2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
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/syscore_ops.h>
25 #include <linux/interrupt.h>
26 #include <linux/msi.h>
27 #include <linux/amd-iommu.h>
28 #include <linux/export.h>
29 #include <linux/iommu.h>
30 #include <asm/pci-direct.h>
31 #include <asm/iommu.h>
33 #include <asm/x86_init.h>
34 #include <asm/iommu_table.h>
35 #include <asm/io_apic.h>
36 #include <asm/irq_remapping.h>
38 #include "amd_iommu_proto.h"
39 #include "amd_iommu_types.h"
40 #include "irq_remapping.h"
43 * definitions for the ACPI scanning code
45 #define IVRS_HEADER_LENGTH 48
47 #define ACPI_IVHD_TYPE 0x10
48 #define ACPI_IVMD_TYPE_ALL 0x20
49 #define ACPI_IVMD_TYPE 0x21
50 #define ACPI_IVMD_TYPE_RANGE 0x22
52 #define IVHD_DEV_ALL 0x01
53 #define IVHD_DEV_SELECT 0x02
54 #define IVHD_DEV_SELECT_RANGE_START 0x03
55 #define IVHD_DEV_RANGE_END 0x04
56 #define IVHD_DEV_ALIAS 0x42
57 #define IVHD_DEV_ALIAS_RANGE 0x43
58 #define IVHD_DEV_EXT_SELECT 0x46
59 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
60 #define IVHD_DEV_SPECIAL 0x48
62 #define IVHD_SPECIAL_IOAPIC 1
63 #define IVHD_SPECIAL_HPET 2
65 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
66 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
67 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
68 #define IVHD_FLAG_ISOC_EN_MASK 0x08
70 #define IVMD_FLAG_EXCL_RANGE 0x08
71 #define IVMD_FLAG_UNITY_MAP 0x01
73 #define ACPI_DEVFLAG_INITPASS 0x01
74 #define ACPI_DEVFLAG_EXTINT 0x02
75 #define ACPI_DEVFLAG_NMI 0x04
76 #define ACPI_DEVFLAG_SYSMGT1 0x10
77 #define ACPI_DEVFLAG_SYSMGT2 0x20
78 #define ACPI_DEVFLAG_LINT0 0x40
79 #define ACPI_DEVFLAG_LINT1 0x80
80 #define ACPI_DEVFLAG_ATSDIS 0x10000000
83 * ACPI table definitions
85 * These data structures are laid over the table to parse the important values
90 * structure describing one IOMMU in the ACPI table. Typically followed by one
91 * or more ivhd_entrys.
103 } __attribute__((packed));
106 * A device entry describing which devices a specific IOMMU translates and
107 * which requestor ids they use.
114 } __attribute__((packed));
117 * An AMD IOMMU memory definition structure. It defines things like exclusion
118 * ranges for devices and regions that should be unity mapped.
129 } __attribute__((packed));
132 bool amd_iommu_irq_remap __read_mostly;
134 static bool amd_iommu_detected;
135 static bool __initdata amd_iommu_disabled;
137 u16 amd_iommu_last_bdf; /* largest PCI device id we have
139 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
141 u32 amd_iommu_unmap_flush; /* if true, flush on every unmap */
143 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
146 /* Array to assign indices to IOMMUs*/
147 struct amd_iommu *amd_iommus[MAX_IOMMUS];
148 int amd_iommus_present;
150 /* IOMMUs have a non-present cache? */
151 bool amd_iommu_np_cache __read_mostly;
152 bool amd_iommu_iotlb_sup __read_mostly = true;
154 u32 amd_iommu_max_pasid __read_mostly = ~0;
156 bool amd_iommu_v2_present __read_mostly;
157 static bool amd_iommu_pc_present __read_mostly;
159 bool amd_iommu_force_isolation __read_mostly;
162 * List of protection domains - used during resume
164 LIST_HEAD(amd_iommu_pd_list);
165 spinlock_t amd_iommu_pd_lock;
168 * Pointer to the device table which is shared by all AMD IOMMUs
169 * it is indexed by the PCI device id or the HT unit id and contains
170 * information about the domain the device belongs to as well as the
171 * page table root pointer.
173 struct dev_table_entry *amd_iommu_dev_table;
176 * The alias table is a driver specific data structure which contains the
177 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
178 * More than one device can share the same requestor id.
180 u16 *amd_iommu_alias_table;
183 * The rlookup table is used to find the IOMMU which is responsible
184 * for a specific device. It is also indexed by the PCI device id.
186 struct amd_iommu **amd_iommu_rlookup_table;
189 * This table is used to find the irq remapping table for a given device id
192 struct irq_remap_table **irq_lookup_table;
195 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
196 * to know which ones are already in use.
198 unsigned long *amd_iommu_pd_alloc_bitmap;
200 static u32 dev_table_size; /* size of the device table */
201 static u32 alias_table_size; /* size of the alias table */
202 static u32 rlookup_table_size; /* size if the rlookup table */
204 enum iommu_init_state {
217 /* Early ioapic and hpet maps from kernel command line */
218 #define EARLY_MAP_SIZE 4
219 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
220 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
221 static int __initdata early_ioapic_map_size;
222 static int __initdata early_hpet_map_size;
223 static bool __initdata cmdline_maps;
225 static enum iommu_init_state init_state = IOMMU_START_STATE;
227 static int amd_iommu_enable_interrupts(void);
228 static int __init iommu_go_to_state(enum iommu_init_state state);
229 static void init_device_table_dma(void);
231 static inline void update_last_devid(u16 devid)
233 if (devid > amd_iommu_last_bdf)
234 amd_iommu_last_bdf = devid;
237 static inline unsigned long tbl_size(int entry_size)
239 unsigned shift = PAGE_SHIFT +
240 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
245 /* Access to l1 and l2 indexed register spaces */
247 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
251 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
252 pci_read_config_dword(iommu->dev, 0xfc, &val);
256 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
258 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
259 pci_write_config_dword(iommu->dev, 0xfc, val);
260 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
263 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
267 pci_write_config_dword(iommu->dev, 0xf0, address);
268 pci_read_config_dword(iommu->dev, 0xf4, &val);
272 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
274 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
275 pci_write_config_dword(iommu->dev, 0xf4, val);
278 /****************************************************************************
280 * AMD IOMMU MMIO register space handling functions
282 * These functions are used to program the IOMMU device registers in
283 * MMIO space required for that driver.
