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.
104 /* Following only valid on IVHD type 11h and 40h */
105 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
107 } __attribute__((packed));
110 * A device entry describing which devices a specific IOMMU translates and
111 * which requestor ids they use.
118 } __attribute__((packed));
121 * An AMD IOMMU memory definition structure. It defines things like exclusion
122 * ranges for devices and regions that should be unity mapped.
133 } __attribute__((packed));
136 bool amd_iommu_irq_remap __read_mostly;
138 static bool amd_iommu_detected;
139 static bool __initdata amd_iommu_disabled;
141 u16 amd_iommu_last_bdf; /* largest PCI device id we have
143 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
145 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
147 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
150 /* Array to assign indices to IOMMUs*/
151 struct amd_iommu *amd_iommus[MAX_IOMMUS];
152 int amd_iommus_present;
154 /* IOMMUs have a non-present cache? */
155 bool amd_iommu_np_cache __read_mostly;
156 bool amd_iommu_iotlb_sup __read_mostly = true;
158 u32 amd_iommu_max_pasid __read_mostly = ~0;
160 bool amd_iommu_v2_present __read_mostly;
161 static bool amd_iommu_pc_present __read_mostly;
163 bool amd_iommu_force_isolation __read_mostly;
166 * List of protection domains - used during resume
168 LIST_HEAD(amd_iommu_pd_list);
169 spinlock_t amd_iommu_pd_lock;
172 * Pointer to the device table which is shared by all AMD IOMMUs
173 * it is indexed by the PCI device id or the HT unit id and contains
174 * information about the domain the device belongs to as well as the
175 * page table root pointer.
177 struct dev_table_entry *amd_iommu_dev_table;
180 * The alias table is a driver specific data structure which contains the
181 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
182 * More than one device can share the same requestor id.
184 u16 *amd_iommu_alias_table;
187 * The rlookup table is used to find the IOMMU which is responsible
188 * for a specific device. It is also indexed by the PCI device id.
190 struct amd_iommu **amd_iommu_rlookup_table;
193 * This table is used to find the irq remapping table for a given device id
196 struct irq_remap_table **irq_lookup_table;
199 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
200 * to know which ones are already in use.
202 unsigned long *amd_iommu_pd_alloc_bitmap;
204 static u32 dev_table_size; /* size of the device table */
205 static u32 alias_table_size; /* size of the alias table */
206 static u32 rlookup_table_size; /* size if the rlookup table */
208 enum iommu_init_state {
221 /* Early ioapic and hpet maps from kernel command line */
222 #define EARLY_MAP_SIZE 4
223 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
224 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
225 static int __initdata early_ioapic_map_size;
226 static int __initdata early_hpet_map_size;
227 static bool __initdata cmdline_maps;
229 static enum iommu_init_state init_state = IOMMU_START_STATE;
231 static int amd_iommu_enable_interrupts(void);
232 static int __init iommu_go_to_state(enum iommu_init_state state);
233 static void init_device_table_dma(void);
235 static int iommu_pc_get_set_reg_val(struct amd_iommu *iommu,
236 u8 bank, u8 cntr, u8 fxn,
237 u64 *value, bool is_write);
239 static inline void update_last_devid(u16 devid)
241 if (devid > amd_iommu_last_bdf)
242 amd_iommu_last_bdf = devid;
245 static inline unsigned long tbl_size(int entry_size)
247 unsigned shift = PAGE_SHIFT +
248 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
253 /* Access to l1 and l2 indexed register spaces */
255 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
259 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
260 pci_read_config_dword(iommu->dev, 0xfc, &val);
264 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
266 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
267 pci_write_config_dword(iommu->dev, 0xfc, val);
268 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
271 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
275 pci_write_config_dword(iommu->dev, 0xf0, address);
276 pci_read_config_dword(iommu->dev, 0xf4, &val);
280 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
282 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
283 pci_write_config_dword(iommu->dev, 0xf4, val);
286 /****************************************************************************
288 * AMD IOMMU MMIO register space handling functions
290 * These functions are used to program the IOMMU device registers in
291 * MMIO space required for that driver.
293 ****************************************************************************/
296 * This function set the exclusion range in the IOMMU. DMA accesses to the
297 * exclusion range are passed through untranslated
299 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
301 u64 start = iommu->exclusion_start & PAGE_MASK;
302 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
305 if (!iommu->exclusion_start)
308 entry = start | MMIO_EXCL_ENABLE_MASK;
309 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
310 &entry, sizeof(entry));
313 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
314 &entry, sizeof(entry));
317 /* Programs the physical address of the device table into the IOMMU hardware */
318 static void iommu_set_device_table(struct amd_iommu *iommu)
322 BUG_ON(iommu->mmio_base == NULL);
324 entry = virt_to_phys(amd_iommu_dev_table);
325 entry |= (dev_table_size >> 12) - 1;
326 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
327 &entry, sizeof(entry));
330 /* Generic functions to enable/disable certain features of the IOMMU. */
331 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
335 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
337 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
340 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
344 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
346 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
349 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
353 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
354 ctrl &= ~CTRL_INV_TO_MASK;
355 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
356 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
359 /* Function to enable the hardware */
360 static void iommu_enable(struct amd_iommu *iommu)
362 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
365 static void iommu_disable(struct amd_iommu *iommu)
367 /* Disable command buffer */
368 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
370 /* Disable event logging and event interrupts */
371 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
372 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
374 /* Disable IOMMU hardware itself */
375 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
379 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
380 * the system has one.
