2 * Copyright (C) 2007-2009 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/list.h>
23 #include <linux/slab.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_proto.h>
29 #include <asm/amd_iommu_types.h>
30 #include <asm/amd_iommu.h>
31 #include <asm/iommu.h>
33 #include <asm/x86_init.h>
36 * definitions for the ACPI scanning code
38 #define IVRS_HEADER_LENGTH 48
40 #define ACPI_IVHD_TYPE 0x10
41 #define ACPI_IVMD_TYPE_ALL 0x20
42 #define ACPI_IVMD_TYPE 0x21
43 #define ACPI_IVMD_TYPE_RANGE 0x22
45 #define IVHD_DEV_ALL 0x01
46 #define IVHD_DEV_SELECT 0x02
47 #define IVHD_DEV_SELECT_RANGE_START 0x03
48 #define IVHD_DEV_RANGE_END 0x04
49 #define IVHD_DEV_ALIAS 0x42
50 #define IVHD_DEV_ALIAS_RANGE 0x43
51 #define IVHD_DEV_EXT_SELECT 0x46
52 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
54 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
55 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
56 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
57 #define IVHD_FLAG_ISOC_EN_MASK 0x08
59 #define IVMD_FLAG_EXCL_RANGE 0x08
60 #define IVMD_FLAG_UNITY_MAP 0x01
62 #define ACPI_DEVFLAG_INITPASS 0x01
63 #define ACPI_DEVFLAG_EXTINT 0x02
64 #define ACPI_DEVFLAG_NMI 0x04
65 #define ACPI_DEVFLAG_SYSMGT1 0x10
66 #define ACPI_DEVFLAG_SYSMGT2 0x20
67 #define ACPI_DEVFLAG_LINT0 0x40
68 #define ACPI_DEVFLAG_LINT1 0x80
69 #define ACPI_DEVFLAG_ATSDIS 0x10000000
72 * ACPI table definitions
74 * These data structures are laid over the table to parse the important values
79 * structure describing one IOMMU in the ACPI table. Typically followed by one
80 * or more ivhd_entrys.
92 } __attribute__((packed));
95 * A device entry describing which devices a specific IOMMU translates and
96 * which requestor ids they use.
103 } __attribute__((packed));
106 * An AMD IOMMU memory definition structure. It defines things like exclusion
107 * ranges for devices and regions that should be unity mapped.
118 } __attribute__((packed));
122 static int __initdata amd_iommu_detected;
123 static bool __initdata amd_iommu_disabled;
125 u16 amd_iommu_last_bdf; /* largest PCI device id we have
127 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
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
134 /* Array to assign indices to IOMMUs*/
135 struct amd_iommu *amd_iommus[MAX_IOMMUS];
136 int amd_iommus_present;
138 /* IOMMUs have a non-present cache? */
139 bool amd_iommu_np_cache __read_mostly;
142 * The ACPI table parsing functions set this variable on an error
144 static int __initdata amd_iommu_init_err;
147 * List of protection domains - used during resume
149 LIST_HEAD(amd_iommu_pd_list);
150 spinlock_t amd_iommu_pd_lock;
153 * Pointer to the device table which is shared by all AMD IOMMUs
154 * it is indexed by the PCI device id or the HT unit id and contains
155 * information about the domain the device belongs to as well as the
156 * page table root pointer.
158 struct dev_table_entry *amd_iommu_dev_table;
161 * The alias table is a driver specific data structure which contains the
162 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
163 * More than one device can share the same requestor id.
165 u16 *amd_iommu_alias_table;
168 * The rlookup table is used to find the IOMMU which is responsible
169 * for a specific device. It is also indexed by the PCI device id.
171 struct amd_iommu **amd_iommu_rlookup_table;
174 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
175 * to know which ones are already in use.