285 ****************************************************************************/
288 * This function set the exclusion range in the IOMMU. DMA accesses to the
289 * exclusion range are passed through untranslated
291 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
293 u64 start = iommu->exclusion_start & PAGE_MASK;
294 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
297 if (!iommu->exclusion_start)
300 entry = start | MMIO_EXCL_ENABLE_MASK;
301 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
302 &entry, sizeof(entry));
305 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
306 &entry, sizeof(entry));
309 /* Programs the physical address of the device table into the IOMMU hardware */
310 static void iommu_set_device_table(struct amd_iommu *iommu)
314 BUG_ON(iommu->mmio_base == NULL);
316 entry = virt_to_phys(amd_iommu_dev_table);
317 entry |= (dev_table_size >> 12) - 1;
318 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
319 &entry, sizeof(entry));
322 /* Generic functions to enable/disable certain features of the IOMMU. */
323 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
327 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
329 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
332 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
336 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
338 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
341 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
345 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
346 ctrl &= ~CTRL_INV_TO_MASK;
347 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
348 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
351 /* Function to enable the hardware */
352 static void iommu_enable(struct amd_iommu *iommu)
354 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
357 static void iommu_disable(struct amd_iommu *iommu)
359 /* Disable command buffer */
360 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
362 /* Disable event logging and event interrupts */
363 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
364 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
366 /* Disable IOMMU hardware itself */
367 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
371 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
372 * the system has one.
374 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
376 if (!request_mem_region(address, end, "amd_iommu")) {
377 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
379 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
383 return (u8 __iomem *)ioremap_nocache(address, end);
386 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
388 if (iommu->mmio_base)
389 iounmap(iommu->mmio_base);
390 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
393 /****************************************************************************
395 * The functions below belong to the first pass of AMD IOMMU ACPI table
396 * parsing. In this pass we try to find out the highest device id this
397 * code has to handle. Upon this information the size of the shared data
398 * structures is determined later.
400 ****************************************************************************/
403 * This function calculates the length of a given IVHD entry
405 static inline int ivhd_entry_length(u8 *ivhd)
407 return 0x04 << (*ivhd >> 6);
411 * This function reads the last device id the IOMMU has to handle from the PCI
412 * capability header for this IOMMU
414 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
418 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
419 update_last_devid(PCI_DEVID(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
425 * After reading the highest device id from the IOMMU PCI capability header
426 * this function looks if there is a higher device id defined in the ACPI table
428 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
430 u8 *p = (void *)h, *end = (void *)h;
431 struct ivhd_entry *dev;
436 find_last_devid_on_pci(PCI_BUS_NUM(h->devid),
442 dev = (struct ivhd_entry *)p;
445 /* Use maximum BDF value for DEV_ALL */
446 update_last_devid(0xffff);
448 case IVHD_DEV_SELECT:
449 case IVHD_DEV_RANGE_END:
451 case IVHD_DEV_EXT_SELECT:
452 /* all the above subfield types refer to device ids */
453 update_last_devid(dev->devid);
458 p += ivhd_entry_length(p);
467 * Iterate over all IVHD entries in the ACPI table and find the highest device
468 * id which we need to handle. This is the first of three functions which parse
469 * the ACPI table. So we check the checksum here.
471 static int __init find_last_devid_acpi(struct acpi_table_header *table)
474 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
475 struct ivhd_header *h;
478 * Validate checksum here so we don't need to do it when
479 * we actually parse the table
481 for (i = 0; i < table->length; ++i)
484 /* ACPI table corrupt */
487 p += IVRS_HEADER_LENGTH;
489 end += table->length;
491 h = (struct ivhd_header *)p;
494 find_last_devid_from_ivhd(h);
506 /****************************************************************************
508 * The following functions belong to the code path which parses the ACPI table
509 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
510 * data structures, initialize the device/alias/rlookup table and also
511 * basically initialize the hardware.
