
1305 static int __init parse_amd_iommu_options(char *str)
1307 for (; *str; ++str) {
1308 if (strncmp(str, "isolate", 7) == 0)
1309 amd_iommu_isolate = true;
1310 if (strncmp(str, "share", 5) == 0)
1311 amd_iommu_isolate = false;
1312 if (strncmp(str, "fullflush", 9) == 0)
1313 amd_iommu_unmap_flush = true;
1319 static int __init parse_amd_iommu_size_options(char *str)
1321 unsigned order = PAGE_SHIFT + get_order(memparse(str, &str));
1323 if ((order > 24) && (order < 31))
1324 amd_iommu_aperture_order = order;
1329 __setup("amd_iommu_dump", parse_amd_iommu_dump);
1330 __setup("amd_iommu=", parse_amd_iommu_options);
1331 __setup("amd_iommu_size=", parse_amd_iommu_size_options);