177 unsigned long *amd_iommu_pd_alloc_bitmap;
179 static u32 dev_table_size; /* size of the device table */
180 static u32 alias_table_size; /* size of the alias table */
181 static u32 rlookup_table_size; /* size if the rlookup table */
183 static inline void update_last_devid(u16 devid)
185 if (devid > amd_iommu_last_bdf)
186 amd_iommu_last_bdf = devid;
189 static inline unsigned long tbl_size(int entry_size)
191 unsigned shift = PAGE_SHIFT +
192 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
197 /****************************************************************************
199 * AMD IOMMU MMIO register space handling functions
201 * These functions are used to program the IOMMU device registers in
202 * MMIO space required for that driver.
204 ****************************************************************************/
207 * This function set the exclusion range in the IOMMU. DMA accesses to the
208 * exclusion range are passed through untranslated
210 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
212 u64 start = iommu->exclusion_start & PAGE_MASK;
213 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
216 if (!iommu->exclusion_start)
219 entry = start | MMIO_EXCL_ENABLE_MASK;
220 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
221 &entry, sizeof(entry));
224 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
225 &entry, sizeof(entry));
228 /* Programs the physical address of the device table into the IOMMU hardware */
229 static void __init iommu_set_device_table(struct amd_iommu *iommu)
233 BUG_ON(iommu->mmio_base == NULL);
235 entry = virt_to_phys(amd_iommu_dev_table);
236 entry |= (dev_table_size >> 12) - 1;
237 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
238 &entry, sizeof(entry));
241 /* Generic functions to enable/disable certain features of the IOMMU. */
242 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
246 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
248 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
251 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
255 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
257 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
260 /* Function to enable the hardware */
261 static void iommu_enable(struct amd_iommu *iommu)
263 printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx\n",
264 dev_name(&iommu->dev->dev), iommu->cap_ptr);
266 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
269 static void iommu_disable(struct amd_iommu *iommu)
271 /* Disable command buffer */
272 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
274 /* Disable event logging and event interrupts */
275 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
276 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
278 /* Disable IOMMU hardware itself */
279 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
283 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
284 * the system has one.
286 static u8 * __init iommu_map_mmio_space(u64 address)
290 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
291 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
293 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
297 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
301 release_mem_region(address, MMIO_REGION_LENGTH);
306 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
308 if (iommu->mmio_base)
309 iounmap(iommu->mmio_base);
310 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
313 /****************************************************************************
315 * The functions below belong to the first pass of AMD IOMMU ACPI table
316 * parsing. In this pass we try to find out the highest device id this
317 * code has to handle. Upon this information the size of the shared data
318 * structures is determined later.
320 ****************************************************************************/
323 * This function calculates the length of a given IVHD entry
325 static inline int ivhd_entry_length(u8 *ivhd)
327 return 0x04 << (*ivhd >> 6);
331 * This function reads the last device id the IOMMU has to handle from the PCI
332 * capability header for this IOMMU
334 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
338 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
339 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
345 * After reading the highest device id from the IOMMU PCI capability header
346 * this function looks if there is a higher device id defined in the ACPI table
348 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
350 u8 *p = (void *)h, *end = (void *)h;
351 struct ivhd_entry *dev;
356 find_last_devid_on_pci(PCI_BUS(h->devid),
362 dev = (struct ivhd_entry *)p;
364 case IVHD_DEV_SELECT:
365 case IVHD_DEV_RANGE_END:
367 case IVHD_DEV_EXT_SELECT:
368 /* all the above subfield types refer to device ids */
369 update_last_devid(dev->devid);
374 p += ivhd_entry_length(p);
383 * Iterate over all IVHD entries in the ACPI table and find the highest device
384 * id which we need to handle. This is the first of three functions which parse
385 * the ACPI table. So we check the checksum here.
387 static int __init find_last_devid_acpi(struct acpi_table_header *table)
390 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
391 struct ivhd_header *h;
394 * Validate checksum here so we don't need to do it when
395 * we actually parse the table
397 for (i = 0; i < table->length; ++i)
400 /* ACPI table corrupt */
401 amd_iommu_init_err = -ENODEV;
405 p += IVRS_HEADER_LENGTH;
407 end += table->length;
409 h = (struct ivhd_header *)p;
412 find_last_devid_from_ivhd(h);
424 /****************************************************************************
426 * The following functions belong the the code path which parses the ACPI table
427 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
428 * data structures, initialize the device/alias/rlookup table and also
429 * basically initialize the hardware.