513 ****************************************************************************/
516 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
517 * write commands to that buffer later and the IOMMU will execute them
520 static int __init alloc_command_buffer(struct amd_iommu *iommu)
522 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
523 get_order(CMD_BUFFER_SIZE));
525 return iommu->cmd_buf ? 0 : -ENOMEM;
529 * This function resets the command buffer if the IOMMU stopped fetching
532 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
534 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
536 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
537 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
539 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
543 * This function writes the command buffer address to the hardware and
546 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
550 BUG_ON(iommu->cmd_buf == NULL);
552 entry = (u64)virt_to_phys(iommu->cmd_buf);
553 entry |= MMIO_CMD_SIZE_512;
555 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
556 &entry, sizeof(entry));
558 amd_iommu_reset_cmd_buffer(iommu);
561 static void __init free_command_buffer(struct amd_iommu *iommu)
563 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
566 /* allocates the memory where the IOMMU will log its events to */
567 static int __init alloc_event_buffer(struct amd_iommu *iommu)
569 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
570 get_order(EVT_BUFFER_SIZE));
572 return iommu->evt_buf ? 0 : -ENOMEM;
575 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
579 BUG_ON(iommu->evt_buf == NULL);
581 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
583 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
584 &entry, sizeof(entry));
586 /* set head and tail to zero manually */
587 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
588 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
590 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
593 static void __init free_event_buffer(struct amd_iommu *iommu)
595 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
598 /* allocates the memory where the IOMMU will log its events to */
599 static int __init alloc_ppr_log(struct amd_iommu *iommu)
601 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
602 get_order(PPR_LOG_SIZE));
604 return iommu->ppr_log ? 0 : -ENOMEM;
607 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
611 if (iommu->ppr_log == NULL)
614 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
616 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
617 &entry, sizeof(entry));
619 /* set head and tail to zero manually */
620 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
621 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
623 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
624 iommu_feature_enable(iommu, CONTROL_PPR_EN);
627 static void __init free_ppr_log(struct amd_iommu *iommu)
629 if (iommu->ppr_log == NULL)
632 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
635 static void iommu_enable_gt(struct amd_iommu *iommu)
637 if (!iommu_feature(iommu, FEATURE_GT))
640 iommu_feature_enable(iommu, CONTROL_GT_EN);
643 /* sets a specific bit in the device table entry. */
644 static void set_dev_entry_bit(u16 devid, u8 bit)
646 int i = (bit >> 6) & 0x03;
647 int _bit = bit & 0x3f;
649 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
652 static int get_dev_entry_bit(u16 devid, u8 bit)
654 int i = (bit >> 6) & 0x03;
655 int _bit = bit & 0x3f;
657 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
661 void amd_iommu_apply_erratum_63(u16 devid)
665 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
666 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
669 set_dev_entry_bit(devid, DEV_ENTRY_IW);
672 /* Writes the specific IOMMU for a device into the rlookup table */
673 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
675 amd_iommu_rlookup_table[devid] = iommu;
679 * This function takes the device specific flags read from the ACPI
680 * table and sets up the device table entry with that information
682 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
683 u16 devid, u32 flags, u32 ext_flags)
685 if (flags & ACPI_DEVFLAG_INITPASS)
686 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
687 if (flags & ACPI_DEVFLAG_EXTINT)
688 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
689 if (flags & ACPI_DEVFLAG_NMI)
690 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
691 if (flags & ACPI_DEVFLAG_SYSMGT1)
692 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
693 if (flags & ACPI_DEVFLAG_SYSMGT2)
694 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
695 if (flags & ACPI_DEVFLAG_LINT0)
696 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
697 if (flags & ACPI_DEVFLAG_LINT1)
698 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
700 amd_iommu_apply_erratum_63(devid);
702 set_iommu_for_device(iommu, devid);
705 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
707 struct devid_map *entry;
708 struct list_head *list;
710 if (type == IVHD_SPECIAL_IOAPIC)
712 else if (type == IVHD_SPECIAL_HPET)
717 list_for_each_entry(entry, list, list) {
718 if (!(entry->id == id && entry->cmd_line))
721 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
722 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
724 *devid = entry->devid;
729 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
734 entry->devid = *devid;
735 entry->cmd_line = cmd_line;
737 list_add_tail(&entry->list, list);
742 static int __init add_early_maps(void)
746 for (i = 0; i < early_ioapic_map_size; ++i) {
747 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
748 early_ioapic_map[i].id,
749 &early_ioapic_map[i].devid,
750 early_ioapic_map[i].cmd_line);
755 for (i = 0; i < early_hpet_map_size; ++i) {
756 ret = add_special_device(IVHD_SPECIAL_HPET,
757 early_hpet_map[i].id,
758 &early_hpet_map[i].devid,
759 early_hpet_map[i].cmd_line);
768 * Reads the device exclusion range from ACPI and initializes the IOMMU with
771 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
773 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
775 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
780 * We only can configure exclusion ranges per IOMMU, not
781 * per device. But we can enable the exclusion range per
782 * device. This is done here
784 set_dev_entry_bit(devid, DEV_ENTRY_EX);
785 iommu->exclusion_start = m->range_start;
786 iommu->exclusion_length = m->range_length;
791 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
792 * initializes the hardware and our data structures with it.
794 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
795 struct ivhd_header *h)
798 u8 *end = p, flags = 0;
799 u16 devid = 0, devid_start = 0, devid_to = 0;
800 u32 dev_i, ext_flags = 0;
802 struct ivhd_entry *e;
806 ret = add_early_maps();
811 * First save the recommended feature enable bits from ACPI
813 iommu->acpi_flags = h->flags;
816 * Done. Now parse the device entries
818 p += sizeof(struct ivhd_header);
823 e = (struct ivhd_entry *)p;
827 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
829 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
830 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
832 case IVHD_DEV_SELECT:
834 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
836 PCI_BUS_NUM(e->devid),
842 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
844 case IVHD_DEV_SELECT_RANGE_START:
846 DUMP_printk(" DEV_SELECT_RANGE_START\t "
847 "devid: %02x:%02x.%x flags: %02x\n",
848 PCI_BUS_NUM(e->devid),
853 devid_start = e->devid;
860 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
861 "flags: %02x devid_to: %02x:%02x.%x\n",
862 PCI_BUS_NUM(e->devid),
866 PCI_BUS_NUM(e->ext >> 8),
867 PCI_SLOT(e->ext >> 8),
868 PCI_FUNC(e->ext >> 8));
871 devid_to = e->ext >> 8;
872 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
873 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
874 amd_iommu_alias_table[devid] = devid_to;
876 case IVHD_DEV_ALIAS_RANGE:
878 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
879 "devid: %02x:%02x.%x flags: %02x "
880 "devid_to: %02x:%02x.%x\n",
881 PCI_BUS_NUM(e->devid),
885 PCI_BUS_NUM(e->ext >> 8),
886 PCI_SLOT(e->ext >> 8),
887 PCI_FUNC(e->ext >> 8));
889 devid_start = e->devid;
891 devid_to = e->ext >> 8;
895 case IVHD_DEV_EXT_SELECT:
897 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
898 "flags: %02x ext: %08x\n",
899 PCI_BUS_NUM(e->devid),
905 set_dev_entry_from_acpi(iommu, devid, e->flags,
908 case IVHD_DEV_EXT_SELECT_RANGE:
910 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
911 "%02x:%02x.%x flags: %02x ext: %08x\n",
912 PCI_BUS_NUM(e->devid),
917 devid_start = e->devid;
922 case IVHD_DEV_RANGE_END:
924 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
925 PCI_BUS_NUM(e->devid),
930 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
932 amd_iommu_alias_table[dev_i] = devid_to;
933 set_dev_entry_from_acpi(iommu,
934 devid_to, flags, ext_flags);
936 set_dev_entry_from_acpi(iommu, dev_i,
940 case IVHD_DEV_SPECIAL: {
946 handle = e->ext & 0xff;
947 devid = (e->ext >> 8) & 0xffff;
948 type = (e->ext >> 24) & 0xff;
950 if (type == IVHD_SPECIAL_IOAPIC)
952 else if (type == IVHD_SPECIAL_HPET)
957 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
963 ret = add_special_device(type, handle, &devid, false);
968 * add_special_device might update the devid in case a
969 * command-line override is present. So call
970 * set_dev_entry_from_acpi after add_special_device.