382 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
384 if (!request_mem_region(address, end, "amd_iommu")) {
385 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
387 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
391 return (u8 __iomem *)ioremap_nocache(address, end);
394 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
396 if (iommu->mmio_base)
397 iounmap(iommu->mmio_base);
398 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
401 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
417 /****************************************************************************
419 * The functions below belong to the first pass of AMD IOMMU ACPI table
420 * parsing. In this pass we try to find out the highest device id this
421 * code has to handle. Upon this information the size of the shared data
422 * structures is determined later.
424 ****************************************************************************/
427 * This function calculates the length of a given IVHD entry
429 static inline int ivhd_entry_length(u8 *ivhd)
431 return 0x04 << (*ivhd >> 6);
435 * After reading the highest device id from the IOMMU PCI capability header
436 * this function looks if there is a higher device id defined in the ACPI table
438 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
440 u8 *p = (void *)h, *end = (void *)h;
441 struct ivhd_entry *dev;
443 u32 ivhd_size = get_ivhd_header_size(h);
446 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
454 dev = (struct ivhd_entry *)p;
457 /* Use maximum BDF value for DEV_ALL */
458 update_last_devid(0xffff);
460 case IVHD_DEV_SELECT:
461 case IVHD_DEV_RANGE_END:
463 case IVHD_DEV_EXT_SELECT:
464 /* all the above subfield types refer to device ids */
465 update_last_devid(dev->devid);
470 p += ivhd_entry_length(p);
479 * Iterate over all IVHD entries in the ACPI table and find the highest device
480 * id which we need to handle. This is the first of three functions which parse
481 * the ACPI table. So we check the checksum here.
483 static int __init find_last_devid_acpi(struct acpi_table_header *table)
486 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
487 struct ivhd_header *h;
490 * Validate checksum here so we don't need to do it when
491 * we actually parse the table
493 for (i = 0; i < table->length; ++i)
496 /* ACPI table corrupt */
499 p += IVRS_HEADER_LENGTH;
501 end += table->length;
503 h = (struct ivhd_header *)p;
506 find_last_devid_from_ivhd(h);
518 /****************************************************************************
520 * The following functions belong to the code path which parses the ACPI table
521 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
522 * data structures, initialize the device/alias/rlookup table and also
523 * basically initialize the hardware.
525 ****************************************************************************/
528 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
529 * write commands to that buffer later and the IOMMU will execute them
532 static int __init alloc_command_buffer(struct amd_iommu *iommu)
534 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
535 get_order(CMD_BUFFER_SIZE));
537 return iommu->cmd_buf ? 0 : -ENOMEM;
541 * This function resets the command buffer if the IOMMU stopped fetching
544 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
546 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
548 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
549 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
551 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
555 * This function writes the command buffer address to the hardware and
558 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
562 BUG_ON(iommu->cmd_buf == NULL);
564 entry = (u64)virt_to_phys(iommu->cmd_buf);
565 entry |= MMIO_CMD_SIZE_512;
567 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
568 &entry, sizeof(entry));
570 amd_iommu_reset_cmd_buffer(iommu);
573 static void __init free_command_buffer(struct amd_iommu *iommu)
575 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
578 /* allocates the memory where the IOMMU will log its events to */
579 static int __init alloc_event_buffer(struct amd_iommu *iommu)
581 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
582 get_order(EVT_BUFFER_SIZE));
584 return iommu->evt_buf ? 0 : -ENOMEM;
587 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
591 BUG_ON(iommu->evt_buf == NULL);
593 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
595 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
596 &entry, sizeof(entry));
598 /* set head and tail to zero manually */
599 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
600 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
602 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
605 static void __init free_event_buffer(struct amd_iommu *iommu)
607 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
610 /* allocates the memory where the IOMMU will log its events to */
611 static int __init alloc_ppr_log(struct amd_iommu *iommu)
613 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
614 get_order(PPR_LOG_SIZE));
616 return iommu->ppr_log ? 0 : -ENOMEM;
619 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
623 if (iommu->ppr_log == NULL)
626 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
628 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
629 &entry, sizeof(entry));
631 /* set head and tail to zero manually */
632 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
633 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
635 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
636 iommu_feature_enable(iommu, CONTROL_PPR_EN);
639 static void __init free_ppr_log(struct amd_iommu *iommu)
641 if (iommu->ppr_log == NULL)
644 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
647 static void iommu_enable_gt(struct amd_iommu *iommu)
649 if (!iommu_feature(iommu, FEATURE_GT))
652 iommu_feature_enable(iommu, CONTROL_GT_EN);
655 /* sets a specific bit in the device table entry. */
656 static void set_dev_entry_bit(u16 devid, u8 bit)
658 int i = (bit >> 6) & 0x03;
659 int _bit = bit & 0x3f;
661 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