431 ****************************************************************************/
434 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
435 * write commands to that buffer later and the IOMMU will execute them
438 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
440 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
441 get_order(CMD_BUFFER_SIZE));
446 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
452 * This function resets the command buffer if the IOMMU stopped fetching
455 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
457 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
459 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
460 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
462 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
466 * This function writes the command buffer address to the hardware and
469 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
473 BUG_ON(iommu->cmd_buf == NULL);
475 entry = (u64)virt_to_phys(iommu->cmd_buf);
476 entry |= MMIO_CMD_SIZE_512;
478 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
479 &entry, sizeof(entry));
481 amd_iommu_reset_cmd_buffer(iommu);
482 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
485 static void __init free_command_buffer(struct amd_iommu *iommu)
487 free_pages((unsigned long)iommu->cmd_buf,
488 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
491 /* allocates the memory where the IOMMU will log its events to */
492 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
494 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
495 get_order(EVT_BUFFER_SIZE));
497 if (iommu->evt_buf == NULL)
500 iommu->evt_buf_size = EVT_BUFFER_SIZE;
502 return iommu->evt_buf;
505 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
509 BUG_ON(iommu->evt_buf == NULL);
511 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
513 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
514 &entry, sizeof(entry));
516 /* set head and tail to zero manually */
517 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
518 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
520 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
523 static void __init free_event_buffer(struct amd_iommu *iommu)
525 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
528 /* sets a specific bit in the device table entry. */
529 static void set_dev_entry_bit(u16 devid, u8 bit)
531 int i = (bit >> 5) & 0x07;
532 int _bit = bit & 0x1f;
534 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
537 static int get_dev_entry_bit(u16 devid, u8 bit)
539 int i = (bit >> 5) & 0x07;
540 int _bit = bit & 0x1f;
542 return (amd_iommu_dev_table[devid].data[i] & (1 << _bit)) >> _bit;
546 void amd_iommu_apply_erratum_63(u16 devid)
550 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
551 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
554 set_dev_entry_bit(devid, DEV_ENTRY_IW);
557 /* Writes the specific IOMMU for a device into the rlookup table */
558 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
560 amd_iommu_rlookup_table[devid] = iommu;
564 * This function takes the device specific flags read from the ACPI
565 * table and sets up the device table entry with that information
567 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
568 u16 devid, u32 flags, u32 ext_flags)
570 if (flags & ACPI_DEVFLAG_INITPASS)
571 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
572 if (flags & ACPI_DEVFLAG_EXTINT)
573 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
574 if (flags & ACPI_DEVFLAG_NMI)
575 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
576 if (flags & ACPI_DEVFLAG_SYSMGT1)
577 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
578 if (flags & ACPI_DEVFLAG_SYSMGT2)
579 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
580 if (flags & ACPI_DEVFLAG_LINT0)
581 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
582 if (flags & ACPI_DEVFLAG_LINT1)
583 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
585 amd_iommu_apply_erratum_63(devid);
587 set_iommu_for_device(iommu, devid);
591 * Reads the device exclusion range from ACPI and initialize IOMMU with
594 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
596 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
598 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
603 * We only can configure exclusion ranges per IOMMU, not
604 * per device. But we can enable the exclusion range per
605 * device. This is done here
607 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
608 iommu->exclusion_start = m->range_start;
609 iommu->exclusion_length = m->range_length;
614 * This function reads some important data from the IOMMU PCI space and
615 * initializes the driver data structure with it. It reads the hardware
616 * capabilities and the first/last device entries
618 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
620 int cap_ptr = iommu->cap_ptr;
623 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
625 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
627 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
630 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
632 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
634 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
638 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
639 * initializes the hardware and our data structures with it.