972 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
980 p += ivhd_entry_length(p);
986 static void __init free_iommu_one(struct amd_iommu *iommu)
988 free_command_buffer(iommu);
989 free_event_buffer(iommu);
991 iommu_unmap_mmio_space(iommu);
994 static void __init free_iommu_all(void)
996 struct amd_iommu *iommu, *next;
998 for_each_iommu_safe(iommu, next) {
999 list_del(&iommu->list);
1000 free_iommu_one(iommu);
1006 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1008 * BIOS should disable L2B micellaneous clock gating by setting
1009 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1011 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1015 if ((boot_cpu_data.x86 != 0x15) ||
1016 (boot_cpu_data.x86_model < 0x10) ||
1017 (boot_cpu_data.x86_model > 0x1f))
1020 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1021 pci_read_config_dword(iommu->dev, 0xf4, &value);
1026 /* Select NB indirect register 0x90 and enable writing */
1027 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1029 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1030 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1031 dev_name(&iommu->dev->dev));
1033 /* Clear the enable writing bit */
1034 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1038 * This function clues the initialization function for one IOMMU
1039 * together and also allocates the command buffer and programs the
1040 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1042 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1046 spin_lock_init(&iommu->lock);
1048 /* Add IOMMU to internal data structures */
1049 list_add_tail(&iommu->list, &amd_iommu_list);
1050 iommu->index = amd_iommus_present++;
1052 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1053 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1057 /* Index is fine - add IOMMU to the array */
1058 amd_iommus[iommu->index] = iommu;
1061 * Copy data from ACPI table entry to the iommu struct
1063 iommu->devid = h->devid;
1064 iommu->cap_ptr = h->cap_ptr;
1065 iommu->pci_seg = h->pci_seg;
1066 iommu->mmio_phys = h->mmio_phys;
1068 /* Check if IVHD EFR contains proper max banks/counters */
1069 if ((h->efr != 0) &&
1070 ((h->efr & (0xF << 13)) != 0) &&
1071 ((h->efr & (0x3F << 17)) != 0)) {
1072 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1074 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1077 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1078 iommu->mmio_phys_end);
1079 if (!iommu->mmio_base)
1082 if (alloc_command_buffer(iommu))
1085 if (alloc_event_buffer(iommu))
1088 iommu->int_enabled = false;
1090 ret = init_iommu_from_acpi(iommu, h);
1094 ret = amd_iommu_create_irq_domain(iommu);
1099 * Make sure IOMMU is not considered to translate itself. The IVRS
1100 * table tells us so, but this is a lie!
1102 amd_iommu_rlookup_table[iommu->devid] = NULL;
1108 * Iterates over all IOMMU entries in the ACPI table, allocates the
1109 * IOMMU structure and initializes it with init_iommu_one()
1111 static int __init init_iommu_all(struct acpi_table_header *table)
1113 u8 *p = (u8 *)table, *end = (u8 *)table;
1114 struct ivhd_header *h;
1115 struct amd_iommu *iommu;
1118 end += table->length;
1119 p += IVRS_HEADER_LENGTH;
1122 h = (struct ivhd_header *)p;
1124 case ACPI_IVHD_TYPE:
1126 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1127 "seg: %d flags: %01x info %04x\n",
1128 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1129 PCI_FUNC(h->devid), h->cap_ptr,
1130 h->pci_seg, h->flags, h->info);
1131 DUMP_printk(" mmio-addr: %016llx\n",
1134 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1138 ret = init_iommu_one(iommu, h);
1154 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1156 u64 val = 0xabcd, val2 = 0;
1158 if (!iommu_feature(iommu, FEATURE_PC))
1161 amd_iommu_pc_present = true;
1163 /* Check if the performance counters can be written to */
1164 if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val, true)) ||
1165 (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2, false)) ||
1167 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1168 amd_iommu_pc_present = false;
1172 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1174 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1175 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1176 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1179 static ssize_t amd_iommu_show_cap(struct device *dev,
1180 struct device_attribute *attr,
1183 struct amd_iommu *iommu = dev_get_drvdata(dev);
1184 return sprintf(buf, "%x\n", iommu->cap);
1186 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1188 static ssize_t amd_iommu_show_features(struct device *dev,
1189 struct device_attribute *attr,
1192 struct amd_iommu *iommu = dev_get_drvdata(dev);
1193 return sprintf(buf, "%llx\n", iommu->features);
1195 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1197 static struct attribute *amd_iommu_attrs[] = {