664 static int get_dev_entry_bit(u16 devid, u8 bit)
666 int i = (bit >> 6) & 0x03;
667 int _bit = bit & 0x3f;
669 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
673 void amd_iommu_apply_erratum_63(u16 devid)
677 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
678 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
681 set_dev_entry_bit(devid, DEV_ENTRY_IW);
684 /* Writes the specific IOMMU for a device into the rlookup table */
685 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
687 amd_iommu_rlookup_table[devid] = iommu;
691 * This function takes the device specific flags read from the ACPI
692 * table and sets up the device table entry with that information
694 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
695 u16 devid, u32 flags, u32 ext_flags)
697 if (flags & ACPI_DEVFLAG_INITPASS)
698 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
699 if (flags & ACPI_DEVFLAG_EXTINT)
700 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
701 if (flags & ACPI_DEVFLAG_NMI)
702 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
703 if (flags & ACPI_DEVFLAG_SYSMGT1)
704 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
705 if (flags & ACPI_DEVFLAG_SYSMGT2)
706 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
707 if (flags & ACPI_DEVFLAG_LINT0)
708 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
709 if (flags & ACPI_DEVFLAG_LINT1)
710 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
712 amd_iommu_apply_erratum_63(devid);
714 set_iommu_for_device(iommu, devid);
717 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
719 struct devid_map *entry;
720 struct list_head *list;
722 if (type == IVHD_SPECIAL_IOAPIC)
724 else if (type == IVHD_SPECIAL_HPET)
729 list_for_each_entry(entry, list, list) {
730 if (!(entry->id == id && entry->cmd_line))
733 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
734 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
736 *devid = entry->devid;
741 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
746 entry->devid = *devid;
747 entry->cmd_line = cmd_line;
749 list_add_tail(&entry->list, list);
754 static int __init add_early_maps(void)
758 for (i = 0; i < early_ioapic_map_size; ++i) {
759 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
760 early_ioapic_map[i].id,
761 &early_ioapic_map[i].devid,
762 early_ioapic_map[i].cmd_line);
767 for (i = 0; i < early_hpet_map_size; ++i) {
768 ret = add_special_device(IVHD_SPECIAL_HPET,
769 early_hpet_map[i].id,
770 &early_hpet_map[i].devid,
771 early_hpet_map[i].cmd_line);
780 * Reads the device exclusion range from ACPI and initializes the IOMMU with
783 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
785 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
787 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
792 * We only can configure exclusion ranges per IOMMU, not
793 * per device. But we can enable the exclusion range per
794 * device. This is done here
796 set_dev_entry_bit(devid, DEV_ENTRY_EX);
797 iommu->exclusion_start = m->range_start;
798 iommu->exclusion_length = m->range_length;
803 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
804 * initializes the hardware and our data structures with it.
806 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
807 struct ivhd_header *h)
810 u8 *end = p, flags = 0;
811 u16 devid = 0, devid_start = 0, devid_to = 0;
812 u32 dev_i, ext_flags = 0;
814 struct ivhd_entry *e;
819 ret = add_early_maps();
824 * First save the recommended feature enable bits from ACPI
826 iommu->acpi_flags = h->flags;
829 * Done. Now parse the device entries
831 ivhd_size = get_ivhd_header_size(h);
833 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
843 e = (struct ivhd_entry *)p;
847 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
849 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
850 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
852 case IVHD_DEV_SELECT:
854 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
856 PCI_BUS_NUM(e->devid),
862 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
864 case IVHD_DEV_SELECT_RANGE_START:
866 DUMP_printk(" DEV_SELECT_RANGE_START\t "
867 "devid: %02x:%02x.%x flags: %02x\n",
868 PCI_BUS_NUM(e->devid),
873 devid_start = e->devid;
880 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
881 "flags: %02x devid_to: %02x:%02x.%x\n",
882 PCI_BUS_NUM(e->devid),
886 PCI_BUS_NUM(e->ext >> 8),
887 PCI_SLOT(e->ext >> 8),
888 PCI_FUNC(e->ext >> 8));
891 devid_to = e->ext >> 8;
892 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
893 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
894 amd_iommu_alias_table[devid] = devid_to;
896 case IVHD_DEV_ALIAS_RANGE:
898 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
899 "devid: %02x:%02x.%x flags: %02x "
900 "devid_to: %02x:%02x.%x\n",
901 PCI_BUS_NUM(e->devid),
905 PCI_BUS_NUM(e->ext >> 8),
906 PCI_SLOT(e->ext >> 8),
907 PCI_FUNC(e->ext >> 8));
909 devid_start = e->devid;
911 devid_to = e->ext >> 8;
915 case IVHD_DEV_EXT_SELECT:
917 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
918 "flags: %02x ext: %08x\n",
919 PCI_BUS_NUM(e->devid),
925 set_dev_entry_from_acpi(iommu, devid, e->flags,
928 case IVHD_DEV_EXT_SELECT_RANGE:
930 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
931 "%02x:%02x.%x flags: %02x ext: %08x\n",
932 PCI_BUS_NUM(e->devid),
937 devid_start = e->devid;
942 case IVHD_DEV_RANGE_END:
944 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
945 PCI_BUS_NUM(e->devid),
950 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
952 amd_iommu_alias_table[dev_i] = devid_to;
953 set_dev_entry_from_acpi(iommu,
954 devid_to, flags, ext_flags);
956 set_dev_entry_from_acpi(iommu, dev_i,
960 case IVHD_DEV_SPECIAL: {
966 handle = e->ext & 0xff;
967 devid = (e->ext >> 8) & 0xffff;
968 type = (e->ext >> 24) & 0xff;
970 if (type == IVHD_SPECIAL_IOAPIC)
972 else if (type == IVHD_SPECIAL_HPET)
977 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
983 ret = add_special_device(type, handle, &devid, false);
988 * add_special_device might update the devid in case a
989 * command-line override is present. So call
990 * set_dev_entry_from_acpi after add_special_device.