641 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
642 struct ivhd_header *h)
645 u8 *end = p, flags = 0;
646 u16 dev_i, devid = 0, devid_start = 0, devid_to = 0;
649 struct ivhd_entry *e;
652 * First set the recommended feature enable bits from ACPI
653 * into the IOMMU control registers
655 h->flags & IVHD_FLAG_HT_TUN_EN_MASK ?
656 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
657 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
659 h->flags & IVHD_FLAG_PASSPW_EN_MASK ?
660 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
661 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
663 h->flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
664 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
665 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
667 h->flags & IVHD_FLAG_ISOC_EN_MASK ?
668 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
669 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
672 * make IOMMU memory accesses cache coherent
674 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
677 * Done. Now parse the device entries
679 p += sizeof(struct ivhd_header);
684 e = (struct ivhd_entry *)p;
688 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
689 " last device %02x:%02x.%x flags: %02x\n",
690 PCI_BUS(iommu->first_device),
691 PCI_SLOT(iommu->first_device),
692 PCI_FUNC(iommu->first_device),
693 PCI_BUS(iommu->last_device),
694 PCI_SLOT(iommu->last_device),
695 PCI_FUNC(iommu->last_device),
698 for (dev_i = iommu->first_device;
699 dev_i <= iommu->last_device; ++dev_i)
700 set_dev_entry_from_acpi(iommu, dev_i,
703 case IVHD_DEV_SELECT:
705 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
713 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
715 case IVHD_DEV_SELECT_RANGE_START:
717 DUMP_printk(" DEV_SELECT_RANGE_START\t "
718 "devid: %02x:%02x.%x flags: %02x\n",
724 devid_start = e->devid;
731 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
732 "flags: %02x devid_to: %02x:%02x.%x\n",
737 PCI_BUS(e->ext >> 8),
738 PCI_SLOT(e->ext >> 8),
739 PCI_FUNC(e->ext >> 8));
742 devid_to = e->ext >> 8;
743 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
744 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
745 amd_iommu_alias_table[devid] = devid_to;
747 case IVHD_DEV_ALIAS_RANGE:
749 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
750 "devid: %02x:%02x.%x flags: %02x "
751 "devid_to: %02x:%02x.%x\n",
756 PCI_BUS(e->ext >> 8),
757 PCI_SLOT(e->ext >> 8),
758 PCI_FUNC(e->ext >> 8));
760 devid_start = e->devid;
762 devid_to = e->ext >> 8;
766 case IVHD_DEV_EXT_SELECT:
768 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
769 "flags: %02x ext: %08x\n",
776 set_dev_entry_from_acpi(iommu, devid, e->flags,
779 case IVHD_DEV_EXT_SELECT_RANGE:
781 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
782 "%02x:%02x.%x flags: %02x ext: %08x\n",
788 devid_start = e->devid;
793 case IVHD_DEV_RANGE_END:
795 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
801 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
803 amd_iommu_alias_table[dev_i] = devid_to;
804 set_dev_entry_from_acpi(iommu,
805 devid_to, flags, ext_flags);
807 set_dev_entry_from_acpi(iommu, dev_i,
815 p += ivhd_entry_length(p);
819 /* Initializes the device->iommu mapping for the driver */
820 static int __init init_iommu_devices(struct amd_iommu *iommu)
824 for (i = iommu->first_device; i <= iommu->last_device; ++i)
825 set_iommu_for_device(iommu, i);
830 static void __init free_iommu_one(struct amd_iommu *iommu)
832 free_command_buffer(iommu);
833 free_event_buffer(iommu);
834 iommu_unmap_mmio_space(iommu);
837 static void __init free_iommu_all(void)
839 struct amd_iommu *iommu, *next;
841 for_each_iommu_safe(iommu, next) {
842 list_del(&iommu->list);
843 free_iommu_one(iommu);
849 * This function clues the initialization function for one IOMMU
850 * together and also allocates the command buffer and programs the
851 * hardware. It does NOT enable the IOMMU. This is done afterwards.