1199 &dev_attr_features.attr,
1203 static struct attribute_group amd_iommu_group = {
1204 .name = "amd-iommu",
1205 .attrs = amd_iommu_attrs,
1208 static const struct attribute_group *amd_iommu_groups[] = {
1213 static int iommu_init_pci(struct amd_iommu *iommu)
1215 int cap_ptr = iommu->cap_ptr;
1216 u32 range, misc, low, high;
1218 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1219 iommu->devid & 0xff);
1223 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1225 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1227 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1230 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1231 amd_iommu_iotlb_sup = false;
1233 /* read extended feature bits */
1234 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1235 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1237 iommu->features = ((u64)high << 32) | low;
1239 if (iommu_feature(iommu, FEATURE_GT)) {
1244 pasmax = iommu->features & FEATURE_PASID_MASK;
1245 pasmax >>= FEATURE_PASID_SHIFT;
1246 max_pasid = (1 << (pasmax + 1)) - 1;
1248 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1250 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1252 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1253 glxval >>= FEATURE_GLXVAL_SHIFT;
1255 if (amd_iommu_max_glx_val == -1)
1256 amd_iommu_max_glx_val = glxval;
1258 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1261 if (iommu_feature(iommu, FEATURE_GT) &&
1262 iommu_feature(iommu, FEATURE_PPR)) {
1263 iommu->is_iommu_v2 = true;
1264 amd_iommu_v2_present = true;
1267 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1270 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1271 amd_iommu_np_cache = true;
1273 init_iommu_perf_ctr(iommu);
1275 if (is_rd890_iommu(iommu->dev)) {
1278 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1282 * Some rd890 systems may not be fully reconfigured by the
1283 * BIOS, so it's necessary for us to store this information so
1284 * it can be reprogrammed on resume
1286 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1287 &iommu->stored_addr_lo);
1288 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1289 &iommu->stored_addr_hi);
1291 /* Low bit locks writes to configuration space */
1292 iommu->stored_addr_lo &= ~1;
1294 for (i = 0; i < 6; i++)
1295 for (j = 0; j < 0x12; j++)
1296 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1298 for (i = 0; i < 0x83; i++)
1299 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1302 amd_iommu_erratum_746_workaround(iommu);
1304 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1305 amd_iommu_groups, "ivhd%d",
1308 return pci_enable_device(iommu->dev);
1311 static void print_iommu_info(void)
1313 static const char * const feat_str[] = {
1314 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1315 "IA", "GA", "HE", "PC"
1317 struct amd_iommu *iommu;
1319 for_each_iommu(iommu) {
1322 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1323 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1325 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1326 pr_info("AMD-Vi: Extended features: ");
1327 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1328 if (iommu_feature(iommu, (1ULL << i)))
1329 pr_cont(" %s", feat_str[i]);
1334 if (irq_remapping_enabled)
1335 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1338 static int __init amd_iommu_init_pci(void)
1340 struct amd_iommu *iommu;
1343 for_each_iommu(iommu) {
1344 ret = iommu_init_pci(iommu);
1349 init_device_table_dma();
1351 for_each_iommu(iommu)
1352 iommu_flush_all_caches(iommu);
1354 ret = amd_iommu_init_api();
1362 /****************************************************************************
1364 * The following functions initialize the MSI interrupts for all IOMMUs
1365 * in the system. It's a bit challenging because there could be multiple
1366 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1369 ****************************************************************************/
1371 static int iommu_setup_msi(struct amd_iommu *iommu)
1375 r = pci_enable_msi(iommu->dev);
1379 r = request_threaded_irq(iommu->dev->irq,
1380 amd_iommu_int_handler,
1381 amd_iommu_int_thread,
1386 pci_disable_msi(iommu->dev);
1390 iommu->int_enabled = true;
1395 static int iommu_init_msi(struct amd_iommu *iommu)
1399 if (iommu->int_enabled)
1402 if (iommu->dev->msi_cap)
1403 ret = iommu_setup_msi(iommu);
1411 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1413 if (iommu->ppr_log != NULL)
1414 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1419 /****************************************************************************
1421 * The next functions belong to the third pass of parsing the ACPI
1422 * table. In this last pass the memory mapping requirements are
1423 * gathered (like exclusion and unity mapping ranges).