992 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1000 p += ivhd_entry_length(p);
1006 static void __init free_iommu_one(struct amd_iommu *iommu)
1008 free_command_buffer(iommu);
1009 free_event_buffer(iommu);
1010 free_ppr_log(iommu);
1011 iommu_unmap_mmio_space(iommu);
1014 static void __init free_iommu_all(void)
1016 struct amd_iommu *iommu, *next;
1018 for_each_iommu_safe(iommu, next) {
1019 list_del(&iommu->list);
1020 free_iommu_one(iommu);
1026 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1028 * BIOS should disable L2B micellaneous clock gating by setting
1029 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1031 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1035 if ((boot_cpu_data.x86 != 0x15) ||
1036 (boot_cpu_data.x86_model < 0x10) ||
1037 (boot_cpu_data.x86_model > 0x1f))
1040 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1041 pci_read_config_dword(iommu->dev, 0xf4, &value);
1046 /* Select NB indirect register 0x90 and enable writing */
1047 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1049 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1050 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1051 dev_name(&iommu->dev->dev));
1053 /* Clear the enable writing bit */
1054 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1058 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1060 * BIOS should enable ATS write permission check by setting
1061 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1063 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1067 if ((boot_cpu_data.x86 != 0x15) ||
1068 (boot_cpu_data.x86_model < 0x30) ||
1069 (boot_cpu_data.x86_model > 0x3f))
1072 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1073 value = iommu_read_l2(iommu, 0x47);
1078 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1079 iommu_write_l2(iommu, 0x47, value | BIT(0));
1081 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1082 dev_name(&iommu->dev->dev));
1086 * This function clues the initialization function for one IOMMU
1087 * together and also allocates the command buffer and programs the
1088 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1090 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1094 spin_lock_init(&iommu->lock);
1096 /* Add IOMMU to internal data structures */
1097 list_add_tail(&iommu->list, &amd_iommu_list);
1098 iommu->index = amd_iommus_present++;
1100 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1101 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1105 /* Index is fine - add IOMMU to the array */
1106 amd_iommus[iommu->index] = iommu;
1109 * Copy data from ACPI table entry to the iommu struct
1111 iommu->devid = h->devid;
1112 iommu->cap_ptr = h->cap_ptr;
1113 iommu->pci_seg = h->pci_seg;
1114 iommu->mmio_phys = h->mmio_phys;
1118 /* Check if IVHD EFR contains proper max banks/counters */
1119 if ((h->efr_attr != 0) &&
1120 ((h->efr_attr & (0xF << 13)) != 0) &&
1121 ((h->efr_attr & (0x3F << 17)) != 0))
1122 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1124 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1128 if (h->efr_reg & (1 << 9))
1129 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1131 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1137 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1138 iommu->mmio_phys_end);
1139 if (!iommu->mmio_base)
1142 if (alloc_command_buffer(iommu))
1145 if (alloc_event_buffer(iommu))
1148 iommu->int_enabled = false;
1150 ret = init_iommu_from_acpi(iommu, h);
1154 ret = amd_iommu_create_irq_domain(iommu);
1159 * Make sure IOMMU is not considered to translate itself. The IVRS
1160 * table tells us so, but this is a lie!
1162 amd_iommu_rlookup_table[iommu->devid] = NULL;
1168 * Iterates over all IOMMU entries in the ACPI table, allocates the
1169 * IOMMU structure and initializes it with init_iommu_one()
1171 static int __init init_iommu_all(struct acpi_table_header *table)
1173 u8 *p = (u8 *)table, *end = (u8 *)table;
1174 struct ivhd_header *h;
1175 struct amd_iommu *iommu;
1178 end += table->length;
1179 p += IVRS_HEADER_LENGTH;
1182 h = (struct ivhd_header *)p;
1184 case ACPI_IVHD_TYPE:
1186 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1187 "seg: %d flags: %01x info %04x\n",
1188 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1189 PCI_FUNC(h->devid), h->cap_ptr,
1190 h->pci_seg, h->flags, h->info);
1191 DUMP_printk(" mmio-addr: %016llx\n",
1194 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1198 ret = init_iommu_one(iommu, h);
1214 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1216 u64 val = 0xabcd, val2 = 0;
1218 if (!iommu_feature(iommu, FEATURE_PC))
1221 amd_iommu_pc_present = true;
1223 /* Check if the performance counters can be written to */
1224 if ((0 != iommu_pc_get_set_reg_val(iommu, 0, 0, 0, &val, true)) ||
1225 (0 != iommu_pc_get_set_reg_val(iommu, 0, 0, 0, &val2, false)) ||
1227 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1228 amd_iommu_pc_present = false;
1232 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1234 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1235 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1236 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1239 static ssize_t amd_iommu_show_cap(struct device *dev,
1240 struct device_attribute *attr,
1243 struct amd_iommu *iommu = dev_get_drvdata(dev);
1244 return sprintf(buf, "%x\n", iommu->cap);
1246 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1248 static ssize_t amd_iommu_show_features(struct device *dev,
1249 struct device_attribute *attr,
1252 struct amd_iommu *iommu = dev_get_drvdata(dev);
1253 return sprintf(buf, "%llx\n", iommu->features);
1255 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1257 static struct attribute *amd_iommu_attrs[] = {
1259 &dev_attr_features.attr,
1263 static struct attribute_group amd_iommu_group = {
1264 .name = "amd-iommu",
1265 .attrs = amd_iommu_attrs,
1268 static const struct attribute_group *amd_iommu_groups[] = {
1273 static int iommu_init_pci(struct amd_iommu *iommu)
1275 int cap_ptr = iommu->cap_ptr;
1276 u32 range, misc, low, high;
1278 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1279 iommu->devid & 0xff);
1283 /* Prevent binding other PCI device drivers to IOMMU devices */
1284 iommu->dev->match_driver = false;