853 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
855 spin_lock_init(&iommu->lock);
857 /* Add IOMMU to internal data structures */
858 list_add_tail(&iommu->list, &amd_iommu_list);
859 iommu->index = amd_iommus_present++;
861 if (unlikely(iommu->index >= MAX_IOMMUS)) {
862 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
866 /* Index is fine - add IOMMU to the array */
867 amd_iommus[iommu->index] = iommu;
870 * Copy data from ACPI table entry to the iommu struct
872 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
876 iommu->cap_ptr = h->cap_ptr;
877 iommu->pci_seg = h->pci_seg;
878 iommu->mmio_phys = h->mmio_phys;
879 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
880 if (!iommu->mmio_base)
883 iommu->cmd_buf = alloc_command_buffer(iommu);
887 iommu->evt_buf = alloc_event_buffer(iommu);
891 iommu->int_enabled = false;
893 init_iommu_from_pci(iommu);
894 init_iommu_from_acpi(iommu, h);
895 init_iommu_devices(iommu);
897 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
898 amd_iommu_np_cache = true;
900 return pci_enable_device(iommu->dev);
904 * Iterates over all IOMMU entries in the ACPI table, allocates the
905 * IOMMU structure and initializes it with init_iommu_one()
907 static int __init init_iommu_all(struct acpi_table_header *table)
909 u8 *p = (u8 *)table, *end = (u8 *)table;
910 struct ivhd_header *h;
911 struct amd_iommu *iommu;
914 end += table->length;
915 p += IVRS_HEADER_LENGTH;
918 h = (struct ivhd_header *)p;
922 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
923 "seg: %d flags: %01x info %04x\n",
924 PCI_BUS(h->devid), PCI_SLOT(h->devid),
925 PCI_FUNC(h->devid), h->cap_ptr,
926 h->pci_seg, h->flags, h->info);
927 DUMP_printk(" mmio-addr: %016llx\n",
930 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
932 amd_iommu_init_err = -ENOMEM;
936 ret = init_iommu_one(iommu, h);
938 amd_iommu_init_err = ret;
953 /****************************************************************************
955 * The following functions initialize the MSI interrupts for all IOMMUs
956 * in the system. Its a bit challenging because there could be multiple
957 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
960 ****************************************************************************/
962 static int iommu_setup_msi(struct amd_iommu *iommu)
966 if (pci_enable_msi(iommu->dev))
969 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
975 pci_disable_msi(iommu->dev);
979 iommu->int_enabled = true;
980 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
985 static int iommu_init_msi(struct amd_iommu *iommu)
987 if (iommu->int_enabled)
990 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
991 return iommu_setup_msi(iommu);
996 /****************************************************************************
998 * The next functions belong to the third pass of parsing the ACPI
999 * table. In this last pass the memory mapping requirements are
1000 * gathered (like exclusion and unity mapping reanges).