1425 ****************************************************************************/
1427 static void __init free_unity_maps(void)
1429 struct unity_map_entry *entry, *next;
1431 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1432 list_del(&entry->list);
1437 /* called when we find an exclusion range definition in ACPI */
1438 static int __init init_exclusion_range(struct ivmd_header *m)
1443 case ACPI_IVMD_TYPE:
1444 set_device_exclusion_range(m->devid, m);
1446 case ACPI_IVMD_TYPE_ALL:
1447 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1448 set_device_exclusion_range(i, m);
1450 case ACPI_IVMD_TYPE_RANGE:
1451 for (i = m->devid; i <= m->aux; ++i)
1452 set_device_exclusion_range(i, m);
1461 /* called for unity map ACPI definition */
1462 static int __init init_unity_map_range(struct ivmd_header *m)
1464 struct unity_map_entry *e = NULL;
1467 e = kzalloc(sizeof(*e), GFP_KERNEL);
1475 case ACPI_IVMD_TYPE:
1476 s = "IVMD_TYPEi\t\t\t";
1477 e->devid_start = e->devid_end = m->devid;
1479 case ACPI_IVMD_TYPE_ALL:
1480 s = "IVMD_TYPE_ALL\t\t";
1482 e->devid_end = amd_iommu_last_bdf;
1484 case ACPI_IVMD_TYPE_RANGE:
1485 s = "IVMD_TYPE_RANGE\t\t";
1486 e->devid_start = m->devid;
1487 e->devid_end = m->aux;
1490 e->address_start = PAGE_ALIGN(m->range_start);
1491 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1492 e->prot = m->flags >> 1;
1494 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1495 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1496 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1497 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1498 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1499 e->address_start, e->address_end, m->flags);
1501 list_add_tail(&e->list, &amd_iommu_unity_map);
1506 /* iterates over all memory definitions we find in the ACPI table */
1507 static int __init init_memory_definitions(struct acpi_table_header *table)
1509 u8 *p = (u8 *)table, *end = (u8 *)table;
1510 struct ivmd_header *m;
1512 end += table->length;
1513 p += IVRS_HEADER_LENGTH;
1516 m = (struct ivmd_header *)p;
1517 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1518 init_exclusion_range(m);
1519 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1520 init_unity_map_range(m);
1529 * Init the device table to not allow DMA access for devices and
1530 * suppress all page faults
1532 static void init_device_table_dma(void)
1536 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1537 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1538 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1542 static void __init uninit_device_table_dma(void)
1546 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1547 amd_iommu_dev_table[devid].data[0] = 0ULL;
1548 amd_iommu_dev_table[devid].data[1] = 0ULL;
1552 static void init_device_table(void)
1556 if (!amd_iommu_irq_remap)
1559 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1560 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1563 static void iommu_init_flags(struct amd_iommu *iommu)
1565 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1566 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1567 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1569 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1570 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1571 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1573 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1574 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1575 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1577 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1578 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1579 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1582 * make IOMMU memory accesses cache coherent
1584 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1586 /* Set IOTLB invalidation timeout to 1s */
1587 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1590 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1593 u32 ioc_feature_control;
1594 struct pci_dev *pdev = iommu->root_pdev;
1596 /* RD890 BIOSes may not have completely reconfigured the iommu */
1597 if (!is_rd890_iommu(iommu->dev) || !pdev)
1601 * First, we need to ensure that the iommu is enabled. This is
1602 * controlled by a register in the northbridge
1605 /* Select Northbridge indirect register 0x75 and enable writing */
1606 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1607 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1609 /* Enable the iommu */
1610 if (!(ioc_feature_control & 0x1))
1611 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1613 /* Restore the iommu BAR */
1614 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1615 iommu->stored_addr_lo);
1616 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1617 iommu->stored_addr_hi);
1619 /* Restore the l1 indirect regs for each of the 6 l1s */
1620 for (i = 0; i < 6; i++)
1621 for (j = 0; j < 0x12; j++)
1622 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1624 /* Restore the l2 indirect regs */
1625 for (i = 0; i < 0x83; i++)
1626 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1628 /* Lock PCI setup registers */
1629 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1630 iommu->stored_addr_lo | 1);
1634 * This function finally enables all IOMMUs found in the system after
1635 * they have been initialized
1637 static void early_enable_iommus(void)
1639 struct amd_iommu *iommu;
1641 for_each_iommu(iommu) {
1642 iommu_disable(iommu);
1643 iommu_init_flags(iommu);
1644 iommu_set_device_table(iommu);
1645 iommu_enable_command_buffer(iommu);
1646 iommu_enable_event_buffer(iommu);
1647 iommu_set_exclusion_range(iommu);
1648 iommu_enable(iommu);
1649 iommu_flush_all_caches(iommu);
1653 static void enable_iommus_v2(void)
1655 struct amd_iommu *iommu;
1657 for_each_iommu(iommu) {
1658 iommu_enable_ppr_log(iommu);
1659 iommu_enable_gt(iommu);
1663 static void enable_iommus(void)
1665 early_enable_iommus();
1670 static void disable_iommus(void)
1672 struct amd_iommu *iommu;
1674 for_each_iommu(iommu)
1675 iommu_disable(iommu);
1679 * Suspend/Resume support
1680 * disable suspend until real resume implemented
1683 static void amd_iommu_resume(void)
1685 struct amd_iommu *iommu;
1687 for_each_iommu(iommu)
1688 iommu_apply_resume_quirks(iommu);
1690 /* re-load the hardware */
1693 amd_iommu_enable_interrupts();
1696 static int amd_iommu_suspend(void)
1698 /* disable IOMMUs to go out of the way for BIOS */
1704 static struct syscore_ops amd_iommu_syscore_ops = {
1705 .suspend = amd_iommu_suspend,
1706 .resume = amd_iommu_resume,
1709 static void __init free_on_init_error(void)
1711 free_pages((unsigned long)irq_lookup_table,
1712 get_order(rlookup_table_size));
1714 kmem_cache_destroy(amd_iommu_irq_cache);
1715 amd_iommu_irq_cache = NULL;
1717 free_pages((unsigned long)amd_iommu_rlookup_table,
1718 get_order(rlookup_table_size));
1720 free_pages((unsigned long)amd_iommu_alias_table,
1721 get_order(alias_table_size));
1723 free_pages((unsigned long)amd_iommu_dev_table,
1724 get_order(dev_table_size));
1728 #ifdef CONFIG_GART_IOMMU
1730 * We failed to initialize the AMD IOMMU - try fallback to GART
1738 /* SB IOAPIC is always on this device in AMD systems */
1739 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1741 static bool __init check_ioapic_information(void)
1743 const char *fw_bug = FW_BUG;
1744 bool ret, has_sb_ioapic;
1747 has_sb_ioapic = false;
1751 * If we have map overrides on the kernel command line the
1752 * messages in this function might not describe firmware bugs
1753 * anymore - so be careful
1758 for (idx = 0; idx < nr_ioapics; idx++) {
1759 int devid, id = mpc_ioapic_id(idx);
1761 devid = get_ioapic_devid(id);
1763 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1766 } else if (devid == IOAPIC_SB_DEVID) {
1767 has_sb_ioapic = true;
1772 if (!has_sb_ioapic) {
1774 * We expect the SB IOAPIC to be listed in the IVRS
1775 * table. The system timer is connected to the SB IOAPIC
1776 * and if we don't have it in the list the system will
1777 * panic at boot time. This situation usually happens
1778 * when the BIOS is buggy and provides us the wrong
1779 * device id for the IOAPIC in the system.