1286 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1288 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1290 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1293 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1294 amd_iommu_iotlb_sup = false;
1296 /* read extended feature bits */
1297 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1298 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1300 iommu->features = ((u64)high << 32) | low;
1302 if (iommu_feature(iommu, FEATURE_GT)) {
1307 pasmax = iommu->features & FEATURE_PASID_MASK;
1308 pasmax >>= FEATURE_PASID_SHIFT;
1309 max_pasid = (1 << (pasmax + 1)) - 1;
1311 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1313 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1315 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1316 glxval >>= FEATURE_GLXVAL_SHIFT;
1318 if (amd_iommu_max_glx_val == -1)
1319 amd_iommu_max_glx_val = glxval;
1321 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1324 if (iommu_feature(iommu, FEATURE_GT) &&
1325 iommu_feature(iommu, FEATURE_PPR)) {
1326 iommu->is_iommu_v2 = true;
1327 amd_iommu_v2_present = true;
1330 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1333 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1334 amd_iommu_np_cache = true;
1336 init_iommu_perf_ctr(iommu);
1338 if (is_rd890_iommu(iommu->dev)) {
1341 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1345 * Some rd890 systems may not be fully reconfigured by the
1346 * BIOS, so it's necessary for us to store this information so
1347 * it can be reprogrammed on resume
1349 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1350 &iommu->stored_addr_lo);
1351 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1352 &iommu->stored_addr_hi);
1354 /* Low bit locks writes to configuration space */
1355 iommu->stored_addr_lo &= ~1;
1357 for (i = 0; i < 6; i++)
1358 for (j = 0; j < 0x12; j++)
1359 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1361 for (i = 0; i < 0x83; i++)
1362 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1365 amd_iommu_erratum_746_workaround(iommu);
1366 amd_iommu_ats_write_check_workaround(iommu);
1368 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1369 amd_iommu_groups, "ivhd%d",
1372 return pci_enable_device(iommu->dev);
1375 static void print_iommu_info(void)
1377 static const char * const feat_str[] = {
1378 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1379 "IA", "GA", "HE", "PC"
1381 struct amd_iommu *iommu;
1383 for_each_iommu(iommu) {
1386 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1387 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1389 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1390 pr_info("AMD-Vi: Extended features: ");
1391 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1392 if (iommu_feature(iommu, (1ULL << i)))
1393 pr_cont(" %s", feat_str[i]);
1398 if (irq_remapping_enabled)
1399 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1402 static int __init amd_iommu_init_pci(void)
1404 struct amd_iommu *iommu;
1407 for_each_iommu(iommu) {
1408 ret = iommu_init_pci(iommu);
1413 init_device_table_dma();
1415 for_each_iommu(iommu)
1416 iommu_flush_all_caches(iommu);
1418 ret = amd_iommu_init_api();
1426 /****************************************************************************
1428 * The following functions initialize the MSI interrupts for all IOMMUs
1429 * in the system. It's a bit challenging because there could be multiple
1430 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1433 ****************************************************************************/
1435 static int iommu_setup_msi(struct amd_iommu *iommu)
1439 r = pci_enable_msi(iommu->dev);
1443 r = request_threaded_irq(iommu->dev->irq,
1444 amd_iommu_int_handler,
1445 amd_iommu_int_thread,
1450 pci_disable_msi(iommu->dev);
1454 iommu->int_enabled = true;
1459 static int iommu_init_msi(struct amd_iommu *iommu)
1463 if (iommu->int_enabled)
1466 if (iommu->dev->msi_cap)
1467 ret = iommu_setup_msi(iommu);
1475 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1477 if (iommu->ppr_log != NULL)
1478 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1483 /****************************************************************************
1485 * The next functions belong to the third pass of parsing the ACPI
1486 * table. In this last pass the memory mapping requirements are
1487 * gathered (like exclusion and unity mapping ranges).
1489 ****************************************************************************/
1491 static void __init free_unity_maps(void)
1493 struct unity_map_entry *entry, *next;
1495 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1496 list_del(&entry->list);
1501 /* called when we find an exclusion range definition in ACPI */
1502 static int __init init_exclusion_range(struct ivmd_header *m)
1507 case ACPI_IVMD_TYPE:
1508 set_device_exclusion_range(m->devid, m);
1510 case ACPI_IVMD_TYPE_ALL:
1511 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1512 set_device_exclusion_range(i, m);
1514 case ACPI_IVMD_TYPE_RANGE:
1515 for (i = m->devid; i <= m->aux; ++i)
1516 set_device_exclusion_range(i, m);
1525 /* called for unity map ACPI definition */
1526 static int __init init_unity_map_range(struct ivmd_header *m)
1528 struct unity_map_entry *e = NULL;
1531 e = kzalloc(sizeof(*e), GFP_KERNEL);
1539 case ACPI_IVMD_TYPE:
1540 s = "IVMD_TYPEi\t\t\t";
1541 e->devid_start = e->devid_end = m->devid;
1543 case ACPI_IVMD_TYPE_ALL:
1544 s = "IVMD_TYPE_ALL\t\t";
1546 e->devid_end = amd_iommu_last_bdf;
1548 case ACPI_IVMD_TYPE_RANGE:
1549 s = "IVMD_TYPE_RANGE\t\t";
1550 e->devid_start = m->devid;
1551 e->devid_end = m->aux;
1554 e->address_start = PAGE_ALIGN(m->range_start);
1555 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1556 e->prot = m->flags >> 1;
1558 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1559 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1560 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1561 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1562 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1563 e->address_start, e->address_end, m->flags);
1565 list_add_tail(&e->list, &amd_iommu_unity_map);
1570 /* iterates over all memory definitions we find in the ACPI table */
1571 static int __init init_memory_definitions(struct acpi_table_header *table)