1002 ****************************************************************************/
1004 static void __init free_unity_maps(void)
1006 struct unity_map_entry *entry, *next;
1008 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1009 list_del(&entry->list);
1014 /* called when we find an exclusion range definition in ACPI */
1015 static int __init init_exclusion_range(struct ivmd_header *m)
1020 case ACPI_IVMD_TYPE:
1021 set_device_exclusion_range(m->devid, m);
1023 case ACPI_IVMD_TYPE_ALL:
1024 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1025 set_device_exclusion_range(i, m);
1027 case ACPI_IVMD_TYPE_RANGE:
1028 for (i = m->devid; i <= m->aux; ++i)
1029 set_device_exclusion_range(i, m);
1038 /* called for unity map ACPI definition */
1039 static int __init init_unity_map_range(struct ivmd_header *m)
1041 struct unity_map_entry *e = 0;
1044 e = kzalloc(sizeof(*e), GFP_KERNEL);
1052 case ACPI_IVMD_TYPE:
1053 s = "IVMD_TYPEi\t\t\t";
1054 e->devid_start = e->devid_end = m->devid;
1056 case ACPI_IVMD_TYPE_ALL:
1057 s = "IVMD_TYPE_ALL\t\t";
1059 e->devid_end = amd_iommu_last_bdf;
1061 case ACPI_IVMD_TYPE_RANGE:
1062 s = "IVMD_TYPE_RANGE\t\t";
1063 e->devid_start = m->devid;
1064 e->devid_end = m->aux;
1067 e->address_start = PAGE_ALIGN(m->range_start);
1068 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1069 e->prot = m->flags >> 1;
1071 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1072 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1073 PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1074 PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1075 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1076 e->address_start, e->address_end, m->flags);
1078 list_add_tail(&e->list, &amd_iommu_unity_map);
1083 /* iterates over all memory definitions we find in the ACPI table */
1084 static int __init init_memory_definitions(struct acpi_table_header *table)
1086 u8 *p = (u8 *)table, *end = (u8 *)table;
1087 struct ivmd_header *m;
1089 end += table->length;
1090 p += IVRS_HEADER_LENGTH;
1093 m = (struct ivmd_header *)p;
1094 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1095 init_exclusion_range(m);
1096 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1097 init_unity_map_range(m);
1106 * Init the device table to not allow DMA access for devices and
1107 * suppress all page faults
1109 static void init_device_table(void)
1113 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1114 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1115 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1120 * This function finally enables all IOMMUs found in the system after
1121 * they have been initialized
1123 static void enable_iommus(void)
1125 struct amd_iommu *iommu;
1127 for_each_iommu(iommu) {
1128 iommu_disable(iommu);
1129 iommu_set_device_table(iommu);
1130 iommu_enable_command_buffer(iommu);
1131 iommu_enable_event_buffer(iommu);
1132 iommu_set_exclusion_range(iommu);
1133 iommu_init_msi(iommu);
1134 iommu_enable(iommu);
1138 static void disable_iommus(void)
1140 struct amd_iommu *iommu;
1142 for_each_iommu(iommu)
1143 iommu_disable(iommu);
1147 * Suspend/Resume support
1148 * disable suspend until real resume implemented
1151 static int amd_iommu_resume(struct sys_device *dev)
1153 /* re-load the hardware */
1157 * we have to flush after the IOMMUs are enabled because a
1158 * disabled IOMMU will never execute the commands we send
1160 amd_iommu_flush_all_devices();
1161 amd_iommu_flush_all_domains();
1166 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
1168 /* disable IOMMUs to go out of the way for BIOS */
1174 static struct sysdev_class amd_iommu_sysdev_class = {
1175 .name = "amd_iommu",
1176 .suspend = amd_iommu_suspend,
1177 .resume = amd_iommu_resume,
1180 static struct sys_device device_amd_iommu = {
1182 .cls = &amd_iommu_sysdev_class,
1186 * This is the core init function for AMD IOMMU hardware in the system.
1187 * This function is called from the generic x86 DMA layer initialization
1190 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1193 * 1 pass) Find the highest PCI device id the driver has to handle.
1194 * Upon this information the size of the data structures is
1195 * determined that needs to be allocated.
1197 * 2 pass) Initialize the data structures just allocated with the
1198 * information in the ACPI table about available AMD IOMMUs
1199 * in the system. It also maps the PCI devices in the
1200 * system to specific IOMMUs
1202 * 3 pass) After the basic data structures are allocated and
1203 * initialized we update them with information about memory
1204 * remapping requirements parsed out of the ACPI table in
1207 * After that the hardware is initialized and ready to go. In the last
1208 * step we do some Linux specific things like registering the driver in
1209 * the dma_ops interface and initializing the suspend/resume support
1210 * functions. Finally it prints some information about AMD IOMMUs and
1211 * the driver state and enables the hardware.