1781 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
1785 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1790 static void __init free_dma_resources(void)
1792 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1793 get_order(MAX_DOMAIN_ID/8));
1799 * This is the hardware init function for AMD IOMMU in the system.
1800 * This function is called either from amd_iommu_init or from the interrupt
1801 * remapping setup code.
1803 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1806 * 1 pass) Find the highest PCI device id the driver has to handle.
1807 * Upon this information the size of the data structures is
1808 * determined that needs to be allocated.
1810 * 2 pass) Initialize the data structures just allocated with the
1811 * information in the ACPI table about available AMD IOMMUs
1812 * in the system. It also maps the PCI devices in the
1813 * system to specific IOMMUs
1815 * 3 pass) After the basic data structures are allocated and
1816 * initialized we update them with information about memory
1817 * remapping requirements parsed out of the ACPI table in
1820 * After everything is set up the IOMMUs are enabled and the necessary
1821 * hotplug and suspend notifiers are registered.
1823 static int __init early_amd_iommu_init(void)
1825 struct acpi_table_header *ivrs_base;
1826 acpi_size ivrs_size;
1830 if (!amd_iommu_detected)
1833 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1834 if (status == AE_NOT_FOUND)
1836 else if (ACPI_FAILURE(status)) {
1837 const char *err = acpi_format_exception(status);
1838 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1843 * First parse ACPI tables to find the largest Bus/Dev/Func
1844 * we need to handle. Upon this information the shared data
1845 * structures for the IOMMUs in the system will be allocated
1847 ret = find_last_devid_acpi(ivrs_base);
1851 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1852 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1853 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1855 /* Device table - directly used by all IOMMUs */
1857 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1858 get_order(dev_table_size));
1859 if (amd_iommu_dev_table == NULL)
1863 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1864 * IOMMU see for that device
1866 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1867 get_order(alias_table_size));
1868 if (amd_iommu_alias_table == NULL)
1871 /* IOMMU rlookup table - find the IOMMU for a specific device */
1872 amd_iommu_rlookup_table = (void *)__get_free_pages(
1873 GFP_KERNEL | __GFP_ZERO,
1874 get_order(rlookup_table_size));
1875 if (amd_iommu_rlookup_table == NULL)
1878 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1879 GFP_KERNEL | __GFP_ZERO,
1880 get_order(MAX_DOMAIN_ID/8));
1881 if (amd_iommu_pd_alloc_bitmap == NULL)
1885 * let all alias entries point to itself
1887 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1888 amd_iommu_alias_table[i] = i;
1891 * never allocate domain 0 because its used as the non-allocated and
1892 * error value placeholder
1894 amd_iommu_pd_alloc_bitmap[0] = 1;
1896 spin_lock_init(&amd_iommu_pd_lock);
1899 * now the data structures are allocated and basically initialized
1900 * start the real acpi table scan
1902 ret = init_iommu_all(ivrs_base);
1906 if (amd_iommu_irq_remap)
1907 amd_iommu_irq_remap = check_ioapic_information();
1909 if (amd_iommu_irq_remap) {
1911 * Interrupt remapping enabled, create kmem_cache for the
1915 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
1916 MAX_IRQS_PER_TABLE * sizeof(u32),
1917 IRQ_TABLE_ALIGNMENT,
1919 if (!amd_iommu_irq_cache)
1922 irq_lookup_table = (void *)__get_free_pages(
1923 GFP_KERNEL | __GFP_ZERO,
1924 get_order(rlookup_table_size));
1925 if (!irq_lookup_table)
1929 ret = init_memory_definitions(ivrs_base);
1933 /* init the device table */
1934 init_device_table();
1937 /* Don't leak any ACPI memory */
1938 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1944 static int amd_iommu_enable_interrupts(void)
1946 struct amd_iommu *iommu;
1949 for_each_iommu(iommu) {
1950 ret = iommu_init_msi(iommu);
1959 static bool detect_ivrs(void)
1961 struct acpi_table_header *ivrs_base;
1962 acpi_size ivrs_size;
1965 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1966 if (status == AE_NOT_FOUND)
1968 else if (ACPI_FAILURE(status)) {
1969 const char *err = acpi_format_exception(status);
1970 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1974 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1976 /* Make sure ACS will be enabled during PCI probe */
1982 /****************************************************************************
1984 * AMD IOMMU Initialization State Machine
1986 ****************************************************************************/
1988 static int __init state_next(void)
1992 switch (init_state) {
1993 case IOMMU_START_STATE:
1994 if (!detect_ivrs()) {
1995 init_state = IOMMU_NOT_FOUND;
1998 init_state = IOMMU_IVRS_DETECTED;
2001 case IOMMU_IVRS_DETECTED:
2002 ret = early_amd_iommu_init();
2003 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2005 case IOMMU_ACPI_FINISHED:
2006 early_enable_iommus();
2007 register_syscore_ops(&amd_iommu_syscore_ops);
2008 x86_platform.iommu_shutdown = disable_iommus;
2009 init_state = IOMMU_ENABLED;
2012 ret = amd_iommu_init_pci();
2013 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2016 case IOMMU_PCI_INIT:
2017 ret = amd_iommu_enable_interrupts();
2018 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2020 case IOMMU_INTERRUPTS_EN:
2021 ret = amd_iommu_init_dma_ops();
2022 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2025 init_state = IOMMU_INITIALIZED;
2027 case IOMMU_INITIALIZED:
2030 case IOMMU_NOT_FOUND:
2031 case IOMMU_INIT_ERROR:
2032 /* Error states => do nothing */
2043 static int __init iommu_go_to_state(enum iommu_init_state state)
2047 while (init_state != state) {
2049 if (init_state == IOMMU_NOT_FOUND ||
2050 init_state == IOMMU_INIT_ERROR)
2057 #ifdef CONFIG_IRQ_REMAP
2058 int __init amd_iommu_prepare(void)
2062 amd_iommu_irq_remap = true;
2064 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2067 return amd_iommu_irq_remap ? 0 : -ENODEV;
2070 int __init amd_iommu_enable(void)
2074 ret = iommu_go_to_state(IOMMU_ENABLED);
2078 irq_remapping_enabled = 1;
2083 void amd_iommu_disable(void)
2085 amd_iommu_suspend();
2088 int amd_iommu_reenable(int mode)
2095 int __init amd_iommu_enable_faulting(void)
2097 /* We enable MSI later when PCI is initialized */
2103 * This is the core init function for AMD IOMMU hardware in the system.