1573 u8 *p = (u8 *)table, *end = (u8 *)table;
1574 struct ivmd_header *m;
1576 end += table->length;
1577 p += IVRS_HEADER_LENGTH;
1580 m = (struct ivmd_header *)p;
1581 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1582 init_exclusion_range(m);
1583 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1584 init_unity_map_range(m);
1593 * Init the device table to not allow DMA access for devices and
1594 * suppress all page faults
1596 static void init_device_table_dma(void)
1600 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1601 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1602 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1606 static void __init uninit_device_table_dma(void)
1610 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1611 amd_iommu_dev_table[devid].data[0] = 0ULL;
1612 amd_iommu_dev_table[devid].data[1] = 0ULL;
1616 static void init_device_table(void)
1620 if (!amd_iommu_irq_remap)
1623 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1624 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1627 static void iommu_init_flags(struct amd_iommu *iommu)
1629 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1630 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1631 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1633 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1634 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1635 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1637 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1638 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1639 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1641 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1642 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1643 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1646 * make IOMMU memory accesses cache coherent
1648 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1650 /* Set IOTLB invalidation timeout to 1s */
1651 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1654 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1657 u32 ioc_feature_control;
1658 struct pci_dev *pdev = iommu->root_pdev;
1660 /* RD890 BIOSes may not have completely reconfigured the iommu */
1661 if (!is_rd890_iommu(iommu->dev) || !pdev)
1665 * First, we need to ensure that the iommu is enabled. This is
1666 * controlled by a register in the northbridge
1669 /* Select Northbridge indirect register 0x75 and enable writing */
1670 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1671 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1673 /* Enable the iommu */
1674 if (!(ioc_feature_control & 0x1))
1675 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1677 /* Restore the iommu BAR */
1678 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1679 iommu->stored_addr_lo);
1680 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1681 iommu->stored_addr_hi);
1683 /* Restore the l1 indirect regs for each of the 6 l1s */
1684 for (i = 0; i < 6; i++)
1685 for (j = 0; j < 0x12; j++)
1686 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1688 /* Restore the l2 indirect regs */
1689 for (i = 0; i < 0x83; i++)
1690 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1692 /* Lock PCI setup registers */
1693 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1694 iommu->stored_addr_lo | 1);
1698 * This function finally enables all IOMMUs found in the system after
1699 * they have been initialized
1701 static void early_enable_iommus(void)
1703 struct amd_iommu *iommu;
1705 for_each_iommu(iommu) {
1706 iommu_disable(iommu);
1707 iommu_init_flags(iommu);
1708 iommu_set_device_table(iommu);
1709 iommu_enable_command_buffer(iommu);
1710 iommu_enable_event_buffer(iommu);
1711 iommu_set_exclusion_range(iommu);
1712 iommu_enable(iommu);
1713 iommu_flush_all_caches(iommu);
1717 static void enable_iommus_v2(void)
1719 struct amd_iommu *iommu;
1721 for_each_iommu(iommu) {
1722 iommu_enable_ppr_log(iommu);
1723 iommu_enable_gt(iommu);
1727 static void enable_iommus(void)
1729 early_enable_iommus();
1734 static void disable_iommus(void)
1736 struct amd_iommu *iommu;
1738 for_each_iommu(iommu)
1739 iommu_disable(iommu);
1743 * Suspend/Resume support
1744 * disable suspend until real resume implemented
1747 static void amd_iommu_resume(void)
1749 struct amd_iommu *iommu;
1751 for_each_iommu(iommu)
1752 iommu_apply_resume_quirks(iommu);
1754 /* re-load the hardware */
1757 amd_iommu_enable_interrupts();
1760 static int amd_iommu_suspend(void)
1762 /* disable IOMMUs to go out of the way for BIOS */
1768 static struct syscore_ops amd_iommu_syscore_ops = {
1769 .suspend = amd_iommu_suspend,
1770 .resume = amd_iommu_resume,
1773 static void __init free_on_init_error(void)
1775 free_pages((unsigned long)irq_lookup_table,
1776 get_order(rlookup_table_size));
1778 kmem_cache_destroy(amd_iommu_irq_cache);
1779 amd_iommu_irq_cache = NULL;
1781 free_pages((unsigned long)amd_iommu_rlookup_table,
1782 get_order(rlookup_table_size));
1784 free_pages((unsigned long)amd_iommu_alias_table,
1785 get_order(alias_table_size));
1787 free_pages((unsigned long)amd_iommu_dev_table,
1788 get_order(dev_table_size));
1792 #ifdef CONFIG_GART_IOMMU
1794 * We failed to initialize the AMD IOMMU - try fallback to GART
1802 /* SB IOAPIC is always on this device in AMD systems */
1803 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1805 static bool __init check_ioapic_information(void)
1807 const char *fw_bug = FW_BUG;
1808 bool ret, has_sb_ioapic;
1811 has_sb_ioapic = false;
1815 * If we have map overrides on the kernel command line the
1816 * messages in this function might not describe firmware bugs
1817 * anymore - so be careful
1822 for (idx = 0; idx < nr_ioapics; idx++) {
1823 int devid, id = mpc_ioapic_id(idx);
1825 devid = get_ioapic_devid(id);
1827 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1830 } else if (devid == IOAPIC_SB_DEVID) {
1831 has_sb_ioapic = true;
1836 if (!has_sb_ioapic) {
1838 * We expect the SB IOAPIC to be listed in the IVRS
1839 * table. The system timer is connected to the SB IOAPIC
1840 * and if we don't have it in the list the system will
1841 * panic at boot time. This situation usually happens
1842 * when the BIOS is buggy and provides us the wrong
1843 * device id for the IOAPIC in the system.