1213 static int __init amd_iommu_init(void)
1218 * First parse ACPI tables to find the largest Bus/Dev/Func
1219 * we need to handle. Upon this information the shared data
1220 * structures for the IOMMUs in the system will be allocated
1222 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1225 ret = amd_iommu_init_err;
1229 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1230 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1231 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1235 /* Device table - directly used by all IOMMUs */
1236 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1237 get_order(dev_table_size));
1238 if (amd_iommu_dev_table == NULL)
1242 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1243 * IOMMU see for that device
1245 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1246 get_order(alias_table_size));
1247 if (amd_iommu_alias_table == NULL)
1250 /* IOMMU rlookup table - find the IOMMU for a specific device */
1251 amd_iommu_rlookup_table = (void *)__get_free_pages(
1252 GFP_KERNEL | __GFP_ZERO,
1253 get_order(rlookup_table_size));
1254 if (amd_iommu_rlookup_table == NULL)
1257 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1258 GFP_KERNEL | __GFP_ZERO,
1259 get_order(MAX_DOMAIN_ID/8));
1260 if (amd_iommu_pd_alloc_bitmap == NULL)
1263 /* init the device table */
1264 init_device_table();
1267 * let all alias entries point to itself
1269 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1270 amd_iommu_alias_table[i] = i;
1273 * never allocate domain 0 because its used as the non-allocated and
1274 * error value placeholder
1276 amd_iommu_pd_alloc_bitmap[0] = 1;
1278 spin_lock_init(&amd_iommu_pd_lock);
1281 * now the data structures are allocated and basically initialized
1282 * start the real acpi table scan
1285 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1288 if (amd_iommu_init_err) {
1289 ret = amd_iommu_init_err;
1293 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1296 if (amd_iommu_init_err) {
1297 ret = amd_iommu_init_err;
1301 ret = sysdev_class_register(&amd_iommu_sysdev_class);
1305 ret = sysdev_register(&device_amd_iommu);
1309 ret = amd_iommu_init_devices();
1315 if (iommu_pass_through)
1316 ret = amd_iommu_init_passthrough();
1318 ret = amd_iommu_init_dma_ops();
1323 amd_iommu_init_api();
1325 amd_iommu_init_notifier();
1327 if (iommu_pass_through)
1330 if (amd_iommu_unmap_flush)
1331 printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
1333 printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n");
1335 x86_platform.iommu_shutdown = disable_iommus;
1343 amd_iommu_uninit_devices();
1345 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1346 get_order(MAX_DOMAIN_ID/8));
1348 free_pages((unsigned long)amd_iommu_rlookup_table,
1349 get_order(rlookup_table_size));
1351 free_pages((unsigned long)amd_iommu_alias_table,
1352 get_order(alias_table_size));
1354 free_pages((unsigned long)amd_iommu_dev_table,
1355 get_order(dev_table_size));
1361 #ifdef CONFIG_GART_IOMMU
1363 * We failed to initialize the AMD IOMMU - try fallback to GART
1373 /****************************************************************************
1375 * Early detect code. This code runs at IOMMU detection time in the DMA
1376 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1379 ****************************************************************************/
1380 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1385 void __init amd_iommu_detect(void)
1387 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
1390 if (amd_iommu_disabled)
1393 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1395 amd_iommu_detected = 1;
1396 x86_init.iommu.iommu_init = amd_iommu_init;
1398 /* Make sure ACS will be enabled */
1403 /****************************************************************************
1405 * Parsing functions for the AMD IOMMU specific kernel command line
1408 ****************************************************************************/
1410 static int __init parse_amd_iommu_dump(char *str)
1412 amd_iommu_dump = true;
1417 static int __init parse_amd_iommu_options(char *str)
1419 for (; *str; ++str) {
1420 if (strncmp(str, "fullflush", 9) == 0)
1421 amd_iommu_unmap_flush = true;
1422 if (strncmp(str, "off", 3) == 0)
1423 amd_iommu_disabled = true;
1429 __setup("amd_iommu_dump", parse_amd_iommu_dump);
1430 __setup("amd_iommu=", parse_amd_iommu_options);