2104 * This function is called from the generic x86 DMA layer initialization
2107 static int __init amd_iommu_init(void)
2111 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2113 free_dma_resources();
2114 if (!irq_remapping_enabled) {
2116 free_on_init_error();
2118 struct amd_iommu *iommu;
2120 uninit_device_table_dma();
2121 for_each_iommu(iommu)
2122 iommu_flush_all_caches(iommu);
2129 /****************************************************************************
2131 * Early detect code. This code runs at IOMMU detection time in the DMA
2132 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2135 ****************************************************************************/
2136 int __init amd_iommu_detect(void)
2140 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2143 if (amd_iommu_disabled)
2146 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2150 amd_iommu_detected = true;
2152 x86_init.iommu.iommu_init = amd_iommu_init;
2157 /****************************************************************************
2159 * Parsing functions for the AMD IOMMU specific kernel command line
2162 ****************************************************************************/
2164 static int __init parse_amd_iommu_dump(char *str)
2166 amd_iommu_dump = true;
2171 static int __init parse_amd_iommu_options(char *str)
2173 for (; *str; ++str) {
2174 if (strncmp(str, "fullflush", 9) == 0)
2175 amd_iommu_unmap_flush = true;
2176 if (strncmp(str, "off", 3) == 0)
2177 amd_iommu_disabled = true;
2178 if (strncmp(str, "force_isolation", 15) == 0)
2179 amd_iommu_force_isolation = true;
2185 static int __init parse_ivrs_ioapic(char *str)
2187 unsigned int bus, dev, fn;
2191 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2194 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2198 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2199 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2204 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2206 cmdline_maps = true;
2207 i = early_ioapic_map_size++;
2208 early_ioapic_map[i].id = id;
2209 early_ioapic_map[i].devid = devid;
2210 early_ioapic_map[i].cmd_line = true;
2215 static int __init parse_ivrs_hpet(char *str)
2217 unsigned int bus, dev, fn;
2221 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2224 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2228 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2229 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2234 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2236 cmdline_maps = true;
2237 i = early_hpet_map_size++;
2238 early_hpet_map[i].id = id;
2239 early_hpet_map[i].devid = devid;
2240 early_hpet_map[i].cmd_line = true;
2245 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2246 __setup("amd_iommu=", parse_amd_iommu_options);
2247 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2248 __setup("ivrs_hpet", parse_ivrs_hpet);
2250 IOMMU_INIT_FINISH(amd_iommu_detect,
2251 gart_iommu_hole_init,
2255 bool amd_iommu_v2_supported(void)
2257 return amd_iommu_v2_present;
2259 EXPORT_SYMBOL(amd_iommu_v2_supported);
2261 /****************************************************************************
2263 * IOMMU EFR Performance Counter support functionality. This code allows
2264 * access to the IOMMU PC functionality.
2266 ****************************************************************************/
2268 u8 amd_iommu_pc_get_max_banks(u16 devid)
2270 struct amd_iommu *iommu;
2273 /* locate the iommu governing the devid */
2274 iommu = amd_iommu_rlookup_table[devid];
2276 ret = iommu->max_banks;
2280 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2282 bool amd_iommu_pc_supported(void)
2284 return amd_iommu_pc_present;
2286 EXPORT_SYMBOL(amd_iommu_pc_supported);
2288 u8 amd_iommu_pc_get_max_counters(u16 devid)
2290 struct amd_iommu *iommu;
2293 /* locate the iommu governing the devid */
2294 iommu = amd_iommu_rlookup_table[devid];
2296 ret = iommu->max_counters;
2300 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2302 int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2303 u64 *value, bool is_write)
2305 struct amd_iommu *iommu;
2309 /* Make sure the IOMMU PC resource is available */
2310 if (!amd_iommu_pc_present)
2313 /* Locate the iommu associated with the device ID */
2314 iommu = amd_iommu_rlookup_table[devid];
2316 /* Check for valid iommu and pc register indexing */
2317 if (WARN_ON((iommu == NULL) || (fxn > 0x28) || (fxn & 7)))
2320 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2322 /* Limit the offset to the hw defined mmio region aperture */
2323 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2324 (iommu->max_counters << 8) | 0x28);
2325 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2326 (offset > max_offset_lim))
2330 writel((u32)*value, iommu->mmio_base + offset);
2331 writel((*value >> 32), iommu->mmio_base + offset + 4);
2333 *value = readl(iommu->mmio_base + offset + 4);
2335 *value = readl(iommu->mmio_base + offset);
2340 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);