1845 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
1849 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1854 static void __init free_dma_resources(void)
1856 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1857 get_order(MAX_DOMAIN_ID/8));
1863 * This is the hardware init function for AMD IOMMU in the system.
1864 * This function is called either from amd_iommu_init or from the interrupt
1865 * remapping setup code.
1867 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1870 * 1 pass) Find the highest PCI device id the driver has to handle.
1871 * Upon this information the size of the data structures is
1872 * determined that needs to be allocated.
1874 * 2 pass) Initialize the data structures just allocated with the
1875 * information in the ACPI table about available AMD IOMMUs
1876 * in the system. It also maps the PCI devices in the
1877 * system to specific IOMMUs
1879 * 3 pass) After the basic data structures are allocated and
1880 * initialized we update them with information about memory
1881 * remapping requirements parsed out of the ACPI table in
1884 * After everything is set up the IOMMUs are enabled and the necessary
1885 * hotplug and suspend notifiers are registered.
1887 static int __init early_amd_iommu_init(void)
1889 struct acpi_table_header *ivrs_base;
1890 acpi_size ivrs_size;
1894 if (!amd_iommu_detected)
1897 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1898 if (status == AE_NOT_FOUND)
1900 else if (ACPI_FAILURE(status)) {
1901 const char *err = acpi_format_exception(status);
1902 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1907 * First parse ACPI tables to find the largest Bus/Dev/Func
1908 * we need to handle. Upon this information the shared data
1909 * structures for the IOMMUs in the system will be allocated
1911 ret = find_last_devid_acpi(ivrs_base);
1915 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1916 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1917 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1919 /* Device table - directly used by all IOMMUs */
1921 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1922 get_order(dev_table_size));
1923 if (amd_iommu_dev_table == NULL)
1927 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1928 * IOMMU see for that device
1930 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1931 get_order(alias_table_size));
1932 if (amd_iommu_alias_table == NULL)
1935 /* IOMMU rlookup table - find the IOMMU for a specific device */
1936 amd_iommu_rlookup_table = (void *)__get_free_pages(
1937 GFP_KERNEL | __GFP_ZERO,
1938 get_order(rlookup_table_size));
1939 if (amd_iommu_rlookup_table == NULL)
1942 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1943 GFP_KERNEL | __GFP_ZERO,
1944 get_order(MAX_DOMAIN_ID/8));
1945 if (amd_iommu_pd_alloc_bitmap == NULL)
1949 * let all alias entries point to itself
1951 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1952 amd_iommu_alias_table[i] = i;
1955 * never allocate domain 0 because its used as the non-allocated and
1956 * error value placeholder
1958 amd_iommu_pd_alloc_bitmap[0] = 1;
1960 spin_lock_init(&amd_iommu_pd_lock);
1963 * now the data structures are allocated and basically initialized
1964 * start the real acpi table scan
1966 ret = init_iommu_all(ivrs_base);
1970 if (amd_iommu_irq_remap)
1971 amd_iommu_irq_remap = check_ioapic_information();
1973 if (amd_iommu_irq_remap) {
1975 * Interrupt remapping enabled, create kmem_cache for the
1979 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
1980 MAX_IRQS_PER_TABLE * sizeof(u32),
1981 IRQ_TABLE_ALIGNMENT,
1983 if (!amd_iommu_irq_cache)
1986 irq_lookup_table = (void *)__get_free_pages(
1987 GFP_KERNEL | __GFP_ZERO,
1988 get_order(rlookup_table_size));
1989 if (!irq_lookup_table)
1993 ret = init_memory_definitions(ivrs_base);
1997 /* init the device table */
1998 init_device_table();
2001 /* Don't leak any ACPI memory */
2002 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2008 static int amd_iommu_enable_interrupts(void)
2010 struct amd_iommu *iommu;
2013 for_each_iommu(iommu) {
2014 ret = iommu_init_msi(iommu);
2023 static bool detect_ivrs(void)
2025 struct acpi_table_header *ivrs_base;
2026 acpi_size ivrs_size;
2029 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
2030 if (status == AE_NOT_FOUND)
2032 else if (ACPI_FAILURE(status)) {
2033 const char *err = acpi_format_exception(status);
2034 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2038 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2040 /* Make sure ACS will be enabled during PCI probe */
2046 /****************************************************************************
2048 * AMD IOMMU Initialization State Machine
2050 ****************************************************************************/
2052 static int __init state_next(void)
2056 switch (init_state) {
2057 case IOMMU_START_STATE:
2058 if (!detect_ivrs()) {
2059 init_state = IOMMU_NOT_FOUND;
2062 init_state = IOMMU_IVRS_DETECTED;
2065 case IOMMU_IVRS_DETECTED:
2066 ret = early_amd_iommu_init();
2067 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2069 case IOMMU_ACPI_FINISHED:
2070 early_enable_iommus();
2071 register_syscore_ops(&amd_iommu_syscore_ops);
2072 x86_platform.iommu_shutdown = disable_iommus;
2073 init_state = IOMMU_ENABLED;
2076 ret = amd_iommu_init_pci();
2077 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2080 case IOMMU_PCI_INIT:
2081 ret = amd_iommu_enable_interrupts();
2082 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2084 case IOMMU_INTERRUPTS_EN:
2085 ret = amd_iommu_init_dma_ops();
2086 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2089 init_state = IOMMU_INITIALIZED;
2091 case IOMMU_INITIALIZED:
2094 case IOMMU_NOT_FOUND:
2095 case IOMMU_INIT_ERROR:
2096 /* Error states => do nothing */
2107 static int __init iommu_go_to_state(enum iommu_init_state state)
2111 while (init_state != state) {
2113 if (init_state == IOMMU_NOT_FOUND ||
2114 init_state == IOMMU_INIT_ERROR)
2121 #ifdef CONFIG_IRQ_REMAP
2122 int __init amd_iommu_prepare(void)
2126 amd_iommu_irq_remap = true;
2128 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2131 return amd_iommu_irq_remap ? 0 : -ENODEV;
2134 int __init amd_iommu_enable(void)
2138 ret = iommu_go_to_state(IOMMU_ENABLED);
2142 irq_remapping_enabled = 1;
2147 void amd_iommu_disable(void)
2149 amd_iommu_suspend();
2152 int amd_iommu_reenable(int mode)
2159 int __init amd_iommu_enable_faulting(void)
2161 /* We enable MSI later when PCI is initialized */
2167 * This is the core init function for AMD IOMMU hardware in the system.
2168 * This function is called from the generic x86 DMA layer initialization
2171 static int __init amd_iommu_init(void)
2175 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2177 free_dma_resources();
2178 if (!irq_remapping_enabled) {
2180 free_on_init_error();
2182 struct amd_iommu *iommu;
2184 uninit_device_table_dma();
2185 for_each_iommu(iommu)
2186 iommu_flush_all_caches(iommu);
2193 /****************************************************************************
2195 * Early detect code. This code runs at IOMMU detection time in the DMA
2196 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2199 ****************************************************************************/
2200 int __init amd_iommu_detect(void)
2204 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2207 if (amd_iommu_disabled)
2210 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2214 amd_iommu_detected = true;
2216 x86_init.iommu.iommu_init = amd_iommu_init;
2221 /****************************************************************************
2223 * Parsing functions for the AMD IOMMU specific kernel command line
2226 ****************************************************************************/
2228 static int __init parse_amd_iommu_dump(char *str)
2230 amd_iommu_dump = true;
2235 static int __init parse_amd_iommu_options(char *str)
2237 for (; *str; ++str) {
2238 if (strncmp(str, "fullflush", 9) == 0)
2239 amd_iommu_unmap_flush = true;
2240 if (strncmp(str, "off", 3) == 0)
2241 amd_iommu_disabled = true;
2242 if (strncmp(str, "force_isolation", 15) == 0)
2243 amd_iommu_force_isolation = true;
2249 static int __init parse_ivrs_ioapic(char *str)
2251 unsigned int bus, dev, fn;
2255 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2258 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2262 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2263 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2268 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2270 cmdline_maps = true;
2271 i = early_ioapic_map_size++;
2272 early_ioapic_map[i].id = id;
2273 early_ioapic_map[i].devid = devid;
2274 early_ioapic_map[i].cmd_line = true;
2279 static int __init parse_ivrs_hpet(char *str)
2281 unsigned int bus, dev, fn;
2285 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2288 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2292 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2293 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2298 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2300 cmdline_maps = true;
2301 i = early_hpet_map_size++;
2302 early_hpet_map[i].id = id;
2303 early_hpet_map[i].devid = devid;
2304 early_hpet_map[i].cmd_line = true;
2309 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2310 __setup("amd_iommu=", parse_amd_iommu_options);
2311 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2312 __setup("ivrs_hpet", parse_ivrs_hpet);
2314 IOMMU_INIT_FINISH(amd_iommu_detect,
2315 gart_iommu_hole_init,
2319 bool amd_iommu_v2_supported(void)
2321 return amd_iommu_v2_present;
2323 EXPORT_SYMBOL(amd_iommu_v2_supported);
2325 /****************************************************************************
2327 * IOMMU EFR Performance Counter support functionality. This code allows
2328 * access to the IOMMU PC functionality.
2330 ****************************************************************************/
2332 u8 amd_iommu_pc_get_max_banks(u16 devid)
2334 struct amd_iommu *iommu;
2337 /* locate the iommu governing the devid */
2338 iommu = amd_iommu_rlookup_table[devid];
2340 ret = iommu->max_banks;
2344 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2346 bool amd_iommu_pc_supported(void)
2348 return amd_iommu_pc_present;
2350 EXPORT_SYMBOL(amd_iommu_pc_supported);
2352 u8 amd_iommu_pc_get_max_counters(u16 devid)
2354 struct amd_iommu *iommu;
2357 /* locate the iommu governing the devid */
2358 iommu = amd_iommu_rlookup_table[devid];
2360 ret = iommu->max_counters;
2364 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2366 static int iommu_pc_get_set_reg_val(struct amd_iommu *iommu,
2367 u8 bank, u8 cntr, u8 fxn,
2368 u64 *value, bool is_write)
2373 /* Check for valid iommu and pc register indexing */
2374 if (WARN_ON((fxn > 0x28) || (fxn & 7)))
2377 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2379 /* Limit the offset to the hw defined mmio region aperture */
2380 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2381 (iommu->max_counters << 8) | 0x28);
2382 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2383 (offset > max_offset_lim))
2387 writel((u32)*value, iommu->mmio_base + offset);
2388 writel((*value >> 32), iommu->mmio_base + offset + 4);
2390 *value = readl(iommu->mmio_base + offset + 4);
2392 *value = readl(iommu->mmio_base + offset);
2397 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);
2399 int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2400 u64 *value, bool is_write)
2402 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
2404 /* Make sure the IOMMU PC resource is available */
2405 if (!amd_iommu_pc_present || iommu == NULL)
2408 return iommu_pc_get_set_reg_val(iommu, bank, cntr, fxn,